Description
Scheduling is the process of deciding how to commit resources between a variety of possible tasks. Time can be specified (scheduling a flight to leave at 8:00) or floating as part of a sequence of events.
ABSTRACT
Title of Document: EFFECTS OF PERFORMANCE SCHEDULES ON
EVENT TICKET SALES
Peggy Hui-Hsing Tseng, Ph.D. 2009
Directed By: Professor Wendy W. Moe, Department of Marketing
Event scheduling is one of many important decisions facing event marketers
in the entertainment industry (i.e., how should multiple performances be scheduled
across markets, across venues, and over time?). While there is ample research
examining the issues of costs and constraints associated with such a decision,
virtually no research exists to examine the impact of these decisions on consumer
demand. Hence, the objective of this dissertation is to examine how consumers
respond to event marketers’ scheduling decisions.
First, a scheduling effect may arise from performances within a market.
When performances are scheduled closely in distance or time, their similarity in
venue locations or performance dates may result in a stronger relationship and
influence ticket sales. This relationship may have a positive effect on ticket sales
because the similarity could signal the quality of an event and suggest the desirability
of these performances. Thus, these performances attract more consumers and sell
more tickets. However, the relationship could be negative. When performances are
close in distance or time, they become direct substitutes and compete for consumer
patronage.
Another effect arises from an event distribution across markets. When an
event travels from one market to another and each market has a different performance
schedule, the word of mouth of this event may accumulate and carry over to later
markets. If so, market sales may be a good proxy of word of mouth. How well (or
poorly) an event sells in preceding markets may affect ticket sales in following
markets.
This dissertation consists of three essays to examine the abovementioned
scheduling effects. We contact a national ticket seller to acquire a dataset containing
ticket sales of a family event traveling across 42 markets. The first essay analyzes a
performance schedule in one metropolitan market and investigates the scheduling
effect on ticket sales. The second essay employs all performance schedules in 42
markets to study heterogeneous market responses and propose explanatory factors.
Finally, the third essay incorporates the distribution sequence of this event and
examines whether ticket sales in preceding markets have a carryover effect to
influence ticket sales in later markets.
EFFECTS OF PERFORMANCE SCHEDULES ON
EVENT TICKET SALES
By
Peggy Hui-Hsing Tseng
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2009
Advisory Committee:
Professor Wendy W. Moe, Chair
Professor Roger R. Betancourt
Professor Yogesh Joshi
Professor P.K. Kannan
Professor Michel Wedel
© Copyright by
Peggy Hui-Hsing Tseng
2009
ii
Table of Contents
Table of Contents .......................................................................................................... ii
List of Tables .................................................................................................................v
List of Figures .............................................................................................................. vi
1 Introduction ............................................................................................................1
2 Essay 1: Scheduling to Sell: Examining the Impact of a Performance Schedule
on Event Ticket Sales .....................................................................................................8
2.1 Introduction .................................................................................................... 8
2.2 Literature Review and Conceptual Framework ........................................... 15
2.2.1 Airline and Movie Scheduling .............................................................. 15
2.2.2 Event Tickets ........................................................................................ 17
2.2.3 Impact of Scheduling on Ticket Sales .................................................. 21
2.2.4 Conceptual Framework ......................................................................... 25
2.3 Model Development ..................................................................................... 28
2.3.1 Model Overview ................................................................................... 28
2.3.2 Scheduling Characteristics .................................................................... 29
2.3.3 Number of Ticket Sales ........................................................................ 30
2.3.4 Timing of Ticket Sales .......................................................................... 34
2.3.5 Endogeneity in Performance Scheduling .............................................. 38
2.4 Data Description ........................................................................................... 39
2.4.1 Description of Events ............................................................................ 40
2.4.2 Description of Ticket Sales ................................................................... 41
2.4.3 Covariate Specifications ....................................................................... 44
2.5 Model Estimation and Benchmark Comparison .......................................... 46
2.5.1 Estimation ............................................................................................. 46
2.5.2 Benchmark Comparison........................................................................ 47
2.6 Results .......................................................................................................... 49
2.6.1 Number of Ticket Sales ........................................................................ 49
2.6.2 Timing of Ticket Sales .......................................................................... 51
2.6.3 Endogeneity in Performance Scheduling .............................................. 53
2.7 Policy Simulation ......................................................................................... 54
2.7.1 Scenario 1: Geographic Change ............................................................ 54
2.7.2 Scenario 2: Temporal Changes ............................................................. 55
2.8 Conclusions .................................................................................................. 56
2.8.1 Summary ............................................................................................... 56
2.8.2 Limitations and Next Steps ................................................................... 57
Table and Figures .................................................................................................... 58
3 Essay 2: Heterogeneous Market Responses to Performance Schedules and Their
Explanatory Factors .....................................................................................................78
3.1 Introduction .................................................................................................. 78
3.2 Preliminary Analysis .................................................................................... 82
3.3 Literature Review and Conceptual Framework ........................................... 84
iii
3.3.1 Literature Review.................................................................................. 84
3.3.2 Conceptual Framework ......................................................................... 85
3.4 Model Development ..................................................................................... 88
3.4.1 Model Overview ................................................................................... 88
3.4.2 Scheduling Characteristics and the Number of Ticket Sales ................ 88
3.4.3 Endogeneity in Performance Scheduling .............................................. 89
3.4.4 Use of the HB Approach for Response Heterogeneity ......................... 90
3.5 Data .............................................................................................................. 90
3.5.1 Definition of Markets ............................................................................ 91
3.5.2 Description of Performance Schedules across Markets ........................ 92
3.5.3 Covariate Specifications ....................................................................... 94
3.6 Model Estimation and Benchmark Comparison .......................................... 97
3.6.1 Estimation ............................................................................................. 97
3.6.2 Benchmark Comparison........................................................................ 98
3.7 Results .......................................................................................................... 99
3.7.1 Heterogeneous Market Responses to Performance Schedules ............. 99
3.7.2 Explanatory Factors for Market Heterogeneity .................................... 99
3.7.3 Endogenous Scheduling Decision....................................................... 101
3.8 Conclusions ................................................................................................ 101
3.8.1 Summary ............................................................................................. 101
3.8.2 Limitations and Next Steps ................................................................. 103
Tables and Figures ................................................................................................ 104
4 Essay 3: Sequential Distribution of a Live Performance Event .........................117
4.1 Introduction ................................................................................................ 117
4.2 Literature Review and Conceptual Framework ......................................... 121
4.2.1 Sequential Distribution ....................................................................... 121
4.2.2 Conceptual Framework ....................................................................... 125
4.3 Model Development ................................................................................... 128
4.3.1 Overview ............................................................................................. 128
4.3.2 Demand Equation: Market Sales ........................................................ 129
4.3.3 Supply Equations: Overall Supply, Venue Usage, and Day Usage .... 130
4.3.4 Variable Specifications ....................................................................... 131
4.3.5 Model Summary.................................................................................. 138
4.4 Data ............................................................................................................ 139
4.4.1 Touring Sequence ............................................................................... 140
4.4.2 Covariates in the Demand Model ....................................................... 141
4.4.3 Covariates in the Supply Model .......................................................... 144
4.5 Estimation and Results ............................................................................... 146
4.5.1 Model Estimation ................................................................................ 146
4.5.2 Results of Demand Equation .............................................................. 147
4.5.3 Results of Supply Equations ............................................................... 149
4.5.4 Correlated Demand and Supply .......................................................... 150
4.6 Conclusions ................................................................................................ 150
4.6.1 Summary ............................................................................................. 150
4.6.2 Conclusion .......................................................................................... 151
Tables and Figures ................................................................................................ 153
iv
5 Conclusion ..........................................................................................................165
5.1 Summary .................................................................................................... 165
5.2 General Discussions ................................................................................... 168
5.3 Contributions .............................................................................................. 170
5.4 Limitation and Future Research ................................................................. 170
Reference ...................................................................................................................172
v
List of Tables
Table 2-1: Description of Data Fields ......................................................................... 58
Table 2-2: Summary of Ticket Sales .......................................................................... 59
Table 2-3: Descriptive Statistics of Covariates ........................................................... 60
Table 2-4: Benchmark Models and Model Fit ............................................................ 61
Table 2-5: Results for the Number of Ticket Sales .................................................... 62
Table 2-6: Results for the Timing of Ticket Sales ...................................................... 63
Table 2-7: Results for Endogenous Performance Scheduling .................................... 64
Table 3-1: Summary of DMA markets ..................................................................... 104
Table 3-2: Summary of Market Responses to Performance Schedules .................... 105
Table 3-3: Venue Locations and their Associated DMAs ........................................ 106
Table 3-4: Summary of Ticket Sales ........................................................................ 107
Table 3-5: Descriptive Statistics of Covariates across Markets ................................ 108
Table 3-6: Descriptive Statistics of Market Characteristics ..................................... 109
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics ............. 110
Table 3-8: Sources of Heterogeneous Market Responses ......................................... 111
Table 3-9: Results of Performance Schedule Model ................................................ 112
Table 4-1: Descriptive Statistics of Covariates (in the log term) .............................. 153
Table 4-2: Correlation Coefficient of Demand Covariates ....................................... 154
Table 4-3: Correlation Coefficient of Dependent Variables ..................................... 155
Table 4-4: Results of the Demand Model ................................................................. 156
Table 4-5: Results of the Supply Model ................................................................... 157
Table 4-6: Correlation between Supply and Demand Models .................................. 158
vi
List of Figures
Figure 2-1: Conceptual Framework of Spatial Decomposition .................................. 65
Figure 2-2: Model Overview ...................................................................................... 66
Figure 2-3: Probability of Ticket Sales over Time ..................................................... 67
Figure 2-4: Sales Distribution by Performances ......................................................... 68
Figure 2-5: Sales Distribution by Days of Week ........................................................ 69
Figure 2-6: Weekly Sales Pattern of a Performance ................................................... 70
Figure 2-7: Heterogeneity in Sales Pattern Across Performances .............................. 71
Figure 2-8: Venue Locations and Driving Distances .................................................. 72
Figure 2-9: Summary of Geographic Distance ........................................................... 73
Figure 2-10: Summary of Temporal Distance ............................................................ 74
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales .................... 75
Figure 2-12: Effect of Schedule Changes on Ticket Sales ......................................... 76
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales ........................ 77
Figure 3-1: Conceptual Framework of Heterogeneous Market Responses .............. 113
Figure 3-2: Example of a DMA and its county information ..................................... 114
Figure 3-3: Locations of DMA Markets and Performing Sequence ......................... 115
Figure 3-4: Heterogeneous Market Responses ......................................................... 116
Figure 4-1: Conceptual Framework .......................................................................... 159
Figure 4-2: Illustration of Market Connectedness .................................................... 160
Figure 4-3: Overview of Model Development .......................................................... 161
Figure 4-4: Venue Locations .................................................................................... 162
Figure 4-5: Touring Dates across Markets ................................................................ 163
Figure 4-6: Capacity-Filled Rate across Markets ..................................................... 164
1
1 Introduction
Scheduling is one of many important decisions facing event marketers in the
entertainment industry. To maximize revenues in a national market, event marketers have to not
only make marketing mix decisions but also schedule performances across markets, across
venues, and over time. Hence, these scheduling decisions result in a performance schedule
within a market or across a national market for an event to perform in different venues and on
various days. In this dissertation, we define a performance schedule as a summary of
performances taking places in various venue locations and on different dates to examine
potential effects of this performance schedule on ticket sales at a performance level and across
markets.
Similar to airline and movie scheduling decisions, event marketers rely on costs and
constraints in their scheduling process. They need to comply with venue availability and seating
capacity to decide when to provide performances and how many performances to provide. They
can minimize the travel distance across markets if it is expensive to move from one market to
another. They can also shorten a performing period in a venue if the cost to rent a facility
outweighs the benefits of having performances for a long period.
While there is ample research examining the issues of costs and constraints associated
with such scheduling decisions, virtually no research exists to examine the impact of these
decisions on consumer demand. In other words, it is unclear to event marketers how consumers
in a market evaluate individual performances and how consumer inferences influence their
purchase decisions such as which performance to buy and when to buy. Hence, at the
2
performance level, consumers’ evaluations about performances in a schedule may affect the
number of ticket sales of these performances and the pattern of these ticket sales over time.
It is important for event marketers to understand potential drivers for the number of ticket
sales sold and the pattern of these ticket sales over time. Once they know the effects of these
drivers, they will be able to estimate expected market demand and schedule performances
accordingly. In this way, they can avoid revenue losses from undersupply and minimizes costs
resulting from potential oversupply. Additionally, event ticket sales are often available for
purchases several months or weeks in advance, and these tickets are not sold at a constant rate
throughout the advance-selling period. If event marketers understand the drivers for the sales
rate, sales pattern can be easily established to monitor actual ticket sales over time and
potentially adjust the marketing strategy as the event approaches.
As such, this dissertation aims to investigate the impact of performance schedules on
ticket sales for a live performance event. Specifically, we propose two types of effects emerging
from the scheduling decisions. One type of effect is likely to exist among performances within a
market because of the way these performances are scheduled across venues and dates. For
example, performances scheduled closely in distance or time might signal the quality of an event,
the desirability of these venues or dates, or just the potential substitutes across venues or dates.
If consumers make any of these inferences and perceive performances to be more or less
favorable, the scheduling effect among performances within a market would influence how well
each performance sells and when ticket sales occur.
The other type of effect might develop across markets where an event performs one after
another. In other words, after an event performs in each market for a period, its word of mouth
may arise and cumulate over time. If the word of mouth travels across markets and consumers
3
have positive feedbacks, how well this event does in previous markets could influence ticket
sales in later markets. Because ticket sales in previous markets can capture the volume of word
of mouth to some extent, it is plausible to observe ticket sales in previous markets to affect sales
in later markets.
This dissertation consists of three essays to examine those possible scheduling effects
mentioned above. For this purpose, we acquired a dataset containing ticket sales of a family
event traveling across 42 markets. We start by characterizing a performance schedule in a single
market and investigating its impact on the number of ticket sales at a performance level and the
pattern of these ticket sales over time (i.e., essay one). From essay one, we conclude that
performances scheduled close to each other in terms of distance can experience more ticket sales,
and these tickets sell at a faster rate. On the other hand, performances scheduled close to each
other in terms of time tend to sell fewer tickets, but do not exhibit any significant changes in
their sale patterns. However, these results are established for one metropolitan market in our
dataset. Hence, we expand the level of analysis to all 42 markets in essay two to ensure
scheduling effects generalizable for this event and examine the heterogeneity across markets.
Essay two confirms consistent scheduling effects across markets and identifies explanatory
factors for the heterogeneity in effect sizes across markets. Finally, essay three proceeds to
investigate scheduling effects at a market level and examines whether an event performing
across markets affects ticket sales in these markets. Specifically, essay three reveals that markets
to which an event travels sell more tickets when event marketers disperse performance dates or
employ multiple venues in these markets. Additionally, these markets do not influence one
another on their ticket sales, but their venues within the same market have such an effect.
Although one may argue that the third essay does not have to be conducted after the first two
4
essays, we choose this sequence because we need to investigate the effect of a touring event
across markets after we can understand and control for the effect of a performance schedule
within a market.
The purpose of the first essay is to lay the foundation of this dissertation and to test
whether scheduling decisions influence ticket sales of performances within a market. We begin
by examining ticket sales at a performance level within a single market and differentiate between
performances based on a set of scheduling characteristics. We derive a set of scheduling
characteristics based on the venue locations and performance dates. According to these
scheduling characteristics, we measure how closely performances are scheduled in terms of
distance or time. A performance of a shorter geographic or temporal distance hence indicates its
similarity in venue location or performance date to other performances.
After examining ticket sales of these performances as a function of their distance
measures, we find that geographic and temporal distances between performances have different
effects on ticket sales. More specifically, a shorter geographic distance between performances
leads to more tickets sold and a faster sales rate for a performance. However, a shorter temporal
distance between performances just causes decreases in the number of ticket sales but does not
influence how fast tickets are sold.
Although it is not clear why consumers process geographic and temporal distances
differently, it is clear that they refer to the way performances are scheduled as a means to make
inferences about these performances. Their inferences about performances of a shorter
geographic distance could be regarding the quality of an event, the desirability of their associated
venues, or others. As a result, these performances sell more tickets and attract consumers to
purchase tickets early. On the other hand, their inferences about performances of a shorter
5
temporal distance could be the high substitutability within a shorter period. Therefore, these
performances compete for consumer patronage and suffer from sales cannibalization.
While the first essay demonstrates the significant effects of geographic and temporal
distances on ticket sales, the results are for one market only. Yet, event marketers often need to
make scheduling decisions for more than one market. It is not clear whether the results in this
first essay are consistent across markets. For this reason, the objective of the second essay is to
use all performance schedules of the same event to investigate heterogeneous market responses
and identify explanatory factors.
To achieve this objective, we follow the same approach as in essay one and analyze all
performance schedules. For performances within their associated market, we characterize them
by their venue locations and performance dates and compute their geographic and temporal
distances to other performances. Then, for each market, we model ticket sales of performances
as a function of their distance measures to understand whether the heterogeneity exists in market
responses. We then model these market specific parameters as a function of their market
characteristics such as market population and additional scheduling characteristics such as travel
sequence along the distribution to explain any differences across markets.
Our results show that market responses to performance schedules are heterogeneous and
can be explained by market and additional scheduling characteristics. Specifically, when a
market has a bigger population, the effects of days of week and baseline attractiveness are
attenuated. Moreover, after an event travels to more markets that are geographically adjacent to
a focal market, the focal market is less responsive to its baseline attractiveness and temporal
schedule. Finally, a current market in a late distribution sequence tends to respond more
favorably to a Sunday performance.
6
Results in essay two hence suggest a possibility that markets where an event travels one
after another may be dependent. In other words, how following markets respond to their
temporal schedule depends on how many geographically adjacent markets an event has visited.
Another possibility is that there may be another means of integrating participating markets along
a touring event such that ticket sales of preceding markets might directly influence ticket sales of
following markets (rather than through the response parameters and geographically adjacent
markets). If so, it is important to incorporate the temporal sequence in an event distribution and
study the impact of preceding markets on following ones.
The primary objective of essay three, therefore, is to examine the impact of a sequentially
distributed event across markets. Additionally, we consider the endogeneity between supply and
demand for an event in case expected market demand influences event marketers’ scheduling
decisions and their schedule further affect ticket sales in a market.
To achieve this objective, we model the supply and demand for an event simultaneously.
We model ticket sales of each market as a function of its performance schedule and the
sequential distribution of this event. In addition to modeling the supply and demand
simultaneously to account for the endogeneity, we use three variables to present the scheduling
influences on market demand and employ the spatially weighted approach to incorporate
different release timing and ticket sales of preceding markets in an event tour.
Our results show that a market experiences more ticket sales when event marketers
disperse performance dates or book multiple venues in this market. Moreover, we show that the
sequentially distributed event has an effect on ticket sales. However, this effect is significant
across venues of the same market but not across markets. When an event performs in more than
7
one venue, its ticket sales in a preceding venue carry over to a later venue and influence its
overall market sales.
The organization of this dissertation is as follows. Chapter 1 introduces the issues of
scheduling facing event marketers and presents a general overview of each essay. Chapters 2, 3,
and 4 discuss the three essays, respectively, in depth. Finally, Chapter 5 provides a brief
summary of each essay, integrates essential results, and points out limitations and future
directions to conclude this dissertation.
8
2 Essay 1: Scheduling to Sell: Examining the Impact of a Performance
Schedule on Event Ticket Sales
2.1 Introduction
Scheduling is one of many important decisions facing event marketers in the
entertainment industry. Besides decisions regarding marketing activities (e.g., pricing and
promotions), event marketers have to schedule performances across markets, across venues, and
over time. Typically, they rely on costs and constraints in their scheduling process (Etschmaier
and Mathaisel 1985; Lohatepanont and Barnhart 2004; Eliashberg et al 2007) and use pricing,
advertising, and days of week to describe how well an event can sell (Weinberg and Shachmut
1978; Putler and Lele 2003; Leslie 2004) or when ticket sales occur (Moe and Fader 2009).
While there is ample research examining the issues of costs and constraints associated
with such scheduling decisions, virtually no research exists to examine the impact of these
decisions on consumer demand. In other words, it is unclear to event marketers how consumers
in a market may evaluate a performance schedule of an event and how consumer inferences
influence their purchase decisions such as which performance to buy and when to buy. Hence, at
the performance level, consumers’ evaluations about performances in a schedule may affect the
number of ticket sales of these performances and the pattern of these ticket sales over time.
From event marketers’ perspective, they need to understand potential drivers for ticket
sales in terms of the number of ticket sales and the pattern of these ticket sales over time. Once
they know how these drivers influence market demand, they can estimate expected market
demand accordingly and schedule performances to meet this market demand. In this way, they
9
can avoid revenue losses from undersupply and prevent decreases in profitability due to
oversupply.
On the other hand, event ticket sales are often available for purchases several months or
weeks in advance, and these ticket sales do not occur at a constant rate throughout an entire
advance-selling period. If event marketers understand drivers for tickets selling at a different
rate, they can portray the sales pattern as a benchmark and monitor actual ticket sales over time.
Consequently, once event marketers find an actual pattern deviating from the benchmark, they
can take actions in time.
In general, scheduling decisions affect the maximum number of tickets that an event can
sell. When event marketers schedule a live performance event, they often allocate multiple
performances across markets, across venues, and over time. Although the number of
performances and the capacities of chosen venues constrain the maximum number of tickets an
event can sell, empirical evidence shows that it is rare for the demand to exceed supply in this
industry. Therefore, one possible impact of scheduling decisions is to constrain the maximum
possible of ticket sales for an event although the supply is usually well beyond the actual demand.
Scheduling decisions might also influence consumer responses in a market. In other
words, when consumers realize performances are scheduled in various venue locations and on
different dates, they may try to rationalize why event marketers schedule performances in this
way and then make inferences about these performances. If so, consumers could formulate
different preferences for these performances to choose one performance to attend and purchase
tickets at their desired time. At a performance level, consumer responses influence how well
individual performances sell and when ticket sales of these performances occur.
10
Therefore, the objective of this essay examines the effects of scheduling characteristics of
performances on ticket sales. Specifically, we use venue locations and performance dates as the
scheduling characteristics of each performance, and we investigate whether performances
scheduled closely in distance or time experience a different number and timing of ticket sales.
We define the timing of ticket sales as tickets sold at different times in an advance-selling
period, and earlier or later timing of ticket sales suggests ticket sales occurred in the earlier or
later advance-selling period. In other words, if scheduling characteristics have an effect,
performances scheduled close in distance or time would experience more (or fewer) ticket sales.
Their ticket sales would occur earlier (or later) than those scheduled farther apart in an advance-
selling period, ceteris paribus.
One possible effect is to see performances scheduled close in distance or time
experiencing more ticket sales and earlier timing of sales than those scheduled farther apart. For
example, if event marketers want to signal the desirability of some venue locations or
performance dates, they could allocate more performances to those specific venues or dates. In
this way, consumers would perceive performances scheduled close in distance or time to be more
attractive (due to the similar scheduling characteristics) and assign higher utilities to these
performances. As a result, these performances could sell more tickets and experience earlier
timing of sales than other performances.
Another possible effect is to observe performances scheduled farther apart in distance or
time experiencing more ticket sales and earlier timing of sales than those scheduled nearby. In
other words, when consumers have higher uncertainty about whether they can attend an event in
a particular venue at a specific time, event marketers could sparsely allocate performances across
venues and dates. In this way, consumers have more alternatives and higher flexibility regarding
11
when and where to attend. Hence, the chance for them to attend this event increases, and
performances scheduled farther apart will be able to sell more tickets and experience earlier
timing of sales. In contrast, performances scheduled nearby merely substitute one another within
certain venues or dates. Consumers do not have to decide which performance they want to buy
and can delay their purchase timing. Hence, these closely scheduled performances compete
against consumer patronage and cannibalize ticket sales. To sum up, scheduling characteristics
might have two possible but contradictory effects on ticket sales. We allow both possibilities
and examine the effects of scheduling characteristics empirically.
Our modeling objective is to measure the scheduling characteristics of each performance
and study the impact of these scheduling characteristics on the number and timing of ticket sales.
We consider the possibility that consumers evaluate venue locations and performance dates
differently. Thus, we differentiate the effect of scheduling across venues from the effect of
scheduling across dates and then investigate these separately effects. Specifically, we refer to the
previous effect as the effect of geographic scheduling or the effect of a geographic schedule and
the later effect as the effect of temporal scheduling or the effect of a temporal schedule.
First, to measure the scheduling characteristics of performances, we characterize each
performance in a schedule of an event by its venue location and performance date. Then, we
compute the geographic and temporal distances between performances to understand how
closely (or distantly) performances are scheduled across venues and dates. For example,
performances of a shorter geographic or temporal distance to others are relatively closer to other
performances than performances of a longer geographic or temporal distance.
Second, to examine the number of ticket sales across performances, we consider the
possibility that some consumers might evaluate the venue locations and performance dates in a
12
schedule but eventually do not attend. To account for consumers who make purchases and those
who miss out the opportunity to attend, we specify the number of ticket sales in the form of sales
share within a potential target market. Then, we examine the share of each performance and the
non-buyer segment within this potential market. We extend the competing destination model
proposed by Fotheringham (1988) and model the share of each performance and the non-buyer
segment as a function of the geographic and temporal distances between performances. By
doing so, we can understand whether scheduling characteristics influence ticket sales at a
performance level. In addition, using sales shares of individual performances and the population
size in a target market as the number of potential buyers, we can obtain the number of ticket
sales at a performance level. We can also obtain a market penetration rate by taking the sum of
sales shares across performances.
Third, to examine the timing of ticket sales across performances, we first consider a
general pattern of ticket sales for a performance. That is, a performance sells fewer tickets in the
beginning of its advance-selling period and obtains more sales over time with the most arriving
in the later period or the week of the performance. Although this is a general pattern over time,
each performance still has a different sales rate. Some performances experience ticket sales
occurred early (i.e., earlier timing of ticket sales) but others experience ticket sales arrived later
(i.e., later timing of ticket sales). To account for variations in sales rate across performances, we
employ a Weibull hazard model to capture the timing of sales over time for individual
performances. We further model the sales rate of each performance as a function of its
geographic and temporal distances to other performances to understand whether these scheduling
characteristics explain the heterogeneity in sales rate.
13
Finally, we consider a possible endogeneity in scheduling decisions. Since the live
performance industry has been established and event marketers have scheduled for a variety of
events, event marketers may have incorporated their experiences into a scheduling process. If so,
a performance schedule is endogenously determined. For example, event marketers might have
scheduled more performances on weekends across all venues because they know these
performances have higher performance attractiveness. It is also likely that they have scheduled
performances based on the responses they expect in this market. Specifically, they may have
scheduled performances closely in distance or time or farther apart, because they know
consumers prefer performances of such scheduling characteristics. In case such an endogeneity
exists in the scheduling process, we control for this possibility in our model development.
We contact a national ticket seller to obtain a dataset of a live performance event and use
its ticket sales to examine the impact of its scheduling characteristics on ticket sales. Although
this event performed across several markets, we use the performances in the New York
metropolitan market as a subset. The reason is that performances scheduled in this market have
richer variations in venue locations and performance dates. In short, this event had 70
performances across four venues in the New York metropolitan market and performed between
March and June 2004.
Our results indicate that the effect of geographic scheduling differs from the effect of
temporal scheduling on the number and timing of ticket sales. Performances scheduled closely
across venues not only sell more tickets but also sell tickets at a faster rate. In contrast,
performances sparsely scheduled across dates sell more tickets but do not have an impact on the
timing of sales.
14
Our explanation for the observed effect of geographic scheduling is that event marketers
may attempt to signal the desirability of some venues by scheduling more performances in those
venues. Although the desirability of these venues may also be owing to population around these
venues and consumers’ shorter travel distance to these venues, we control for this possibility in
our benchmark models and still find a significant effect of geographic scheduling. Therefore, in
contrast with performances scheduled in distant venues, performances scheduled in nearby
venues signal higher desirability or popularity to consumers. As a result, more consumers are
interested in these performances and are more willing to purchase tickets earlier.
On the other hand, our explanation for the observed effect of temporal scheduling is that
event marketers try to accommodate consumers’ uncertain preferences to performance dates by
scheduling performances sparsely across dates. In this way, performances on dispersed dates (or
of a longer temporal distance) provide consumers higher flexibility and further sell more tickets.
In contrast, performances within a short time span (or of a shorter temporal distance) substitute
one another and cannibalize ticket sales.
In addition to the impact of scheduling characteristics on ticket sales, our results also
indicate that there exists some endogeneity in the scheduling process. Event marketers consider
how attractive performances are when they schedule performances across dates. However, our
results show that event marketers have not yet incorporated the effects of geographic and
temporal scheduling when they allocate performances across venues and dates.
The organization of this essay is as follows. We first review past works relevant to
scheduling and event ticket sales. Then, we present our conceptual framework and model
development. After a detailed data description, we provide our results and discuss possible
rationale behind the scheduling effects. We also conduct two policy simulations to show how re-
15
allocating performances to a different venue or date would result different number and timing of
ticket sales. Finally, we conclude this essay with limitations and next steps.
2.2 Literature Review and Conceptual Framework
2.2.1 Airline and Movie Scheduling
Because scheduling a live performance event is an important yet understudied research
stream, we search literature in other contexts where scheduling is also critical to managers. We
find airline and movie scheduling literature a good fit because managers have a common
objective to schedule a series of flights, screens, or performances to meet the market demand.
Therefore, we discuss studies in airline and movie scheduling in turn.
Airline Scheduling
Airline scheduling is a complex system. It involves demand estimation, pricing for
different segments, flight scheduling for various routes, fleet assignments for individual flights,
crew scheduling, aircraft rotation, flight gate assignments, and many other decisions (Etschmaier
and Mathaisel 1985; Dobson and Lederer 1993; Jarrah et al 2000; Lohatepanont and Barnhart
2004; Dorndorf et al 2007). Therefore, any small changes require a series of adjustments in the
entire system.
Traditionally, airline scheduling has been a constrained-optimization decision.
Researchers use historical data to estimate demand and consider the expected demand to
construct flight schedules (Dobson and Lederer 1993; Lohatepanont and Barnhart 2004). After
schedule constructions, other departments evaluate proposed schedules to set airfares, assign
aircrafts, and make other operational related decisions (Etschmaier and Mathaisel 1985; Dobson
and Lederer 1993; Jarrah et al 2000; Dorndorf et al 2007). Finally, they examine associated
16
profits and revise flight schedules to ensure profit maximization. Therefore, in this iterative
decision process, scheduling is primarily constraint-driven. An airline managers’ objective is to
maximize profitability while minimizing operational costs within feasible boundaries.
The similarity between airline scheduling and performance scheduling is the common
scheduling nature. Airline managers and event marketers have to decide when and where flights
or performances have to take place. Their conceptual objective is the same because they aim to
launch a schedule to extract the most demand in a given market. However, the demand in the
airline schedule is either treated as exogenous (McGill and van Ryzin 1999), based on historical
data (Etschmaier and Mathaisel 1985), or dependent on price and departure and arrival times
(Dobson and Lederer 1993). Researchers assume that dropping flights always leads to losses in
revenues (Lohatepanont and Barnhart 2004) and have not yet investigated how scheduling
density (e.g., frequencies of flights) influences ticket demand.
Movie Scheduling
In the movie industry, movie scheduling takes place after movie distributors release
movies to exhibitors. The primary task that exhibitors perform is to allocate a number of screens
within a theater to meet the local demand for this movie (Swami, Eliashberg, and Weinberg 1999;
Eliashberg et al 2009). Compared with advertising effect on box office revenues, movie
scheduling is a relatively new research direction in this industry (Eliashberg, Elberse, and
Leenderss 2006).
When it comes to the implementation of movie scheduling, exhibitors have to select
movies that might contribute higher revenues because they have a limited number of screens in a
theater. They usually start with estimating demand for a variety of movies and then select a
smaller set of movies to play (Swami et al 1999; Eliashberg et al 2007; Eliashberg et al 2009).
17
After they choose movies of interest, they allocate available screens to these movies (Elberse and
Eliashberg 2003). Although the general principle is that the longer exhibitors play a movie, the
more revenues they get, the decreasing demand over time and the contract with distributors also
determine how long a movie shows in a theater (Swami et al 1999). Finally, exhibitors refer to
box office revenues in a previous week, movie genres, days of week, and times of day to revise
their scheduling decision on a weekly basis (Elberse and Eliashberg 2003; Eliashberg et al 2007).
Movie scheduling is similar to airline scheduling in the sense that both contexts heavily
rely on the operational constraints, and profitability maximization is the top priority. It is also
similar to performance scheduling because within these contexts the purpose is to serve local
demand by offering movies or performances at various days of week or times of day. However,
researchers in movie scheduling usually assumes demand to be exogenous (Swami et al 1999;
Eliashberg et al 2007; Eliashberg et al 2009) or influenced by marketing activities (Elberse and
Eliashberg 2003). They have not yet considered the competition between theaters (Eliashberg et
al 2006) or the impact of scheduling density on box office revenues.
To sum up, the focus of airline and movie scheduling is constraint optimization and profit
maximization. Demand is often assumed exogenous and influences scheduling decisions.
Whether these scheduling decisions influence demand deserves further investigation. Therefore,
the objective of this essay is to examine the impact of scheduling on ticket sales in the context of
a live performance event. Next, we review current research in event tickets to understand
existing drivers for ticket sales before we discuss how scheduling could affect ticket sales.
2.2.2 Event Tickets
Live performance events typically refer to concerts, musicals, or circus acts, etc. that
perform live in front of an audience. Because an event often provides multiple performances
18
across venues and dates, understanding how well each performance can sell is an important issue
for event marketers and researchers. Hence, we review relevant works and discuss factors that
influence the number and timing of ticket sales.
Number of Ticket Sales
Identifying drivers of ticket sales has been a common topic studied in marketing,
economics, and performing arts literature. Researchers have used product related drivers,
consumer characteristics, and seasonality to explain consumer attendance (Weinberg and
Shachmut 1978; Currim, Weinberg, and Wittink 1981; Venkatesh and Mahajan 1993; Reddy,
Swaminathan, and Motley 1998; Putler and Lele 2003; Leslie 2004). According to the number
of events involved in the drivers, we further classify product-related drivers into (1) assortment-
related drivers, (2) event-related drivers, and (3) performance-related drivers. We discuss these
drivers and their importance in turn.
Assortment-related drivers often refer to factors related with a bundle of events. In other
words, this type of drivers exists when multiple events are grouped together in a product offering.
Researchers have shown that different bundle size, event types in a bundle, and associated
seating benefits can attract varying degrees of demand because consumers have heterogeneous
preferences to genres (e.g., operas, musicals), language types (e.g., Italian, English), and
willingness to pay (Currim, Weinberg, and Wittink 1981; Havlena and Holak 1988; Venkatesh
and Mahajan 1993). On the other hand, event-specific drivers refer to information of a specific
event (e.g., pricing or critics’ reviews for a Broadway show). While genre, pricing, and
advertising are important for all types of events (Weinberg and Shachmut 1978; Weinberg 1986;
Reddy et al 1998; Corning and Levy 2002; Putler and Lele 2003; Leslie 2004), previews and
critics’ reviews are more common for Broadway shows or theatrical events (Reddy et al 1998;
19
Corning and Levy 2002). Because a venue manager’s objective is to increase ticket sales within
a venue, they usually achieve their objective by bundling various events in a subscription
package or scheduling a variety of events to attract consumer attendance. Therefore, the first
two types of drivers are important from a venue manager’s perspective.
Performance-specific drivers, on the other hand, do not limit the number of events
needed but focus on the lower level of characteristics such as days of week or times of day
(Weinberg and Shachmut 1978; Corning and Levy 2002; Putler and Lele 2003; Leslie 2004). In
contrast with the first two types of drivers, this type of drivers is important to event marketers,
especially when they promote a single event that tours across venues and dates. In other words,
when an event has multiple performances across venues and dates, the assortment-related drivers
are not applicable to a single event. The event-related drivers are important yet pricing and
advertising are often planned at a market level and result in a constant effect across all
performances. Consequently, event marketers can only rely on days of week and times of day as
descriptive drivers to differentiate ticket sales of multiple performances.
The fourth type of driver is consumer characteristics. They often refer to consumers’
income levels, willingness to pay, driving distances to venues, and tastes for genres (Moore 1966;
Currim et al 1981; Venkatesh and Mahajan 1993; Leslie 2004). Although this type of drivers
allow event marketers and venue managers to understand consumer preferences better, it is
relatively difficult for event marketers to identify their potential consumers, especially when an
event travels to a new market and there is no previous consumer information available to event
marketers. The last type of driver is seasonality. It generally refers to the season (i.e., spring,
summer, fall, and winter) that events or performances takes place and is commonly used as a
control variable (Weinberg and Shachmut 1978; Weinberg 1986; Corning and Levy 2002).
20
As such, although there are five types of descriptive drivers for event tickets, there are
not plenty of drivers useful for event marketers to understand variations in ticket sales at a
performance level. Therefore, it is important for researchers to investigate additional drivers to
explain such a variation.
Timing of Ticket Sales
To attend an event, consumers have to purchase tickets no later than its performance date.
Throughout an advance-selling period, their purchase timing may range from very early (i.e.,
advance purchase) to the last minute (i.e., spot purchase). Historically, there are few studies in
event ticket purchases. However, there are many in other contexts such as airline ticket
purchases. Thus, we refer to studies in other contexts to discuss firms’ motivation to advance
sell and consumers’ motivation to advance purchase tickets for an event.
Generally, advance selling is common in the service-related industry or in a long lead-
time manufacturing industry. Although it is not necessary to charge lower prices in an advance-
selling period (Xie and Shugan 2001), firms still tend to use a two-stage pricing (i.e., charge
discounted rate in the early market but a regular rate in the spot market) as the primary tool to
attract consumers’ early purchases and secure some demand well in advance (Desiraju and Xie
1999; Shugan and Xie 2000; Cachon 2004; McCardle, Rajaram, and Tang 2004; Tang et al
2004). Two good examples of advance selling are that, first, a venue manager offers a bundle of
events at a lower rate to attract early arrivals of subscription ticket sales (Currim, Weinberg, and
Wittink 1981; Havlena and Holak 1988; Venkatesh and Mahajan 1993) and, second, an airline
company charges a cheaper airfare to attract leisure travelers’ advance purchases (Weatherford et
al 1993; Gallego and van Ryzin 1994; Talluri and van Ryzin 2004).
21
Regarding consumers’ motivation to advance purchase, extant literature has suggested
various reasons for consumers’ early versus spot purchases. In addition to reservation prices,
consumers’ uncertainty toward the consumption state also determines their purchase timing
decisions (Desiraju and Shugan 1999; Shugan and Xie 2000). Specifically, they tend to
procrastinate when they have higher uncertainty about whether they can attend an event in the
future. In contrast, they tend to purchase early when they are more certain to attend in the future.
Other psychological drivers include consumption utility and personal characteristics. For
example, consumers may want to savor their vacation experience better by paying earlier (Prelec
and Loewenstein 1998). Their tendency of being an innovator in their group versus being a
follower also affects their purchase timing decision. Innovators tend to purchase earlier and
influence followers in the later period (Moe and Fader 2002).
Although pricing has been a major factor to affect consumers’ purchase timing, a recent
study by Moe and Fader (2009) illustrated a need to re-evaluate the impact of pricing on the
timing of ticket sales. Specifically, they examined the timing of ticket sales across different
price tiers in the context of event tickets and found that consumers who purchase in advance are
not affected by the price discounts or face values of tickets. This result is very different from
airline tickets. Perhaps it is because there are no so-called “business” or “leisure” buyers in the
context of event tickets. Hence, the reason for consumers to advance purchase event tickets is
not clear and deserves further investigation.
2.2.3 Impact of Scheduling on Ticket Sales
The objective of this essay is to examine the impact of scheduling on ticket sales of
performances of the same event. Because performances are scheduled with different frequencies
across venues and dates, the similarity in these scheduling characteristics varies across
22
performances. We refer to literature in context effect, signaling, retail location, and distribution
services for possible effects of these scheduling characteristics.
Context Effect
Context effect has been extensively studied by behavioral literature (Huber, Payne, and
Puto 1982; Huber and Puto 1983; Simonson 1989) where researchers investigate how
alternatives of dominated or dominating attributes influence consumer choices. In general, this
stream of literature proposes violation of proportionality (Luce 1959). Researchers examine
situations when products of similar attributes are more attractive (i.e., attraction effect) and when
they are substitutable (i.e., substitution effect) and where alternatives in the middle level of
attributes are more favorable (i.e., compromise effect) (Huber et al 1982; Huber and Puto 1983;
Simonson 1989). In other words, when consumers have uncertain preferences to product
attributes, they would choose alternatives based on various reasons (Simonson 1989). A
dominating alternative may have a higher choice share under the attraction effect although it may
have a lower share under the substitution effect (Huber at al 1982; Huber and Puto 1983).
However, it is also likely that consumers prefer the new added alternative that has compromised
attributes under the compromise effect.
The relevance between the context effect and the scheduling effect is that similarity
between alternatives (i.e., performances in this case) could influence consumer perception and
choice decisions. As consumers pay more attention to alternatives that share a similar attribute
level (Huber et al 1982), they might focus on evaluating performances in the same or near
venues (or on the same or near dates) and find these performances more attractive or substitutive.
At the aggregate level, the similarity in these scheduling characteristics could further determine
how well these performances sell. On the other hand, performances of compromised attributes
23
(e.g., in a preferred venue but on a less preferred date or vice versa) might also sell differently
from other dominating or dominated performances.
Signaling
Signaling has been studied in marketing to address the issue of product quality. The
assumption for signaling to work is under the separating equilibrium (Chu and Chu 1994) that
manufacturers are credible and have high transaction costs to signal (Moorthy and Srinivasan
1995). Therefore, credible manufacturers can use money-back guarantees (Moorthy and
Srinivasan 1995) or an extended warranty (Padmanabhan and Rao 1993; Lutz and Padmanabhan
1995; Soberman 2003) to signal the quality of their product. They can also sell their products in
reputable retailers for consumers to infer the reputation of manufacturers, especially when their
reputation is not directly observable to consumers (Chu and Chu 1994; Purohit and Srivastava
2001).
We relate signaling with performance scheduling because it is likely that event marketers
want to signal the performing quality of an event to a market. Because sending a false signal can
be expensive (e.g., incremental costs for multiple performances and revenue losses from empty
seats), we assume event marketers are credible. They could increase the number of
performances in a market to signal event popularity. They could also allocate these
performances densely in some venues or on particular dates to suggest popularity or desirability
of these venues or dates. As a result, depending on which signal event marketers want to send,
they will schedule performances differently. After consumers receive the signal and believe the
credibility, they could adjust their preferences and make purchase decisions accordingly.
24
Retail Location
In the retail industry, store locations influence consumers’ shopping destination and store
choices (Fotheringham 1988). Therefore, collocating or keeping some distance away from
primary competitors has been an important issue in retail location literature (Mazzeo 2002;
Vitorino 2007; Zhu and Singh 2009). Although some studies show that shorter geographic
distance between retailers leads to direct competition and decreases in store profitability (Watson
2005; Orhun 2005; Seim 2006; Thomadsen 2007; Zhu and Singh 2009; Zhu, Singh, and
Manuszak 2009), other studies indicate benefits for retailers to locate closely (Mazzeo 2002;
Vitorino 2007; Zhu, Singh, and Dukes 2007). For example, retailers collocating in a shopping
center provide consumers a larger product assortment (across stores) such that these retailers can
attract more consumers to the shopping center (Vitorino 2007; Zhu et al 2007). Motels, on the
other hand, collocate in a highway exit to facilitate consumer search and attract more traffic
(Mazzeo 2002).
The relevance between retail location and performance scheduling is that event marketers,
like retailers, have to consider how far in distance or time they should keep performances apart
to avoid substitution within the same event (i.e., cannibalization). On the other hand, event
marketers might also consider scheduling performances closely enough to provide more
alternatives or signal the popularity to attract more demand. Therefore, if the substitution
between performances outweighs the collocating synergy, performances scheduled closely could
suffer fewer ticket sales and slow sales arrival (because consumers can always wait until the last
minute). Otherwise, densely scheduled performances would experience more ticket sales and
faster sales arrival.
25
Distribution Services
In addition to abovementioned spatial differentiation in retail location, retailers also try to
differentiate themselves from other competitors by improving their distribution services
(Betancourt 2004). Distribution services generally refer to the extent of services that retailers are
able to deliver to consumers. These services include product assortment, assurance of product
delivery at consumers’ desired time or in the desired form, spatial accessibility between stores
and consumer residences, and so on (Betancourt 2004). For example, retailers can provide a
larger assortment of products, extend business hours, or open more store locations to enhance
their distribution services.
In the live performance industry, similarly, both venue managers and event marketers
may aim to enhance their distribution services. To accommodate consumers who have different
tastes, venue managers may improve their assortment by scheduling a variety of events in their
venues. On the other hand, event marketers may consider consumers who have state uncertainty
about to which venue they want to go and which date they can attend. In this case, they might
schedule performances in multiple venues and disperse performance dates to allow consumers to
attend at their own convenience. In other words, if event marketers schedule performances
sparsely across venues and dates, these performances could accommodate more consumers at
different times and in different venues. In this way, sparsely scheduled performances will sell
more than densely scheduled performances.
2.2.4 Conceptual Framework
Although there may be several reasons to explain event marketers’ scheduling decisions
and consumers’ decision process, this information is unfortunately unobservable to researchers.
Hence, we can only use the abovementioned literature to speculate potential scheduling effects.
26
To sum up, if event marketers want to signal quality, they will increase the number of
performances and schedule these performances closely in distance or time to attract more
consumers. Then, consumers will perceive these performances to be more popular.
Consequently, more consumers will attend these performances and these consumers will be more
likely to purchase tickets in advance. On the other hand, if event marketers schedule
performances to enhance their distribution services, they will schedule performances across
multiple venues and disperse performance dates farther apart. In this way, performances
scheduled farther apart are able to accommodate more consumers and sell more tickets. Yet, it is
still possible for consumers to procrastinate given their uncertainty for the consumption state. As
such, based on different streams of literature, we can anticipate different scheduling mechanisms
and consumer responses.
However, if event marketers do not have a specific scheduling strategy but schedule
performances to comply with operational constraints such as venue availability and seating
capacity, their scheduling process is similar to airline and movie scheduling. Then, the impact of
scheduling could be positive or negative. According to the retail location literature, event
marketers may expect performances in close distance or time to have a substitutive relationship
and cannibalize ticket sales. They may also expect a collocation synergy between performances
to attract more demand. From consumers’ perspective, they may make their own inferences
about these performances based on different contexts. According to the attraction effect, they
may perceive performances of similar scheduling characteristics to be more attractive. Hence,
these performances will sell more tickets and these ticket sales will arrive earlier. In contrast,
consumers may perceive these performances to be highly substitutable (i.e., substitution effect)
27
and delay their purchase timing. If so, these performances will sell less and sell more slowly
than do those of dissimilar scheduling characteristics.
However, event marketers may not just simultaneously schedule performances densely
(or sparsely) across venues and dates. They may sometimes schedule performances densely in
certain venues to signal venue popularity yet disperse performance dates to accommodate
consumers’ uncertainty of attendance timing. Similarly, they may schedule performances
densely on certain dates as popular leisure activities but allocate these performances in several
distant venues to increase spatial accessibility. Consequently, if event marketers have separate
objectives and schedule accordingly, consumers will find performances scheduled closely in
distance (or time) but distantly in time (or distance). In this way, performances have different
scheduling characteristics across venue locations and performance dates, and consumers will
evaluate a geographic and temporal schedule separately. Hence, the effect of geographic
scheduling may differ from the effect of temporal scheduling. We allow this possibility and
empirically test these scheduling effects.
On the other hand, to study the impact of performance schedules on ticket sales, we also
have to control for the attractiveness of performances on different days of week and the potential
endogeneity between scheduling decisions and expected market response. In other words, if
event marketers have some knowledge about how a market responds to a performance schedule
and then use this knowledge to schedule performances, the performance schedule will be
endogenously set and the scheduling impact will be biased. In case the endogeneity exists, we
propose to examine the effect of performance schedule on ticket sales and control for the
endogeneity simultaneously. Figure 2-1 below summarizes our conceptual framework.
Figure 2-1: Conceptual Framework
28
2.3 Model Development
2.3.1 Model Overview
To test the potential impact of scheduling, our modeling objective is to measure the
scheduling characteristics of each performance and study the impact of these scheduling
characteristics on the number and timing of ticket sales. Thus, our model development consists
of four steps. First, we measure the scheduling characteristics to capture the similarity or
dissimilarity in venue locations and performance dates across performances. Second, we
examine the number of ticket sales across performances. Because it is possible that some
consumers evaluate the venue locations and performance dates but do not attend (i.e., non-
buyers), we incorporate the impact of scheduling characteristics on the size of non-buyer
segment. In this way, event marketers can understand how much market potential they have
captured and how much they have missed out. Third, we examine the timing of ticket sales
across performances. Because each performance sells tickets at a different rate and experience
different timing of ticket sales in an advance-selling period, it is important to capture
heterogeneous sales patterns and explain the differences. Finally, we consider a possible
endogeneity between performance scheduling and market response. If event marketers know the
effects of scheduling on the number of ticket sales, they could allocate performances based on
the positive or negative effect and expected performance attractiveness. Under this situation,
performance scheduling is endogenous with market response (Manchanda, Rossi, and
Chintagunta 2004). It is important to control for this endogeneity to ensure unbiased model
results. Figure 2-2 below summarizes our model development and we discuss each modeling
element in turn.
Figure 2-2: Model Overview
29
2.3.2 Scheduling Characteristics
To capture the similarity or dissimilarity in scheduling characteristics of performances,
we refer to a performance schedule of an event and differentiate between performances based on
their venue locations and performance dates. We create two measures to represent the
scheduling similarity in this performance schedule to understand how performances are
scheduled closely or distantly across venues and dates.
Specifically, we take the inverse geographic distance (in miles) between performances as
the geographic density measure and the inverse temporal distance (in days) between performance
dates as the temporal density measure (Fotheringham 1988). In this way, performances
scheduled in the same or proximate venues will have a higher value in geographic density to
represent similarity in the geographic schedule. Performances scheduled on the same or near
dates will have a higher value in temporal density to show similarity in the temporal schedule.
Therefore, our specifications for the density measures are as follows:
(1) uE0
j
=
1
} - 1
1
miles
jji
]
j
|
=1,j=ji
(2) TNP
j
=
1
} - 1
1
(1 +uays
jji
)
]
j
|
=1,j=ji
where GEO
j
is the geographic density for performance j (j=1,2,.., J), TMP
j
is the temporal
density for performance j, miles
jj’
represents the geographic distance between venues of
performance j and j’ (j?j’), and days
jj’
represents the temporal distance between performance
dates of j and j’.
To compute the distance between venue locations and performance dates, we use driving
distance (in miles) between the venues of j and j’ as miles
jj’
. We also specify miles
jj’
=1 for
30
performances in the same location to avoid the denominator equal to zero. In addition, we use
the absolute value of distance (in days) between performances j and j’ as days
jj’
. However, we
specify the denominator as (1+days
jj’
) in equation (2) to avoid performances on the same date
having a zero temporal distance.
Therefore, with respect to a target performance j, after taking its average (inverse)
geographic and temporal distance to other performances, GEO
j
and TMP
j
suggest its average
geographic and temporal density. The higher GEO
j
and TMP
j
of this performance, the closer this
performance is to other performances. As such, this performance is densely scheduled around
other performances and has higher similarity in venue locations and performance dates.
2.3.3 Number of Ticket Sales
To examine the number of ticket sales across performances and understand how much
market potential event marketers miss out, we apply a competing destination model by
Fotheringham (1988) and specify the share of each performance and the non-buyer segment as a
function of scheduling characteristics. By doing so, we can understand whether geographic and
temporal scheduling influence ticket sales at a performance level. In addition, using sales shares
of individual performances and the population size in a target market, we can obtain the number
of ticket sales expected at a performance level. We can also obtain a market penetration rate by
summing sales shares across performances.
Competing Destination Model
Among various extensions of the logit model, we consider the competing destination
model proposed by Fotheringham (1988) as a good alternative. The competing destination
model extends the traditional logit model by releasing the property of independence from
irrelevant alternatives (IIA) (Luce 1959; McFadden 1974). It examines consumers’ spatial
31
choice set as a function of geographic distance between stores. Then, the composition of the
spatial choice set further influences consumers’ store choices. If stores of shorter geographic
distance have higher chance to be in consumers’ choice set, these stores will attract more
consumers and have higher choice shares. However, if stores of shorter geographic distance
have a lower chance to be in consumers’ choice set, these stores will substitute one another and
have lower choice shares. The specification of the competing destination model is as follows
(Fotheringham 1988):
(3)
P
ì]
=
exp(I
ì]
) · l
ì
(] e H)
? exp(I
ì]?
) · l
ì
(]? e H)
]
j
?
=1
where P
ij
denotes the probability that consumer i shops in retail outlet j, V
ij
represents the
deterministic utility of retailer j to consumer i, and l
I
(j e N) is the likelihood that retailer j is in
consumer’s spatial choice set. After aggregating P
ij
across consumers, P
j
represents the market
share of a retailer j in a studied market of interest (González-Benito 2005).
Moreover, to measure the likelihood of spatial choice set, Fotheringham specified the
likelihood as a function of average inverse geographic distance between retail stores
(Fotheringham 1983; Fotheringham 1988) and empirically test the role of geographic distance:
(4)
l
ì
(] e H) =
`
Í
Í 1
} -1
w
j?
u
jj?
]
j?=1
j=j?
/
¹
¹
?
where d
jj’
is the geographic distance between a target store j and a competing store j’, w
j’
is the
weight for the competing store j’, and ? is the parameter indicating the role of geographic
distance. More specifically, if ?> 0, stores of shorter geographic distance to other stores will
have higher chance to be included in consumers’ choice set. If ?< 0, in contrast, stores will have
32
lower chance to be included. Finally, if ?= 0, geographic distance does not affect the
composition of consumers’ spatial choice set.
Extension of Competing Destination Model
To examine the separate effects of geographic and temporal scheduling, we extend the
competing destination model by incorporating the geographic and temporal density values and
allowing separate parameter values for these scheduling characteristics. In addition, we include
the non-buyer segment in a market as another alternative for potential buyers. In this way, we
can understand how scheduling characteristics affect the sales share of performances and the size
of non-buyer segment. Our adaption is as follows:
(5) P(]) =
exp(I
]
) · 0E0
]
0
1
IHP
]
0
2
1 + ? exp(I
]?
) · 0E0
]?
0
1
IHP
]?
0
2
]
j
?
=1
where P(j) denotes the sales share of performance j, V
j
represents the attractiveness of this
performance (which we will discuss later), GEO
j
and TMP
j
are the geographic and temporal
density values of performance j, and parameters ?
1
and ?
2
represent the scheduling effects.
If ?
1
or ?
2
> 0, geographic or temporal scheduling has a positive effect. Performances of
similar venue locations or performance dates will enjoy higher sales share. In contrast, if ?
1
or ?
2
< 0, geographic or temporal scheduling has a negative effect. Performances of similar venue
locations or performance dates will substitute one another and suffer from sales cannibalization.
However, if ?
1
or ?
2
= 0, scheduling has no impact on sales share. Moreover, we can use the
population size in a target market (which we will discuss in the data section), POP, to calculate
the expected number of ticket sales for any performance: Sales(j) = P0P × P(j). We can also
obtain the market penetration rate by summing sales share across performances [i. e. , ? P(j)
]
j=1
¸
and understand the size of non-buyer segment as 1 - ? P(j)
]
j=1
.
33
To measure and control the effect of performance attractiveness (V
j
), event ticket
literature has suggested various drivers for ticket sales that can be a good proxy for attractiveness.
However, because we focus on analyzing multiple performances of the same event, only
performance-related drivers are applicable. Therefore, we specify V
j
as a function of days of
week indicators (Friday, Saturday, and Sunday). The reason for us to choose these three days of
week indicators is that past studies indicate that performances on those days have higher
attractiveness (Corning and Levy 2002; Putler and Lele 2003). Hence, we specify performance
attractiveness as a control covariate for the number of ticket sales and incorporate a random error
term for unobserved attractiveness:
(6) v
j
= ?
0
+ ?
1
FRIBAY
j
+ ?
2
SAT0RBAY
j
+ ?
3
S0NBAY
j
+ ?
j
where ?
j
~N(u, ?
2
)
where ?
0
reflects the baseline attractiveness, ?
1
, ?
2
, and ?
3
suggest the effects of days of week on
incremental performance attractiveness for a Friday, Saturday, or Sunday performance, and E[V
j
]
represents the expected performance attractiveness.
To sum up, we adapt the competing destination model to understand the effects of
geographic and temporal scheduling on the sales share of performance and the size of non-buyer
segment. We also control for the performance attractiveness due to the days of week effects.
Although a linear regression or spatial model can also examine the number of ticket sales across
performances, our model specification is more appropriate than a regression or spatial model. A
linear regression is commonly used in event tickets literature (Moore 1966; Weinberg and
Shachmut 1978; Weinberg 1986; Reddy et al 1998; Corning and Levy 2002). However, it
cannot examine the sales share of each performance and the non-buyer segment at the same time
In other words, a linear regression does not allow us to understand how much of the market
performances have captured and how much they have left untapped. In contrast, our proposed
34
model can accomplish all of these limitations of a linear regression model. In comparison to a
spatial model, the model proposed in this dissertation specifically measures the effects of
performance schedules as drivers of ticket sales and not just modeling the spatial correlation
between performances (Bradlow et al 2005).
2.3.4 Timing of Ticket Sales
To examine the timing of ticket sales across performances, we first consider a typical
sales pattern for a performance. In general, a performance starts selling tickets 12 weeks prior to
its performing date. It usually sells fewer tickets in the beginning of its advance-selling period
and obtains more sales over time with the most arriving in the later period or the week of the
performance. Given this typical pattern, however, each performance still has a different sales
rate. Some performances experience ticket sales arrived earlier (i.e., earlier timing of ticket sales)
but others experience ticket sales arrived later (i.e., later timing of ticket sales).
Figure 2-3 below illustrates three patterns of ticket sales in an advance-selling period.
First, the solid line in Figure 2-3 (Case 1) is one common pattern, where consumers have low
probability to purchase well in advance. As time passes, the probability of a ticket transaction
slowly increases and peaks at the week of performance. However, there are some situations
where consumers expect performances to be of greater performance attractiveness or higher
popularity. As a result, they are more willing to purchase earlier and result in more ticket sales
arriving in the middle of an advance-selling period, as shown in the dotted line of Figure 2-3
(Case 2). Yet, there is another case when consumers think performances of lower performance
attractiveness and/or of higher substitutability. In this instance, they do not want to commit early
and wait until the week (or the day) of a performance. Therefore, ticket sales for such a
35
performance are very low in the entire advance-selling period and only peak in the spot market.
The broken line in Figure 2-3 (Case 3) represents this pattern.
Figure 2-3: Pattern of Ticket Sales Over Time
Weibull Hazard Model
Although performances usually follow a similar pattern as seen in Figure 2-3 (Case 1),
there still exists variability of sales pattern among performances (e.g., Case 2 and Case 3 in
Figure 2-3). To account for variations in sales rate across performances, we need a model that is
flexible enough to capture various sales patterns and examine the performance-specific sales
rates. Hence, we specify a Weibull hazard model to fit the timing of ticket sales because of its
flexibility in capturing various sales patterns, as shown in Figure 2-3. A Weibull hazard process
has the following properties:
(7) b
]
(t) = z
]
c
]
t
c
]
-1
S
]
(t) = c
-x
]
t
c
]
F
]
(t) = 1 -S
]
(t) = 1 -c
-x
]
t
c
]
where, with respect to a performance j, h
j
(t) is the instantaneous hazard rate for a ticket purchase
made at time t given this transaction has not yet been made, S
j
(t) is the survival rate for a ticket
purchase that has not yet occurred up to time t, and F
j
(t) is the cumulative probability for sales to
arrive over time. More specifically, ?
j
is the slope parameter for performance j to represent how
fast ticket sales arrive (?
j
>0), and c
j
is the shape parameter to capture an overall pattern of ticket
sales (c
j
>0).
For any discrete time t (e.g., week) in an advance-selling period, the probability of a
ticket purchase becomes:
36
(8)
P
]
(t) = F
]
(t) -F
]
(t -1) = c
-x
]
(t-1)
c
]
-c
-x
]
t
c
]
However, ticket sales in the context of a live performance event have to arrive no later
than the performance date. We adjust the probability of a ticket purchase in the spot market (i.e.,
the week of the performance) as follows:
(9)
P
]
(I
]
) = 1 -c
-x
]
(1
]
-1)
c
]
where T
j
is the number of advance selling weeks for performance j.
In addition, because we often observe seasonality (e.g., Thanksgiving, Christmas, etc.) or
marketing activities in an advance-selling period, we can include a time-varying covariate to
control for resulting sales bumps. Therefore, we rewrite equations (7), (8), and (9) as follows:
(10)
b
]
(t) = z
]
c
]
t
c
]
-1
c
[
]
X
]t
S
]
(t) = exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_
F
]
(t) = 1 -exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_
P
]
(t) = exp_-z
]
|u
c
]
- (u -1)
c
]
]
t-1
u=1
c
[
]
X
]u
_ -
exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_ vt = 1·(I
]
-1)
P
]
(I) = exp_-z
]
|u
c
]
- (u -1)
c
]
]
t-1
u=1
c
[
]
X
]u
_
where X
jt
is a time-varying covariate or seasonality indicator and ?
j
is its associated parameter.
Consequently, the timing of ticket sales for each performance changes with its parameters ?
j
, c
j
,
and ?
j
.
37
Heterogeneity in Sales Patterns
To capture and explain the heterogeneity in sales pattern, we further specify the sales rate
of each performance as a function of its geographic and temporal density measures. We also
include two control covariates to ensure unbiased effects of these scheduling characteristics. The
first covariate is the length of advance-selling period because Moe and Fader (2002, 2009) found
that sales tend to arrive more slowly under a longer advance-selling period. In addition, because
consumers may purchase tickets much earlier when they expect performance to be more
attractive, we incorporate the expected performance attractiveness as the second control
covariate. Consequently, we specify the Weibull parameters (?
j
and c
j
) and the covariate effect
(?
j
) to follow the multivariate normal distribution. We take the log transformation for the
Weibull parameters to ensure positive values:
(11)
_
log(z
]
)
log(c
]
)
[
]
_ ~HIN(µ
j
, L
1
)
where
µ
j
= y
û
+y
1
0E0
]
+y
2
IHP
]
+y
3
I
]
+y
4
E|I
]
]
where ?
j
is the vector of expected Weibull parameters and covariate effect, GEO
j
and TMP
j
are
the geographic and temporal density measures in equations (1) and (2), T
j
is the number of
advance selling weeks for performance j, E[V
j
] is the expected performance attractiveness in
equation (6), and ?
0
, ?
1
, ?
2
, ?
3
, and ?
4
are the vectors of parameters for these covariates.
Therefore, using the parameter results in equation (11), event marketers can understand why
some performances experience earlier timing of ticket sales while other experience later timing
of ticket sales.
38
2.3.5 Endogeneity in Performance Scheduling
Finally, we consider a possible endogeneity in scheduling decisions. Since the live
performance industry has been established and event marketers have scheduled for a variety of
events, event marketers might have incorporated their experiences into a scheduling process. In
other words, how densely event marketers allocate performances across venues may be
dependent on the effect of geographic scheduling, and how densely event marketers allocate
performances across dates may be dependent on the effect of temporal scheduling. Moreover, it
is also likely that event marketers increase the total number of performances and schedule those
on weekend to increase the performance attractiveness. If so, the geographic and temporal
density values vary with the scheduling impact and the expected performance attractiveness.
To control for this type of endogeneity, we refer to a modeling approach proposed by
Manchanda et al (2004). In their research of pharmaceutical detailing, they mentioned that sales
representatives visit various doctors with different frequencies, and sales representatives
determine the frequencies based on how many prescriptions a doctor writes without any detailing
and how strong the effect of detailing is if they visit this doctor. They pointed out the
endogeneity between the decision of detailing and the effect of detailing, and further proposed a
model to correct this endogeneity.
In other words, they specified the expected prescription volume from a doctor as a
function of its baseline volume, the magnitude of detailing, and the detailing effect:
(12) ln(p
ì
-
) ÷ _
[
0ì
(1 -[
2ì
)
_ +_
[
1ì
(1 -[
2ì
)
_ Ðct
where ln(p
ì
-
) is the expected volume of prescription, j
[
0i
(1-[
2i
)
[ is the baseline prescription volume,
Det is the frequency of office visits, and j
[
1i
(1-[
2i
)
[ is the effect of detailing.
39
To model the endogenous detailing behavior, they specify the expected value of detailing
(ln(p
ì
)) as a function of the baseline prescription volume and the effect of detailing:
(13) ln(p
ì
) = y
0
+y
1
_
[
0ì
(1 -[
2ì
)
_ +y
2
_
[
1ì
(1 -[
2ì
)
_
In this way, if detailing is indeed endogenous, the parameter ?
1
or ?
2
will be significantly
different from zero, and the endogeneity between detailing and its effect is under control.
To control for the possible endogenous scheduling behaviors, we take the same modeling
approach as Manchanda et al (2004). We specify geographic and temporal density measures
(GEO
j
and TMP
j
) as a function of expected performance attractiveness (E[V
j
] in equation 6), and
the effects of scheduling characteristics (?
1
and ?
2
in equation 5). Because GEO
j
and TMP
j
in
equation (1) and (2) are between 0 and 1, we take the logit transformation for these density
measures and specify them following the multivariate normal distribution:
(14)
_
logit(0E0
]
)
logit(IHP
]
)
_ ~HIN(m
j
, L
2
)
where
j
m
1]
m
2]
[ = ç
û
+ç
1
E |I
]
] +ç
2
_
0
1
0
2
_
As such, if scheduling decisions are indeed endogenous and reflect on the density
measures, the parameters ?
1
or ?
2
will be significantly different from zero and this endogeneity
will be taken into account.
2.4 Data Description
We contact a national ticket seller to obtain a dataset of live performance events.
Because of the confidentiality agreement with our data provider, we cannot disclose our data
40
provider or the names of events. However, we will describe the nature of the events and the
behaviors observed in the dataset.
In this dataset, there were two events touring across several cities in the U.S. Each event
had a different number of performances in a city and lasted for a different period. For each
performance, we observe its venue location and performance date. In addition, we also have
detailed information regarding when tickets were purchased, for how much money, at which
price levels, and through which channels. Moreover, we are also able to observe the pattern of
ticket sales because transactions were recorded at a daily level. Table 2-1 provides a detailed
description of each field in our data set, which contains abundant information about the live
performance event and has many research opportunities for marketing researchers.
Table 2-1: Description of Variables in the Dataset
2.4.1 Description of Events
The events we have in the dataset are two popular family events. They are live
entertainments that targets on families with young children. In general, there are several types of
family events such as children’s music and theater (e.g., The Wiggles and Dora the Explorer
Live!), circus (e.g., Ringling Brothers and Barnum & Bailey and UniverSoul Circus), ice shows
(e.g., Disney on Ice series), magic shows (e.g., Xtreme Magic and Steve Wyrick ), and so on.
They usually travel across the U.S. or stay in a local market such as Las Vegas. The family
events we have are within the abovementioned categories.
For the two events we have, one sold 2.2 millions of tickets between January and June
2004 and travelled across 50 cities on 245 dates for 449 performances. The other sold 0.8
millions of tickets between January and May 2004 and travelled across 17 cities on 85 dates for
157 performances. The reason for the sales discrepancy is that one event had multiple
41
performing groups touring across cities simultaneously while the other had only one group
performing in our observed time span. Although these two events travelled to numerous cities
during their tours and had three cities in common (i.e., Jacksonville, FL; Miami, FL; Phoenix,
AZ), these events did not perform in those cities at the same time but at least 2.5 months apart.
Therefore, we assume there was no direct competition between these events to affect ticket sales.
Although both events are representative in terms of their ticket sales and the number of
performances, we take only one event in this dissertation to keep the event attractiveness
constant and examine merely the scheduling effect on ticket sales. In this way, once we confirm
a significant scheduling effect, we can further incorporate the renown of different events as an
extension. As such, we choose the event that had more performances. Among 50 cities that this
event toured, we also find some cities were within the same metropolitan markets (e.g., New
York and other metropolitan markets). This observation suggests that the event had multiple
stops in some markets and had higher variations in both geographic and temporal schedules.
Hence, we further select performances in the New York metropolitan market
1
and examine the
impact of geographic and temporal schedules on ticket sales in essay one. In summary, this
event had performances in Continental Arena in East Rutherford, NJ, Nassau Coliseum in
Uniondale, NY, Madison Square Garden in New York, NY, and Sovereign Bank Arena in
Trenton, NJ, respectively between March and June 2004 for 70 performances.
2.4.2 Description of Ticket Sales
Because a dataset of event tickets is not commonly available in marketing, we first
examine ticket sales by price levels and channel types to describe how much money people
usually pay and through which channel. Then, we examine the distribution of ticket sales across
1
We follow the Census Bureau data to define the boundary of a metropolitan market.
42
performances to understand how many tickets each performance sells and when ticket sales
arrive.
Ticket Sales by Price Levels
We first aggregate ticket sales by price levels and performances to examine any different
sales distributions across price levels. On average, the admission fees to a performance
(including face value, facility fees, and service charges) are $30.44 and there are about six price
tiers for consumers to choose. Although seating quality in a venue determines the price levels,
83% of ticket sales are contributed by mid-priced levels (i.e., price levels 2, 3 and 4) with
average price ranges from $20 to $50.
Specifically, price level 3 (mean price=$25.31, std= 3.66) represents 50% of ticket sales,
and price level 4 (mean price= $20.02; std= 3.17) and price level 2 (mean price= $50.16; std=
6.89) contribute 21% and 12% of ticket sales, respectively. In addition, we find the average
admission fees are relatively equal across venues and days of week. In other words, price
variations are within a performance (via price levels) but not across venues or days of week.
Ticket Sales by Channel Types
Next, we aggregate ticket sales by channel types and performances to examine ticket
sales across channels. Although consumers can purchase tickets through any of the six available
channels (i.e., primary box office, secondary box office, Hermes (automatic phone), Internet,
outlet, and phone), we find majority of ticket sales are made through the primary box office
(62% of ticket sales), following by the Internet (22% of ticket sales) and a ticket outlet (11% of
ticket sales). A possible reason for a primary box office to be a dominating channel choice is
that consumers do not have to pay for the convenience charges when they buy tickets in a box
office.
43
Ticket Sales by Performances
We also aggregate ticket sales across performances. As Figure 2-4 shows, a performance
on average sells 8,316 tickets but has its standard deviation being 3,525. Upon a closer look of
the sales distribution by days of week (based on performance dates), we find that weekend
performances tend to have more ticket sales than weekdays. Yet, the variation of ticket sales on
the same day of week is still prominent. The boxplot in Figure 2-5 summarizes the sales
distributions by days of week and indicate a clear variation even on the same day of week. For
example, a Friday performance sells 8,112 tickets on average but has a big standard deviation of
2,509, and a Saturday performance has average ticket sales of 9,552 but has the standard
deviation being 3,808. According to Figure 2-5 and the observation that each performance has
similar price levels, it is convincing that there must be additional factors to explain the variations
in ticket sales. Although one can argue that ticket sales are due to venue capacities, we find the
sizes of capacity in the four venues are similar and there are no sold out for any performance.
Hence, we do not consider the impact of venue capacity on ticket sales in this essay.
Figure 2-4: Sales Distribution by Performances
Figure 2-5: Sales Distribution by Days of Week
Finally, we aggregate daily ticket sales into weekly sales to examine the sales pattern for
each performance. On average, ticket sales arrive up to 15 weeks prior with the range between
11 and 19 weeks. Table 2-2 presents the ticket sales across performances and the breakdown of
weekly sales throughout an advance-selling period in Table 2-2. On average, a performance sells
8,316 tickets with 24% of sales arrived one month prior, 33% of sales arrived 2 to 4 weeks prior,
and 42% of sales arrived in the week of performances. Figure 2-6 shows a sales pattern for a
randomly chosen performance and it is a common pattern in our dataset. However, given this
44
similar pattern across performances, some performances experience ticket sales much earlier
than other performances (e.g., 59% vs. 3% of total sales arrived in the early stage) yet other
performances do not have as many ticket sales arrived in the last week (e.g., 75% vs. 11% of
total sales arrived in the spot stage). The boxplot in Figure 2-7 demonstrates the heterogeneity in
the timing of ticket sales across performances throughout an advance-selling period.
Table 2-2: Summary of Ticket Sales
Figure 2-6: Weekly Sales Pattern of a Performance
Figure 2-7: Heterogeneity in Timing of Ticket Sales across Performances
2.4.3 Covariate Specifications
Before we estimate the proposed model, there are several covariates not directly provided
in the dataset that require our attention. They are the geographic and temporal density measures
(GEO
j
and TMP
j
), the days of week indicators (FRIDAY
j
, SATURDAY
j
, SUNDAY
j
), the
estimated population size in the target market (POP), the length of advance-selling period (T
j
),
and the time-varying covariate (X
jt
). We discuss and specify these covariates in turn.
To compute the density measures, we first refer to venue locations and use the Google
TM
maps to find the driving distance (in miles) between venues. The numbers in Figure 2-8 indicate
the venue locations in the New York metropolitan market and represent the travel sequence
across venues. In addition, the numeric values between venues represent the mileage between
venues (i.e., mile
jj’
). We summarize the geographic distance between venues and the number of
performances in each venue in Figure 2-9 and follow equation (1) to calculate the geographic
density for each performance. Next, we refer to performance dates to calculate the temporal
distance (in absolute values) between performances (day
jj’
). According to the performance dates
45
and their temporal distance to others in Figure 2-10, we apply equation (2) to compute the
temporal density for every performance.
Figure 2-8: Venue Locations and Driving Distances
Figure 2-9: Summary of Geographic Distance
Figure 2-10: Summary of Temporal Distance
To compute the days of week indicators, we refer to performance dates to identify on
which days of week performances take place (i.e., FRIDAY, SATURDAY and SUNDAY).
Although some performances are scheduled on the same date, unfortunately our data does not
indicate time of day for the performances.
Next, we compute the population size in the target market (POP). Given that the event
targets families with young children, we define the target market as the population of families
with children under 10 years old. To compute the market size, we refer to the U.S. Census
Bureau for the 2000 data to find the total number of families with children under 18 years old
and the percentage of all children who are under 10 years old. We multiply these two numbers
to get the family population with children under 10 years old and then multiply the average
family size to get the population size in the target market. According to these calculations, there
are 4,082,615 potential consumers in the New York metropolitan market. We use this market
size and ticket sales across all performances to find the market penetration rate to be 14%, which
means the non-buyer segment represents 86% of the target market.
To measure the number of advance selling weeks (T
j
in equation 11) as a control
covariate, we compute the difference between the first sale date and the performance date for
each performance. Then, we divide this number by seven to convert the advance-selling period
46
to weeks. Although tickets may be available for sales prior to the first sale date, we think our
approach a good proxy given very few sales arrived in the early selling period.
Finally, because we observe Christmas within the advance-selling period for some
performances, we incorporate a time-varying indicator (X
jt
) in equation (10) to control for a
possible pre-Christmas shopping and resulting sales bumps. For each performance j, we code
X
jt
=1 if the advance selling week t is consistent with the pre-Christmas shopping week (i.e., 7
days prior to Christmas). Otherwise, X
jt
=0. Table 2-3 summarizes the descriptive statistics of
covariates.
Table 2-3: Descriptive Statistics of Covariates
2.5 Model Estimation and Benchmark Comparison
2.5.1 Estimation
We choose the Bayesian statistics approach to estimate the number of ticket sales, the
timing of ticket sales, and the endogeneity in a performance schedule simultaneously. We
specify appropriate and diffuse priors for our parameters in the WinBUGS program and estimate
the model over 40,000 iterations. After checking the convergence criteria, we check the
autocorrelation plots for all covariates, discarded 30,000 iterations for burn-in, and use the
remaining iterations as the posterior distribution. We specified the prior distribution of
parameters below:
Priors for modeling the number of ticket sales:
o
0
~N(-6,1u) For the baseline performance attractiveness
o
ì
~N(u,1uu) For the effects of days of week (where i=1, 2,3)
0
ì
~N(u,1uu) For the effect of scheduling characteristics (where i=1, 2)
o
2
~I0(u.1,u.1) For the variance of the performance attractiveness
47
Priors for modeling the timing of ticket sales:
y
ìk
~N(u, 1uu) For the Weibull parameters and the covariate effect
(where i=0, 1, 2, 3, 4 and k= 0, 1, 2, 3)
L
1
-1
~Weibull(I
3
, S)
For the variance-covariance of the Weibull parameters and the covariate
effect
Priors for modeling the endogeneity in performance schedule:
¢
ìk
~N(u,1uu) For the expected geographic and temporal density values
(where i=0, 1, 2 and k= 1, 2)
L
2
-1
~Weibull(I
2
, 2)
2.5.2 Benchmark Comparison
Before presenting our model results, we specify benchmark models to compare with our
proposed model to rule out alternative explanations for our proposed scheduling effects. First,
because the central focus of our modeling efforts is to examine the impact of scheduling
characteristics, one ideal benchmark model is to exclude any scheduling effect but only
incorporate performance attractiveness (i.e., Benchmark 1). Second, some venue locations are
more popular than others. For example, a venue in the New York city might be more attractive
than another venue in Uniondale. Therefore, we consider the second benchmark that
incorporates the venue-specific indicator variables. Finally, Population density around the venue
locations could also explain ticket sales. In other words, event marketers may schedule more
performances in a specific venue because the population density in this venue is high and the
scheduling decision is simply to meet potential market demand in this venue. To rule out this
alternative explanation, we extend our proposed model by including the population density
around each venue location of performances as another explanatory variable (i.e., Benchmark 3).
We estimate our proposed and benchmark models to compare the model fit using the
deviance information criteria (i.e., DIC, Spiegelhalter et al 2002):
48
(15) ÐIC = Ð(ç
) +2pÐ
where Ð(ç
) is the deviance evaluated at the posterior means ç
and pD is the effective number of
parameters in a model, calculated as the difference between the posterior deviance and the
deviance of the posterior mean.
Table 2-4 summarizes the model fit across the benchmark and proposed models.
According to the DIC reported for every model, we find that incorporating the scheduling effect
is definitely superior. Although adding city effects improves the DIC from 157,134 (Benchmark
1) to 156,459 (Benchmark 2), city effects cannot explain the ticket sales as well as does the
proposed model (DIC= 156,221). Moreover, we find our proposed model has a similar fit to
Benchmark 3 (DIC=156,220). Although Benchmark 3 has a smaller DIC value by one unit,
Ntzoufras (2009, p.220) suggests that a model performs better than another does if the DIC
difference is greater than 2. Therefore, we conclude that our proposed model is as good as
Benchmark 3.
Finally, we compare the parameter results between these two models and find significant
and consistent effects of scheduling characteristics on ticket sales. Although the third benchmark
model also shows that the population density and travel distance from consumers’ residences to
venues is relevant, results in this benchmark model still indicate a significant geographic effect.
In other words, although consumers may prefer a venue nearest to them, it is still very likely that
consumers are willing to travel to a farther venue that has a stronger association with a leisure
activity (Okada 2005).
Therefore, we are confident that there are scheduling effects to influence consumer
decisions and ticket sales across performances. We discuss the parameter results and their
implications in the next section.
49
Table 2-4: Benchmark Models and Model Fit
2.6 Results
2.6.1 Number of Ticket Sales
Table 2-5 summarizes our parameter results for the number of ticket sales. First, we find
that performances on Saturday and Sunday have higher attractiveness to increase ticket sales
than those on other days of week (?
2
= 0.35; ?
3
= 0.34). This result is consistent with prior
literature (Corning and Levy 2002; Putler and Lele 2003) and shows the importance of
controlling for performance attractiveness when examining the impact of scheduling
characteristics.
Second, we find that scheduling indeed influences how many tickets each performance
can sell. When performances are scheduled closely in venues and have a shorter geographic
distance to other performances, they attract more consumers and sell more tickets (?
1
= 0.32).
Additionally, when performances are scheduled sparsely along a time span and have a longer
temporal distance to others, they attract more consumers and sell more tickets (?
2
= -0.14).
Because the geographic and temporal density measures have different effects on ticket sales,
these results suggest that consumers evaluate geographic and temporal schedules separately and
have different responses.
Table 2-5: Results for the Number of Ticket Sales
As we mentioned earlier, event marketers’ scheduling and consumers’ decision making
are both unobserved processes to researchers. Hence, we can only speculate possible underlying
mechanisms based on prior literature and our results. One way to interpret different consumer
responses to geographic and temporal schedules is that event marketers have several objectives
when they schedule performances. These objectives influence their scheduling decisions and
50
consumers’ reactions. For example, their objectives may be to signal venue popularity and
enhance assurance of product delivery at the desired time. If so, they will schedule performances
closely in venues of interest yet sparsely across performance dates. From consumers’
perspective, after they see such geographic and temporal schedules, they receive the signal of
venue popularity and find the flexibility in attendance timing. Then, they shape their preferences
to favor performances of such scheduling characteristics and further influence their purchase
decisions. As a result, these performances sell more tickets than other performances.
Alternatively, event marketers may not have a predetermined scheduling strategy in mind.
The geographic and temporal schedules are the consequences of constrained optimization. If so,
consumers will make their own inferences about these performances. For consumers who do not
have specific preferences to venue locations and performance dates, they might evaluate
performances differently based on different contexts. For instance, they may find an event
highly associated with some venues because these venues are close to each other and offer more
performances. Due to the similar venue locations and shorter geographic distance between
venues, these venues may catch consumers’ attention better and become consumers’ preferred
venues when consumers attend an event. Performances in these venues hence share this
common advantage to attract more consumers and sell more tickets.
On the other hand, consumers may perceive an event highly associated with some days of
week because many performances are scheduled around those days. Therefore, these days of
week would catch more attention and become more salient when consumers consider when to
attend. However, consumers usually have uncertainty for the future and prefer a wider range of
dates for selections. Closely scheduled performances at any time could merely substitute one
another and suffer from sales cannibalization.
51
According to these two interpretations, one implication for event marketers is that they
should keep their scheduling strategy (if they indeed have such a strategy) to schedule
performances densely in venue locations but sparsely across performance dates. Even if they do
not have such a strategy but only practice constrained optimization, our results provide them
another useful scheduling implication. That is, they should incorporate the scheduling effects
(i.e., a positive geographic effect but a negative temporal effect) into their decision process as
new constraints to find the most optimal solution.
2.6.2 Timing of Ticket Sales
Table 2-6 describes the parameter estimates and indicates that geographic density and the
number of advance selling weeks have significant effects on the Weibull parameters (?
11
= -7.56;
?
12
= 0.93). Because the objective of this paper is to examine the scheduling effect while
controlling for the number of advance selling weeks, we discuss the geographic effect on the
timing of ticket sales more in details.
Table 2-6: Results for Timing of Ticket Sales
However, it is less straightforward to observe the net effect of geographic density on the
timing of ticket sales based on the parameter results. We proceed to simulate performances of
different levels of geographic density to visually show their effects on the timing of ticket sales.
Therefore, Figure 2-11 presents three hypothetical performances of different levels of geographic
density but of the same performance attractiveness. The solid line illustrates the expected timing
of ticket sales resulting from geographic density being the mean value observed in our dataset
(GEO). The two dashed lines show the expected timing of ticket sales for two performances
where their geographic density values are one standard deviation higher or lower than the mean,
respectively. We can observe from Figure 2-11 that once the geographic density increases by
52
one standard deviation from the mean (GEO +1SD), the cumulative ticket sales after 12 weeks of
advance selling increase from 11% to 25% of its expected total amount. In contrast, when the
geographic density decreases by one standard deviation from the mean (GEO -1SD), only 7%
arrived after 12 weeks. Therefore, we conclude that performances in densely scheduled venues
have shorter geographic distance to other performances such that they sell tickets at a faster rate
than those of longer geographic distance.
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales
Our intuition for this result is that venues where these performances are closely scheduled
share the similarity in venue locations and geographic density. These venues can catch
consumers’ attention and lead to an attraction effect. In other words, even after controlling for
performance attractiveness, consumers still think performances in these venues more attractive.
As a result, they are willing to purchase tickets much earlier.
The implication of this result is that event marketers can monitor when and how fast
ticket sales arrive based on the geographic density information across performances. They can
use the expected timing of ticket sales as benchmark measures to compare with realized sales. In
this way, they can be aware of possible sales deviation in an advance-selling period rather than in
the week of performance. Additionally, it is also important for operational and financial
planning because event marketers can adjust their concession and security throughout an
advance-selling period to make sure a performance is not over or under staffed. Moreover, they
can have a better knowledge with the cash flows based on the expected timing of ticket sales.
Regarding the sales bump as a result of pre-Christmas shopping effect (?
j
), we find that
geographic density, number of advance selling weeks, and expected performance attractiveness
all contribute positive effects (?
13
= 4.84; ?
33
= 0.97; ?
43
= 2.45). Yet, temporal density does not
53
have such an effect. In other words, performances of higher attractiveness tend to experience
earlier timing of ticket sales. The attractiveness comes from the expected individual
performance attractiveness (due to the days of week effect) and the geographic density. Because
performances of shorter geographic distance are perceived more attractive even after controlling
for individual performance attractiveness, these performances are more salient to consumers as
Christmas gifts
2
.
2.6.3 Endogeneity in Performance Scheduling
As we have mentioned, the scheduling decision is likely endogenous and dependent on
expected performance attractiveness or effects of geographic and temporal scheduling. As such,
the results discussed above are only managerially meaningful if we accommodate the potential
endogenous scheduling decisions. Table 2-7 indicates some evidence about the endogeneity in
performance scheduling. Specifically, event marketers consider the expected performance
attractiveness when designing a temporal schedule (?
12
= 0.26). When they expect a performance
to be more attractive, they schedule more performances similar to this one. Consequently, there
are more performances scheduled temporally close to each other, resulting in higher temporal
density. However, event marketers neither incorporate performance attractiveness when
designing a geographic schedule, nor do they incorporate the geographic and temporal effects.
This result implies that event marketers may primarily rely on performance attractiveness in their
scheduling decision whether they are aware of the scheduling effects.
Table 2-7: Results for Endogenous Performance Scheduling
2
We considered other covariates, such as the cumulative sales of earlier performances, but found that it had no
significant impact on the timing of ticket sales.
54
2.7 Policy Simulation
To demonstrate the scheduling effects on ticket sales, we conduct a policy simulation by
varying the geographic schedule (Scenario 1) or temporal schedule (Scenario 2) to compare with
the current setting. To make a fair comparison and restrict a new schedule within the same
geographic and temporal range, we only re-allocate one performance, keep the rest unchanged,
and evaluate the differences in ticket sales for the target performance as well as the entire market.
2.7.1 Scenario 1: Geographic Change
According to results of the number of ticket sales, performances generate more sales
volume when they are scheduled in the same or proximate venues and have a higher geographic
density. Hence, in Scenario 1 we reschedule a performance from Venue 2 (Nassau Coliseum) to
Venue 3 (Madison Square Garden) in Figure 2-9, the most densely scheduled venue in our
dataset. As a result, geographic density of this performance increases. We also keep the same
performance date to ensure unchanged performance attractiveness and temporal density.
Figure 2-12 presents the impact of schedule changes on ticket sales. After relocating a
performance from Nassau Coliseum to Madison Square Garden, its ticket sales increase from
5,102 to 7,132, resulting in a difference of 2,031 tickets (which is 24% of average ticket sales per
performance). On the other hand, the overall market sales increase from 532,285 under the
existing schedule to 535,424 tickets in Scenario 1. Note that the 3,140 increases in market sales
are greater than 2,031 increases in a target performance. This increase in market sales provides
the evidence of market expansion. Hence, a geographic schedule after minor changes can attract
more consumers to the rescheduled performance and increase the market penetration.
Figure 2-12: Effect of Schedule Changes on Ticket Sales
55
Relocating the target performance to a different venue also changes the timing of ticket
sales. Under the modified schedule, the cumulative ticket sales of the target performance reach
60% of total sales after 12 weeks, yet the same performance only sells 28% under the original
schedule. Figure 2-13 shows that weekly sales of this performance arrive at a different rate and
results in different patterns. This implies that any monitoring or benchmarking of early ticket
sales needs to incorporate the geographic density in a schedule.
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales
2.7.2 Scenario 2: Temporal Changes
According to our results from studying the number of ticket sales, the second learning is
to disperse performance dates to accommodate more consumers and increase ticket sales.
Therefore, in Scenario 2, we reschedule the same target performance in Scenario 1 to two weeks
earlier. Yet, we keep its venue location and day of week constant. In this way, this target
performance has the same level of performance attractiveness and geographic density, yet with a
lower level of temporal density.
In contrast with Scenario 1 where we see substantial changes in the number and timing of
ticket sales, we observe much smaller changes in Scenario 2. This is consistent with the smaller
parameter values for temporal density compared to those for geographic density (see Table 2-5
and Table 2-6). Specifically, ticket sales for the rescheduled performance increase very slightly
by 385 tickets under the modified temporal schedule and the difference in total sales in the
market is also quite small (772 tickets). Figure 2-12 also shows minimal differences in the
percentage of tickets sold in the first 12 weeks. Hence, these results imply that geographic
scheduling is more important than temporal scheduling. A geographic schedule has a greater
influence on the number and timing of ticket sales than does a temporal schedule.
56
2.8 Conclusions
2.8.1 Summary
In the live entertainment industry, scheduling performances and estimating ticket demand
are two primary tasks facing event marketers. Because these tasks have been treated as two
independent problems by event marketers and marketing researchers, this essay aims to bridge
performance scheduling and demand estimation by examining the potential impact of scheduling
on ticket sales.
According to different streams of literatures, we find that it is possible to see closely
scheduled performances selling more than distantly scheduled performances, yet it is also
possible to see the opposite effect. Therefore, we allow these two possibilities and empirically
test the effect of scheduling characteristics on the number and timing of ticket sales. Specifically,
we characterize performances of the same event by their venue locations and performance dates.
Using their scheduling characteristics, we construct two density measures (i.e., geographic and
temporal density) to capture how close in distance or time performances are scheduled to each
other. Then, we model the number and timing of ticket sales as a function of these density
measures. In addition, we also control for a possible endogeneity in case event marketers
incorporate market responses in their scheduling process.
We contact a national ticket seller to obtain a dataset of a live performance event and use
its ticket sales to examine the impact of its scheduling characteristics on ticket sales. This event
had 70 performances across four venues in the New York metropolitan market and performed
between March and June 2004.
Our results indicate that performances of different scheduling characteristics sell
differently in terms of their number and timing of ticket sales. Specifically, we find that the
57
effect of geographic scheduling differs from the effect of temporal scheduling. Performances
scheduled closely in distance not only sell more tickets but also sell tickets at a faster rate. In
contrast, performances scheduled sparsely in time sell more tickets but do not have an impact on
the timing of sales.
Our explanation for the observed effects is that event marketers may schedule
performances to signal the desirability of venues and accommodate consumers’ uncertain
attendance timing. In this way, performances in the desired venues and along a wider temporal
stretch are more attractive to consumers (even after we have controlled the individual
performance attractiveness).
2.8.2 Limitations and Next Steps
Although this essay shows significant effects of geographic and temporal scheduling on
ticket sales, the results are for one market only. However, event marketers often need to make
scheduling decisions for more than one market. Thus, it is not clear whether the results in essay
one hold in other markets. Hence, the objective of essay two is to use all performance schedules
of the same event to investigate heterogeneous market responses and identify explanatory factors.
We discuss essay two in the next chapter.
58
Table and Figures
Table 2-1: Description of Data Fields
Category in the Data Fields Description
Event Name of event
Identification number Used to differentiate repeat performances of the same event
Performance date Month-Date-Year
Venue location Name of a venue and its location (City and State)
Transaction types Indicate individual purchases, group purchases, school
purchases and so on
Sales date Month-Date-Year
Channel types Six channel types: Primary Box Office, Secondary Box Office,
Hermes (Automatic phone), Internet, Outlet, and Phone.
Price levels Label of price levels
Price paid Indicates the face value, facility charges, service charges, and
the total price paid
Daily tickets Number of tickets sold
59
Table 2-2: Summary of Ticket Sales
Mean Std Dev Min Max
Total Ticket Sales 8,316 3,525 1,827 15,810
Ticket Sales by Stage
Early Sales
(one month prior)
24% 14% 3% 59%
Late Sales
(2-4 weeks prior)
33% 9% 15% 53%
Spot Sales
(performance week)
42% 14% 11% 75%
60
Table 2-3: Descriptive Statistics of Covariates
Description Mean Std Dev Min Max
GEO Geographic Density 0.368 0.185 0.128 0.540
TMP Temporal Density 0.118 0.022 0.063 0.144
FRIDAY Friday performance 0.157 0.367 0 1
SATURDAY Saturday performance 0.300 0.462 0 1
SUNDAY Sunday performance 0.286 0.455 0 1
T
Number of advance-selling weeks 15 2.044 11 19
61
Table 2-4: Benchmark Models and Model Fit
Benchmark Model Proposed Model
1 2 3
Performance attractiveness 9 9 9 9
City effects 9
Population around venues 9
Model Fit
DIC 157,134 156,459 156,220 156,221
62
Table 2-5: Results for the Number of Ticket Sales
Parameter Description Median (STD)
Scheduling Effect
?
1
Effect of geographic density 0.32 (0.05)**
?
2
Effect of temporal density -0.14 (0.05)**
Expected performance attractiveness: E[V
j
]
?
0
Baseline value of event -6.32 (0.12)**
?
1
Friday effect 0.15 (0.13)
?
2
Saturday effect 0.35 (0.12)**
?
3
Sunday effect 0.34 (0.13)**
?
2
Variance of performance attractiveness 5.45 (1.02)
** significant at the 95% highest posterior density
63
Table 2-6: Results for the Timing of Ticket Sales
Parameter Description Median (STD)
Weibull slope parameter: log(?
j
)
?
01
Intercept 3.11 (3.62)
?
11
Effect of geographic density -7.56 (2.20)**
?
21
Effect of temporal density 0.67 (2.99)
?
31
Number of advance-selling weeks -1.08 (0.21)**
?
41
Expected performance attractiveness -0.63 (0.77)
Weibull shape parameter: log(c
j
)
?
02
Intercept -0.64 (1.03)
?
12
Effect of geographic density 0.93 (0.24)**
?
22
Effect of temporal density 0.23 (0.98)
?
32
Number of advance-selling weeks 0.05 (0.02)**
?
42
Expected performance attractiveness -0.16 (0.17)
Time-varying pre-Christmas shopping effect: ?
j
?
03
Intercept 2.09 (3.24)
?
13
Effect of geographic density 4.84 (2.51)*
?
23
Effect of temporal density 1.63 (3.04)
?
33
Number of advance-selling weeks 0.97 (0.28)**
?
43
Expected performance attractiveness 2.45 (0.81)**
Variance-covariance matrix: ?
1
log(?
j
) log(c
j
) ?
j
log(?
j
) 17.38 -1.3 -11.49
log(c
j
) 0.13 0.77
? 12.58
** significant at the 95% highest posterior density
* significant at the 90% highest posterior density
64
Table 2-7: Results for Endogenous Performance Scheduling
Parameter Description Median (STD)
Expected geographic density: m
1
?
01
Intercept 0.18 (0.91)
?
11
Expected performance attractiveness 0.22 (0.15)
?
21
Effect of geographic density 0.04 (0.97)
Expected temporal density: m
2
?
02
Intercept -0.61 (0.75)
?
12
Expected value of performance 0.26 (0.12)**
?
22
Effect of temporal density -0.03 (1.01)
Variance-covariance matrix: ?
2
m
1
m
2
m
1
0.36 0.07
m
2
0.06
** significant at the 95% highest posterior density
65
Figure 2-1: Conceptual Framework of Spatial Decomposition
Geographic
Density
Temporal
Density
Scheduling Characteristics:
Ticket Sales:
(1) Number of Tickets Sold
(2) Timing of Ticket Sales
Attractiveness of
Individual Performances
Geographic
Effect
Temporal
Effect
Endogeneity
66
Figure 2-2: Model Overview
Number of Ticket Sales Timing of Ticket Sales
Scheduling Characteristics
(Geographic and Temporal Density)
67
Figure 2-3: Probability of Ticket Sales over Time
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
r
o
b
a
b
i
l
i
t
y
Advance Selling Week
Case 1
Case 2
Case 3
68
Figure 2-4: Sales Distribution by Performances
sales
20000.00 15000.00 10000.00 5000.00 0.00
F
r
e
q
u
e
n
c
y
12
10
8
6
4
2
0
Mean =8315.70
Std. Dev. =3525.
329
N =70
69
Figure 2-5: Sales Distribution by Days of Week
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
T
i
c
k
e
t
S
a
l
e
s
70
Figure 2-6: Weekly Sales Pattern of a Performance
0
1000
2000
3000
4000
5000
6000
7000
T
i
c
k
e
t
s
a
l
e
s
Advance Selling Week (t)
71
Figure 2-7: Heterogeneity in Sales Pattern Across Performances
0%
10%
20%
30%
40%
50%
60%
70%
80%
1 month prior 2?4 weeks prior last week
%
o
f
t
o
t
a
l
t
i
c
k
e
t
s
a
l
e
s
Advance Selling Period
72
Figure 2-8: Venue Locations and Driving Distances
Note: the number in Figure 2-8 indicates the sequence that the event travelled.
That is, the event went to Venue 1, 2, 3, and 4, respectively.
46
29
90
16
67
67
Note: distance in miles
Source: Google Maps
1
2
3
4
73
Figure 2-9: Summary of Geographic Distance
Continental
Arena
Nassau Coliseum
Madison Square
Garden
Sovereign Bank
Arena
0
5
10
15
20
25
30
35
40
0 10 20 30 40 50 60
N
u
m
b
e
r
o
f
p
e
r
f
o
r
m
a
n
c
e
s
i
n
a
v
e
n
u
e
Average distance to other venues (in miles)
74
Figure 2-10: Summary of Temporal Distance
0
10
20
30
40
50
60
70
2/23/2004 3/14/2004 4/3/2004 4/23/2004 5/13/2004 6/2/2004 6/22/2004
A
v
e
r
a
g
e
d
i
s
t
a
n
c
e
t
o
o
t
h
e
r
p
e
r
f
o
r
m
a
n
c
e
d
a
t
e
s
(
i
n
d
a
y
s
)
Performance date
75
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
%
o
f
t
o
t
a
l
t
i
c
k
e
t
s
Advance Selling Week
mean GEO +1 SD
mean GEO
mean GEO ?1 SD
Cumulative ticket sales= 25%
Cumulative ticket sales= 11%
Cumulative ticket sales= 7%
76
Figure 2-12: Effect of Schedule Changes on Ticket Sales
Current Schedule:
Target performance=5,102 Market sales=532,285
0
500
1000
1500
2000
2500
3000
3500
Scenario 1 Scenario 2
C
h
a
n
g
e
s
i
n
T
i
c
k
e
t
S
a
l
e
s
target performance market sales
+2031
+385
+772
+3140
77
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
1 2 3 4 5 6 7 8 9 10 11 12
%
o
f
T
o
t
a
l
S
a
l
e
s
A
r
r
i
v
e
d
Advance Selling Week
current scenario 1 scenario 2
Cumulative ticket sales= 28% of total
Cumulative ticket sales= 60% of total
78
3 Essay 2: Heterogeneous Market Responses to Performance Schedules and
Their Explanatory Factors
3.1 Introduction
In the first essay, we characterize multiple performances of a single event by their venue
locations and performance dates to understand how their geographic and temporal scheduling
characteristics influence their ticket sales. Using ticket sales of a live performing event in the
New York metropolitan market, we find that the effect of geographic scheduling differs from the
effect of temporal scheduling. Performances scheduled in nearby venues not only sell more
tickets but also sell tickets at a faster rate. In contrast, performances scheduled distantly in time
sell more tickets but do not have an impact on the timing of sales.
However, event marketers often need to make scheduling decisions for more than one
market. Although our finding in essay one has rich implications for event marketers, it is unclear
whether event marketers can apply this finding to all markets. For example, the event analyzed
in essay one had 70 performances in four venues for 32 days in the New York metropolitan
market. However, when it travelled to other markets such as Norfolk, Virginia, it had 19
performances in two venues for 10 days. It also had 21 performances in one venue for 10 days in
the Atlanta area. Thus, performance schedules can vary across markets and these schedules may
not have the same effect on ticket sales across markets. Even if a schedule is the same across
markets, these markets may not respond to their schedule identically. This limitation in essay
one hence motivates our second essay to examine the effect of performance schedules across
markets.
79
Although several studies in the event tickets literature examine descriptive drivers for
event ticket sales, very limited research focuses on examining heterogeneous market responses
(Moore 1966; Weinberg and Shachmut 1978; Havlena and Holak 1988; Reddy et al 1998).
Hence, the objective of essay two is to use all performance schedules of the same event to
investigate heterogeneous market responses and identify explanatory factors. It is important
because events do not always go to the same set of markets when they are on tour. A long
lasting event may go on tour several times and travels to a different set of markets each time.
Hence, once the heterogeneity in market responses and explanatory drivers are known, event
marketers could infer a likely response in a new market or select markets for touring based on
expected market responses.
To accomplish our research objective, we first conduct a preliminary analysis to analyze
all performance schedules and examine their impact on the number of ticket sales. After
applying the model developed in essay one to the entire dataset and estimating market responses
iteratively across markets, we find that the effect of geographic scheduling differs from the effect
of temporal scheduling and this difference is consistent across markets. In terms of the
geographic scheduling, performances scheduled in close venues sell more ticket sales than
sparsely scheduled ones, but this result only hold for markets that use multiple venues.
Regarding the temporal scheduling, in contrast, performances scheduled distantly in time sell
more than the densely scheduled ones.
Although we find consistent scheduling effects from the preliminary results, we also
observe the market responses are of different magnitudes. In other words, some markets are
more responsive to geographic (temporal) schedules than other markets. To identify the factors
that explain these differences across markets, we extend the model developed in essay one to not
80
only examine market-specific response parameters but also investigate observed and unobserved
heterogeneity via the hierarchical Bayesian approach. In addition, we also control for the
possible endogeneity in the performance scheduling process. Among several marketing
characteristics, we choose the size of market population as the first explanatory factor. We also
examine characteristics of a touring event to understand whether participating markets that are
adjacent to each other and the order that an event travels across affect the magnitude of
scheduling effects.
We use the same family event mentioned in essay one and analyze all of its performance
schedules in the dataset to test our model. Because this event sequentially performed 449 times
in 50 cities in the U.S. domestic market between January and June 2004, we aggregate these 50
cities to 42 designated market areas (which will be discussed in detail in the data section). As a
result, there are six markets where the event performed in multiple venues yet all 42 markets
have some variations in their temporal schedules.
Our results show that market and additional scheduling characteristics can both explain
the differences across market responses. First, when a market has a bigger population, the
effects of days of week and baseline attractiveness are attenuated. Our explanation is that
usually there are more events offered to a bigger market than to a smaller one. Consumers in a
bigger market may be used to seeing several competing events offered simultaneously and
having a variety of events to choose from. Hence, they are less responsive to an event no matter
on which day of week it may be on as compared to consumers in a smaller market.
Second, we find the additional scheduling characteristics can partly explain
heterogeneous market responses. With respect to a current market, after an event travels to more
of its geographically adjacent markets, the current market is less responsive to its baseline
81
attractiveness and temporal schedule. A possible reason is that after an event has gone to more
geographically adjacent markets, its newness wears out but its reputation might accumulate over
time. As a result, consumers may refer to other measures such as word of mouth to make their
purchase decisions rather than refer to the baseline attractiveness and its temporal schedule.
Moreover, after an event perform in several markets (whether these markets are adjacent
or not), a market in which an event performs later tends to respond more favorably to a Sunday
performance. Our explanation is that after an event has lasted longer and gone to more markets,
its reputation, or word of mouth, accumulates over time (Reddy et al 1981) even though its
newness may wear out. Because a Sunday performance tends to be the last performance in a
market (at least in the case of our dataset), consumers might think Sunday as their “last
opportunity” to enjoy this event before it leaves for another market. As a result, a market in a
later temporal sequence has a stronger Sunday effect.
We also find that the nature of performance schedules is endogenous. Different from
essay one that endogeneity is found only in a temporal schedule, we find in essay two that
geographic and temporal schedules are both done endogenously after we pool all performances
across markets for analysis. When event marketers expect high performance attractiveness, they
schedule more performances in all venues and tend to allocate those performances around
weekends or along a limited time span. As a result, on average, performances have a shorter
geographic and temporal distance to others and have higher density values. Moreover, when
event marketers understand that consumers prefer performances on dispersed dates because of
uncertain timing of attendance, they decrease the number of weekend performances and/or
disperse performance dates. Consequently, performances in a temporal schedule have lower
82
temporal density values. Since we have accounted for this endogeneity in the estimation process,
the results we present here are unbiased.
The rest of essay two is organized as follows. First, we conduct a preliminary analysis
using all performances observed across markets to show heterogeneous market responses to
performance schedules. Second, we review extant literature to find possible reasons for
heterogeneity and propose our conceptual framework. In the next section, we present the
modeling structure extended from the first essay and discuss the dataset in details. After the
model estimation and benchmark comparison, we present our results and conclude this essay
with limitations and next steps.
3.2 Preliminary Analysis
As described in essay one, a family event went to 50 cities on 245 dates for 449
performances between January and June 2004 (see §2.4.1 Description of Events for more details).
Among which, several cities are within the same metropolitan area and show richer variations in
geographic and temporal schedules. To test whether markets have heterogeneous responses to
their performance schedules, we conduct a preliminary analysis to run the model in equations (5),
(6), and (14) for one market at a time and examine their market specific parameters.
More specifically, we use the designated market area (DMA) to aggregate 50 cities into
42 markets (see the data section for full descriptions) and summarize the market information in
Table 3-1. As Table 3-1 shows, we sort markets by their first performance date and assign a
unique market identification number. Every market has its information listed regarding its first
and last performance dates, number of performances, length of performing period, and venue
usage. Hence, for performances within a market, we characterize these performances by their
83
venue locations and performance dates. Then, we follow the equations (1) and (2) in essay one
to compute their geographic and temporal density and understand how densely or sparsely these
performances are scheduled. Consequently, for each market, we model ticket sales of its
performances as a function of their density measures and performance attractiveness while
controlling for a possible endogeneity in performance scheduling.
Table 3-1: Summary of Performance Schedules across Markets
After iteratively estimating the model across 42 markets, we find that the days of week
represent performance attractiveness and increase ticket sales (?
0
ranges from -3.82 to -8.23; ?
1
ranges from 0.10 to 0.95; ?
2
ranges from 0.24 to 1.48; ?
3
ranges from 0.13 to 1.09). Moreover,
the effect of geographic scheduling differs from the effect of temporal scheduling. Performances
scheduled in nearby venues sell more tickets than those scheduled in distant venues (?
1
ranges
from 0.92 to 4.29). However, this result only holds for markets that use multiple venues. On the
other hand, performances scheduled dispersed across dates sell more than the those scheduled
close in time (?
2
ranges from -0.04 to -2.92). We summarize these market-specific parameters in
Table 3-2.
Table 3-2: Summary of Market Responses to Performance Schedules
According to our preliminary analysis, we confirm consistent yet heterogeneous market
responses to performance schedules. To understand why market responses are different, we
review extant literature that suggests potential explanatory factors for this heterogeneity.
84
3.3 Literature Review and Conceptual Framework
3.3.1 Literature Review
Consumers are different individuals and their preferences and decisions often differ from
one another. However, as the proverb says, “birds of a feather flock together.” It is very
common to observe consumers who have similar tastes living in similar areas and making similar
purchase decisions. Accordingly, when marketers offer a variety of products to consumers, they
often expect heterogeneous consumer responses across zip codes, counties, states, or
metropolitan markets.
Heterogeneity in the unit of analysis is commonly studied in marketing and the entities
include individual consumers, products, firms, markets, and countries. For example, consumers
who have different demographic characteristics or live in different zip codes make different
choices regarding automobiles (Yang and Allenby 2002), book formats (i.e., Print vs. PDF; Jank
and Kannak 2005), or adoptions of online grocer (Choi, Hui, and Bell 2009). Their shopping
behaviors also differ across product types (e.g., motels of high, medium, or low quality, Mazzeo
2002; department stores of upscale, midscale, or discount, Vitorino 2007), store formats (e.g.,
supermarkets, hypermarkets, and discount stores, González-Benito 2005; discounted or regular
retailers, Zhu et al 2007), and brand names (e.g., Wal-mart, K-mart, Target, Zhu and Singh 2009;
Zhu et al 2009). Besides examining the heterogeneity at an individual level, researchers can also
summarize consumer responses across zip codes (Yang and Allenby 2002; Jank and Kannan
2005), metropolitan statistical areas (Zhu and Singh 2009), or countries (Elberse and Eliashberg
2003) to study heterogeneity at an aggregate level.
One way to account for heterogeneity in empirical analyses is to specify individual-
specific parameters (Corning and Levy 2002; González-Benito 2005; Mazzeo 2002; Vitorino
85
2007; Moe and Fader 2009; Zhu and Singh 2009; Zhu et al 2009). For example, Corning and
Levy (2002) specified venue-specific parameters when examining ticket sales across venues to
understand whether consumers of those venues have different responses to product offerings. In
the retail locations, Mazzeo (2002) specified type-specific effects of spatial competition to study
whether motels of low, medium, or high quality types have different spatial effects on
profitability. In addition, Zhu and Singh (2009) used brand-specific parameters to examine
asymmetric competition effects among Wal-mart, K-mart, and Target.
Although heterogeneity has been extensively studied in many contexts, most of prior
studies in the event tickets literature have not yet examined heterogeneous market responses. In
other words, researchers assume the effects of days of week, prices, and promotions are
homogeneous across venues, events, or performances (Moore 1966; Weinberg and Shachmut
1978; Havlena and Holak 1988; Reddy et al 1998). Although Corning and Levy (2002) and Moe
and Fader (2009) are the two exceptions where Corning and Levy (2002) allowed parameters to
be venue specific and Moe and Fader (2009) specified parameters varied with events and price
tiers, they did not identify explanatory factors for their proposed heterogeneity. Given that we
have found the heterogeneous market responses in the preliminary analysis, the objective of
essay two is to identify explanatory factors to explain the differences across markets.
3.3.2 Conceptual Framework
In our conceptual framework, we first discuss possible market characteristics that may
explain differences across markets. Then, we discuss characteristics of a touring event that may
provide context dependent reasons for response heterogeneity.
86
Market Characteristics
When the analysis is made at a market level rather than at an individual level, the first
issue is to define what a market is. Depending on the context of interest, a market can be a
metropolitan statistical area (MSA; Zhu et al 2007), a designated market area defined by A.C.
Nielsen (DMA; Carlyle, Slater, and Chakroff 2008), or a retail trade area (Bronnenberg and Mela
2004). Then, researchers try to find the market characteristics that may explain the difference to
some extent.
In general, metropolitan areas are assumed more similar to other MSAs than to rural
areas, and the similarity or differences may be due to the population size, population density,
income, education, household size, household values, commute time to work, etc. (Mazzeo 2002;
Vitorino 2007; Zhu et al 2007; Zhu and Singh 2009; Zhu et al 2009). For this reason, we
propose that market characteristics can explain the heterogeneous market responses in our
preliminary analysis.
Additional Scheduling Characteristics
We refer to additional scheduling characteristics as characteristics of a touring event. For
example, one characteristic is that its performing group travels from one market to another.
Because this distribution mechanism follows the sequential distribution approach (Elberse and
Eliashberg 2003; Bronnenberg and Mela 2004), we think sequential distribution literature may
provide context dependent characteristics to explain why consumers in different markets react to
performance schedules differently.
Extant works in sequential distribution have focused on the effect of geographic
adjacency on market adoption (Bronnenberg and Mela 2004) and the effect of release timing on
box-office revenues (Elberse and Eliashberg 2003). When Bronnenberg and Mela (2004)
87
studied the spatial evolution of a new product adoption across markets, they found manufacturers
tend to enter markets that are geographically adjacent to a current market. In other words, the
initial market serves as a lead market and its lead market effect rolls out sequentially to
geographically adjacent markets.
On the other hand, the release timing in the distribution also influences how well a
product sells. For example, Elberse and Eliashberg (2003) studied motion pictures to investigate
the issue of release timing between the U.S. market and foreign markets. Although they only
examined the effect of release timing between the initial market and following foreign markets
rather than the effect of release timing along the entire sequence, they still found that shortening
the time lag between two markets increases the revenues of a later market.
Therefore, the sequential distribution literature has traditionally discussed the roles of
geographic adjacency and release timing as important covariates. Whether the geographic
adjacency and release timing in the distribution explains different market responses still remains
unknown and deserves further investigation. Similarly, a live performance event follows a
temporal sequence to travel across markets. Each market along the sequence has different
release timing and some of these markets are geographically adjacent to one another. It is
important to evaluate its geographic adjacency and temporal sequence to understand whether
these additional scheduling characteristics explain different markets responses to performance
schedules. Therefore, we incorporate market characteristics and additional scheduling
characteristics to explain different market responses across markets.
In short, we summarize our conceptual framework in Figure 3-1. Similar to essay one,
we examine the impact of geographic and temporal scheduling on the number of ticket sales and
control for the performance attractiveness through its days of week. In addition, we also control
88
for the possible endogeneity in performance scheduling. Finally, we examine whether and how
market and additional scheduling characteristics explain response heterogeneity across markets.
Figure 3-1: Conceptual Framework
3.4 Model Development
3.4.1 Model Overview
Our model development consists of four steps. First, we use the geographic and temporal
density measures to capture the scheduling characteristics of performances in their associated
markets and understand how densely or sparsely these performances are scheduled. Second,
within each market, we specify market-specific parameters and model ticket sales at a
performance level as a function of these density measures and performance attractiveness. Third,
we control for a possible endogeneity between performance schedules and expected market
responses. Finally, we employ a hierarchical Bayesian (HB) approach to incorporate the
heterogeneity in market responses. Among these four steps, the first three steps are adapted from
essay one, yet the fourth step is the model extension in essay two.
Although the HB approach is not the only method to study heterogeneous market
responses and the latent class analysis (Kamakura and Russell 1989) may be another appropriate
alternative, we choose the HB approach because it can accommodate unobserved heterogeneity
across markets (Rossi and Allenby 2003) in addition to the heterogeneity explained by market
characteristics and additional scheduling characteristics.
3.4.2 Scheduling Characteristics and the Number of Ticket Sales
To begin with, we refer to equations (1) and (2) in essay one to capture the scheduling
characteristics by their geographic and temporal density for all performances in their markets.
89
Then, we refer to equation (5) to rewrite the sales share of performances (and the share of non-
buyers) within a market with market-specific parameters:
(16)
P
m
(]) =
exp(I
]m
) · 0E0
]m
0
1m
IHP
]m
0
2m
1 + ? exp(I
]mi
) · 0E0
]m?
0
1m
IHP
]m?
0
2m
]
m
j
|
=1
where
I
]m
= ?
0m
+?
1m
FRIBAY
jm
+?
2m
SAT0RBAY
jm
+?
3m
S0NBAY
jm
+ ?
jm
; ?
jm
~N(u, ?
s
2
)
where P
m
(j) is the sales share of performance j in market m, V
jm
is its performance attractiveness
(which is a function of days of week), GEO
jm
and TMP
jm
represent the geographic and temporal
density measures, and ?
0m
, ?
1m
, ?
2m
, ?
3m
, ?
1m
, and ?
2m
are market specific parameters. Therefore,
among J
m
performances in market m, their parameters are homogeneous within a market but
heterogeneous across markets. Using the population size in a target market (POP
m
) and the sales
share of a performance, we can calculate the expected ticket sales of a performance (i.e.,
Solcs
m
(]) = P0P
m
× P
m
(])), the market penetration rate of this event (i.e., ? P
m
(])
]
m
]=1
), and the
size of non-buyer segment (i.e., 1 - ? P
m
(])
]
m
]=1
).
3.4.3 Endogeneity in Performance Scheduling
Next, we account for a possible endogeneity in performance scheduling by revising
equation (14) as follows:
(17)
_
logit(0E0
]m
)
logit(IHP
]m
)
_ = ç
û
+ç
1
E |I
]m
] +ç
2
_
0
1m
0
2m
_ +q where p
ì
~N(u, o
q
i
2
I), i = 1,2
where E[V
jm
] is the expected performance attractiveness of performance j in market m, ?
1m
, and
?
2m
represent the effects of geographic and temporal scheduling, and ¢
0
, ¢
1
, and ¢
2
are the
parameters of interest. When ¢
1
or ¢
2
is significant, scheduling is endogenous but has been
taken into account. Note that the specification of equation (17) is the same as equation (14)
90
except that we specify an independent relationship between geographic and temporal density. In
this way, when an event performs in a single venue within a market and only has variation in its
temporal schedule, we can directly drop GEO
jm
in equations (16) and (17) and just investigate a
possible endogeneity in temporal scheduling.
3.4.4 Use of the HB Approach for Response Heterogeneity
Finally, we take the HB approach to explain differences of response parameters across
markets and specify these market-specific parameters as a function of their unobserved
heterogeneity and observed market characteristics and additional scheduling characteristics:
(18)
l
l
l
l
l
l
o
0m
o
1m
o
2m
o
3m
0
1m
0
2m
1
1
1
1
1
1
= A
m
= µX
m
+e where e~N(u, o
c
i
2
I), i = 1,2,S,4,S,6
where ?
m
is a 6-by-1 vector that contains the market-specific parameters (?
0m
, ?
1m
, ?
2m
, ?
3m
, ?
1m
,
and ?
2m
) in equation (16), X
m
is a k-by-1 vector that includes market characteristics and
additional scheduling characteristics, ? is a 6-by-k matrix that represents the effects for these
characteristics, and e is a vector of random errors for the unobserved heterogeneity.
3.5 Data
We contact a national ticket seller to obtain a dataset of two family events and analyze
ticket sales for one event in this dissertation (see §2.4.1 for more information about this event).
In short, this event sequentially performed 449 times in 50 cities in the U.S. domestic market
between January and June 2004. In essay one, we only used ticket sales in the New York
metropolitan market. In essay two, we analyze all performances in the dataset to investigate
heterogeneity in market responses.
91
3.5.1 Definition of Markets
Among 50 venues in the dataset, some venues are in the same MSAs, some are the only
venues in their MSAs, and others are in rural areas (i.e., non-MSAs). To avoid information
losses after aggregating venues to MSAs, we aggregate venues by their designated market areas
(DMAs) for urban and rural cities. The definition and classification of DMAs are proposed by
Nielsen Media Research (Weiner 2000) where each DMA consists of several counties and
consumers in the same DMA receive the same TV broadcasting and media messages (Carlyle,
Slater, and Chakroff 2008). The advantage of aggregating data by DMAs is that, suppose
marketing activities for an event are planned at a DMA level, consumers within the same DMA
are potentially aware of this event even though they may live in a rural area far away from a
venue.
Hence, we refer to a website by the Truck Ads
®
(www.truckads.com) that disaggregates
the U.S. market into several DMAs and lists all individual counties within each DMA. Figure 3-
2 provides an example of the Orlando DMA in Florida. As the map shows, there are nine
counties within this DMA. According to the venue locations in our dataset (i.e., names of venues,
cities, and states), we can identify in which DMA a venue is and which counties are in this DMA.
As a result, we aggregate 50 venues into 42 DMAs in Table 3-3. Figure 3-3 also shows the
locations of these DMAs. Each shaded area represents a DMA and a number in a box is its
market identification number assigned in Table 3-1.
Figure 3-2: Example of a DMA and its county information
Table 3-3: Venue Locations and their Associated DMAs
Figure 3-3: DMA locations
92
3.5.2 Description of Performance Schedules across Markets
According to the DMAs and performance information in our dataset, we summarize the
temporal schedule in each market into Table 3-1 by its first and last dates, length of performing
period, total number of performances, and venue usage in a market. On average, there were
10.69 performances in a market, lasted for 5.79 days, and used 1.19 venues.
Temporal Sequence of the Event Distribution
In Table 3-1, we sort the temporal schedules by their first performance date and observe
the temporal sequence of the event distribution. In general, this event first performed in
Tallahassee in January (market ID= 1), Atlanta in February (market ID= 12), New York in
March (market ID= 16), St. Paul in May (market ID= 33), and Tucson in June (market ID = 42)
3
.
Table 3-1 also indicates a few incidences where two or three markets started performances on the
same date or one to two days apart. For instance, the event had performances in Providence
(market ID= 26) and Worcester (market ID= 27) between May 1 and May 9 while having
performances in La Crosse (market ID= 25) between May 4 and May 5. Therefore, it is likely
that there were up to three performing groups touring in the same period. Because we cannot
identify which groups performed in which markets, we assume that the performing quality is
constant across performing groups and do not affect how markets respond to their performance
schedules. Hence, we analyze all 42 markets together regardless of their performing groups.
Venue Usage of the Event across Markets
In terms of the venue usage across markets, among 42 DMAs, only six markets had more
than one venue in use. These markets were the Greenville-Asheville, Raleigh-Fayetteville,
Norfolk-Hampton, Champaign-Springfield, New York, and Philadelphia DMAs. Moreover,
3
This event kept on travelling after performing in Tucson. However, the performance schedules available in the
dataset were truncated up to June 2004. Thus, we examine the performance schedules of these 42 markets only.
93
except for the New York market that scheduled performances across four venues, the rest had
performances scheduled in two venues. We also find that the event did not always perform in
these venues consecutively within a market. Sometimes the event offered all performances in
one market and then left for a new market, but sometimes it offered a few performances in one
venue and then provided more after a period. Specifically, this event completed all scheduled
performances consecutively in the markets of Norfolk-Hampton, Champaign-Springfield, and
Philadelphia. However, in the market of Greenville-Asheville and Raleigh-Fayetteville, after the
event performed in one venue, it left for other markets and then came back three to four months
later. In the New York market, it performed in three venues consecutively, left for other markets,
and then returned seven weeks later.
Geographic Adjacency of Markets
Upon a closer look of these markets and their locations in Figure 3-3, we also find that
this event had an extensive tour in the East Coast and some of the Mid-West markets. Moreover,
this event sometimes went to near or adjacent markets but occasionally traveled to an isolated
market. For example, this event went to Orlando (market ID=2) and Tampa (market ID=3) in a
consecutive order but went to an isolated market in Tucson (market ID=42).
Descriptions of Ticket Sales
Table 3-4 summarizes the average ticket sales at a performance level and its total ticket
sales at a market level. Across all markets, a performance can sell an average of 3,825 tickets
with a standard deviation being 1,162 tickets. However, depending on in which market a
performance is, it can sell as many as 8,316 tickets in New York (market ID= 16) or as little as
585 tickets in Madison (market ID= 28). In terms of ticket sales at a market level, an event can
sell an average of 51,905 tickets with a standard deviation being 90,992. Although it seems that
94
markets that use multiple venues experience more ticket sales (e.g., New York and Philadelphia
are ranked as the top two best selling markets), markets of a single venue usage do not
necessarily sell less. For example, the event performed in only one venue in Atlanta (market
ID=12) had market sales ranked at the third place. Moreover, Miami is also a single venue DMA
but its market sales exceeded the DMAs of Greenville-Ashville (market ID=9) and Champaign-
Springfield (market ID=20). Because the focus of essay two is on sales at performance level and
response heterogeneity across markets, we will address the issue of market sales in essay three.
Table 3-4: Summary of Ticket Sales across Markets
3.5.3 Covariate Specifications
Before we estimate our model, we still have to measure geographic and temporal density
as scheduling characteristics, create the days of week indicators for performance attractiveness,
and compute the population size of the target markets. Moreover, we need to select market
characteristics and measure the geographic adjacency and temporal sequence for the additional
scheduling characteristics. Hence, we discuss each covariate in turn.
Geographic and Temporal Density Measures
First, we calculate the geographic density for performances in markets (GEO
jm
) where
more than one venue is used. These markets are Greenville-Asheville, Raleigh-Fayetteville,
Norfolk-Hampton, Champaign-Springfield, New York, and Philadelphia. We start with
identifying venue locations on the Google Maps to compute the geographic distance (in miles)
between venues in the same market. Then, we apply the equation (1) in essay one to compute
the geographic density for each performance. As a result, depending on in which market a
performance is, the average geographic density of a performance ranges between 0.368 and
0.556 with the average across markets being 0.476 and the standard deviation being 0.09.
95
Next, we use the equation (2) in essay one to calculate the temporal density for
performances in their associated markets (TMP
jm
). On average, a performance in a market has
its temporal density ranging from 0.118 to 0.667 with the average across markets being 0.476
and the standard deviation being 0.09. Table 3-5 provides the descriptive statistics of these
covariates.
Table 3-5: Descriptive Statistics of Covariates across Markets
Days of Week Indicators and Market Population
According to observed performance dates in the dataset, we further create the days of
week indicators (FRIDAY
jm
, SATURDAY
jm
, and SUNDAY
jm
) to measure performance
attractiveness. On average, a market has 17% of performances on Friday, 38% on Saturday, and
25% on Sunday. However, there are markets without any Friday, Saturday, or Sunday
performances, as shown in those blank cells in Table 3-5. We also summarize the descriptive
statistics of days of week covariates in Table 3-5.
To compute a population size (POP
m
) in a target market (i.e., family population with
children under 10 years of age) across all DMAs, we refer to the Census Bureau to collect
relevant population information at the county level and then aggregate the population size by
DMAs, the same approach used in essay one. As a result, the average population size in a target
market is 439,662 with a standard deviation being 647,091. Table 3-6 presents the summary
information of the population size in each DMA.
Table 3-6: Descriptive Statistics of Market Characteristics
Market Characteristics
In addition to population size in a target market, we consider population density, family
income, and average family size as other potential market characteristics to explain response
96
heterogeneity. Different from the extant literature, we choose the information at a family level
rather than at a household level because our event targets at families with young children. Hence,
we first download the U.S. 2000 Census data at the county level to match the counties of interest
in our 42 DMAs. Because the population size in each county differs, we weight the market
characteristics in each county by its population size to calculate the average value for each DMA.
Table 3-6 presents the summary statistics across these DMAs.
Additional Scheduling Characteristics
We consider two additional characteristics of a performance schedule to explain the
heterogeneity across markets. The first characteristic we examine reflects the order of markets
that appear in a schedule. The second is the number of geographically adjacent markets with
respect to a focal market. We refer to the previous covariate as the temporal sequence and the
later as geographic adjacency.
To measure the temporal sequence, we refer to Table 3-1 that sorts markets by their first
day of performance to check which market is in the earliest distribution timeline (Order=1).
Then, we go down the list to assign an increasing number to markets in a later distribution
timeline. For example, according to Table 3-1, Tallahassee is the first market and Orlando is the
second market. We assign Order
1
=1 and Order
2
=2. For markets that had the first performance
on the same date (e.g., Columbia and Greenville-Asheville), we assigned an equal rank to these
markets (i.e., Order
8
= Order
9
= 8). Table 3-7 summarizes the descriptive statistics of temporal
sequence across markets.
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics
According to the lead market effect (Bronnenberg and Mela 2004), markets that are
adjacent and adopt a new product first have an impact on adjacent markets that have not yet been
97
adopted. We follow this logic to measure the geographic adjacency among our participating
DMAs. Hence, we first refer to Figure 3-3 to locate these 42 DMAs and check which markets
are adjacent to one another. Then, we refer to the order of each market to count how many
adjacent markets an event went to before it arrives to a focal market. Finally, we use this number
to represent the extent of geographic adjacency of a participating market. As a result, an event
went to an average of 0.95 spatially adjacent markets before it goes to a focal market. Table 3-7
summarizes the descriptive statistics of geographic adjacency across markets
4
.
The purpose of essay two is to explain heterogeneous market responses and propose
explanatory factors. Thus, we use the same set of covariates in the HB approach in equation (18)
and choose the population size in a target market, the order of markets that an event travels, and
the number of geographically adjacent markets for a focal market. We also examined other
market characteristics as shown in Table 3-6. Although population density is another significant
explanatory factor, it has the same effect as the population size. On the other hand, we find the
average family income and family size cannot explain any response heterogeneity across markets.
3.6 Model Estimation and Benchmark Comparison
3.6.1 Estimation
We choose the hierarchical Bayesian approach to estimate the number of ticket sales,
heterogeneous market responses, and endogeneity in performance scheduling simultaneously.
We specify appropriate and diffuse priors for our parameters in the WinBUGS program and
estimate the model over 10,000 iterations. After checking the convergence criteria, we examine
the autocorrelation plots for all covariates, discard the first 5,000 iterations for burn-in, and use
4
We also count how many adjacent markets a focal market has regardless of when the event performed in these
markets. However, this alternative measure did not explain any of the response heterogeneity.
98
the remaining iterations as the posterior distribution. We specify the prior distributions of
parameters below:
Priors for modeling performance attractiveness:
?
s
2
~I0(u.1,u.1)
Priors for modeling the heterogeneity in market responses:
[
kì
~N(u,1u) where k= 1, 2, 3, 4 and i=1, 2, 3, 4, 5, 6
o
c
i
2
~I0(u.1,u.1) where i=1, 2, 3, 4, 5, 6
Priors for modeling the endogeneity in performance scheduling:
¢
kì
~N(u,1u) where k= 0, 1, 2 and i=1, 2
p
ì
~N(u,1u) where k=1, 2
3.6.2 Benchmark Comparison
Before presenting our model results, we compare the model fit between the proposed and
benchmark models. Because our proposed model aims to explain the heterogeneous market
responses to performance schedules, we think a homogeneous model that does not account for
any heterogeneity to be an appropriate benchmark (i.e., Benchmark 1). In addition, we also
compare the proposed model with the preliminary analysis (i.e., Benchmark 2). After comparing
the model fit using the DIC measure (equation 15), we find that the DIC of the proposed model
is 5590.63. This fit measure is lower than the heterogeneous model without explanatory factors
(DIC=5613.94) and the homogeneous model (DIC=6927.14). Hence, our proposed model has
incorporated the market heterogeneity and explained the differences across markets.
99
3.7 Results
3.7.1 Heterogeneous Market Responses to Performance Schedules
According to the market-specific parameters, we find that markets respond to
performance schedules consistently. Although the effect sizes vary from markets to markets, all
expected values of parameters (i.e., effects of baseline performance attractiveness, days of week,
geographic density, and temporal density) are consistently positive or negative. Figure 3-4
visually presents these heterogeneous parameter values. Specifically, the parameter values of the
baseline performance attractiveness is consistently negative across markets (E[?
0m
] ranges from
-4.43 to -7.10) and performances on Friday, Saturday, or Sunday have incremental attractiveness
to generate more ticket sales (E[?
1m
] ranges from 0.08 to 0.39; E[?
2m
] ranges from 0.16 to 0.66;
E[?
3m
] rangers from 0.18 to 0.44).
Figure 3-4: Heterogeneous Market Responses
Moreover, markets have consistent responses to their geographic and temporal schedules.
For markets where multiple venues are in use, densely scheduled performances across venues
sell more tickets than sparsely scheduled performances (E[?
1m
] ranges from 0.81 to 2.13). In
contrast, sparsely scheduled performances along a time span sell more than densely scheduled
performances (E[?
2m
] ranges from -0.10 to -0.59). These results are consistent with results in
essay one and our preliminary analysis. Hence, we confirm the consistent (yet heterogeneous)
market responses to performance schedules.
3.7.2 Explanatory Factors for Market Heterogeneity
Our results in Table 3-8 report explanatory factors for the heterogeneity in market
responses. First, when a market has a bigger population, the effects of days of week and baseline
attractiveness are attenuated (?
11
= -0.25; ?
12
= -0.03; ?
13
= -0.05; ?
14
= -0.02). Our explanation is that
100
usually there are more events offered to a bigger market than to a smaller one. Consumers in a
bigger market may be used to seeing several competing events offered simultaneously and have a
variety of events to choose. Hence, they are less responsive to an event (no matter on which
days of week it is) than are consumers in a smaller market.
Table 3-8: Explanatory Factors for Market Heterogeneity
Second, we find that additional scheduling characteristics can partly explain
heterogeneous market responses. With respect to a current market, after an event travels to more
of its geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule (?
21
= -0.14; ?
26
= 0.10). A possible reason is that after an
event has gone to more spatially adjacent markets, its newness wears out but its reputation might
accumulate over time. As a result, consumers may refer to other measures such as word of
mouth to make their purchase decisions rather than refer to the baseline attractiveness and its
temporal schedule.
Moreover, after an event follows its temporal sequence to perform in several markets
(whether these markets are adjacent or not), a current market in a late distribution sequence tends
to respond more favorably to a Sunday performance (?
34
=0.10). This result is also graphically
shown in Figure 3-4 (d). Our explanation is that after an event has lasted longer and gone to
more markets, its reputation or word of mouth accumulates over time (Reddy et al 1981) even
though its newness may wear out. Because a Sunday performance tends to be the last
performance in a market (at least it is the case in our dataset), consumers might think Sunday as
their “last opportunity” to enjoy this event before it leaves for another market. As a result, a
market in a later temporal sequence has a stronger Sunday effect.
101
3.7.3 Endogenous Scheduling Decision
We also find an endogeneity in performance schedules. Different from essay one where
only endogeneity is found in a temporal schedule, in essay two we find that geographic and
temporal schedules are both done endogenously after we pool all performances across markets
for analysis. When event marketers expect high performance attractiveness, they schedule more
performances in all venues (?
11
= 0.18) and tend to allocate those performances around weekends
or along a limited time span (?
12
= 0.84). As a result, on average, performances have a shorter
geographic and temporal distance to others and have higher density values. Moreover, when
event marketers understand that consumers prefer performances on dispersed dates because of
uncertain timing of attendance, event marketers decrease the number of weekend performances
and/or disperse performance dates (?
22
= 1.30). Consequently, performances in a temporal
schedule have lower temporal density values. Since we have accounted for this endogeneity in
the estimation process, the results we present here are unbiased. Table 3-9 summarizes the
results of the endogeneity in performance scheduling.
Table 3-9: Endogenous Performance Schedules
3.8 Conclusions
3.8.1 Summary
In the first essay, we observe multiple performances of a single event and examine them
by their venue locations and performance dates to understand how their geographic and temporal
scheduling characteristics influence their ticket sales. However, event marketers often need to
make scheduling decisions for more than one market. Although our finding in essay one has rich
implications for event marketers, it is unclear whether event marketers can apply this finding to
102
all markets. Hence, the objective of essay two is to use all performance schedules of the same
event to investigate heterogeneous market responses and identify explanatory factors.
To accomplish our research objective, we first conduct a preliminary analysis and find
consistent scheduling effects. However, we also observe the market responses are of different
magnitudes. To identify the factors that explain these differences across markets, we extend the
model developed in essay one to not only examine market-specific response parameters but also
investigate observed and unobserved heterogeneity via the hierarchical Bayesian approach.
Among several marketing characteristics, we choose the size of market population as the first
explanatory factor. We also use additional scheduling characteristics along the distribution of
this event to examine whether geographic adjacency between markets and temporal sequence
along the distribution affect the magnitude of scheduling effects.
We use the same family event mentioned in essay one and aggregate the 50 cities it
travelled to 42 designated market areas. Our results show that market characteristics and
additional scheduling characteristics can both explain the differences across market responses.
First, when a market has a bigger population, the effects of days of week and baseline
attractiveness are attenuated. Second, with respect to a current market, after an event travels to
more of its geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule. Moreover, after an event follows its temporal sequence to
perform in several markets, a current market in a late distribution sequence tends to respond
more favorably to a Sunday performance. We also find an endogeneity in performance
schedules. Since we have accounted for this endogeneity in the estimation process, the results
we present here are unbiased.
103
3.8.2 Limitations and Next Steps
This essay examines heterogeneous market responses to performance schedules and
contributes to the event tickets literature by investigating explanatory factors. Results of this
essay provide a more generalizable scheduling guideline for event marketers and assist event
marketers in anticipating potential market response based on market characteristics and
additional scheduling characteristics.
However, the limitation of this essay is that we allow the additional scheduling
characteristics to explain the heterogeneity in market responses (i.e., parameter effects) but have
not yet directly examined whether additional scheduling characteristics in an event distribution
affect ticket sales across markets. As the sequential distribution literature suggests, preceding
markets tend to have some effects on later markets through their spatial adjacency (Bronnenberg
and Mela 2004) or time lag between release timing (Elberse and Eliashberg 2003). It is
important to examine whether sales in different markets are independent or not. We continue
discussing this issue in essay three.
104
Tables and Figures
Table 3-1: Summary of DMA markets
Market
ID
Market
First
Date
Last
Date
Number of
Performances
Number of
Show Dates
Number of
Venues
1 Tallahassee 1/1 1/4 6 4 1
2 Orlando 1/2 1/4 6 3 1
3 Tampa 1/7 1/11 8 5 1
4 Miami 1/8 1/18 16 10 1
5 Jacksonville 1/14 1/18 9 5 1
6 Birmingham 1/21 1/25 10 5 1
7 Nashville 1/22 1/25 8 4 1
8 Columbia 1/28 2/1 9 5 1
9 Greenville-Asheville 1/28 6/13 19 10 2
10 Raleigh-Fayetteville 2/5 5/23 17 9 2
11 Richmond 2/11 2/16 11 6 1
12 Atlanta 2/12 2/22 21 10 1
13 Norfolk-Hampton 2/18 2/29 19 10 2
14 Cincinnati 2/25 2/29 9 5 1
15 Charlotte 3/3 3/7 11 5 1
16 New York 3/3 6/6 70 32 4
17 Wheeling 3/17 3/20 7 4 1
18 Louisville 3/24 3/28 7 5 1
19 Terre Haute 4/6 4/7 4 2 1
20 Champaign-Springfield 4/9 4/25 11 6 2
21 Carbondale 4/13 4/14 3 2 1
22 Philadelphia 4/14 5/2 31 16 2
23 Charleston 4/15 4/18 8 4 1
24 Albany 4/29 5/2 8 4 1
25 La Crosse 5/4 5/5 4 2 1
26 Providence 5/5 5/9 9 5 1
27 Worcester 5/5 5/9 8 5 1
28 Madison 5/7 5/9 6 3 1
29 Hartford 5/12 5/16 9 5 1
30 Dayton 5/13 5/16 7 4 1
31 Mankato 5/18 5/19 3 2 1
32 Rochester 5/19 5/23 8 5 1
33 St. Paul 5/21 5/23 6 3 1
34 Cedar Rapids 5/25 5/26 4 2 1
35 Hershey 5/26 5/31 9 6 1
36 Sedalia 5/29 5/31 5 3 1
37 Memphis 6/2 6/5 6 4 1
38 Evansville 6/3 6/6 6 4 1
39 Wilkes-Barre 6/9 6/13 8 5 1
40 North Little Rock 6/16 6/20 8 5 1
41 Macon 6/23 6/27 8 5 1
42 Tucson 6/24 6/27 7 4 1
Average (STD) 10.69 (10.81) 5.79 (4.97) 1.19 (0.55)
105
Table 3-2: Summary of Market Responses to Performance Schedules
Market ID ?
0
?
1
?
2
?
3
?
1
?
2
1 -5.5 ** 0.3 1.32 ** 0.15 -1.8 **
2 -5.76 ** 1.19 ** 0.58 ** -0.79 **
3 -6.32 ** 0.75 ** 1.48 ** 0.88 ** -1.33 **
4 -6.97 ** 0.3 0.79 ** 0.61 ** -1.24 **
5 -5.96 ** 0.9 ** 1.33 ** 0.7 ** -1.69 **
6 -5.42 ** 0.62 ** 0.89 ** 0.22 -2.04 **
7 -5.99 ** 0.86 ** 1.42 ** 0.73 ** -1.4 **
8 -5.92 ** 0.95 ** 1.36 ** 0.59 ** -1.81 **
9 -7.17 ** 0.66 ** 0.94 ** 0.6 ** 0.96 ** -1.8 **
10 -6.16 ** 0.5 ** 0.81 ** 0.41 ** 4.29 ** -2.92 **
11 -5.66 ** 0.9 ** 1.3 ** 1.09 ** -2.38 **
12 -6.72 ** 0.24 0.5 ** 0.34 ** -1.2 **
13 -7.23 ** 0.54 ** 0.75 ** 0.48 ** 0.92 ** -0.84 **
14 -5.68 ** 0.4 * 0.91 ** 0.41 ** -1.74 **
15 -5.78 ** 0.69 ** 0.96 ** 0.53 ** -1.67 **
16 -8.23 ** 0.1 0.24 ** 0.2 ** 0.51 ** -1.16 **
17 -5.4 ** 0.94 ** 1.19 ** -2.22 **
18 -6.17 ** 0.56 * 1.21 ** 0.22 -1.18 **
19 -3.82 ** -2.22 **
20 -6.03 ** 0.40 ** 0.35 ** 2.09 ** -2.04 **
21 -4.43 ** -1.17 **
22 -7.18 ** 0.28 ** 0.37 ** 0.24 ** 0.99 ** -1.3 **
23 -5.68 ** 0.83 ** 1.27 ** 0.61 ** -1.8 **
24 -5.62 ** 0.89 ** 1.4 ** 0.79 ** -2.1 **
25 -4.79 ** -0.66 **
26 -5.63 ** 0.34 * 0.85 ** 0.27 * -1.71 **
27 -6.82 ** 0.53 ** 1.27 ** 0.71 ** -0.37 **
28 -6.1 ** 0.98 ** 0.29 -0.04
29 -5.86 ** 0.35 0.84 ** 0.29 * -1.32 **
30 -6.2 ** 0.7 ** 1.46 ** 0.9 ** -1.25 **
31 -3.83 ** -2.12 **
32 -6.19 ** 0.66 ** 1.34 ** 0.87 ** -1.41 **
33 -6.44 ** 1.05 ** 0.5 ** -0.59 *
34 -4.59 ** -0.98 **
35 -6.83 ** 0.45 ** 1.11 ** 0.71 ** -0.34 **
36 -6.35 ** 0.65 ** 0.77 ** 0.08
37 -6.13 ** 0.59 * 1.24 ** -1.08 **
38 -6.04 ** 0.35 1.16 ** 0.13 -0.88 **
39 -6 ** 0.51 ** 1.13 ** 0.6 ** -1.47 **
40 -6.06 ** 0.58 ** 1.19 ** 0.6 ** -1.51 **
41 -6.14 ** 0.48 * 1.22 ** 0.57 ** -1.34 **
42 -5.79 ** 0.54 ** 1.36 ** 0.82 ** -1.7 **
**: significant at the 95% highest posterior density
* : significant at the 90% highest posterior density
106
Table 3-3: Venue Locations and their Associated DMAs
Venue City State DMA Market
Pepsi Arena Albany NY Albany
Philips Arena Atlanta GA Atlanta
Bjcc Arena Birmingham AL Birmingham
Southern Illinois University Carbondale IL Carbondale
Us Cellular Center Cedar Rapids IA Cedar Rapids
Assembly Hall Champaign IL Champaign-Springfield
Prairie Capital Convention Center Springfield IL Champaign-Springfield
Charleston Civic Center Charleston WV Charleston
Charlotte Coliseum Charlotte NC Charlotte
Us Bank Arena Cincinnati OH Cincinnati
Colonial Center Columbia SC Columbia
Ej Nutter Center Dayton OH Dayton
Roberts Municipal Stadium Evansville IN Evansville
Asheville Civic Center Asheville NC Greenville-Asheville
Bi-Lo Center Greenville SC Greenville-Asheville
Hartford Civic Center Hartford CT Hartford
Giant Center Hershey PA Hershey
Jacksonville Veterans Memorial Arena Jacksonville FL Jacksonville
La Crosse Center Lacrosse WI Lacrosse
Louisville Gardens Louisville KY Louisville
Macon Centreplex Macon GA Macon
Alliant Energy Center Coliseum Madison WI Madison
Midwest Wireless Civic Center Mankato MN Mankato
Pyramid Arena Memphis TN Memphis
American Airlines Arena Miami FL Miami
Gaylord Entertainment Ctr Nashville TN Nashville
Continental Arena East Rutherford NJ New York
Madison Square Garden New York NY New York
Sovereign Bank Arena Trenton NJ New York
Nassau Coliseum Uniondale NY New York
Hampton Coliseum Hampton VA Norfolk-Hampton
Norfolk Scope Arena Norfolk VA Norfolk-Hampton
Alltel Arena North Little Rock AK North Little Rock
Td Waterhouse Centre Orlando FL Orlando
Boardwalk Hall Atlantic City NJ Philadelphia
Wachovia Spectrum Philadelphia PA Philadelphia
Dunkin' Donuts Center Providence RI Providence
Crown Center Of Cumberland Fayetteville NC Raleigh-Fayetteville
Rbc Center Raleigh NC Raleigh-Fayetteville
Richmond Coliseum Richmond VA Richmond
Blue Cross Arena Rochester NY Rochester
Mathewson Center Sedalia MO Sedalia
Xcel Energy Center St Paul MN St Paul
Tallahassee Leon County Civic Center Tallahassee FL Tallahassee
St Pete Times Forum Tampa FL Tampa
Hulman Center Terre Haute IN Terre Haute
Tcc Arena Tucson AZ Tucson
Wesbanco Arena Wheeling WV Wheeling
Wachovia Arena Wilkes-Barre PA Wilkes-Barre
Worcester Centrum Centre Worcester MA Worcester
107
Table 3-4: Summary of Ticket Sales
Market ID Market
Ticket Sales per Performance
Market Sales
Mean STD Min Max
1 Tallahassee 2362 1236 1659 4858 14174
2 Orlando 4348 396 3893 4952 26088
3 Tampa 6660 2056 2644 9294 53277
4 Miami 5192 2499 2150 10635 83075
5 Jacksonville 5038 1148 3357 6766 45341
6 Birmingham 8072 2048 5237 11214 80720
7 Nashville 6359 1944 3897 9274 50869
8 Columbia 3571 1147 2511 5707 32136
9 Greenville-Asheville 3475 2306 800 7716 66024
10 Raleigh-Fayetteville 5481 3480 1376 12747 93171
11 Richmond 6355 1240 3863 7703 69901
12 Atlanta 6611 1908 2499 9969 138827
13 Norfolk-Hampton 5209 1433 3095 8093 98977
14 Cincinnati 7788 2501 2638 10228 70092
15 Charlotte 7457 2647 2935 11459 82030
16 New York 8316 3525 1827 15810 582099
17 Wheeling 1848 450 993 2469 12939
18 Louisville 2734 317 2326 3173 19137
19 Terre Haute 3031 422 2545 3563 12124
20 Champaign-Springfield 1045 527 491 1910 11500
21 Carbondale 3182 1321 1838 4478 9546
22 Philadelphia 4497 2182 1548 8915 139421
23 Charleston 4407 651 3434 5442 35258
24 Albany 6489 1489 3456 8084 51914
25 La Crosse 907 219 603 1096 3629
26 Providence 5088 1065 2803 6216 45795
27 Worcester 3233 995 1521 4838 25864
28 Madison 585 210 278 914 3508
29 Hartford 5137 1258 2768 6925 46237
30 Dayton 2704 614 2011 3669 18928
31 Mankato 869 321 646 1237 2608
32 Rochester 2314 588 1472 3322 18513
33 St. Paul 1855 337 1516 2324 11130
34 Cedar Rapids 2037 287 1689 2356 8149
35 Hershey 758 147 574 982 6825
36 Sedalia 1088 203 855 1304 5442
37 Memphis 3135 321 2624 3539 18811
38 Evansville 972 429 507 1780 5834
39 Wilkes-Barre 2921 933 2018 4583 23369
40 North Little Rock 2879 827 2044 4281 23029
41 Macon 1183 369 735 1691 9461
42 Tucson 3465 824 2437 4379 24257
Market average 3825 1162 2098 5712
51905
(STD= 90992)
108
Table 3-5: Descriptive Statistics of Covariates across Markets
Market
ID
FRIDAY SATURDAY SUNDAY GEO TMP Population of
Target Market Mean STD Mean STD Mean STD Mean STD Mean STD
1 0.167 0.41 0.500 0.55 0.167 0.41 0.539 0.14 104326
2 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 477402
3 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 578505
4 0.125 0.34 0.375 0.50 0.250 0.45 0.299 0.04 741879
5 0.222 0.44 0.333 0.50 0.222 0.44 0.476 0.10 282479
6 0.200 0.42 0.300 0.48 0.200 0.42 0.470 0.07 321845
7 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 430888
8 0.222 0.44 0.333 0.50 0.222 0.44 0.476 0.10 169819
9 0.211 0.42 0.316 0.48 0.263 0.45 0.482 0.03 0.232 0.05 344151
10 0.235 0.44 0.353 0.49 0.176 0.39 0.492 0.09 0.245 0.05 463117
11 0.182 0.40 0.273 0.47 0.273 0.47 0.457 0.10 183972
12 0.190 0.40 0.286 0.46 0.286 0.46 0.307 0.04 1074597
13 0.211 0.42 0.316 0.48 0.263 0.45 0.497 0.03 0.296 0.04 345541
14 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 435717
15 0.182 0.40 0.273 0.47 0.273 0.47 0.473 0.07 466459
16 0.157 0.37 0.300 0.46 0.286 0.46 0.368 0.18 0.118 0.02 4100325
17 0.286 0.49 0.429 0.53 0.520 0.11 53729
18 0.143 0.38 0.429 0.53 0.143 0.38 0.470 0.13 298428
19 0.667 0.00† 66486
20 0.273 0.47 0.545 0.52 0.182 0.40 0.461 0.05 0.318 0.06 156173
21 0.667 0.14 155322
22 0.129 0.34 0.290 0.46 0.258 0.44 0.556 0.17 0.221 0.03 1321626
23 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 211135
24 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 225057
25 0.667 0.00† 86726
26 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 287211
27 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 1126579
28 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 147178
29 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 463979
30 0.143 0.38 0.429 0.53 0.286 0.49 0.532 0.12 240286
31 0.667 0.14 17074
32 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 186215
33 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 802661
34 0.667 0.00† 140908
35 0.111 0.33 0.333 0.50 0.222 0.44 0.438 0.11 316530
36 0.400 0.55 0.400 0.55 0.567 0.09 423202
37 0.167 0.41 0.500 0.55 0.506 0.13 347835
38 0.167 0.41 0.500 0.55 0.167 0.41 0.539 0.14 128297
39 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 227157
40 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 230179
41 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 109799
42 0.143 0.38 0.429 0.53 0.286 0.49 0.532 0.12 174998
Market
Average
0.172 0.40 0.382 0.51 0.250 0.46 0.476 0.09 0.476 0.09 439662
Note:
1. Markets without any observations in days of week are shown in blank cells.
2. Markets without any variation in their geographic schedules are shown in blank cells.
† Performances in market 19, 25, and 34 had equal temporal distance to other performances in the same market.
109
Table 3-6: Descriptive Statistics of Market Characteristics
Market Characteristics N Mean STD Min Max
POP
m
Population of target market 42 439662 647091 17074 4100325
POP_Density
m
Population density (in the log term) 42 5.717 0.85 3.787 7.615
Income
m
Family income (in the log term) 42 10.594 0.15 10.260 10.883
Family_Size
m
Average family size 42 3.028 0.09 2.857 3.276
110
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics
Additional Scheduling Characteristics N Mean STD Min Max
ADJ
m
Number of geographically adjacent
markets performing prior to a focal
market
42 0.95 1.00 0 4
ORDER
m
Distribution order in the temporal
sequence
42 21.43 12.3 1 42
111
Table 3-8: Sources of Heterogeneous Market Responses
Parameter Description Median (Std Dev)
Effect of Baseline performance attractiveness: E[?
0m
]
?
01
Intercept -4.30 (0.21)**
?
11
Population size in a target market -0.25 (0.06)**
?
21
Num. of geographically contiguous markets -0.14 (0.09)*
?
31
Num. of preceding markets in temporal sequence -0.12 (0.16)
Effect of Friday performances: E[?
1m
]
?
02
Intercept 0.31 (0.73)**
?
12
Population size in a target market -0.03 (0.01)**
?
22
Num. of geographically contiguous markets 0.00 (0.03)
?
32
Num. of preceding markets in temporal sequence 0.03 (0.06)
Effect of Saturday performances: E[?
2m
]
?
03
Intercept 0.51 (0.07)**
?
13
Population size in a target market -0.05 (0.01)**
?
23
Num. of geographically contiguous markets 0.01 (0.02)
?
33
Num. of preceding markets in temporal sequence 0.06 (0.05)
Effect of Sunday performances: E[?
3m
]
?
04
Intercept 0.21 (0.07)**
?
14
Population size in a target market -0.02 (0.01)**
?
24
Num. of geographically contiguous markets 0.00 (0.03)
?
34
Num. of preceding markets in temporal sequence 0.10 (0.05)**
Effect of Geographic density: E[?
1m
]
?
05
Intercept 1.22 (0.85)*
?
15
Population size in a target market -0.18 (0.32)
?
25
Num. of geographically contiguous markets -0.25 (1.19)
?
35
Num. of preceding markets in temporal sequence 0.86 (1.53)
Effect of Temporal density: E[?
2m
]
?
06
Intercept -0.61 (0.15)**
?
16
Population size in a target market 0.04 (0.04)
?
26
Num. of geographically contiguous markets 0.10 (0.06)**
?
36
Num. of preceding markets in temporal sequence 0.09 (0.11)
* significant at the 90% highest posterior density
** significant at the 95% highest posterior density
112
Table 3-9: Results of Performance Schedule Model
Parameter Description Median (Std Dev)
Expected geographic density: m
1
?
01
Intercept -0.06 (0.47)
?
11
Expected performance attractiveness 0.18 (0.06)**
?
21
Effect of geographic density 0.09 (0.14)
Expected temporal density: m
2
?
02
Intercept 3.35 (0.27)**
?
12
Expected value of performance 0.84 (0.05)**
?
22
Effect of temporal density 1.30 (0.18)**
** significant at the 95% highest posterior density
113
Figure 3-1: Conceptual Framework of Heterogeneous Market Responses
Attractiveness of
Performances
(1) Geographic
Density
At a performance level:
number of ticket sales
Temporal
Effect
Geographic
Effect
Scheduling Characteristics:
(2) Temporal
Density
Days of Week
Effects
Touring
Characteristics
Market
Characteristics
114
Figure 3-2: Example of a DMA and its county information
115
Figure 3-3: Locations of DMA Markets and Performing Sequence
116
Figure 3-4: Heterogeneous Market Responses
(a) Expected baseline effect: E[?
0
]
(b) Expected Friday effect: E[?
1
]
(c) Expected Saturday effect: E[?
2
]
(d) Expected Sunday effect: E[?
3
]
(e) Expected geographic effect: E[?
1
]
(f) Expected temporal effect: E[?
2
]
117
4 Essay 3: Sequential Distribution of a Live Performance Event
4.1 Introduction
When and where to schedule performances are two of the most important decisions
facing event marketers in the live entertainment industry. When event marketers schedule a tour
for an event, they have to design a performance schedule within each participating market and
determine an overall travel sequence across markets. Therefore, their scheduling decisions are
within and across markets and may have different effects on ticket sales.
In the first two essays, we have shown the effect of within-market scheduling and
identified explanatory factors for heterogeneous market responses. Specifically, we find that
venue locations in a geographic schedule influences ticket sales differently from do performance
dates in a temporal schedule. Densely scheduled performances across venues sell more tickets,
yet densely scheduled performances across times sell fewer tickets. Moreover, the population
size, geographic adjacency between markets, and temporal sequence in an event distribution can
explain heterogeneous market responses to some extent.
Because essays one and two have studied the effect of within-market scheduling and left
the impact of across-market scheduling unknown, essay three examines an event distribution
across markets and its impact on market sales. Specifically, an event distribution involves
scheduling across markets. Event marketers first decide a touring sequence at one time. Then, a
performing group follows this sequence to travel from one market to another. This group
performs in one market at a time, provides a few shows within a venue, and then leaves for
another market. Therefore, the mechanism of an event distribution is the same as the sequential
distribution.
118
Sequential distribution has been studied in marketing literature where researchers study
the market roll out of a new product (Bronnenberg and Mela 2004) and movie releases across
markets or channels (Lehmann and Weinberg 2000; Elberse and Eliashberg 2003; Chintagunta,
Gopinath, and Venkataraman 2009). These works show a dependent relationship between
preceding and following markets and indicate the effect of sequential distribution on sales or
profitability. In addition, they suggest the underlying reasons for the effect of sequential
distribution to be the lead market effect from geographically adjacent markets (Bronnenberg and
Mela 2004), word of mouth effect from previous markets (Elberse and Eliashberg 2003;
Chintagunta et al 2009), or effect of release timing between channels (Lehmann and Weinberg
2000). Hence, it is common to observe how well a new product sells in previous markets to
influence whether other markets adopt this product, when following markets launch this product,
and how well this product sells.
Similarly, when an event is distributed across markets, it is likely to see preceding
markets influencing following markets. This influence may come from geographic adjacency,
word of mouth, or release timing. Although essay two uses geographical adjacency between
markets and temporal sequence in a distribution to explain heterogeneous market responses, it
has not yet explored the possibility that markets may have a more direct dependent relationship.
Consequently, the objective of essay three is to examine whether an event distributed
sequentially across markets has an effect such that ticket sales in preceding markets can
influence sales in those following markets. We refer to such an effect as the carryover effect in
this essay.
To achieve this objective, we model ticket sales of each market as a function of its
performance schedule within a market and potential carryovers from an event distribution.
119
However, one modeling challenge is that scheduling decisions are correlated and endogenous
with the demand. For instance, the number of performances, the number of venues booked, and
the number of days scheduled may be highly correlated with one another. If we simply use these
scheduling decisions to explain market demand, these scheduling variables will be highly
correlated and suffer from the issue of collinearity.
To solve this issue, in addition to modeling supply and demand simultaneously to account
for the endogeneity, we have to use variables that are independent of one another yet still
represent the scheduling influences. Thus, in the demand model for event tickets, we use the size
of trading areas of booked venue(s) in a market as one instrumental variable for the scheduling
influence. We also use the distribution of performance dates in a schedule as the second
instrumental variable for the supply of an event on different dates.
On the other hand, we capture carryovers from an event distribution from the beginning
of its tour. Because an event travels to markets at different times and each market along the
distribution has different release timing, we employ the spatially weighted approach to account
for ticket sales of preceding markets as well as their release timing.
To specify the supply decisions simultaneously with the demand, we assume that event
scheduling within a market consists of three related decisions. First, event marketers consider
how many seats they have to provide in order to sell an expected number of tickets. We call this
decision an overall supply in a market. Next, they decide how many venues they need to reserve
given the number of seats needed in a market. This is the decision of venue usage. In the
meantime, they have to determine how many days they need to book given the number of seats
needed and the size of a venue capacity. We refer to this decision as the day usage decision. In
120
this way, we take into account the influence of expected demand on the supply decisions and use
variables to incorporate the scheduling influences on demand.
We contact a national ticket seller to obtain a dataset of live performance events and use
ticket sales for one event to test our model. This particular event sequentially performed 449
times in 50 cities in the U.S. domestic market between January and June 2004. Because we
aggregated these cities into 42 DMAs in essay two, we proceed to analyze ticket sales at a
market level in essay three.
The first finding of our demand model is that an event experiences more market sales
when more consumers are within the primary trading area of its venue(s). The intuition behind is
that when an event is more accessible to consumers and has more consumers within its primary
trading area, it provides more convenience to consumers due to a shorter travel distance. As a
result, an event with a larger trading area because of using more venues in a market can
accommodate more consumers and increase ticket sales.
Second, we show that an event sells more tickets when it has performances on various
dates in a market. In other words, when an event has more performances available to a market
and has a dispersed temporal schedule, it provides more flexibility to consumers especially for
those who have higher uncertainty about whether they could attend at a particular time.
Therefore, an event with a bigger variance in the distribution of performance dates sells more
tickets in a market.
Third, we show that an event distribution has a carryover effect on ticket sales. However,
the influence is across multiple venues within the same market but not across different markets.
In other words, when an event performs from one market to another, its ticket sales in preceding
markets do not affect sales in following markets. Yet, when this event performs in more than
121
one venue, its ticket sales in a preceding venue carry over to a later venue and influence its
overall market sales.
We think the nature of the family event analyzed in this essay is the underlying reason for
a carryover effect significant within a market but not across markets. Because this family event
targets young children and their parents, it is easier to observe children and parents discussing
event information within a market than across markets. Moreover, this family event travels
within the U.S. and often goes to a similar set of markets after a year or longer. Thus, these
markets do not have to depend on other participating markets but can rely on their own historical
experience to determine the quality of this event.
Finally, we show the endogeneity in the supply of an event. Event marketers use the
expected market demand to determine how many seats they need to provide, and this overall
supply further influences the number of venues and days they schedule the event.
The rest of essay three is organized as follows. We start with reviewing relevant
literature and constructing the conceptual framework. Next, we introduce our model
development and describe our data. After the model estimation and benchmark comparisons, we
discuss our results and conclude this essay.
4.2 Literature Review and Conceptual Framework
Essay three centers on literature in sequential distribution. Although we have reviewed
some relevant works in essay two, we discuss this literature in depth to show its mechanism and
possible effects in turn.
4.2.1 Sequential Distribution
Sequential distribution has been studied in movie and retailing contexts. The concept of
sequential distribution is that a new product starts in one channel or market and then gradually
122
distributes to another. Hence, as time passes by, the product availability increases and reaches
more consumers (Lehmann and Weinberg 2000). Based on where sequential distribution takes
place, we categorize extant works into two mechanisms: (1) sequential distribution across
channels and (2) sequential distribution across markets.
Sequential Distribution across Channels
Sequential distribution across channels refers to a new product released from one
channel to another, and it is a very common mechanism in the movie industry. One key
objective of research in this stream is to understand the impact of release timing of a new movie
title on its box-office revenues (Lehmann and Weinberg 2000; Hennig-Thurau et al 2006;
Hennig-Thurau, Houston, and Walsh 2007). In this way, researchers can suggest the optimal
release timing of a movie to another channel.
Specifically, Lehmann and Weinberg (2000) examined the optimal release timing from
movie theaters to video rental stores, and they found shortening the release timing (compared
with current practice) leads to increases in profits. On the other hand, Hennig-Thurau et al (2006)
studied revenue drivers in different movie channels (i.e., theater and video). They found that
release timing has a stronger influence on short-term box-office revenues than on long-term
revenues. Yet, release timing does not affect how well a movie sells on the video channel.
Another work by Hennig-Thurau et al (2007) examined the optimal release timing across
four distribution channels (i.e., theater, video purchase, video rental, and video on demand) and
further indicated that changing the order of distribution or shortening the release timing increases
profits for movie studios. To sum up, studies in this category emphasize the effect of time lags
between channels and show higher profitability due to a shortened time lag between channels.
123
Sequential Distribution across Markets
Sequential distribution across markets refers to a new product distributed from one
market to another, and it is observed in the movie and retailing industries. A primary objective
of this research stream is to investigate whether and how previous market success carries over to
later markets and influences market entry decisions or product sales in later markets (Elberse and
Eliashberg 2003; Bronnenberg and Mela 2004; Chintagunta et al 2009).
In terms of the movie industry, no matter whether movies are of limited or wide release
within a market, a common approach is to distribute movies sequentially across markets. For
instance, Elberse and Eliashberg (2003) investigated the relationship between the U.S. domestic
and a foreign market. Within a market, they concluded that box-office revenues of a movie title
and the number of screens in a theater in a preceding week affect the supply and demand in a
following week. They also found that total box-office revenues in the U.S market tend to carry
over to a foreign market and influence the supply and demand for the same movie in the opening
week. However, when they tested an interaction effect between the U.S. box-office revenues and
the time lag on a foreign market, they found this carryover effect only significant for the screen
management decision but insignificant for market demand.
Similarly, Chintagunta et al (2009) studied a sequentially released movie across the U.S.
local markets and examined the effect of online word of mouth (measured by the valance of
online reviews) from previous markets on the box-office revenues on the opening day in a new
market. They found that a time lag between an initial market and a current market negatively
affects sales, yet the average user rating of online reviews positively influences sales. However,
reviews are accumulated from the opening of a movie up to a current market. Researchers did
124
not incorporate the joint effect between the release timing and user rating generated in different
markets.
On the other hand, in the retailing context, Bronnenberg and Mela (2004) examined
manufacturers’ market entry decisions and retailers’ chain adoption decisions for newly
introduced frozen pizzas. They suggested that past market entry decisions influence whether a
manufacturer enters a new market, and past chain adoption decisions affect whether a chain in a
new market adopts this product. However, because the focus of this paper was on the supply
side, researchers did not investigate the effect of previous market entry or retail adoption on
demand in following markets.
Rationale for the Effect of Sequential Distribution
One primary reason for retailers or movie studios to practice sequential distribution is to
prevent financial losses of a failing launch (Lehmann and Weinberg 2000). Moreover, there are
several underlying reasons to explain why sequential distribution would be effective and
influence sales in later markets or channels.
For example, the success-breed-success effect (Elberse and Eliashberg 2003; Hennig-
Thuran et al 2006) shows a previous success in preceding markets or channels may be replicated
more easily in later entities. The word-of-mouth effect (Elberse and Eliashberg 2003; Hennig-
Thuran et al 2007; Chintagunta et al 2009) suggests that people exchange opinions and their
experiences influence how other people think. Moreover, the lead market effect (Bronnenberg
and Mela 2004) posits that similar behaviors tend to take place in spatially adjacent markets.
Thus, it is easier for adjacent markets to observe a focal market and imitate behaviors in this
focal market.
125
Summary
To sum up, studies in sequential distribution show a dependent relationship between
preceding and following markets and indicate the effect of sequential distribution on sales or
profitability. However, one limitation is that some, if not all, of these papers assume the decision
of sequential distribution is made one at a time rather than simultaneously (Lehmann and
Weinberg 2000; Elberse and Eliashberg 2003; Bronnenberg and Mela 2004; Hennig-Thuran et al
2006; Hennig-Thuran et al 2007). In other words, decisions of release timing, market entry,
chain adoption, and screen management are made sequentially after managers observe outcomes
(i.e., adoption decisions, box-office revenues, or profitability) from previous adoptions.
When managers have to plan a new product launch simultaneously for all participating
markets, it is not clear whether these markets still have a dependent relationship such that sales
in preceding markets carry over to following markets and influence their sales. Accordingly,
essay three contributes to this literature by studying an event distribution and its impact while a
touring sequence has to be planned at one time for an event to travel sequentially across markets.
4.2.2 Conceptual Framework
The objective of essay three is to examine whether markets along an event distribution
have a dependent relationship such that ticket sales of preceding markets have a carryover effect
to influence ticket sales in following markets. Hence, in our conceptual framework, we first
discuss how a performance schedule in a market influences its ticket sales. Then, we discuss
why preceding markets along an event distribution could influence following markets and what
the possible impact might be. Finally, we discuss the endogeneity between supply and demand
for an event.
126
Effects of Performance Schedule within a Market
When an event provides many performances in a market, its performance schedule is of a
relatively larger scale, compared with an event providing fewer performances. Among these
performances, if event marketers choose to book multiple venues and each venue is surrounded
by densely populated consumers, this event will be able to reach more consumers and have a
bigger primary trading area (Huff 1964) due to its enhanced spatial accessibility (Betancourt
2004). On the other hand, if these performances are at different times of day across various days
of week, this event will provide more flexibility to consumers and can deliver the performing
contents at consumers’ desired times (Betancourt 2004). In this way, the supply of a
performance schedule within a market influences how well an event sells in this market.
Effects of Carryovers from an Event Distribution
Moreover, as an event travels across markets and incurs a varying number of ticket sales,
it is possible to see preceding markets influencing following markets due to the effect of event
distribution. In other words, how well an event sells in a market may influence its sales in
adjacent markets via the lead market effect (Bronnenberg and Mela 2004). Addition, it is
possible that when an event sells well in one market, event marketers try to replicate this success
in another market because of the success-breed-success effect (Elberse and Eliashberg 2003;
Hennig-Thuran et al 2006).
On the other hand, it is also likely that consumers who have attended an event talk about
this event online or offline thus influencing people who have not yet attended. In this way, the
word of mouth of this event may travel across markets and influence people in different areas
(Elberse and Eliashberg 2003; Chingatunga et al 2009). Hence, the more ticket sales an event
127
experiences from previous markets, the higher is its volume of word of mouth, and the more
likely an event will sell well.
One special case occurs when an event travels to multiple venues of the same market. It
is possible that people who have gone to an event in a preceding venue express their opinions
about this event and influence other people in the same market. If so, an event distribution will
not only influence ticket sales across markets but also affect ticket sales across venues within the
same market. Consequently, the carryover effect from an event distribution may be across
markets as well as within a market (but across multiple venues).
Finally, although the population size of a market may influence the baseline market
demand, once we control for this market characteristic, a sequentially distributed event may still
influence its market sales through its performance schedule within a market and carryovers from
an event distribution.
Endogenous Supply and Demand of an Event Distribution
Meanwhile, because supply is often endogenous with demand, it is likely that the supply
of an event in a market influences its market sales, and the expected demand in this market
affects the supply of the same market. Therefore, our conceptual framework for an event
distribution must consider supply and demand simultaneously and allow the expected demand
and other scheduling constraints to influence the supply decision. Figure 4-1 below indicates the
endogenous relationship and summarizes the effects of performance schedule and event
distribution on market sales.
Figure 4-1: Conceptual Framework of Essay Three
Specifically, we assume event marketers make several scheduling decisions for a
touring event. Similar to the screen management decision in the movie industry (Elberse and
128
Eliashberg 2003; Basuroy, Desai, and Talukdar 2006; Eliashberg et al 2007), the first scheduling
decision is to determine the total number of seats an event needs to provide in individual markets.
Because the total number of seats is the maximum possible seats that an event can sell within a
market, we call this an overall supply decision. To endogenize the overall supply with market
sales, we assume event marketers rely on a size of market population and expected market
demand to set a desirable amount of supply. Hence, they may increase the overall supply when
they expect higher demand in a bigger market. In addition, after setting an overall supply of an
event, other scheduling issues are deciding how many venues to use and how many days to book
for this event. We call these decisions as venue usage and day usage and assume these decisions
are as a result of the market characteristics (e.g., population size or population density) and
scheduling constraints (e.g., venue capacity, venue availability, or facility rental fees).
4.3 Model Development
4.3.1 Overview
To test our conceptual framework, we model the supply and demand for an event
simultaneously. At the supply side, we model the total number of seats, venues, and days needed.
At the demand side, we specify ticket sales of each market as a function of its performance
schedule within a market and potential carryovers from an event distribution. However, one
modeling challenge is that scheduling decisions are correlated and endogenous with market
demand. For instance, the number of performances, the number of venues booked, and the
number of days scheduled in a market may be highly correlated with one another. If we simply
use these scheduling decisions to explain market demand, these covariates will be highly
correlated and suffer from the issue of collinearity.
129
To solve this issue, in addition to modeling supply and demand simultaneously to account
for the endogeneity, we have to use variables that are independent of one another yet still
represent the scheduling influences. Thus, in the demand model for event tickets, we use the size
of trading areas of booked venue(s) in a market as one variable for the scheduling influence. We
also use the distribution of performance dates in a schedule as the second variable for the supply
of an event on different dates.
Although our modeling approach is similar to that of Elberse and Eliashberg (2003) who
specified the number of screens and box-office revenues simultaneously for a sequentially
distributed movie, our approach differs from theirs because we model several elements in the
scheduling decisions in additional to just the capacity decision.
As such, we model the linerized supply and demand to follow the multivariate normal
distribution.
(19)
Y =
l
l
l
l
ln(Solcs
m
)
ln(Scots
m
)
ln(Icnucs
m
)
ln(Ðoys
m
) 1
1
1
1
~HIN__
y
1m
y
2m
y
3m
y
4m
_ , L_
where Sales
m
is the number of ticket sales in market m (i.e., market demand), Seats
m
is the total
number of seats supplied in market m (i.e., overall supply), Venues
m
is the number of venues
used in a market (i.e., venue usage), and Days
m
is the number of days available in a schedule (i.e.,
day usage). These dependent variables have expected values y
m
and a variance-covariance
matrix ?. In this way, the correlations between supply decisions and the correlation between
supply and demand are controlled in the variance-covariance matrix.
4.3.2 Demand Equation: Market Sales
According to equation (19), we model expected market sales as a function of its
performance schedule within a market and potential carryovers from an event distribution:
130
(20) y
1m
= o
0
+ X
1m
u
1
+X
2m
u
2
+Z
m
u
3
where y
1m
is the expected ticket sales (in the log term) in market m, X
1
is a vector of variables to
represent the scheduling influences, X
2
represents potential carryovers from an event distribution,
Z
m
is a vector of market characteristics used as control covariates, and ? is the vector of
associated parameter effects. We discuss the operationalization of these covariates in variable
specification.
4.3.3 Supply Equations: Overall Supply, Venue Usage, and Day Usage
Although some studies assume an exogenous supply decision (Swami et al 1999;
Eliashberg et al 2005; Eliashberg et al 2007; Chintagunta et al 2009), we propose an endogenous
and positive relationship between overall supply and market demand (Elberse and Eliashberg
2003; Basuroy et al 2006).
We assume that event scheduling within a market consists of several related decisions.
First, event marketers consider how many seats they have to provide in order to sell an expected
number of tickets. Hence, we specify the expected number of seats as a function of its expected
market demand and market characteristics:
(21) y
2m
= [
0
+?
1
y
1m
+Z
m
µ
2
where y
2m
is the expected number of seats (in the log term) provided, y
1m
is the expected market
sales (in the log term), Z
m
is a vector of market characteristics used as control covariates, and ?
is the vector of associated parameter effects.
Second, we assume the following decisions are to decide how many venues to reserve
and how many days to book. Specifically, given the amount of supply event marketers have to
provide, they can also refer to market characteristics such as market land area and population
density in a market to determine how many venues to book. They can also refer to the size of
131
venue capacity and the cost of venue rental to decide how many days are needed in order to be
cost effective. For example, if a market is densely populated and its land area is big, event
marketers may consider booking more venues to increase the spatial accessibility of this event.
On the other hand, if the average venue capacity is small and the daily cost of renting a venue is
high, event marketers may consider booking fewer days but scheduling more performances in a
day. We specify the expected venue and day usage as follows:
(22) y
3m
= y
0
+?
1
y
2m
+Z
m
y
2
y
4m
= 0
0
+ ?
1
y
2m
+Z
m
6
2
where y
2m
is the expected number of seats (in the log term) provided, y
2m
is the expected number
of venues (in the log term) needed, y
3m
is the expected number of days (in the log term) needed,
Z
m
is a vector of market characteristics, and ? and ? are the vectors of associated parameter
effects.
Similar to Elberse and Eliashberg (2003), the advantage of specifying supply and demand
equations in the way above is that parameters in equations (20), (21), and (22) represent
elasticity of covariates. These parameter values suggest how changes in their covariates result in
changes in demand and supply. We can also compare ? in equation (20) to rank the importance
of these covariates on market sales.
4.3.4 Variable Specifications
Because of the endogenous and correlated scheduling decisions, we aim to find
representative variables that are correlated with a performance schedule but do not have a severe
collinearity in the demand equation. Among several possible measures, we find the size of
trading area of venues and the flexibility in a temporal schedule may serve the modeling purpose.
The rationale is that when an event uses multiple venues in its performance schedule, it has a
132
bigger primary trading area and is more accessible for consumers within this area. In addition,
when this event has more performances at different times of day and on various days of week, it
provides consumers with higher flexibility to attend at their own convenience. Since these
benefits are relevant with a performance schedule but are not highly collinear, we propose three
variables and discuss how we operationalize these variables in turn.
Size of Trading Area
We follow the retail trading area in the retail location literature (Huff 1964; Applebaum
1966; Cliquet 1998) to compute the size of trading area of booked venue(s) in a market.
Specifically, we measure the size of trading area by referring to the population density around
each venue of a performance schedule. Because population density can determine the size of a
potential market (Huff 1964) yet it is not uniformly distributed (Donthu and Rust 1989), we
assume the zip code of each venue as its primary trading area and consumers living in the same
zip code to be the potential consumers. Therefore, the proxy of total consumers within the
primary trading areas is as follows:
(23)
Sizc
m
= Ðcnsity
¡
vcnuc
m
¡=1
where Size
m
is the size of trading area of selected venues, x
1m
is the total number of consumers
within the primary trading areas and Density
v
is the population density around the zip code of
venue v.
However, the level of people’s willingness to travel can expand or shrink the trading area
of a venue (Huff 1964; Applebaum 1966). In other words, when people have higher tolerance to
travel, a venue is able to reach more people and has a bigger trading area. To consider the factor
of travel tolerance, we further include an adjustment term and rewrite equation (23) as follows:
133
(24)
Sizc
m
= Jcnsity
¡
· z
¡
vcnucs
m
¡=1
where ?
v
is an adjustment term to represent consumers’ willingness to travel.
To approximate consumers’ willingness to travel, we use the average commute time as a
proxy. Our assumption is that consumers who spend more time commuting on a daily basis are
more willing to travel and have higher travel tolerance. Hence, we compute ?
v
as an index
relative to the average (Mazzeo 2002):
(25) z
¡
=
commutc
¡
mcon(commutc
¡
)
; v : e {m = 1,2, …, H]
where commute
v
is the average commute time (in minutes) for people living in zip code v and
mean(commute
v
) is the average. In this way, if ?
v
> 1, people are willing to travel farther, and
venue v has an expended trading area to reach more people. In contrast, if ?
v
< 1, the trading
area of venue v shrinks. Finally, if ?
v
= 1, the trading area of a venue is as it is.
Flexibility in a Temporal Schedule
Because the flexibility in a temporal schedule represents how easily consumers can attend
an event at different times of day and across various days of week, we use the distribution of
performance dates in a schedule as other variables. Specifically, we compute the average
number of performances per day for the flexibility during a day. In addition, among all
performances available to a market, we use its distribution and calculate its variance to represent
the flexibility during a week. Therefore, if an event has more performances per day and has a
bigger variance, consumers will have more flexibility to attend this event at their own
convenience.
134
Carryovers from an Event Distribution
As we mentioned in the conceptual framework, carryovers from an event distribution
may affect sales in later markets. Moreover, when an event performs in multiple venues within a
market, the carryover effect may also exist across venues but within this market. Therefore, we
measure across-market carryovers and within-market carryovers in this section.
To measure across-market carryovers since the beginning of an event distribution, we use
ticket sales from preceding markets as a proxy (Elberse and Eliashberg 2003; Bronnenberg and
Mela 2004). In this way, various amounts of ticket sales from preceding markets represent
different magnitudes of carryovers from these markets. However, because an event travels to
markets at a different time and each market along the distribution has different release timing, we
employ the spatially weighted approach to account for ticket sales of preceding markets as well
as their release timing (Yang and Allenby 2003; Bronnenberg and Mela 2004; Choi, Hui, and
Bell 2009).
(26)
AC
m
= W
1
Sales
m
= w
1mmi
Solcs
mi
M
m
|
=1
mi=m
where
w
1mmi
=
1
exp(J
1
(m, m'))
where AC
m
represents the amount of across-market carryovers up to market m, Sales
m
is a vector
of ticket sales that market m’ has occurred up to the beginning of market m (m’?m), and W
1
is a
spatial weight matrix where each element (W
1mm’
) is an exponentially weighted distance measure
between a preceding market m’ and a current market m (d
1
(m,m’)).
Although it is arguable that our across-market measure only considers the temporal
sequence of participating markets but not considers the geographic adjacency between markets,
we think our measure is better because this essay focuses on understanding the effect of an entire
135
distribution rather than the effect of spatially adjacent markets only. However, it is possible to
include the geographic adjacency between markets as an extension and study the effect of spatial
adjacency.
Therefore, before measuring d
1
in equation (26), we observe markets are temporally
adjacent in several ways. Figure 4-2 below illustrates various possibilities of market
connectedness, where a darker color represents a preceding market m’, a lighter color indicates a
current market m, and the width of a box is the length of a performing period. In our dataset, we
observe several ways of market connectedness. We classify them into three cases and discuss
each one in turn.
Figure 4-2: Illustration of Market Connectedness
Specifically, in the first case, Apart, performances in two markets are apart from each
other in a few days or are tightly connected. Case 1.a and Case 1.b in Figure 4-2 illustrates these
situations, respectively. In the second case, Overlap, performances in two markets have an
overlapped performing period. This overlapped period may be for a few days or for a longer
period. Case 2.a and Case 2.b in Figure 4-2 illustrates these situations, respectively. Note that
in the first two cases, an event only employs one venue to provide performances. If there are
multiple performing groups touring in the U.S. at the same time, it is possible to observe markets
without any time lag in between or markets have performances at the same time. Case 1.b, Case
2.a, and Case 2.b in Figure 4-2 represent these situations.
On the other hand, when an event performs in two markets and uses multiple venues in
one of the markets, it is likely that performances in a current market are in between two venues
of another market or overlaps with one of the venues. Case 3.a, In between a break, and Case
3.b, Overlap with one venue, in Figure 4-2 are examples of these situations. However, when an
136
event performs in two markets and both employ multiple venues, it is possible to see performing
periods overlapping in some venues yet apart in others (i.e., Figure 4-2, Case 3.c: between and
overlap). Therefore, Case 3.b and Case 3.c in Figure 4-2 exist if an event has more than one
performing group touring at the same time.
According to these different situations, we measure d
1
(in weeks) differently. In Case 1
(Apart), we measure d
1
by calculating the time lag (in days) between two markets and then
converting this value to week:
(27) J
1
(m, m
i
) = #oport 7 ?
where #apart indicates the time lag between two markets in days.
Case 2 represents a situation that markets have an overlapped performing period.
Because there are performances showing concurrently in two markets, we assume the carryovers
from these markets should have no decay but a stronger effect. In other words, markets of some
overlapped performing periods should have stronger influence on each other than markets that
are apart from each other. Therefore, overlapped markets should have higher spatial weights.
To do this, we allow d
1
(m,m’) in equation (26) to be negative such that its associated spatial
weight (W
1mm’
) is larger:
(28) J
1
(m, m
i
) = -1 × #o:crlop 7 ?
where #overlap indicates the number of days that two markets have overlapped performances.
This specification ensures overlapped markets in Case 2 have higher weights than markets in
Case 1.
Finally, in Case 3, venues are either in between a break (Case 3.a), overlap with one
venue (Case 3.b), or between and overlap (Case 3.c). We consider each condition separately and
measure d
1
accordingly:
137
• Case 3.a In between a break
Case 3.a is when performances in market m are in between the first and second venue of
market m’. Because by the time of market m, sales in the second venue of market m’ have not
yet occurred. Hence, d
1
(m,m’) is the time lag between performances in the first venue of market
m’ and those in the venue of market m. That is, Case 3.a is similar to Case 1.a. We measure d
1
using equation (27). However, the associated market sales (Sales
m’
) in equation (26) are up to
the end of first venue of market m’ only.
• Case 3.b Overlap with one venue
Case 3.b is when performances in market m overlap with the first venue of market m’.
This is a similar case to Case 2. Therefore, we measure d
1
using equation (28) yet sales for
market m’ are up to the end of first venue only.
• Case 3.c Between and overlap
Because Case 3.c is a combination of Case 1 and Case 2, we need to consider the number
of days markets are apart and overlapped at the same time. Hence, we allow #apart and #overlap
to cancel each other and measure d
1
accordingly:
(29) J
1
(m, m
i
) = |#oport + (-1 × #o:crlop)] 7 ?
To measure within-market carryovers from preceding venues in a market, we take the
same spatially weighted approach and assign a different weight based on the time lag to a current
venue v:
(30)
wC
m
= W
2
Sales
v
= w
2¡¡i
Solcs
¡i
¡,¡iem
where
w
2¡¡i
=
1
exp(J
2
, :'))
138
where WC
m
is the within-market carryovers for market m (if it is a market that uses multiple
venues), Sales
v’
is a vector of ticket sales that a preceding venue v’ has occurred up to the
beginning of venue v, and W
2
is a spatial weight matrix to measure the temporal difference (d
2
)
between a current venue v and its preceding venue v’ in market m.
In a market where an event employs multiple venues, performances tend to take place
sequentially with a number of days apart from one another. Hence, we measure d
2
in a similar
way to Case 1 in equation (27):
(31) J
2
, :
i
) = #oport 7 ?
4.3.5 Model Summary
To sum up, we model the supply and demand simultaneously. At the demand side, we
use three variables and two spatially weighted measures to represent the scheduling influences
from a performance schedule and carryovers from an event distribution. We also use the
population size as a market characteristic to control for the baseline effect on market demand. At
the supply side, we model the total number of seats, venues, and days an event needs to provide
for each market. We allow the expected market demand to influence the total number of seats
and assume event marketers book venues and days based on their overall supply. Meanwhile,
we take into account the potential effects of market characteristics and scheduling constraints in
the supply decision.
Finally, in case there are omitted yet correlated variables to affect supply and demand
jointly, we allow correlated error terms and estimate supply and demand simultaneously. We
rewrite our model specification as follow and present the relationships among these covariates in
Figure 4-3:
139
(32) Y =
l
l
l
l
ln(Solcs
m
)
ln(Scots
m
)
ln(Icnucs
m
)
ln(Ðoys
m
) 1
1
1
1
~HIN__
y
1m
y
2m
y
3m
y
4m
_ , L_
y
1m
= o
0
+o
1
ln(Sizc
m
) +o
2
ln(Intcnsity
m
) +o
3
ln(Iorioncc
m
)
+o
4
ln(AC
m
) +o
5
ln(wC
m
) +o
6
ln(P0P
m
)
y
2m
= [
0
+[
1
y
1m
+[
2
ln(P0P
m
)
y
3m
= y
0
+y
1
y
2m
+y
2
ln(Arco
m
) +y
3
ln(Ðcnsity
m
)
y
4m
= 0
0
+0
1
y
2m
+0
2
ln(Copocity
m
) +0
3
ln(Fccs
m
)
where
M = market 1,2,…,M market
Sales
m
= Number of ticket sales (market demand)
Seats
m
= Number of seats supplied (overall supply)
Venues
m
= Number of venues booked (venue usage)
Days
m
= Number of days scheduled (day usage)
Size
m
= Size of trading area of venues
Intenstiy
m
= Average number of performances per show day
Variance
m
= Variance in the distribution of performance dates
AC
m
= Across-market carryovers up to market m
WC
m
= Within-market carryovers in market m
POP
m
= Size of market population
Area
m
= Size of market land area (in square miles)
Density
m
= Population density in market m
Capacity
m
= Avg. capacity of selected venues in market m
Rental
m
= Avg. rental fees in market m
Figure 4-3: Overview of Model Development
4.4 Data
We contact a national ticket seller to obtain a dataset of live performance events and use
ticket sales for one event to test our model. This event sequentially performed 449 times in 50
140
cities in the U.S. domestic market between January and June 2004. We have aggregated the data
into 42 DMAs in essay two and analyze this aggregated dataset in essay three
5
.
4.4.1 Touring Sequence
We have discussed how the event travels across markets in essay two. Table 3-1 lists the
first and last dates of performances and the venue usage for each DMA and Figure 3-3 shows the
touring sequence graphically. Among these 42 DMAs, the event performed in multiple venues
in six DMAs where we observe two ways of within-market touring.
One approach is that the event first performed in one venue in a focal DMA, left for other
DMAs, and then returned to the focal DMA again but performed in a different venue.
Specifically, the event took this approach in the Greenville-Asheville DMA (market ID= 9), the
Raleigh-Fayetteville DMA (market ID= 10), and the Champaign-Springfield DMA (market
ID=20). In the Greenville-Asheville DMA, the event first performed in Greenville between
January 28 and February 1, left for other DMAs, and then returned to Asheville on June 9, 2004.
In the Raleigh-Fayetteville DMA, it performed in Raleigh between February 5 and February 9,
left for other DMAs, and then returned to Fayetteville on May 20, 2004. Finally, in the
Champaign-Springfield DMA, it performed in Springfield between April 9 and April 11 and then
in Champaign on April 23.
The other approach is that the event performed in venues within a market consecutively
and then left for other DMAs. For example, when it performed in the Norfolk-Hampton DMA
(market ID= 13), it first stayed in Norfolk between February 18 and February 22 and then went
to Hampton on February 25. In the Philadelphia DMA (market ID= 22), it first performed in
Philadelphia between April 14 and April 25 and then went to Atlantic City on April 28.
5
Essay one provides detailed discussion about the event and essay two describes how we aggregate sales data into
42 DMAs.
141
However, the tour for the New York DMA (market ID= 16) is a combination of these two
approaches. The event first performed in East Rutherford, Uniondale, and New York
consecutively between March 3 and April 11. Then, it left for other DMAs and finally returned
to Trenton on June 3.
4.4.2 Covariates in the Demand Model
Venue Usage and Size of Trading Area
In terms of the venue usage within a market, we find that a selected venue may not be in
the center of a DMA but it is often located in a densely populated area. For example, Figure 4-4
illustrates the venue locations of the six DMAs discussed above and shows the population
density around each venue. Therefore, it is reasonable to assume that the zip code of a venue is
the primary trading area for this venue and consumers living in this zip code are the potential
consumers in the trading area. As such, it is meaningful to use the population density to measure
the size of potential market.
Figure 4-4: Venue Locations and Surrounding Population Density
Specifically, to calculate the size of trading area for selected venues in a market, we refer
to the data collected from the U.S. Census Bureau for population, land area (in square miles), and
consumers’ travel time to work (in minutes) at a zip code level for all venues observed in the
dataset. We further divide the population by the size of land area for the population density of
each venue (Venue
v
). On average, there are 3,289 consumers in the same zip code of a venue
with the standard deviation being 4,239.
In addition, the average travel time to work across all observed venues is 20.70 minutes
with the standard deviation being 5.71. We divide travel time to work for each venue
(Commute
v
) by the sample average to get the index of travel tolerance (?
v
). Hence, the mean
142
travel tolerance across venues is 1 with its standard deviation being 0.28. Using the new
information above, we compute the size of trading area at a venue level. Then, for DMAs that
have multiple venues, we aggregate across venues to conclude that an average size of trading
area in a market is about 4,291 consumers with the standard deviation being 9,968. Table 4-1
indicates the descriptive statistics of the size of trading area across markets (in the log term).
Table 4-1: Descriptive Statistics of Covariates (in the log term)
Day Usage and Flexibility Measures
We summarize the touring dates across markets in Figure 4-5. On average, it took an
event 5.79 days performing in a market with the number of show days ranging from two to 32.
In terms of days of week for performances, we find that most markets (32 out of 42 DMAs) tend
to offer the last performance on Sunday. However, depending on the number of performances
offered in a market, some markets started the first performance on Wednesday (22 out of 42
DMAs), some on Thursday (10 out of 42 DMAs), and the rest on Tuesday or Friday. In other
words, the flexibility that an event provides with consumers to attend on various days of week is
heterogeneous across markets. Descriptively, nine DMAs have performances for two to three
days in a week, nine DMAs have performances across four days, 14 DMAs have performances
across five days, and 10 DMAs have performances more than six days.
Figure 4-5: Touring Dates across Markets
After examining the distribution of performance dates for each market, we find an
average market has its variance in the distribution of performance dates to be 199.71 days with
the standard deviation being 831.68. This skewed distribution is due to four DMAs that have
performances in multiple venues and have a long lag between venues. Therefore, if we exclude
143
these four DMAs, the average value of variance is 2.325 days with its standard deviation being
3.37.
In addition, on average, there are 1.79 performances available during a show day with the
standard deviation being 0.21. Although we do not know the specific times for day of our
performances, we still find the flexibility that an event provides with consumers to attend at
different times of day to be heterogeneous across markets. Table 4-1 indicates the descriptive
statistics of the flexibility in days of week and for times of day (in the log term).
Across-Market and Within-Market Carryovers
Figure 4-5 also illustrates how markets are temporally connected with one another.
According to venue usage and the first and last dates of performances, we classify 42 DMAs into
markets that are completely apart from one another (Case 1 in Figure 4-2), overlapped (Case 2),
or are in between multiple venues (Case 3). We further apply equations (27), (28), or (29) to
compute the time lag between markets to ensure all preceding markets have different spatial
weights in equation (26) and calculate across-market carryovers accordingly. Hence, the average
across-market carryovers are 73,963.88 with the standard deviation being 65,688.67.
Figure 4-5: Touring Dates across Markets
Similarly, we use equation (31) to calculate the time lag between venues within a market
for those six DMAs that have multiple venues. As a result, we have an average within-market
carryovers being 51,564.62 and its standard deviation being 73,673.07. Table 4-1 indicates the
descriptive statistics of the across-market and within-market carryovers (in the log term).
Population Size
The last covariate in the demand equation is the population size in each market. Because
we have computed the target market population in essay two, we include the descriptive statistics
144
in Table 4-1. The average population size in a market is 439,662 and the standard deviation is
647091.
4.4.3 Covariates in the Supply Model
Venue Capacity and Market Capacity
Before we study the overall supply in each market (Seat
m
) as the first supply decision, we
need to know the number of performances and the seating capacity for each venue. However, we
do not have information regarding venue capacity in our dataset, so we refer to venue websites
and the Wikipedia to collect seating capacity data.
According to venue websites, venues have various configurations for different events
(e.g., basketball games, hockey games, concerts, etc.). Hence, we choose the format that is the
closest to the setting of a family event and record the associated capacity as the venue capacity
(Capacity
v
). Using all venue capacity in the same market, we further compute the average venue
capacity in a market (Capacity
m
) for equation (32) and observe heterogeneity in venue capacity.
On average, a venue has 13,612 seats with standard deviation being 4,682. Fifty percent of
venues have capacity between 10,423 seats (quartile 1) and 17,315 seats (quartile 3).
Next, we multiply the venue capacity by the number of performances in this venue to
know the total number of seats supplied by this venue and then sum across all venues in the same
market to get the market capacity (Seats
m
). On average, an event offers 156,747 seats in a
market with its standard deviation being 184,537. Some markets only supplied a few seats (e.g.,
Mankato DMA offered 14,496 seats in a total) but some markets provided as many as 1,159,059
seats (i.e., New York DMA).
According to the market sales and market capacity, we find that at most 50% of market
capacity was filled. On average, only 24% of total capacity was filled in a market. Figure 4-6
145
presents the rate of filled capacity across markets. Table 4-1 also indicates the descriptive
statistics of the overall market supply (in the log term).
Figure 4-6: Capacity-Filled Rate across Markets
Market Land Area and Population Density
To study the number of venues a market needs to book in equation (32), we still need
information about the market land area and population density at a market level. Hence, we use
the U.S. Census Bureau statistics to obtain the market land area (in square miles) and the size of
population at a county level.
After we divide the county population by its land area, we understand the population
density at a county level. Then, we sum across all counties within the same DMA to get the
population density (Density
m
) at a market level. We also compute the land area in a market
(Area
m
) by summing the land area across counties in the same DMA. Therefore, an average
market has 3,846 square miles of land and its population density is 422 people per square mile.
Table 4-1 indicates the descriptive statistics of the market land area and population density (in
the log term).
Facility Rental Fees
We also use the facility rental fees (Fees
m
) to study the number of days event marketers
need to book in a market. However, because this information is not publicly available, we
assume rental fees of a venue are positively correlated with housing values in the same zip code.
In other words, when a median value of a single-family house is high, it is very likely that rental
fees of a venue in this zip code are also high. Hence, we collect housing market data from the
U.S. Census Bureau based on this assumption. The distribution of the housing value across
markets is skewed. The average value (in thousands) is 2055 and the standard deviation is 4855.
146
The median of the housing value is 94.2. Table 4-1 also summarizes the descriptive statistics of
this covariate (in the log term).
Correlation
Finally, before we estimate the proposed model, we also check the correlation among our
covariates and the correlation among all dependent variables. Table 4-2 reports the correlation
between covariates in the demand model and Table 4-3 indicates that dependent variables are
moderate or highly correlated. Hence, our proposed model that correlates all error terms among
dependent variables has accounted for this issue.
Table 4-2: Correlation Coefficient of Demand Covariates
Table 4-3: Correlation Coefficient of Dependent Variables
4.5 Estimation and Results
4.5.1 Model Estimation
We estimate demand and supply equations simultaneously using the Bayesian approach
where parameters are specified to follow diffuse prior distribution as follows:
Priors for model of market sales:
o
0
~N(1u,1u) and o
ì
~N(u,1uu) where i=1, 2, 3, 4, 5, 6
Priors for model of overall supply:
[
0
~N(11,1u) and [
ì
~N(u,1uu) where i=1, 2
Priors for model of venue usage:
y
0
~N(u.11,1u) and y
ì
~N(u,1uu) where i=1, 2, 3
Priors for model of day usage:
0
0
~N(1.S,1u) and 0
ì
~N(u,1uu) where i=1, 2, 3
Priors for the variance-covariance of Supply and Demand:
L
-1
~wcibull(I
4
, 4)
147
We run 30,000 iterations in WinBUGs. After checking the convergence criteria, we
check the autocorrelation plots for all covariates, discard the first 20,000 iterations for burn-in,
and use the remaining iterations as the posterior distribution.
4.5.2 Results of Demand Equation
Scheduling Influence in a Performance Schedule
First, we show that the size of trading area of selected venues in a market has a positive
effect on market sales. When an event performs in multiple venues in a market or when its
venues are located in densely populated areas, this event sells more tickets in this market
(?
1
=0.160). Our explanation is that scheduling performances in multiple venues or selecting
venues in densely populated area can improve the spatial accessibility of this event (Huff 1964;
Donthu and Rust 1989; Betancourt 2004). In this way, consumers living within the trading area
do not have to travel a longer distance to attend an event. When there are more consumers
within the trading area of a market, it is more likely for this market to sell more tickets (Huff
1964; Applebaum 1966; Cliquet 1988).
Second, we find the flexibility of performance dates in a schedule has a positive effect on
ticket sales. When the variance in the distribution of performance dates increases, an event
provides consumers with higher flexibility to attend on various days of week (Betancourt 2004)
and sell more tickets in a market (?
3
=0.235). However, when there are more performances in a
show day, the flexibility for times of day does not contribute to ticket sales.
Carryovers in an Event Distribution
Moreover, we show that a sequentially distributed event has an effect on ticket sales.
However, the influence is across multiple venues within the same market but not across different
markets. In other words, when an event performs from one market to another, its ticket sales in
148
preceding markets do not affect sales in following markets. Yet, when this event performs in
more than one venue within a market, its ticket sales in a preceding venue can carry over to a
later venue and influence its overall market sales (?
5
=0.083).
We think the nature of the family event analyzed in this essay is the underlying reason for
a carryover effect significant within a market but not across markets. Because this family event
targets young children and their parents, it is easier to observe children and parents discussing
event information within a market than across markets. Moreover, this family event travels
within the U.S. and often goes to a similar set of markets after a year or longer. Thus, these
markets do not have to depend on other participating markets but can rely on their own historical
experience to determine the quality of this event.
Our results are consistent with the results in Elberse and Eliashberg (2003). Specifically,
they used sales in a preceding week as the volume of word of mouth and found a significant
effect on box-office revenues in a following week within the same market. This is similar to our
carryover effect within a market except that our carryover effect is across venues but not over
time. On the other hand, when Elberse and Eliashberg (2003) measured the interaction effect
between market sales in a domestic market and the time lag between a domestic and foreign
market, they found this interaction, or weighted word of mouth, effect insignificant across
markets. Although our spatially weighted measure for the carryovers across markets is similar
to their measure except that we take into account all of the participating markets instead of just
the initial market, our results are consistent with their work and support their results.
Elasticity
Another advantage of our model specification is that parameter estimates in our demand
model suggest the elasticity for all covariates. After comparing elasticity across significant
149
covariates, we conclude that the variance in the distribution of performance dates is more
important than the size of trading area of selected venues and the carryovers of an event
distribution within a market, respectively in this order. Although market population has a higher
elasticity (?
6
= 0.264), this is a market characteristic that event marketers cannot manipulate in
their scheduling decisions. Furthermore, these elasticity values suggest event marketers which
factors to strengthen. If trade-offs among these factors have to be made, they can make rational
decisions based on the elasticity. We summarize the parameter estimates in our demand model
in Table 4-4.
Table 4-4: Results of the Demand Model
4.5.3 Results of Supply Equations
From the results of the supply model, we confirm the endogeneity between the supply
and demand for an event. Event marketers use the expected market demand to determine how
many seats they need to provide (?
1
=0.771) but not the size of market population. Moreover, in
terms of the venue usage, when event marketers need to increase their overall supply, they tend
to schedule performances in more venues (?
1
=0.473) but do not consider the size of market land
area or population density in this market.
On the other hand, when event marketers evaluate how many days to book for an event,
they consider how many seats they need to supply (?
1
= 0.929) and the average venue capacity in
a market (?
2
= -0.703). In other words, when they need to provide more seats to a market, they
book more days for this event and have a longer performing duration. However, the number of
days needed decreases with the venue capacity. Although we do not find average housing value
influences the decision of day usage, it is possible that this variable is not an appropriate proxy
150
for the rental facility fees in a market. Table 4-5 below summarizes the parameter values in the
supply equations.
Table 4-5: Results of the Supply Model
4.5.4 Correlated Demand and Supply
Finally, we find correlated error terms in the supply and demand models. In other words,
there are unspecified covariates affecting both supply and demand at the same time. The
correlation between the market sales and overall supply is 0.549 and the correlation between the
market sales and day usage is 0.499. On the other hand, the decisions of overall supply and day
usage are also correlated (corr(?
Seats
, ?
Days
)=0.823)). Hence, it is necessary to assume correlated
error terms in our model to avoid biased parameter estimates. Table 4-6 below indicates the
correlation coefficients among our supply and demand models.
Table 4-6: Correlation between Supply and Demand Models
4.6 Conclusions
4.6.1 Summary
When an event travels across markets and has its distribution sequence planned at one
time, it is not clear whether and how the sequential distribution of this event influences ticket
sales in each participating market. The objective of essay three is to examine whether markets
along an event distribution have a dependent relationship and whether ticket sales of preceding
markets have a carryover effect to influence ticket sales in following markets.
To achieve this objective, we model ticket sales of each market as a function of its
performance schedule within a market and potential carryovers from an event distribution.
Specifically, we employ three variables to represent the scheduling influences from various
venues and performance dates, and take the spatially weighted approach to capture carryovers of
151
participating markets that have different ticket sales and release timing. We also specify the
supply and demand for this event simultaneously to account for a likely endogeneity. At the
supply side, we model separate but correlated decisions of overall supply, venue usage, and day
usage. In this way, our proposed model provides better understanding of scheduling effects on
demand and control for the endogenous supply and demand.
We contact a national ticket seller to obtain a dataset of a live performance event and
analyze ticket sales at a market level. The first result indicates that an event experiences more
market sales when it plays in several venues and has a bigger trading area to accommodate more
potential consumers. Second, an event sells more tickets when its performances are dispersed
across days of week but not during many times a day. In other words, the flexibility in a
temporal schedule is along the days of week to attract more consumer attendance.
Third, we find a significant effect of carryovers from an event distribution. When an
event performs in multiple venues within a market, ticket sales in a preceding market carry over
to later venues and influence its market sales although this carryover effect does not exist across
participating markets. Finally, we find supply and demand for an event to be endogenous. Event
marketers use the expected market demand to determine how many seats they need to provide,
and this overall supply further influences how many venues they reserve and how many days
they book in a schedule.
4.6.2 Conclusion
This essay contributes to the sequential distribution literature by studying an event
distribution where its touring sequence is set at one time rather than sequentially. We show that
the impact of sequential distribution exists locally but not across markets. In other words,
although markets along a tour do not have a dependent relationship, venues of the same market
152
have a dependent relationship and preceding venues can influence ticket sales in later venues and
ticket sales in this market.
The methodological contribution of this essay is to employ variables for an endogenous
and correlated performance schedule. By modeling the supply and demand simultaneously and
having the variables in the demand equation, we ensure unbiased scheduling effects and provide
actionable implications for event marketers.
To sum up, the first implication of essay three lies in the decision of venue usage. After
event marketers decide how many venues to book, they can refer to the size of trading area of
each alternative venue and select among these venues accordingly. Moreover, they can add or
drop venues based on the size of trading area if the desirable number of venues is not feasible in
a market. The second implication lies in the decision of day usage. Event marketers should
consider not to allocate multiple performances in a day but to disperse performances across
various days of week. However, event marketers should still evaluate the overall costs for such
scheduling changes in any venue or day usage.
Finally, when a touring sequence has to be planned simultaneously, it is not necessary for
event markets to consider any dependent relationship across markets. However, if this event
performs in multiple venues in a specific market, it is preferred that event marketers schedule
this event in a more important venue prior to other venues. In other words, the scheduling
objective for a touring event should be to minimize the travel distance across markets but
maximize the within-market carryovers for venues in the same market.
153
Tables and Figures
Table 4-1: Descriptive Statistics of Covariates (in the log term)
Variable Label N Mean Std Dev Min Max
Covariates in the demand equation
Sales Number of ticket sales
42 10.18 1.15 7.87 13.27
Size Size of trading area
42 7.54 1.26 4.78 11.08
Intensity Avg. number of performances in a
show day
42 0.58 0.12 0.34 0.79
Variance Variance in the distribution of
performance dates
42 1.02 2.20 -1.10 8.44
AC Across-market carryovers
42 10.93 0.74 9.52 12.75
WC Within-market carryovers
6 5.16 8.33 -7.45 12.13
POP Population size
42 12.46 0.92 9.75 15.23
Covariates in the supply equations
Seats Number of seats supplied
42 11.59 0.85 9.58 13.96
Venues Number of venues booked
42 0.12 0.30 0 1.39
Days Number of days supplied
42 1.57 0.56 0.69 3.47
Area Size of market land area (in square
miles)
42 7.90 0.80 6.55 9.81
Density Population density of a market
42 5.72 0.85 3.79 7.62
Capacity Avg. capacity of selected venues in a
market
42 9.45 0.42 8.06 10.09
Rental Avg. housing value of a market (in
thousands)
42 5.24 1.85 3.60 9.71
154
Table 4-2: Correlation Coefficient of Demand Covariates
ln(Size) ln(Intensity) ln(Variance) ln(AC) ln(WC) ln(POP)
ln(Size) --
ln(Intensity) 0.41 --
ln(Variance) 0.52 0.22 --
ln(AC) -0.02 0.08 -0.00 --
ln(WC) 0.44 0.22 0.20 0.11 --
ln(POP) 0.63 0.28 0.45 -0.23 0.16 --
155
Table 4-3: Correlation Coefficient of Dependent Variables
ln(Sales) ln(Seats) ln(Venues) ln(Days)
ln(Sales) --
ln(Seats) 0.87 --
ln(Venues) 0.51 0.57 --
ln(Days) 0.80 0.86 0.73 --
156
Table 4-4: Results of the Demand Model
Parameter Effect Median (Std Dev)
ln(Sales): Expected ticket sales in a market
?
0
Intercept 5.329 (1.164) **
?
1
Effect of geographic coverage 0.161 (0.073) *
?
2
Effect of intensity 0.128 (0.168)
?
3
Effect of variance 0.235 (0.038) **
?
4
Effect of across-market carryover 0.002 (0.063)
?
5
Effect of within-market carryover 0.083 (0.021) **
?
6
Effect of population size 0.264 (0.086) **
** significant at the 95% highest posterior density
157
Table 4-5: Results of the Supply Model
Parameters Median (Std Dev)
ln(Seats): Expected number of seats supplied in a market
?
0
Intercept 4.766 (0.832) **
?
1
Effect of expected market sales 0.771 (0.086) **
?
2
Effect of population -0.073 (0.050)
ln(Venues): Expected number of venues needed
?
0
Intercept -4.621 (0.709) **
?
1
Effect of planned overall supply 0.473 (0.067) **
?
2
Effect of market land area -0.042 (0.046)
?
3
Effect of market population density -0.087 (0.050)
ln(Days): Expected number of days needed
?
0
Intercept -2.169 (1.097)
**
?
1
Effect of planned overall supply 0.929 (0.070)
**
?
2
Effect avg. venue capacity -0.703 (0.098)
**
?
3
Effect of housing value -0.024 (0.028)
** significant at the 95% highest posterior density
158
Table 4-6: Correlation between Supply and Demand Models
?
?
Sales
?
Seats
?
Venues
?
Sales
--
?
Seats
0.549
**
--
?
Venues
-0.169
-0.280 --
?
Days
0.499
**
0.823
**
-0.273
** significant at the 95% highest posterior density
159
Figure 4-1: Conceptual Framework
Market
Characteristics
Scheduling costs
& constraints
Carryovers across
and within a market
Endogeneity
Sales in Previous
Markets and Venues
Demand for an Event
Ticket sales
at a market level
Supply of an Event
(Scheduling Decisions)
1. Overall supply
2. Venue usage
3. Day usage
Figure 4
Case (1):
(1.a) A fe
Case (2):
(2.a) Littl
Case (3):
(3.a) In be
(3.c) Betw
4-2: Illustrati
Apart
w days apart
Overlap
e overlapped
In between a
etween a brea
ween and over
ion of Marke
a market of m
ak
rlap
et Connected
multiple ven
160
dness
(1.b) Tig
(2.b) Mu
nues
(3.b) Ov
ghtly connect
uch overlappe
verlap with on
t
ed
ne venue
previ
curren
ious market m
nt market m
’
161
Figure 4-3: Overview of Model Development
Size of
Trading Area
Market
population
Venue
capacity
Rental
facility fees
Population
density
Market land
area
Within?market
carryover
Sales
Across?market
carryover
Seats
Venues
Days
Intensity
Variance
?
1
?
2
?
3
?
4
?
5
?
6
?
1
?
2
?
1
?
2
?
3
?
1
?
2
?
3
162
Figure 4-4: Venue Locations
(a) Greenville-Asheville
(b) Raleigh-Fayetteville
(c) Norfolk-Hampton
(d) New York
(e) Champaign-Springfield
(f) Philadelphia
163
Figure 4-5: Touring Dates across Markets
12/30/2003
1/19/2004
2/8/2004
2/28/2004
3/19/2004
4/8/2004
4/28/2004
5/18/2004
6/7/2004
6/27/2004
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41
P
e
r
f
o
r
m
a
n
c
e
D
a
t
e
s
Market ID
164
Figure 4-6: Capacity-Filled Rate across Markets
0%
10%
20%
30%
40%
50%
60%
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41
%
C
a
p
a
c
i
t
y
F
i
l
l
e
d
Market ID
165
5 Conclusion
5.1 Summary
Scheduling is an important decision facing event marketers in the live entertainment
industry. When they schedule a tour for an event, they have to design a performance schedule
within each participating market and determine an overall travel sequence across markets.
Therefore, their scheduling decisions are within and across markets, which may have different
effects on ticket sales.
Although marketing research in the live entertainment industry has focused on
identifying drivers for ticket sales, researchers have not evaluated whether the scheduling
decisions influence how markets respond. In other words, researchers treat the supply and
demand for an event as two separate problems and they have not yet investigated the relationship
between supply and demand.
As such, this dissertation analyzes performance schedules of a live performance event
and examines the effects on ticket sales within and across markets. Specifically, essay one
investigates whether and how performances of similar scheduling characteristics sell differently
in terms of how many tickets each performance sells and when ticket sales arrive. We use the
venue locations and performance dates as the scheduling characteristics for each performance
and measure the similarity in these scheduling characteristics by the geographic distance between
venues and the temporal distance between performance dates.
Methodologically, we use the competing destination model to examine the number of
ticket sales and the Weibull hazard model for the timing of ticket sales. In addition, we also
control for a possible endogeneity between a performance schedule and it demand effect. Using
166
70 performances in the New York market, we show that performances sell differently depending
on how similar their venue locations and performance dates are.
In other words, when performances are in the same or nearby venues, they have higher
similarity in a geographic schedule. From consumers’ perspective, they may perceive nearby
venues to be more attractive due to this similarity in venue locations. As a result, more
consumers purchase tickets for those performances and they are more willing to purchase these
tickets early. On the other hand, when performances are on the same or closer dates, they have
higher similarity in a temporal schedule. Because consumers often have uncertainty about their
consumption state and prefer various attendance timing for choices, shorter temporal distance
between performances result in higher competition and sales cannibalization.
Since results in essay one are limited in the New York metropolitan market, essay two
analyzes performance schedules across 42 markets and examines whether the results in essay one
are heterogeneous across markets and if there are any explanatory factors to explain differences
across markets. We first conduct a preliminary analysis using the same model developed in
essay one and compare the effects of scheduling characteristics across markets. After confirming
the heterogeneous market responses to performance schedules, we employ the hierarchical
Bayesian approach to identify explanatory factors for differences across markets.
Our results show that market population, geographic adjacency between markets, and
temporal sequence in an event distribution can explain different market responses. First, when a
market has a bigger population, the effects of days of week and baseline attractiveness are
attenuated. Second, with respect to a current market, after an event travels to more of its
geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule. Third, after an event follows its temporal sequence to
167
perform in several markets, a current market in a late distribution sequence tends to respond
more favorably to a Sunday performance.
As such, essay one examines the impact of a performance schedule in a single market,
and essay two analyzes all performance schedules observed in the dataset and uses market and
additional scheduling characteristics to explain different market responses. However, the event
in our dataset is sequentially distributed across its participating markets. It is important to
understand whether the carryover effect exists due to an event distribution and how this
carryover effect influences ticket sales of participating markets.
Therefore, essay three analyzes the distribution of this live performance event and
examines whether ticket sales of preceding markets carry over to following markets and
influence ticket sales in those following markets. Besides controlling for the effect of a
performance schedule within a market, we model ticket sales of each market as a function of its
potential carryovers from an event distribution. We also model the supply and demand for an
event simultaneously to account for a possible endogeneity. Specifically, we use the size of
trading area(s) of scheduled venue(s) in a market and flexibility in the distribution of
performance dates in a schedule to represent the scheduling influences and employ the spatially
weighted approach to incorporated carryovers of preceding markets and their different release
timing.
Our results show that when an event has a larger trading area in a market and/or offers
more performances along a dispersed time span, it tends to sell more tickets in this market.
Moreover, we show that an event distribution has an effect on ticket sales. However, the effect
of carryovers is significant across venues of the same market but not across markets. When an
168
event performs in more than one venue, its ticket sales in a preceding venue carry over to a later
venue and influence its overall market sales.
5.2 General Discussions
In general, this dissertation indicates several scheduling effects. We classify these effects
into effect of within-market scheduling and effect of across-market scheduling. Moreover, we
study the effect of within-market scheduling into two aspects. Essays one and two focus on the
effect of within-market scheduling at a performance level, yet essay three addresses the same
effect at a market level. We sum up these scheduling effects and discuss the differences in turn.
First, for the effect of within-market scheduling at a performance level in essays one and
two, we find that the effect of a geographic schedule differs from the effect of a temporal
schedule. The similarity in venue locations benefit ticket sales at a performance level but the
similarity in performance dates cannibalizes ticket sales. Moreover, market responses to
performance schedules are heterogeneous and can be explained via the market and additional
scheduling characteristics. Second, for the effect of within-market scheduling at a market level
in essay three, we find that an event sells more tickets when it employs more venues in its
geographic schedule and serve a bigger trading area. It also sells more when the distribution of
its performance dates in a temporal schedule has a big variance to offer consumers greater
flexibility in attendance timing. Third, for the effect of across-market scheduling in essay three,
we find that markets are not dependent on one another, but venues within the same market have a
dependent relationship to influence ticket sales in this market.
Although the effect of a geographic schedule at a performance level seems contradictory
to the effect at a market level, these results are in fact complimentary. Although the first two
essays suggest event marketers to decrease the geographic distance between performances and
169
increase the similarity in venue locations, this suggestion is for markets that employing multiple
venues only. When an event performs in just one single venue, event marketers can follow the
learning in essay three to select a venue that has the biggest trading area. In other words, if event
marketers decide to schedule an event in multiple venues, they can apply their learning in essay
three to select venues and then allocate performances to these venues based on the first two
essays.
Consequently, the order of these essays allows us to understand the scheduling effects
from a performance level to a market level. Essay one starts with examining the effect of a
performance schedule at a performance level in a single market and concludes that performances
scheduled closely in distance but distantly in time can experience more ticket sales. In addition,
essay one also suggests performances experiencing earlier timing of ticket sales when these
performances are scheduled in nearby venues. To test the generalizability of these results and
explain the heterogeneity across markets, essay two expands the scope of analysis and confirms
the scheduling impact in all participating markets of a touring event.
Finally, essay three examines whether an event has a carryover effect when it
sequentially distributes across markets. This essay concludes that markets do not influence one
another on their ticket sales yet their venues within the same market have such an effect.
Although one may argue that the third essay does not have to be conducted after the first two
essays, we choose this sequence to investigate the carryover effect after we can understand and
control for the effect of a performance schedule within a market.
Finally, although it is arguable that essay three could have used the density measures
developed in essay one, we choose to use three variables for the following reasons. First, the
scheduling characteristics represent the similarity between performances in a schedule. They do
170
not represent how well an event is able to serve its trading area at a market level or express the
flexibility in attendance timing in a temporal schedule. Second, the measure of geographic
density in essay one is applicable only when an event performs in multiple venues in a market.
For markets where an event performs in one venue only, there is no variation in its geographic
distance yet the trading area of this single venue could still influence ticket sales. Therefore, it is
necessary to use different measures to differentiate the effect of scheduling characteristics at a
performance level from the effect of a performance schedule at a market level.
5.3 Contributions
This dissertation has both empirical and academic contributions to the marketing field.
Empirically, we show that performance schedules do affect ticket sales. Managers can use
performance schedules to estimate ticket sales at a performance or market level. Event
marketers can use these estimates as benchmarks to monitor a pattern of ticket sales and even
allocate marketing resources accordingly.
Academically, the findings in this dissertation enrich literature in event tickets and
sequential distribution. We introduce new drivers of ticket sales to the event tickets literature
such that researchers can use new differential measures to explain variations in the number and
timing of ticket sales. Moreover, we examine a sequential distribution problem in a new context
where simultaneous planning is needed and find the effect of sequential distribution only within
a local market but not across markets.
5.4 Limitation and Future Research
The primary limitation in this dissertation is that we do not have access to consumer
identification data. Although individual transactions are observed, we cannot model a
consumer’s decision process to understand the effect of a performance schedule at a finer level.
171
To resolve this issue, we could apply an agent based modeling approach. Using our model
results as aggregated parameter values, we can further simulate individual consumers in a market
using the U.S. Census data and allow variations in agents’ preferences. This future direction will
better assist event marketers in performance scheduling and allow researchers to study marketing
problems using a complexity system.
Another future research lies in the pricing structure of event tickets. As our data suggest,
the total price that a consumer pays includes the face value, facility fees, and convenience
charges, where face value represents the highest share in the total price paid, followed by
convenience charges and facility fees. When summarizing consumers’ channel usage, we find
that consumers tend to purchase in box offices to avoid paying for convenience fees. However,
convenience fees are the major revenue source for ticket sellers. If ticket sellers and event
promoters could collaborate and re-structure the pricing breakdowns (e.g., the merger between
Ticketmaster, a primary ticket seller, and Live Nation, an event promoter), it is likely that
consumers’ ticket purchases will migrate to the Internet or other channels. This new topic
involves pricing and channel strategies and we leave it for a future direction.
172
Reference
Applebaum, William (1966), “Methods for Determining Store Trade Areas, Market Penetration,
and Potential Sales,” Journal of Marketing Research, 3 (2), pp.127-141.
Basuroy, Suman, Kalpesh Kaushik Desai, and Debabrata Talukdar (2006), “An Empirical
Investigation of Signaling in the Motion Picture Industry,” Journal of Marketing Research, 43
(2), pp.287-295.
Betancourt, Roger R. (2004), Economics of Retailing and Distribution, Edward Elgar Publishing
Ltd.
Bradlow, Eric T., Bart Bronnenberg, Gary J. Russell, Neeraj Arora, Sri Devi Duvvuri, Frankel
Ter Hofstede, Catarina Sismeiro, Raphael Thomadsen, and Sha Yang (2005), “Spatial Models in
Marketing,” Marketing Letters, 16 (3-4) pp.267-278.
Bronnenberg, Bart J. and Carl F. Mela (2004), “Market Roll-Out and Retailer Adoption for New
Brands,” Marketing Science, 23 (4), pp.500-518.
Cachon, Gérard P. (2004), “The Allocation of Inventory Risk in a Supply Chain: Push, Pull, and
Advance-Purchase Discount Contracts,” Management Science, 50 (2), p.222-238.
Carlyle, Kellie E., Michael D. Slater, and Jennifer L. Chakroff (2008), “Newspaper Coverage of
Intimate Partner Violence: Skewing Representations of Risk,” Journal of Communication, 58 (1),
pp168-186.
Chintagunta, Pradeep K., Shyam Gopinath, and Sriram Venkataraman (2009), “Online Word-of-
mouth Effecs on the Offline Sales of Sequentially released new products: An Application to the
Movie Market,” working paper.
Choi, Jeonghye, Sam K. Hui, David R. Bell (2009), “Spatio-Temporal Analysis of Imitation
Behavior in Adoption of an Online Grocery Retailer,” Journal of Marketing Research,
forthcoming.
Chu, Wujin and Woosik Chu (1994), “Signaling Quality by Selling through a Reputable Retailer:
an Example of Renting the Reputation of Another Agent,” Marketing Science, 13 (2), pp.177-
189.
Cliquet, Gerard (1998), “Integration and territory coverage of the hypermarket industry in France:
A relative entropy measure,” International Review of Retail, Distribution and Consumer
Research, 8 (2), pp.205-224.
Corning, Jonathan and Armando Levy (2002), “Demand for Live Theater with Market
Segmentation and Seasonality,” Journal of Cultural Economics, 26 (3), pp.217-235.
173
Currim, Imran S., Charles B. Weinberg, Dick R. Wittink (1981), “Design of Subscription
Programs for a Performing Arts Series,” Journal of Consumer Research, 8 (2), pp.67-75
Desiraju, Ramarao and Steven M. Shugan (1999), “Strategic Service Pricing and Yield
Management,” Journal of Marketing, 63 (1), pp.44-56.
Dobson, Gregory and Phillip J. Lederer Lederer (1993), “Airline Scheduling and Routing in a
Hub-and-Spoke System,” Transportation Science, 27 (3), pp.281-297.
Donthu, Naveen and Roland T. Rust (1989), “Estimating Geographic Customer Densities Using
Kernel Density Estimation,” Marketing Science, 8 (2), pp.191-203.
Dorndorf, Ulrich, Florian Jaehn, Chen Lin, Hui Ma, and Erwin Pesch (2007), “Disruption
Management in Flight Gate Scheduling,” Statistica Neerlandica, 61 (1), pp.92-114.
Elberse, Anita and Jehoshua Eliashberg (2003), “Demand and Supply Dynamics for Sequentially
Released Products in International Markets: The Case of Motion Pictures,” Marketing Science,
22 (3), pp.329-354.
Eliashberg, Jehoshua, Sanjeev Swami, Charles B. Weinberg, and Berend Wierenga (2009),
“Evolutionary Approach o the Development of Decision Support Systems in the Movie Industry,”
Decision Support Systems, 47 (1), p.1-12.
____, Anita Elberse, and Mark A.A.M. Leenders (2006), “The Motion Picture Industry: Critical
Issues in Practice, Current Research, and New Research Directions,” Marketing Science, 25 (6),
pp.638-661.
____, Quintus Hegie, Jason Ho, Dennis Huisman, Steven J. Miller, Sanjeev Swami, Charles B.
Weinberg, and Berend Wierenga (2007), “Demand-Driven Scheduling of Movies in a Multiplex,”
working paper.
Etschmaier, Maximilian M. and Dennis F. X. Mathaisel (1985), “Airline Scheduling: An
Overview,” Transportation Science, 19 (2), pp.127-138.
Fotheringham, A. Stewart (1983), “Some theoretical aspects of destination choice and their
relevance to production-constrained gravity models,” Environment and Planning A, 15 (8),
pp.1121-1132.
____ (1988), “Consumer Store Choice and Choice Set Definition,” Marketing Science, 7 (3),
pp.299-310.
Gallego, Guillerimo and Garrett van Ryzin (1994), “Optimal Dynamic Pricing of Inventories
with Stochastic Demand over Finite Horizons,” Management Science, 40 (8), pp.999-1020.
González-Benito, Óscar (2005), “Spatial Competitive Interaction of Retail Store Formats:
Modeling Proposal and Empirical Results,” Journal of Business Research, 58 (4), pp.457-466.
174
Havlena, William J. and Susan L. Holak (1988), “The Influence of Variety on the Demand for
Bundles of Musical Performances,” Advances in Consumer Research, 15, pp.22-26.
Hennig-Thurau, Thorsten, Mark B. Houston, Gianfranco Walsh (2006), “The Differing Roles of
Success Drivers Across Sequential Channels: An Application to the Motion Picture Industry,”
Journal of the Academy of Marketing Science, 34 (4), pp.559-575.
____, Victor Henning, Henrik Sattler, Felix Eggers, and Mark B. Houston (2007), “The Last
Picture Show? Timing and Order of Movie Distribution Channels,” Journal of Marketing, 71 (4),
pp.63-83.
Huber, Joel, John W. Payne, and Christopher Puto (1982), “Adding Asymmetrically Dominated
Alternatives: Violations of Regularity and the Similarity Hypothesis,” Journal of Consumer
Research, 9 (1), pp.90-98.
____ and Christopher Puto (1983), “Market Boundaries and Product Choice: Illustrating
Attraction and Substitution Effects,” Journal of Consumer Research, 10 (1), pp.31-44.
Huff, David L. (1964), “Defining and Estimating a Trading Area,” Journal of Marketing, 28 (3),
pp.34-38.
Jank, Wolfgang and P.K. Kannan (2005), “Understanding Geographic Markets of Online Firms
Using Spatial Models of Consumer Choice,” Marketing Science, 24 (4), pp.623-634.
Jarrah, Ahmad I., Jon Goodstein, and Ram Narasimhan (2000), “An Efficient Airline Re-
Fleeting Model for the Incremental Modification of Planned Fleet Assignments,” Transportation
Science, 34 (4), pp.349-362.
Kamakura, Wagner A. and Gary J. Russell (1989), “A Probabilistic Choice Model for Market
Segmentation and Elasticity Structure,” Journal of Marketing Research, 26 (4), pp. 379-390.
Lehmann, Donald R. and Charles B. Weinberg (2000), “Sales Through Sequential Distribution
Channels: An Application to Movies and Videos,” Journal of Marketing, 64 (3), pp.18-33.
Leslie, Phillip (2004), “Price Discrimination in Broadway Theater,” RAND Journal of
Economics, 35 (3), pp.520-541.
Lohatepanont, Manoj and Cynthia Barnhart (2004), “Airline Schedule Planning: Integrated
Models and Algorithms for Schedule Design and Fleet Assignment,” Transportation Science, 38
(1), pp.19-32.
Luce, R. Duncan (1959), Individual Choice Behavior, New York: John Wiley & Sons, Inc.
Lutz, Nancy A. and V. Padmanabhan (1995), “Why do We Observe Minimal Warranties?”
Marketing Science, 14 (4), pp.417-441.
175
Manchanda, Puneet, Peter E. Rossi, and Pradeep K. Chintagunta (2004), “Response Modeling
with Nonrandom Marketing-Mix Variables,” Journal of Marketing Research, 41 (4), pp.467-478.
Mazzeo, Michael J. (2002), “Product Choice and Oligopoly Market Structure,” RAND Journal of
Economics, 33 (2), pp.221-242.
McCardle, Kevin, Kumar Rajaram, Christopher S. Tang (2004), “Advance Booking Discount
Programs Under Retail Competition,” Management Science, 50 (5), pp.701-708.
McFadden, Daniel L. (1974), “Conditional Logit Analysis of Qualitative Choice Behavior,” in
P.Zarembka, ed., Frontiers in Econometrics, Academic Press, New York, pp.105-142.
McGill, Jeffrey I. and Garrett J. van Ryzin (1999), “Revenue Management: Research Overview
and Prospects,” Transportation Science, 33 (2), pp.233-256.
Moe, Wendy W. and Peter S. Fader (2002), “Using Advance Purchase Orders to Forecast New
Product Sales,” Marketing Science, 21 (3), pp.347-364.
____ and ____ (2009), “The Role of Price Tiers in Advance Purchasing of Event Tickets,”
Journal of Service Research, forthcoming.
Moore, Thomas Gale (1966), “The Demand for Broadway Theatre Tickets,” The Review of
Economics and Statistics, 48 (1), pp.79-87.
Moorthy, Sridhar and Kannan Srinivasan (1995), “Signaling Quality with a Money-Back
Guarantee: The Role of Transaction Costs,” Marketing Science, 14 (4), pp.442-466.
Ntzoufras, Ioannis (2009), Bayesian Modeling Using WinBUGS, John Wiley & Sons, Inc.,
Hoboken, New Jersey.
Okada, Erica Mina (2005), “Justification Effects on Consumer Choice of Hedonic and Utilitarian
Goods,” Journal of Marketing Research, 42 (1), pp.43-53.
Orhun, A. Yesim (2005), “Spatial Differentiation in the Supermarket Industry,” working paper.
Padmanabhan, V. and Ram C. Rao (1993), “Warranty Policy and Extended Service Contracts:
Theory and an Application to Automobiles,” Marketing Science, 12 (3), pp.230-247.
Prelec, Drazen and George Loewenstein (1998), “The Red and the Black: Mental Accounting of
Savings and Debt,” Marketing Science, 17 (1), pp.4-28.
Purohit, Devavrat and Joydeep Srivastava (2001), “Effect of Manufacturer Reputation, Retailer
Reputation, and Product Warranty on Consumer Judgments of Product Quality: A Cue
Diagnosticity Framework,” Journal of Consumer Psychology, 10 (3), pp.123-134.
176
Putler, Daniel S. and Shilpa Lele (2003), “An Easily Implemented Framework for Forecasting
Ticket Sales to Performing Arts Events,” Marketing Letters, 14 (4), pp.307-320.
Reddy, Srinivas K., Vanitha Swaminathan, and Carol M. Motley (1998), “Exploring the
Determinants of Broadway Show Success,” Journal of Marketing Research, 35 (3), pp.370-383.
Rossi, Peter E. and Greg M. Allenby (2003), “Bayesian Statistics and Marketing,” Marketing
Science, 22 (3), pp.304-328.
Seim, Katja (2006), “An Empirical Model for Firm Entry with Endogenous Product-Type
Choices,” RAND Journal of Economics, 37 (3), pp.619-640.
Shugan, Steven M. and Jinhong Xie (2000), “Advance Pricing of Services and Other
Implications of Separating Purchase and Consumption,” Journal of Service Research, 2 (3),
pp.227-239.
Simonson (1989), “Choice Based on Reasons: The Case of Attraction and Compromise Effects,”
Journal of Consumer Research, 16 (3), pp.158-174.
Soberman, David A. (2003), “Simultaneous Signaling and Screening with Warranties,” Journal
of Marketing Research, 40 (2), pp.176-192.
Spiegelhalter, David J., Nicola G. Best, Bradley P. Carlin, and Angelika van der Linde (2002),
“Bayesian Measures of Model Complexity and Fit,” Journal of the Royal Statistical Society:
Series B (Statistical Methodology) , 64 (4), pp.583-639.
Swami, Sanjeev, Jehoshua Eliashberg, and Charles B. Weinberg (1999), “SliverScreener: a
Modeling Approach to Movie Screens Management,” Marketing Science, 18 (3), pp.352-372.
Talluri, Kalyan and Garrett van Ryzin (2004), “Revenue Management Under a General Discrete
Choice Model of Consumer Behavior,” Management Science, 50 (1), pp.15-33.
Tang, Christopher S., Kumar Rajaram, Aydin Alptekinoglu, and Jihong Ou (2004), “The
Benefits of Advance Booking Discount Program: Model and Analysis,” Management Science,
50 (4), pp.465-478.
Thomadsen, Raphael (2007), “Product Positioning and Competition: The Role of Location in the
Fast Food Industry,” Marketing Science, 26 (6), pp.792-804.
Venkatesh R. and Vijay Mahajan (1993), “A Probabilistic Approach to Pricing a Bundle of
Products or Services,” Journal of Marketing Research, 30 (4), pp.494-508.
Vitorino, Maria Ana (2007), “Empirical Entry Games with complementarities: an Application to
the Shopping Center Industry,” working paper.
Watson, Randal (2005), “Entry and Location Choice in Eyewear Retailing,” working paper.
177
Weinberg, Charles B. and Kenneth M. Shachmut (1978), “ARTS PLAN: a Model Based System
for Use in Planning a Performing Arts Series,” Management Science, 24 (6), pp.654-664.
___ (1986), “ARTS PLAN: Implementation, Evolution, and Usage,” Marketing Science, 5 (2),
pp.143-158.
Weatherford, Lawrence R., Samuel E. Bodily, and Phillip E. Pfeifer (1993), “Modeling the
Customer Arrival Process and Comparing Decision Rules in Perishable Asset Revenue
Management Situations,” Transportation Science, 27 (3), pp.239-251.
Weiner , Richard (2000), “Of Bulldog Editions and Matters,” Public Relations Tactics, 7 (9),
pp.7.
Xie, Jinhong and Steven M. Shugan (2001), “Electronic Tickets, Smart Cards, and Online
Prepayments: When and How to Advance Sell,” Marketing Science, 20 (3), pp.219-243.
Yang, Sha and Greg M. Allenby (2003), “Modeling Interdependent Consumer Preferences,”
Journal of Marketing Research, 40 (3), pp.282-294.
Zhu, Ting and Vishal Singh (2009), “Spatial Competition with Endogenous Location Choices: an
Application to Discount Retailing,” Quantitative Marketing and Economics, 7 (1), pp.1-35.
___, ___, and Mark Manuszak (2009), “Market Structure and Competition in the Retail Discount
Industry,” Journal of Marketing Research, forthcoming.
__, __, and Anthony Dukes (2007), “Local Competition, Entry, and Agglomeration,” working
paper.
doc_855010833.pdf
Scheduling is the process of deciding how to commit resources between a variety of possible tasks. Time can be specified (scheduling a flight to leave at 8:00) or floating as part of a sequence of events.
ABSTRACT
Title of Document: EFFECTS OF PERFORMANCE SCHEDULES ON
EVENT TICKET SALES
Peggy Hui-Hsing Tseng, Ph.D. 2009
Directed By: Professor Wendy W. Moe, Department of Marketing
Event scheduling is one of many important decisions facing event marketers
in the entertainment industry (i.e., how should multiple performances be scheduled
across markets, across venues, and over time?). While there is ample research
examining the issues of costs and constraints associated with such a decision,
virtually no research exists to examine the impact of these decisions on consumer
demand. Hence, the objective of this dissertation is to examine how consumers
respond to event marketers’ scheduling decisions.
First, a scheduling effect may arise from performances within a market.
When performances are scheduled closely in distance or time, their similarity in
venue locations or performance dates may result in a stronger relationship and
influence ticket sales. This relationship may have a positive effect on ticket sales
because the similarity could signal the quality of an event and suggest the desirability
of these performances. Thus, these performances attract more consumers and sell
more tickets. However, the relationship could be negative. When performances are
close in distance or time, they become direct substitutes and compete for consumer
patronage.
Another effect arises from an event distribution across markets. When an
event travels from one market to another and each market has a different performance
schedule, the word of mouth of this event may accumulate and carry over to later
markets. If so, market sales may be a good proxy of word of mouth. How well (or
poorly) an event sells in preceding markets may affect ticket sales in following
markets.
This dissertation consists of three essays to examine the abovementioned
scheduling effects. We contact a national ticket seller to acquire a dataset containing
ticket sales of a family event traveling across 42 markets. The first essay analyzes a
performance schedule in one metropolitan market and investigates the scheduling
effect on ticket sales. The second essay employs all performance schedules in 42
markets to study heterogeneous market responses and propose explanatory factors.
Finally, the third essay incorporates the distribution sequence of this event and
examines whether ticket sales in preceding markets have a carryover effect to
influence ticket sales in later markets.
EFFECTS OF PERFORMANCE SCHEDULES ON
EVENT TICKET SALES
By
Peggy Hui-Hsing Tseng
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2009
Advisory Committee:
Professor Wendy W. Moe, Chair
Professor Roger R. Betancourt
Professor Yogesh Joshi
Professor P.K. Kannan
Professor Michel Wedel
© Copyright by
Peggy Hui-Hsing Tseng
2009
ii
Table of Contents
Table of Contents .......................................................................................................... ii
List of Tables .................................................................................................................v
List of Figures .............................................................................................................. vi
1 Introduction ............................................................................................................1
2 Essay 1: Scheduling to Sell: Examining the Impact of a Performance Schedule
on Event Ticket Sales .....................................................................................................8
2.1 Introduction .................................................................................................... 8
2.2 Literature Review and Conceptual Framework ........................................... 15
2.2.1 Airline and Movie Scheduling .............................................................. 15
2.2.2 Event Tickets ........................................................................................ 17
2.2.3 Impact of Scheduling on Ticket Sales .................................................. 21
2.2.4 Conceptual Framework ......................................................................... 25
2.3 Model Development ..................................................................................... 28
2.3.1 Model Overview ................................................................................... 28
2.3.2 Scheduling Characteristics .................................................................... 29
2.3.3 Number of Ticket Sales ........................................................................ 30
2.3.4 Timing of Ticket Sales .......................................................................... 34
2.3.5 Endogeneity in Performance Scheduling .............................................. 38
2.4 Data Description ........................................................................................... 39
2.4.1 Description of Events ............................................................................ 40
2.4.2 Description of Ticket Sales ................................................................... 41
2.4.3 Covariate Specifications ....................................................................... 44
2.5 Model Estimation and Benchmark Comparison .......................................... 46
2.5.1 Estimation ............................................................................................. 46
2.5.2 Benchmark Comparison........................................................................ 47
2.6 Results .......................................................................................................... 49
2.6.1 Number of Ticket Sales ........................................................................ 49
2.6.2 Timing of Ticket Sales .......................................................................... 51
2.6.3 Endogeneity in Performance Scheduling .............................................. 53
2.7 Policy Simulation ......................................................................................... 54
2.7.1 Scenario 1: Geographic Change ............................................................ 54
2.7.2 Scenario 2: Temporal Changes ............................................................. 55
2.8 Conclusions .................................................................................................. 56
2.8.1 Summary ............................................................................................... 56
2.8.2 Limitations and Next Steps ................................................................... 57
Table and Figures .................................................................................................... 58
3 Essay 2: Heterogeneous Market Responses to Performance Schedules and Their
Explanatory Factors .....................................................................................................78
3.1 Introduction .................................................................................................. 78
3.2 Preliminary Analysis .................................................................................... 82
3.3 Literature Review and Conceptual Framework ........................................... 84
iii
3.3.1 Literature Review.................................................................................. 84
3.3.2 Conceptual Framework ......................................................................... 85
3.4 Model Development ..................................................................................... 88
3.4.1 Model Overview ................................................................................... 88
3.4.2 Scheduling Characteristics and the Number of Ticket Sales ................ 88
3.4.3 Endogeneity in Performance Scheduling .............................................. 89
3.4.4 Use of the HB Approach for Response Heterogeneity ......................... 90
3.5 Data .............................................................................................................. 90
3.5.1 Definition of Markets ............................................................................ 91
3.5.2 Description of Performance Schedules across Markets ........................ 92
3.5.3 Covariate Specifications ....................................................................... 94
3.6 Model Estimation and Benchmark Comparison .......................................... 97
3.6.1 Estimation ............................................................................................. 97
3.6.2 Benchmark Comparison........................................................................ 98
3.7 Results .......................................................................................................... 99
3.7.1 Heterogeneous Market Responses to Performance Schedules ............. 99
3.7.2 Explanatory Factors for Market Heterogeneity .................................... 99
3.7.3 Endogenous Scheduling Decision....................................................... 101
3.8 Conclusions ................................................................................................ 101
3.8.1 Summary ............................................................................................. 101
3.8.2 Limitations and Next Steps ................................................................. 103
Tables and Figures ................................................................................................ 104
4 Essay 3: Sequential Distribution of a Live Performance Event .........................117
4.1 Introduction ................................................................................................ 117
4.2 Literature Review and Conceptual Framework ......................................... 121
4.2.1 Sequential Distribution ....................................................................... 121
4.2.2 Conceptual Framework ....................................................................... 125
4.3 Model Development ................................................................................... 128
4.3.1 Overview ............................................................................................. 128
4.3.2 Demand Equation: Market Sales ........................................................ 129
4.3.3 Supply Equations: Overall Supply, Venue Usage, and Day Usage .... 130
4.3.4 Variable Specifications ....................................................................... 131
4.3.5 Model Summary.................................................................................. 138
4.4 Data ............................................................................................................ 139
4.4.1 Touring Sequence ............................................................................... 140
4.4.2 Covariates in the Demand Model ....................................................... 141
4.4.3 Covariates in the Supply Model .......................................................... 144
4.5 Estimation and Results ............................................................................... 146
4.5.1 Model Estimation ................................................................................ 146
4.5.2 Results of Demand Equation .............................................................. 147
4.5.3 Results of Supply Equations ............................................................... 149
4.5.4 Correlated Demand and Supply .......................................................... 150
4.6 Conclusions ................................................................................................ 150
4.6.1 Summary ............................................................................................. 150
4.6.2 Conclusion .......................................................................................... 151
Tables and Figures ................................................................................................ 153
iv
5 Conclusion ..........................................................................................................165
5.1 Summary .................................................................................................... 165
5.2 General Discussions ................................................................................... 168
5.3 Contributions .............................................................................................. 170
5.4 Limitation and Future Research ................................................................. 170
Reference ...................................................................................................................172
v
List of Tables
Table 2-1: Description of Data Fields ......................................................................... 58
Table 2-2: Summary of Ticket Sales .......................................................................... 59
Table 2-3: Descriptive Statistics of Covariates ........................................................... 60
Table 2-4: Benchmark Models and Model Fit ............................................................ 61
Table 2-5: Results for the Number of Ticket Sales .................................................... 62
Table 2-6: Results for the Timing of Ticket Sales ...................................................... 63
Table 2-7: Results for Endogenous Performance Scheduling .................................... 64
Table 3-1: Summary of DMA markets ..................................................................... 104
Table 3-2: Summary of Market Responses to Performance Schedules .................... 105
Table 3-3: Venue Locations and their Associated DMAs ........................................ 106
Table 3-4: Summary of Ticket Sales ........................................................................ 107
Table 3-5: Descriptive Statistics of Covariates across Markets ................................ 108
Table 3-6: Descriptive Statistics of Market Characteristics ..................................... 109
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics ............. 110
Table 3-8: Sources of Heterogeneous Market Responses ......................................... 111
Table 3-9: Results of Performance Schedule Model ................................................ 112
Table 4-1: Descriptive Statistics of Covariates (in the log term) .............................. 153
Table 4-2: Correlation Coefficient of Demand Covariates ....................................... 154
Table 4-3: Correlation Coefficient of Dependent Variables ..................................... 155
Table 4-4: Results of the Demand Model ................................................................. 156
Table 4-5: Results of the Supply Model ................................................................... 157
Table 4-6: Correlation between Supply and Demand Models .................................. 158
vi
List of Figures
Figure 2-1: Conceptual Framework of Spatial Decomposition .................................. 65
Figure 2-2: Model Overview ...................................................................................... 66
Figure 2-3: Probability of Ticket Sales over Time ..................................................... 67
Figure 2-4: Sales Distribution by Performances ......................................................... 68
Figure 2-5: Sales Distribution by Days of Week ........................................................ 69
Figure 2-6: Weekly Sales Pattern of a Performance ................................................... 70
Figure 2-7: Heterogeneity in Sales Pattern Across Performances .............................. 71
Figure 2-8: Venue Locations and Driving Distances .................................................. 72
Figure 2-9: Summary of Geographic Distance ........................................................... 73
Figure 2-10: Summary of Temporal Distance ............................................................ 74
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales .................... 75
Figure 2-12: Effect of Schedule Changes on Ticket Sales ......................................... 76
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales ........................ 77
Figure 3-1: Conceptual Framework of Heterogeneous Market Responses .............. 113
Figure 3-2: Example of a DMA and its county information ..................................... 114
Figure 3-3: Locations of DMA Markets and Performing Sequence ......................... 115
Figure 3-4: Heterogeneous Market Responses ......................................................... 116
Figure 4-1: Conceptual Framework .......................................................................... 159
Figure 4-2: Illustration of Market Connectedness .................................................... 160
Figure 4-3: Overview of Model Development .......................................................... 161
Figure 4-4: Venue Locations .................................................................................... 162
Figure 4-5: Touring Dates across Markets ................................................................ 163
Figure 4-6: Capacity-Filled Rate across Markets ..................................................... 164
1
1 Introduction
Scheduling is one of many important decisions facing event marketers in the
entertainment industry. To maximize revenues in a national market, event marketers have to not
only make marketing mix decisions but also schedule performances across markets, across
venues, and over time. Hence, these scheduling decisions result in a performance schedule
within a market or across a national market for an event to perform in different venues and on
various days. In this dissertation, we define a performance schedule as a summary of
performances taking places in various venue locations and on different dates to examine
potential effects of this performance schedule on ticket sales at a performance level and across
markets.
Similar to airline and movie scheduling decisions, event marketers rely on costs and
constraints in their scheduling process. They need to comply with venue availability and seating
capacity to decide when to provide performances and how many performances to provide. They
can minimize the travel distance across markets if it is expensive to move from one market to
another. They can also shorten a performing period in a venue if the cost to rent a facility
outweighs the benefits of having performances for a long period.
While there is ample research examining the issues of costs and constraints associated
with such scheduling decisions, virtually no research exists to examine the impact of these
decisions on consumer demand. In other words, it is unclear to event marketers how consumers
in a market evaluate individual performances and how consumer inferences influence their
purchase decisions such as which performance to buy and when to buy. Hence, at the
2
performance level, consumers’ evaluations about performances in a schedule may affect the
number of ticket sales of these performances and the pattern of these ticket sales over time.
It is important for event marketers to understand potential drivers for the number of ticket
sales sold and the pattern of these ticket sales over time. Once they know the effects of these
drivers, they will be able to estimate expected market demand and schedule performances
accordingly. In this way, they can avoid revenue losses from undersupply and minimizes costs
resulting from potential oversupply. Additionally, event ticket sales are often available for
purchases several months or weeks in advance, and these tickets are not sold at a constant rate
throughout the advance-selling period. If event marketers understand the drivers for the sales
rate, sales pattern can be easily established to monitor actual ticket sales over time and
potentially adjust the marketing strategy as the event approaches.
As such, this dissertation aims to investigate the impact of performance schedules on
ticket sales for a live performance event. Specifically, we propose two types of effects emerging
from the scheduling decisions. One type of effect is likely to exist among performances within a
market because of the way these performances are scheduled across venues and dates. For
example, performances scheduled closely in distance or time might signal the quality of an event,
the desirability of these venues or dates, or just the potential substitutes across venues or dates.
If consumers make any of these inferences and perceive performances to be more or less
favorable, the scheduling effect among performances within a market would influence how well
each performance sells and when ticket sales occur.
The other type of effect might develop across markets where an event performs one after
another. In other words, after an event performs in each market for a period, its word of mouth
may arise and cumulate over time. If the word of mouth travels across markets and consumers
3
have positive feedbacks, how well this event does in previous markets could influence ticket
sales in later markets. Because ticket sales in previous markets can capture the volume of word
of mouth to some extent, it is plausible to observe ticket sales in previous markets to affect sales
in later markets.
This dissertation consists of three essays to examine those possible scheduling effects
mentioned above. For this purpose, we acquired a dataset containing ticket sales of a family
event traveling across 42 markets. We start by characterizing a performance schedule in a single
market and investigating its impact on the number of ticket sales at a performance level and the
pattern of these ticket sales over time (i.e., essay one). From essay one, we conclude that
performances scheduled close to each other in terms of distance can experience more ticket sales,
and these tickets sell at a faster rate. On the other hand, performances scheduled close to each
other in terms of time tend to sell fewer tickets, but do not exhibit any significant changes in
their sale patterns. However, these results are established for one metropolitan market in our
dataset. Hence, we expand the level of analysis to all 42 markets in essay two to ensure
scheduling effects generalizable for this event and examine the heterogeneity across markets.
Essay two confirms consistent scheduling effects across markets and identifies explanatory
factors for the heterogeneity in effect sizes across markets. Finally, essay three proceeds to
investigate scheduling effects at a market level and examines whether an event performing
across markets affects ticket sales in these markets. Specifically, essay three reveals that markets
to which an event travels sell more tickets when event marketers disperse performance dates or
employ multiple venues in these markets. Additionally, these markets do not influence one
another on their ticket sales, but their venues within the same market have such an effect.
Although one may argue that the third essay does not have to be conducted after the first two
4
essays, we choose this sequence because we need to investigate the effect of a touring event
across markets after we can understand and control for the effect of a performance schedule
within a market.
The purpose of the first essay is to lay the foundation of this dissertation and to test
whether scheduling decisions influence ticket sales of performances within a market. We begin
by examining ticket sales at a performance level within a single market and differentiate between
performances based on a set of scheduling characteristics. We derive a set of scheduling
characteristics based on the venue locations and performance dates. According to these
scheduling characteristics, we measure how closely performances are scheduled in terms of
distance or time. A performance of a shorter geographic or temporal distance hence indicates its
similarity in venue location or performance date to other performances.
After examining ticket sales of these performances as a function of their distance
measures, we find that geographic and temporal distances between performances have different
effects on ticket sales. More specifically, a shorter geographic distance between performances
leads to more tickets sold and a faster sales rate for a performance. However, a shorter temporal
distance between performances just causes decreases in the number of ticket sales but does not
influence how fast tickets are sold.
Although it is not clear why consumers process geographic and temporal distances
differently, it is clear that they refer to the way performances are scheduled as a means to make
inferences about these performances. Their inferences about performances of a shorter
geographic distance could be regarding the quality of an event, the desirability of their associated
venues, or others. As a result, these performances sell more tickets and attract consumers to
purchase tickets early. On the other hand, their inferences about performances of a shorter
5
temporal distance could be the high substitutability within a shorter period. Therefore, these
performances compete for consumer patronage and suffer from sales cannibalization.
While the first essay demonstrates the significant effects of geographic and temporal
distances on ticket sales, the results are for one market only. Yet, event marketers often need to
make scheduling decisions for more than one market. It is not clear whether the results in this
first essay are consistent across markets. For this reason, the objective of the second essay is to
use all performance schedules of the same event to investigate heterogeneous market responses
and identify explanatory factors.
To achieve this objective, we follow the same approach as in essay one and analyze all
performance schedules. For performances within their associated market, we characterize them
by their venue locations and performance dates and compute their geographic and temporal
distances to other performances. Then, for each market, we model ticket sales of performances
as a function of their distance measures to understand whether the heterogeneity exists in market
responses. We then model these market specific parameters as a function of their market
characteristics such as market population and additional scheduling characteristics such as travel
sequence along the distribution to explain any differences across markets.
Our results show that market responses to performance schedules are heterogeneous and
can be explained by market and additional scheduling characteristics. Specifically, when a
market has a bigger population, the effects of days of week and baseline attractiveness are
attenuated. Moreover, after an event travels to more markets that are geographically adjacent to
a focal market, the focal market is less responsive to its baseline attractiveness and temporal
schedule. Finally, a current market in a late distribution sequence tends to respond more
favorably to a Sunday performance.
6
Results in essay two hence suggest a possibility that markets where an event travels one
after another may be dependent. In other words, how following markets respond to their
temporal schedule depends on how many geographically adjacent markets an event has visited.
Another possibility is that there may be another means of integrating participating markets along
a touring event such that ticket sales of preceding markets might directly influence ticket sales of
following markets (rather than through the response parameters and geographically adjacent
markets). If so, it is important to incorporate the temporal sequence in an event distribution and
study the impact of preceding markets on following ones.
The primary objective of essay three, therefore, is to examine the impact of a sequentially
distributed event across markets. Additionally, we consider the endogeneity between supply and
demand for an event in case expected market demand influences event marketers’ scheduling
decisions and their schedule further affect ticket sales in a market.
To achieve this objective, we model the supply and demand for an event simultaneously.
We model ticket sales of each market as a function of its performance schedule and the
sequential distribution of this event. In addition to modeling the supply and demand
simultaneously to account for the endogeneity, we use three variables to present the scheduling
influences on market demand and employ the spatially weighted approach to incorporate
different release timing and ticket sales of preceding markets in an event tour.
Our results show that a market experiences more ticket sales when event marketers
disperse performance dates or book multiple venues in this market. Moreover, we show that the
sequentially distributed event has an effect on ticket sales. However, this effect is significant
across venues of the same market but not across markets. When an event performs in more than
7
one venue, its ticket sales in a preceding venue carry over to a later venue and influence its
overall market sales.
The organization of this dissertation is as follows. Chapter 1 introduces the issues of
scheduling facing event marketers and presents a general overview of each essay. Chapters 2, 3,
and 4 discuss the three essays, respectively, in depth. Finally, Chapter 5 provides a brief
summary of each essay, integrates essential results, and points out limitations and future
directions to conclude this dissertation.
8
2 Essay 1: Scheduling to Sell: Examining the Impact of a Performance
Schedule on Event Ticket Sales
2.1 Introduction
Scheduling is one of many important decisions facing event marketers in the
entertainment industry. Besides decisions regarding marketing activities (e.g., pricing and
promotions), event marketers have to schedule performances across markets, across venues, and
over time. Typically, they rely on costs and constraints in their scheduling process (Etschmaier
and Mathaisel 1985; Lohatepanont and Barnhart 2004; Eliashberg et al 2007) and use pricing,
advertising, and days of week to describe how well an event can sell (Weinberg and Shachmut
1978; Putler and Lele 2003; Leslie 2004) or when ticket sales occur (Moe and Fader 2009).
While there is ample research examining the issues of costs and constraints associated
with such scheduling decisions, virtually no research exists to examine the impact of these
decisions on consumer demand. In other words, it is unclear to event marketers how consumers
in a market may evaluate a performance schedule of an event and how consumer inferences
influence their purchase decisions such as which performance to buy and when to buy. Hence, at
the performance level, consumers’ evaluations about performances in a schedule may affect the
number of ticket sales of these performances and the pattern of these ticket sales over time.
From event marketers’ perspective, they need to understand potential drivers for ticket
sales in terms of the number of ticket sales and the pattern of these ticket sales over time. Once
they know how these drivers influence market demand, they can estimate expected market
demand accordingly and schedule performances to meet this market demand. In this way, they
9
can avoid revenue losses from undersupply and prevent decreases in profitability due to
oversupply.
On the other hand, event ticket sales are often available for purchases several months or
weeks in advance, and these ticket sales do not occur at a constant rate throughout an entire
advance-selling period. If event marketers understand drivers for tickets selling at a different
rate, they can portray the sales pattern as a benchmark and monitor actual ticket sales over time.
Consequently, once event marketers find an actual pattern deviating from the benchmark, they
can take actions in time.
In general, scheduling decisions affect the maximum number of tickets that an event can
sell. When event marketers schedule a live performance event, they often allocate multiple
performances across markets, across venues, and over time. Although the number of
performances and the capacities of chosen venues constrain the maximum number of tickets an
event can sell, empirical evidence shows that it is rare for the demand to exceed supply in this
industry. Therefore, one possible impact of scheduling decisions is to constrain the maximum
possible of ticket sales for an event although the supply is usually well beyond the actual demand.
Scheduling decisions might also influence consumer responses in a market. In other
words, when consumers realize performances are scheduled in various venue locations and on
different dates, they may try to rationalize why event marketers schedule performances in this
way and then make inferences about these performances. If so, consumers could formulate
different preferences for these performances to choose one performance to attend and purchase
tickets at their desired time. At a performance level, consumer responses influence how well
individual performances sell and when ticket sales of these performances occur.
10
Therefore, the objective of this essay examines the effects of scheduling characteristics of
performances on ticket sales. Specifically, we use venue locations and performance dates as the
scheduling characteristics of each performance, and we investigate whether performances
scheduled closely in distance or time experience a different number and timing of ticket sales.
We define the timing of ticket sales as tickets sold at different times in an advance-selling
period, and earlier or later timing of ticket sales suggests ticket sales occurred in the earlier or
later advance-selling period. In other words, if scheduling characteristics have an effect,
performances scheduled close in distance or time would experience more (or fewer) ticket sales.
Their ticket sales would occur earlier (or later) than those scheduled farther apart in an advance-
selling period, ceteris paribus.
One possible effect is to see performances scheduled close in distance or time
experiencing more ticket sales and earlier timing of sales than those scheduled farther apart. For
example, if event marketers want to signal the desirability of some venue locations or
performance dates, they could allocate more performances to those specific venues or dates. In
this way, consumers would perceive performances scheduled close in distance or time to be more
attractive (due to the similar scheduling characteristics) and assign higher utilities to these
performances. As a result, these performances could sell more tickets and experience earlier
timing of sales than other performances.
Another possible effect is to observe performances scheduled farther apart in distance or
time experiencing more ticket sales and earlier timing of sales than those scheduled nearby. In
other words, when consumers have higher uncertainty about whether they can attend an event in
a particular venue at a specific time, event marketers could sparsely allocate performances across
venues and dates. In this way, consumers have more alternatives and higher flexibility regarding
11
when and where to attend. Hence, the chance for them to attend this event increases, and
performances scheduled farther apart will be able to sell more tickets and experience earlier
timing of sales. In contrast, performances scheduled nearby merely substitute one another within
certain venues or dates. Consumers do not have to decide which performance they want to buy
and can delay their purchase timing. Hence, these closely scheduled performances compete
against consumer patronage and cannibalize ticket sales. To sum up, scheduling characteristics
might have two possible but contradictory effects on ticket sales. We allow both possibilities
and examine the effects of scheduling characteristics empirically.
Our modeling objective is to measure the scheduling characteristics of each performance
and study the impact of these scheduling characteristics on the number and timing of ticket sales.
We consider the possibility that consumers evaluate venue locations and performance dates
differently. Thus, we differentiate the effect of scheduling across venues from the effect of
scheduling across dates and then investigate these separately effects. Specifically, we refer to the
previous effect as the effect of geographic scheduling or the effect of a geographic schedule and
the later effect as the effect of temporal scheduling or the effect of a temporal schedule.
First, to measure the scheduling characteristics of performances, we characterize each
performance in a schedule of an event by its venue location and performance date. Then, we
compute the geographic and temporal distances between performances to understand how
closely (or distantly) performances are scheduled across venues and dates. For example,
performances of a shorter geographic or temporal distance to others are relatively closer to other
performances than performances of a longer geographic or temporal distance.
Second, to examine the number of ticket sales across performances, we consider the
possibility that some consumers might evaluate the venue locations and performance dates in a
12
schedule but eventually do not attend. To account for consumers who make purchases and those
who miss out the opportunity to attend, we specify the number of ticket sales in the form of sales
share within a potential target market. Then, we examine the share of each performance and the
non-buyer segment within this potential market. We extend the competing destination model
proposed by Fotheringham (1988) and model the share of each performance and the non-buyer
segment as a function of the geographic and temporal distances between performances. By
doing so, we can understand whether scheduling characteristics influence ticket sales at a
performance level. In addition, using sales shares of individual performances and the population
size in a target market as the number of potential buyers, we can obtain the number of ticket
sales at a performance level. We can also obtain a market penetration rate by taking the sum of
sales shares across performances.
Third, to examine the timing of ticket sales across performances, we first consider a
general pattern of ticket sales for a performance. That is, a performance sells fewer tickets in the
beginning of its advance-selling period and obtains more sales over time with the most arriving
in the later period or the week of the performance. Although this is a general pattern over time,
each performance still has a different sales rate. Some performances experience ticket sales
occurred early (i.e., earlier timing of ticket sales) but others experience ticket sales arrived later
(i.e., later timing of ticket sales). To account for variations in sales rate across performances, we
employ a Weibull hazard model to capture the timing of sales over time for individual
performances. We further model the sales rate of each performance as a function of its
geographic and temporal distances to other performances to understand whether these scheduling
characteristics explain the heterogeneity in sales rate.
13
Finally, we consider a possible endogeneity in scheduling decisions. Since the live
performance industry has been established and event marketers have scheduled for a variety of
events, event marketers may have incorporated their experiences into a scheduling process. If so,
a performance schedule is endogenously determined. For example, event marketers might have
scheduled more performances on weekends across all venues because they know these
performances have higher performance attractiveness. It is also likely that they have scheduled
performances based on the responses they expect in this market. Specifically, they may have
scheduled performances closely in distance or time or farther apart, because they know
consumers prefer performances of such scheduling characteristics. In case such an endogeneity
exists in the scheduling process, we control for this possibility in our model development.
We contact a national ticket seller to obtain a dataset of a live performance event and use
its ticket sales to examine the impact of its scheduling characteristics on ticket sales. Although
this event performed across several markets, we use the performances in the New York
metropolitan market as a subset. The reason is that performances scheduled in this market have
richer variations in venue locations and performance dates. In short, this event had 70
performances across four venues in the New York metropolitan market and performed between
March and June 2004.
Our results indicate that the effect of geographic scheduling differs from the effect of
temporal scheduling on the number and timing of ticket sales. Performances scheduled closely
across venues not only sell more tickets but also sell tickets at a faster rate. In contrast,
performances sparsely scheduled across dates sell more tickets but do not have an impact on the
timing of sales.
14
Our explanation for the observed effect of geographic scheduling is that event marketers
may attempt to signal the desirability of some venues by scheduling more performances in those
venues. Although the desirability of these venues may also be owing to population around these
venues and consumers’ shorter travel distance to these venues, we control for this possibility in
our benchmark models and still find a significant effect of geographic scheduling. Therefore, in
contrast with performances scheduled in distant venues, performances scheduled in nearby
venues signal higher desirability or popularity to consumers. As a result, more consumers are
interested in these performances and are more willing to purchase tickets earlier.
On the other hand, our explanation for the observed effect of temporal scheduling is that
event marketers try to accommodate consumers’ uncertain preferences to performance dates by
scheduling performances sparsely across dates. In this way, performances on dispersed dates (or
of a longer temporal distance) provide consumers higher flexibility and further sell more tickets.
In contrast, performances within a short time span (or of a shorter temporal distance) substitute
one another and cannibalize ticket sales.
In addition to the impact of scheduling characteristics on ticket sales, our results also
indicate that there exists some endogeneity in the scheduling process. Event marketers consider
how attractive performances are when they schedule performances across dates. However, our
results show that event marketers have not yet incorporated the effects of geographic and
temporal scheduling when they allocate performances across venues and dates.
The organization of this essay is as follows. We first review past works relevant to
scheduling and event ticket sales. Then, we present our conceptual framework and model
development. After a detailed data description, we provide our results and discuss possible
rationale behind the scheduling effects. We also conduct two policy simulations to show how re-
15
allocating performances to a different venue or date would result different number and timing of
ticket sales. Finally, we conclude this essay with limitations and next steps.
2.2 Literature Review and Conceptual Framework
2.2.1 Airline and Movie Scheduling
Because scheduling a live performance event is an important yet understudied research
stream, we search literature in other contexts where scheduling is also critical to managers. We
find airline and movie scheduling literature a good fit because managers have a common
objective to schedule a series of flights, screens, or performances to meet the market demand.
Therefore, we discuss studies in airline and movie scheduling in turn.
Airline Scheduling
Airline scheduling is a complex system. It involves demand estimation, pricing for
different segments, flight scheduling for various routes, fleet assignments for individual flights,
crew scheduling, aircraft rotation, flight gate assignments, and many other decisions (Etschmaier
and Mathaisel 1985; Dobson and Lederer 1993; Jarrah et al 2000; Lohatepanont and Barnhart
2004; Dorndorf et al 2007). Therefore, any small changes require a series of adjustments in the
entire system.
Traditionally, airline scheduling has been a constrained-optimization decision.
Researchers use historical data to estimate demand and consider the expected demand to
construct flight schedules (Dobson and Lederer 1993; Lohatepanont and Barnhart 2004). After
schedule constructions, other departments evaluate proposed schedules to set airfares, assign
aircrafts, and make other operational related decisions (Etschmaier and Mathaisel 1985; Dobson
and Lederer 1993; Jarrah et al 2000; Dorndorf et al 2007). Finally, they examine associated
16
profits and revise flight schedules to ensure profit maximization. Therefore, in this iterative
decision process, scheduling is primarily constraint-driven. An airline managers’ objective is to
maximize profitability while minimizing operational costs within feasible boundaries.
The similarity between airline scheduling and performance scheduling is the common
scheduling nature. Airline managers and event marketers have to decide when and where flights
or performances have to take place. Their conceptual objective is the same because they aim to
launch a schedule to extract the most demand in a given market. However, the demand in the
airline schedule is either treated as exogenous (McGill and van Ryzin 1999), based on historical
data (Etschmaier and Mathaisel 1985), or dependent on price and departure and arrival times
(Dobson and Lederer 1993). Researchers assume that dropping flights always leads to losses in
revenues (Lohatepanont and Barnhart 2004) and have not yet investigated how scheduling
density (e.g., frequencies of flights) influences ticket demand.
Movie Scheduling
In the movie industry, movie scheduling takes place after movie distributors release
movies to exhibitors. The primary task that exhibitors perform is to allocate a number of screens
within a theater to meet the local demand for this movie (Swami, Eliashberg, and Weinberg 1999;
Eliashberg et al 2009). Compared with advertising effect on box office revenues, movie
scheduling is a relatively new research direction in this industry (Eliashberg, Elberse, and
Leenderss 2006).
When it comes to the implementation of movie scheduling, exhibitors have to select
movies that might contribute higher revenues because they have a limited number of screens in a
theater. They usually start with estimating demand for a variety of movies and then select a
smaller set of movies to play (Swami et al 1999; Eliashberg et al 2007; Eliashberg et al 2009).
17
After they choose movies of interest, they allocate available screens to these movies (Elberse and
Eliashberg 2003). Although the general principle is that the longer exhibitors play a movie, the
more revenues they get, the decreasing demand over time and the contract with distributors also
determine how long a movie shows in a theater (Swami et al 1999). Finally, exhibitors refer to
box office revenues in a previous week, movie genres, days of week, and times of day to revise
their scheduling decision on a weekly basis (Elberse and Eliashberg 2003; Eliashberg et al 2007).
Movie scheduling is similar to airline scheduling in the sense that both contexts heavily
rely on the operational constraints, and profitability maximization is the top priority. It is also
similar to performance scheduling because within these contexts the purpose is to serve local
demand by offering movies or performances at various days of week or times of day. However,
researchers in movie scheduling usually assumes demand to be exogenous (Swami et al 1999;
Eliashberg et al 2007; Eliashberg et al 2009) or influenced by marketing activities (Elberse and
Eliashberg 2003). They have not yet considered the competition between theaters (Eliashberg et
al 2006) or the impact of scheduling density on box office revenues.
To sum up, the focus of airline and movie scheduling is constraint optimization and profit
maximization. Demand is often assumed exogenous and influences scheduling decisions.
Whether these scheduling decisions influence demand deserves further investigation. Therefore,
the objective of this essay is to examine the impact of scheduling on ticket sales in the context of
a live performance event. Next, we review current research in event tickets to understand
existing drivers for ticket sales before we discuss how scheduling could affect ticket sales.
2.2.2 Event Tickets
Live performance events typically refer to concerts, musicals, or circus acts, etc. that
perform live in front of an audience. Because an event often provides multiple performances
18
across venues and dates, understanding how well each performance can sell is an important issue
for event marketers and researchers. Hence, we review relevant works and discuss factors that
influence the number and timing of ticket sales.
Number of Ticket Sales
Identifying drivers of ticket sales has been a common topic studied in marketing,
economics, and performing arts literature. Researchers have used product related drivers,
consumer characteristics, and seasonality to explain consumer attendance (Weinberg and
Shachmut 1978; Currim, Weinberg, and Wittink 1981; Venkatesh and Mahajan 1993; Reddy,
Swaminathan, and Motley 1998; Putler and Lele 2003; Leslie 2004). According to the number
of events involved in the drivers, we further classify product-related drivers into (1) assortment-
related drivers, (2) event-related drivers, and (3) performance-related drivers. We discuss these
drivers and their importance in turn.
Assortment-related drivers often refer to factors related with a bundle of events. In other
words, this type of drivers exists when multiple events are grouped together in a product offering.
Researchers have shown that different bundle size, event types in a bundle, and associated
seating benefits can attract varying degrees of demand because consumers have heterogeneous
preferences to genres (e.g., operas, musicals), language types (e.g., Italian, English), and
willingness to pay (Currim, Weinberg, and Wittink 1981; Havlena and Holak 1988; Venkatesh
and Mahajan 1993). On the other hand, event-specific drivers refer to information of a specific
event (e.g., pricing or critics’ reviews for a Broadway show). While genre, pricing, and
advertising are important for all types of events (Weinberg and Shachmut 1978; Weinberg 1986;
Reddy et al 1998; Corning and Levy 2002; Putler and Lele 2003; Leslie 2004), previews and
critics’ reviews are more common for Broadway shows or theatrical events (Reddy et al 1998;
19
Corning and Levy 2002). Because a venue manager’s objective is to increase ticket sales within
a venue, they usually achieve their objective by bundling various events in a subscription
package or scheduling a variety of events to attract consumer attendance. Therefore, the first
two types of drivers are important from a venue manager’s perspective.
Performance-specific drivers, on the other hand, do not limit the number of events
needed but focus on the lower level of characteristics such as days of week or times of day
(Weinberg and Shachmut 1978; Corning and Levy 2002; Putler and Lele 2003; Leslie 2004). In
contrast with the first two types of drivers, this type of drivers is important to event marketers,
especially when they promote a single event that tours across venues and dates. In other words,
when an event has multiple performances across venues and dates, the assortment-related drivers
are not applicable to a single event. The event-related drivers are important yet pricing and
advertising are often planned at a market level and result in a constant effect across all
performances. Consequently, event marketers can only rely on days of week and times of day as
descriptive drivers to differentiate ticket sales of multiple performances.
The fourth type of driver is consumer characteristics. They often refer to consumers’
income levels, willingness to pay, driving distances to venues, and tastes for genres (Moore 1966;
Currim et al 1981; Venkatesh and Mahajan 1993; Leslie 2004). Although this type of drivers
allow event marketers and venue managers to understand consumer preferences better, it is
relatively difficult for event marketers to identify their potential consumers, especially when an
event travels to a new market and there is no previous consumer information available to event
marketers. The last type of driver is seasonality. It generally refers to the season (i.e., spring,
summer, fall, and winter) that events or performances takes place and is commonly used as a
control variable (Weinberg and Shachmut 1978; Weinberg 1986; Corning and Levy 2002).
20
As such, although there are five types of descriptive drivers for event tickets, there are
not plenty of drivers useful for event marketers to understand variations in ticket sales at a
performance level. Therefore, it is important for researchers to investigate additional drivers to
explain such a variation.
Timing of Ticket Sales
To attend an event, consumers have to purchase tickets no later than its performance date.
Throughout an advance-selling period, their purchase timing may range from very early (i.e.,
advance purchase) to the last minute (i.e., spot purchase). Historically, there are few studies in
event ticket purchases. However, there are many in other contexts such as airline ticket
purchases. Thus, we refer to studies in other contexts to discuss firms’ motivation to advance
sell and consumers’ motivation to advance purchase tickets for an event.
Generally, advance selling is common in the service-related industry or in a long lead-
time manufacturing industry. Although it is not necessary to charge lower prices in an advance-
selling period (Xie and Shugan 2001), firms still tend to use a two-stage pricing (i.e., charge
discounted rate in the early market but a regular rate in the spot market) as the primary tool to
attract consumers’ early purchases and secure some demand well in advance (Desiraju and Xie
1999; Shugan and Xie 2000; Cachon 2004; McCardle, Rajaram, and Tang 2004; Tang et al
2004). Two good examples of advance selling are that, first, a venue manager offers a bundle of
events at a lower rate to attract early arrivals of subscription ticket sales (Currim, Weinberg, and
Wittink 1981; Havlena and Holak 1988; Venkatesh and Mahajan 1993) and, second, an airline
company charges a cheaper airfare to attract leisure travelers’ advance purchases (Weatherford et
al 1993; Gallego and van Ryzin 1994; Talluri and van Ryzin 2004).
21
Regarding consumers’ motivation to advance purchase, extant literature has suggested
various reasons for consumers’ early versus spot purchases. In addition to reservation prices,
consumers’ uncertainty toward the consumption state also determines their purchase timing
decisions (Desiraju and Shugan 1999; Shugan and Xie 2000). Specifically, they tend to
procrastinate when they have higher uncertainty about whether they can attend an event in the
future. In contrast, they tend to purchase early when they are more certain to attend in the future.
Other psychological drivers include consumption utility and personal characteristics. For
example, consumers may want to savor their vacation experience better by paying earlier (Prelec
and Loewenstein 1998). Their tendency of being an innovator in their group versus being a
follower also affects their purchase timing decision. Innovators tend to purchase earlier and
influence followers in the later period (Moe and Fader 2002).
Although pricing has been a major factor to affect consumers’ purchase timing, a recent
study by Moe and Fader (2009) illustrated a need to re-evaluate the impact of pricing on the
timing of ticket sales. Specifically, they examined the timing of ticket sales across different
price tiers in the context of event tickets and found that consumers who purchase in advance are
not affected by the price discounts or face values of tickets. This result is very different from
airline tickets. Perhaps it is because there are no so-called “business” or “leisure” buyers in the
context of event tickets. Hence, the reason for consumers to advance purchase event tickets is
not clear and deserves further investigation.
2.2.3 Impact of Scheduling on Ticket Sales
The objective of this essay is to examine the impact of scheduling on ticket sales of
performances of the same event. Because performances are scheduled with different frequencies
across venues and dates, the similarity in these scheduling characteristics varies across
22
performances. We refer to literature in context effect, signaling, retail location, and distribution
services for possible effects of these scheduling characteristics.
Context Effect
Context effect has been extensively studied by behavioral literature (Huber, Payne, and
Puto 1982; Huber and Puto 1983; Simonson 1989) where researchers investigate how
alternatives of dominated or dominating attributes influence consumer choices. In general, this
stream of literature proposes violation of proportionality (Luce 1959). Researchers examine
situations when products of similar attributes are more attractive (i.e., attraction effect) and when
they are substitutable (i.e., substitution effect) and where alternatives in the middle level of
attributes are more favorable (i.e., compromise effect) (Huber et al 1982; Huber and Puto 1983;
Simonson 1989). In other words, when consumers have uncertain preferences to product
attributes, they would choose alternatives based on various reasons (Simonson 1989). A
dominating alternative may have a higher choice share under the attraction effect although it may
have a lower share under the substitution effect (Huber at al 1982; Huber and Puto 1983).
However, it is also likely that consumers prefer the new added alternative that has compromised
attributes under the compromise effect.
The relevance between the context effect and the scheduling effect is that similarity
between alternatives (i.e., performances in this case) could influence consumer perception and
choice decisions. As consumers pay more attention to alternatives that share a similar attribute
level (Huber et al 1982), they might focus on evaluating performances in the same or near
venues (or on the same or near dates) and find these performances more attractive or substitutive.
At the aggregate level, the similarity in these scheduling characteristics could further determine
how well these performances sell. On the other hand, performances of compromised attributes
23
(e.g., in a preferred venue but on a less preferred date or vice versa) might also sell differently
from other dominating or dominated performances.
Signaling
Signaling has been studied in marketing to address the issue of product quality. The
assumption for signaling to work is under the separating equilibrium (Chu and Chu 1994) that
manufacturers are credible and have high transaction costs to signal (Moorthy and Srinivasan
1995). Therefore, credible manufacturers can use money-back guarantees (Moorthy and
Srinivasan 1995) or an extended warranty (Padmanabhan and Rao 1993; Lutz and Padmanabhan
1995; Soberman 2003) to signal the quality of their product. They can also sell their products in
reputable retailers for consumers to infer the reputation of manufacturers, especially when their
reputation is not directly observable to consumers (Chu and Chu 1994; Purohit and Srivastava
2001).
We relate signaling with performance scheduling because it is likely that event marketers
want to signal the performing quality of an event to a market. Because sending a false signal can
be expensive (e.g., incremental costs for multiple performances and revenue losses from empty
seats), we assume event marketers are credible. They could increase the number of
performances in a market to signal event popularity. They could also allocate these
performances densely in some venues or on particular dates to suggest popularity or desirability
of these venues or dates. As a result, depending on which signal event marketers want to send,
they will schedule performances differently. After consumers receive the signal and believe the
credibility, they could adjust their preferences and make purchase decisions accordingly.
24
Retail Location
In the retail industry, store locations influence consumers’ shopping destination and store
choices (Fotheringham 1988). Therefore, collocating or keeping some distance away from
primary competitors has been an important issue in retail location literature (Mazzeo 2002;
Vitorino 2007; Zhu and Singh 2009). Although some studies show that shorter geographic
distance between retailers leads to direct competition and decreases in store profitability (Watson
2005; Orhun 2005; Seim 2006; Thomadsen 2007; Zhu and Singh 2009; Zhu, Singh, and
Manuszak 2009), other studies indicate benefits for retailers to locate closely (Mazzeo 2002;
Vitorino 2007; Zhu, Singh, and Dukes 2007). For example, retailers collocating in a shopping
center provide consumers a larger product assortment (across stores) such that these retailers can
attract more consumers to the shopping center (Vitorino 2007; Zhu et al 2007). Motels, on the
other hand, collocate in a highway exit to facilitate consumer search and attract more traffic
(Mazzeo 2002).
The relevance between retail location and performance scheduling is that event marketers,
like retailers, have to consider how far in distance or time they should keep performances apart
to avoid substitution within the same event (i.e., cannibalization). On the other hand, event
marketers might also consider scheduling performances closely enough to provide more
alternatives or signal the popularity to attract more demand. Therefore, if the substitution
between performances outweighs the collocating synergy, performances scheduled closely could
suffer fewer ticket sales and slow sales arrival (because consumers can always wait until the last
minute). Otherwise, densely scheduled performances would experience more ticket sales and
faster sales arrival.
25
Distribution Services
In addition to abovementioned spatial differentiation in retail location, retailers also try to
differentiate themselves from other competitors by improving their distribution services
(Betancourt 2004). Distribution services generally refer to the extent of services that retailers are
able to deliver to consumers. These services include product assortment, assurance of product
delivery at consumers’ desired time or in the desired form, spatial accessibility between stores
and consumer residences, and so on (Betancourt 2004). For example, retailers can provide a
larger assortment of products, extend business hours, or open more store locations to enhance
their distribution services.
In the live performance industry, similarly, both venue managers and event marketers
may aim to enhance their distribution services. To accommodate consumers who have different
tastes, venue managers may improve their assortment by scheduling a variety of events in their
venues. On the other hand, event marketers may consider consumers who have state uncertainty
about to which venue they want to go and which date they can attend. In this case, they might
schedule performances in multiple venues and disperse performance dates to allow consumers to
attend at their own convenience. In other words, if event marketers schedule performances
sparsely across venues and dates, these performances could accommodate more consumers at
different times and in different venues. In this way, sparsely scheduled performances will sell
more than densely scheduled performances.
2.2.4 Conceptual Framework
Although there may be several reasons to explain event marketers’ scheduling decisions
and consumers’ decision process, this information is unfortunately unobservable to researchers.
Hence, we can only use the abovementioned literature to speculate potential scheduling effects.
26
To sum up, if event marketers want to signal quality, they will increase the number of
performances and schedule these performances closely in distance or time to attract more
consumers. Then, consumers will perceive these performances to be more popular.
Consequently, more consumers will attend these performances and these consumers will be more
likely to purchase tickets in advance. On the other hand, if event marketers schedule
performances to enhance their distribution services, they will schedule performances across
multiple venues and disperse performance dates farther apart. In this way, performances
scheduled farther apart are able to accommodate more consumers and sell more tickets. Yet, it is
still possible for consumers to procrastinate given their uncertainty for the consumption state. As
such, based on different streams of literature, we can anticipate different scheduling mechanisms
and consumer responses.
However, if event marketers do not have a specific scheduling strategy but schedule
performances to comply with operational constraints such as venue availability and seating
capacity, their scheduling process is similar to airline and movie scheduling. Then, the impact of
scheduling could be positive or negative. According to the retail location literature, event
marketers may expect performances in close distance or time to have a substitutive relationship
and cannibalize ticket sales. They may also expect a collocation synergy between performances
to attract more demand. From consumers’ perspective, they may make their own inferences
about these performances based on different contexts. According to the attraction effect, they
may perceive performances of similar scheduling characteristics to be more attractive. Hence,
these performances will sell more tickets and these ticket sales will arrive earlier. In contrast,
consumers may perceive these performances to be highly substitutable (i.e., substitution effect)
27
and delay their purchase timing. If so, these performances will sell less and sell more slowly
than do those of dissimilar scheduling characteristics.
However, event marketers may not just simultaneously schedule performances densely
(or sparsely) across venues and dates. They may sometimes schedule performances densely in
certain venues to signal venue popularity yet disperse performance dates to accommodate
consumers’ uncertainty of attendance timing. Similarly, they may schedule performances
densely on certain dates as popular leisure activities but allocate these performances in several
distant venues to increase spatial accessibility. Consequently, if event marketers have separate
objectives and schedule accordingly, consumers will find performances scheduled closely in
distance (or time) but distantly in time (or distance). In this way, performances have different
scheduling characteristics across venue locations and performance dates, and consumers will
evaluate a geographic and temporal schedule separately. Hence, the effect of geographic
scheduling may differ from the effect of temporal scheduling. We allow this possibility and
empirically test these scheduling effects.
On the other hand, to study the impact of performance schedules on ticket sales, we also
have to control for the attractiveness of performances on different days of week and the potential
endogeneity between scheduling decisions and expected market response. In other words, if
event marketers have some knowledge about how a market responds to a performance schedule
and then use this knowledge to schedule performances, the performance schedule will be
endogenously set and the scheduling impact will be biased. In case the endogeneity exists, we
propose to examine the effect of performance schedule on ticket sales and control for the
endogeneity simultaneously. Figure 2-1 below summarizes our conceptual framework.
Figure 2-1: Conceptual Framework
28
2.3 Model Development
2.3.1 Model Overview
To test the potential impact of scheduling, our modeling objective is to measure the
scheduling characteristics of each performance and study the impact of these scheduling
characteristics on the number and timing of ticket sales. Thus, our model development consists
of four steps. First, we measure the scheduling characteristics to capture the similarity or
dissimilarity in venue locations and performance dates across performances. Second, we
examine the number of ticket sales across performances. Because it is possible that some
consumers evaluate the venue locations and performance dates but do not attend (i.e., non-
buyers), we incorporate the impact of scheduling characteristics on the size of non-buyer
segment. In this way, event marketers can understand how much market potential they have
captured and how much they have missed out. Third, we examine the timing of ticket sales
across performances. Because each performance sells tickets at a different rate and experience
different timing of ticket sales in an advance-selling period, it is important to capture
heterogeneous sales patterns and explain the differences. Finally, we consider a possible
endogeneity between performance scheduling and market response. If event marketers know the
effects of scheduling on the number of ticket sales, they could allocate performances based on
the positive or negative effect and expected performance attractiveness. Under this situation,
performance scheduling is endogenous with market response (Manchanda, Rossi, and
Chintagunta 2004). It is important to control for this endogeneity to ensure unbiased model
results. Figure 2-2 below summarizes our model development and we discuss each modeling
element in turn.
Figure 2-2: Model Overview
29
2.3.2 Scheduling Characteristics
To capture the similarity or dissimilarity in scheduling characteristics of performances,
we refer to a performance schedule of an event and differentiate between performances based on
their venue locations and performance dates. We create two measures to represent the
scheduling similarity in this performance schedule to understand how performances are
scheduled closely or distantly across venues and dates.
Specifically, we take the inverse geographic distance (in miles) between performances as
the geographic density measure and the inverse temporal distance (in days) between performance
dates as the temporal density measure (Fotheringham 1988). In this way, performances
scheduled in the same or proximate venues will have a higher value in geographic density to
represent similarity in the geographic schedule. Performances scheduled on the same or near
dates will have a higher value in temporal density to show similarity in the temporal schedule.
Therefore, our specifications for the density measures are as follows:
(1) uE0
j
=
1
} - 1
1
miles
jji
]
j
|
=1,j=ji
(2) TNP
j
=
1
} - 1
1
(1 +uays
jji
)
]
j
|
=1,j=ji
where GEO
j
is the geographic density for performance j (j=1,2,.., J), TMP
j
is the temporal
density for performance j, miles
jj’
represents the geographic distance between venues of
performance j and j’ (j?j’), and days
jj’
represents the temporal distance between performance
dates of j and j’.
To compute the distance between venue locations and performance dates, we use driving
distance (in miles) between the venues of j and j’ as miles
jj’
. We also specify miles
jj’
=1 for
30
performances in the same location to avoid the denominator equal to zero. In addition, we use
the absolute value of distance (in days) between performances j and j’ as days
jj’
. However, we
specify the denominator as (1+days
jj’
) in equation (2) to avoid performances on the same date
having a zero temporal distance.
Therefore, with respect to a target performance j, after taking its average (inverse)
geographic and temporal distance to other performances, GEO
j
and TMP
j
suggest its average
geographic and temporal density. The higher GEO
j
and TMP
j
of this performance, the closer this
performance is to other performances. As such, this performance is densely scheduled around
other performances and has higher similarity in venue locations and performance dates.
2.3.3 Number of Ticket Sales
To examine the number of ticket sales across performances and understand how much
market potential event marketers miss out, we apply a competing destination model by
Fotheringham (1988) and specify the share of each performance and the non-buyer segment as a
function of scheduling characteristics. By doing so, we can understand whether geographic and
temporal scheduling influence ticket sales at a performance level. In addition, using sales shares
of individual performances and the population size in a target market, we can obtain the number
of ticket sales expected at a performance level. We can also obtain a market penetration rate by
summing sales shares across performances.
Competing Destination Model
Among various extensions of the logit model, we consider the competing destination
model proposed by Fotheringham (1988) as a good alternative. The competing destination
model extends the traditional logit model by releasing the property of independence from
irrelevant alternatives (IIA) (Luce 1959; McFadden 1974). It examines consumers’ spatial
31
choice set as a function of geographic distance between stores. Then, the composition of the
spatial choice set further influences consumers’ store choices. If stores of shorter geographic
distance have higher chance to be in consumers’ choice set, these stores will attract more
consumers and have higher choice shares. However, if stores of shorter geographic distance
have a lower chance to be in consumers’ choice set, these stores will substitute one another and
have lower choice shares. The specification of the competing destination model is as follows
(Fotheringham 1988):
(3)
P
ì]
=
exp(I
ì]
) · l
ì
(] e H)
? exp(I
ì]?
) · l
ì
(]? e H)
]
j
?
=1
where P
ij
denotes the probability that consumer i shops in retail outlet j, V
ij
represents the
deterministic utility of retailer j to consumer i, and l
I
(j e N) is the likelihood that retailer j is in
consumer’s spatial choice set. After aggregating P
ij
across consumers, P
j
represents the market
share of a retailer j in a studied market of interest (González-Benito 2005).
Moreover, to measure the likelihood of spatial choice set, Fotheringham specified the
likelihood as a function of average inverse geographic distance between retail stores
(Fotheringham 1983; Fotheringham 1988) and empirically test the role of geographic distance:
(4)
l
ì
(] e H) =
`
Í
Í 1
} -1
w
j?
u
jj?
]
j?=1
j=j?
/
¹
¹
?
where d
jj’
is the geographic distance between a target store j and a competing store j’, w
j’
is the
weight for the competing store j’, and ? is the parameter indicating the role of geographic
distance. More specifically, if ?> 0, stores of shorter geographic distance to other stores will
have higher chance to be included in consumers’ choice set. If ?< 0, in contrast, stores will have
32
lower chance to be included. Finally, if ?= 0, geographic distance does not affect the
composition of consumers’ spatial choice set.
Extension of Competing Destination Model
To examine the separate effects of geographic and temporal scheduling, we extend the
competing destination model by incorporating the geographic and temporal density values and
allowing separate parameter values for these scheduling characteristics. In addition, we include
the non-buyer segment in a market as another alternative for potential buyers. In this way, we
can understand how scheduling characteristics affect the sales share of performances and the size
of non-buyer segment. Our adaption is as follows:
(5) P(]) =
exp(I
]
) · 0E0
]
0
1
IHP
]
0
2
1 + ? exp(I
]?
) · 0E0
]?
0
1
IHP
]?
0
2
]
j
?
=1
where P(j) denotes the sales share of performance j, V
j
represents the attractiveness of this
performance (which we will discuss later), GEO
j
and TMP
j
are the geographic and temporal
density values of performance j, and parameters ?
1
and ?
2
represent the scheduling effects.
If ?
1
or ?
2
> 0, geographic or temporal scheduling has a positive effect. Performances of
similar venue locations or performance dates will enjoy higher sales share. In contrast, if ?
1
or ?
2
< 0, geographic or temporal scheduling has a negative effect. Performances of similar venue
locations or performance dates will substitute one another and suffer from sales cannibalization.
However, if ?
1
or ?
2
= 0, scheduling has no impact on sales share. Moreover, we can use the
population size in a target market (which we will discuss in the data section), POP, to calculate
the expected number of ticket sales for any performance: Sales(j) = P0P × P(j). We can also
obtain the market penetration rate by summing sales share across performances [i. e. , ? P(j)
]
j=1
¸
and understand the size of non-buyer segment as 1 - ? P(j)
]
j=1
.
33
To measure and control the effect of performance attractiveness (V
j
), event ticket
literature has suggested various drivers for ticket sales that can be a good proxy for attractiveness.
However, because we focus on analyzing multiple performances of the same event, only
performance-related drivers are applicable. Therefore, we specify V
j
as a function of days of
week indicators (Friday, Saturday, and Sunday). The reason for us to choose these three days of
week indicators is that past studies indicate that performances on those days have higher
attractiveness (Corning and Levy 2002; Putler and Lele 2003). Hence, we specify performance
attractiveness as a control covariate for the number of ticket sales and incorporate a random error
term for unobserved attractiveness:
(6) v
j
= ?
0
+ ?
1
FRIBAY
j
+ ?
2
SAT0RBAY
j
+ ?
3
S0NBAY
j
+ ?
j
where ?
j
~N(u, ?
2
)
where ?
0
reflects the baseline attractiveness, ?
1
, ?
2
, and ?
3
suggest the effects of days of week on
incremental performance attractiveness for a Friday, Saturday, or Sunday performance, and E[V
j
]
represents the expected performance attractiveness.
To sum up, we adapt the competing destination model to understand the effects of
geographic and temporal scheduling on the sales share of performance and the size of non-buyer
segment. We also control for the performance attractiveness due to the days of week effects.
Although a linear regression or spatial model can also examine the number of ticket sales across
performances, our model specification is more appropriate than a regression or spatial model. A
linear regression is commonly used in event tickets literature (Moore 1966; Weinberg and
Shachmut 1978; Weinberg 1986; Reddy et al 1998; Corning and Levy 2002). However, it
cannot examine the sales share of each performance and the non-buyer segment at the same time
In other words, a linear regression does not allow us to understand how much of the market
performances have captured and how much they have left untapped. In contrast, our proposed
34
model can accomplish all of these limitations of a linear regression model. In comparison to a
spatial model, the model proposed in this dissertation specifically measures the effects of
performance schedules as drivers of ticket sales and not just modeling the spatial correlation
between performances (Bradlow et al 2005).
2.3.4 Timing of Ticket Sales
To examine the timing of ticket sales across performances, we first consider a typical
sales pattern for a performance. In general, a performance starts selling tickets 12 weeks prior to
its performing date. It usually sells fewer tickets in the beginning of its advance-selling period
and obtains more sales over time with the most arriving in the later period or the week of the
performance. Given this typical pattern, however, each performance still has a different sales
rate. Some performances experience ticket sales arrived earlier (i.e., earlier timing of ticket sales)
but others experience ticket sales arrived later (i.e., later timing of ticket sales).
Figure 2-3 below illustrates three patterns of ticket sales in an advance-selling period.
First, the solid line in Figure 2-3 (Case 1) is one common pattern, where consumers have low
probability to purchase well in advance. As time passes, the probability of a ticket transaction
slowly increases and peaks at the week of performance. However, there are some situations
where consumers expect performances to be of greater performance attractiveness or higher
popularity. As a result, they are more willing to purchase earlier and result in more ticket sales
arriving in the middle of an advance-selling period, as shown in the dotted line of Figure 2-3
(Case 2). Yet, there is another case when consumers think performances of lower performance
attractiveness and/or of higher substitutability. In this instance, they do not want to commit early
and wait until the week (or the day) of a performance. Therefore, ticket sales for such a
35
performance are very low in the entire advance-selling period and only peak in the spot market.
The broken line in Figure 2-3 (Case 3) represents this pattern.
Figure 2-3: Pattern of Ticket Sales Over Time
Weibull Hazard Model
Although performances usually follow a similar pattern as seen in Figure 2-3 (Case 1),
there still exists variability of sales pattern among performances (e.g., Case 2 and Case 3 in
Figure 2-3). To account for variations in sales rate across performances, we need a model that is
flexible enough to capture various sales patterns and examine the performance-specific sales
rates. Hence, we specify a Weibull hazard model to fit the timing of ticket sales because of its
flexibility in capturing various sales patterns, as shown in Figure 2-3. A Weibull hazard process
has the following properties:
(7) b
]
(t) = z
]
c
]
t
c
]
-1
S
]
(t) = c
-x
]
t
c
]
F
]
(t) = 1 -S
]
(t) = 1 -c
-x
]
t
c
]
where, with respect to a performance j, h
j
(t) is the instantaneous hazard rate for a ticket purchase
made at time t given this transaction has not yet been made, S
j
(t) is the survival rate for a ticket
purchase that has not yet occurred up to time t, and F
j
(t) is the cumulative probability for sales to
arrive over time. More specifically, ?
j
is the slope parameter for performance j to represent how
fast ticket sales arrive (?
j
>0), and c
j
is the shape parameter to capture an overall pattern of ticket
sales (c
j
>0).
For any discrete time t (e.g., week) in an advance-selling period, the probability of a
ticket purchase becomes:
36
(8)
P
]
(t) = F
]
(t) -F
]
(t -1) = c
-x
]
(t-1)
c
]
-c
-x
]
t
c
]
However, ticket sales in the context of a live performance event have to arrive no later
than the performance date. We adjust the probability of a ticket purchase in the spot market (i.e.,
the week of the performance) as follows:
(9)
P
]
(I
]
) = 1 -c
-x
]
(1
]
-1)
c
]
where T
j
is the number of advance selling weeks for performance j.
In addition, because we often observe seasonality (e.g., Thanksgiving, Christmas, etc.) or
marketing activities in an advance-selling period, we can include a time-varying covariate to
control for resulting sales bumps. Therefore, we rewrite equations (7), (8), and (9) as follows:
(10)
b
]
(t) = z
]
c
]
t
c
]
-1
c
[
]
X
]t
S
]
(t) = exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_
F
]
(t) = 1 -exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_
P
]
(t) = exp_-z
]
|u
c
]
- (u -1)
c
]
]
t-1
u=1
c
[
]
X
]u
_ -
exp_-z
]
|u
c
]
-(u -1)
c
]
]
t
u=1
c
[
]
X
]u
_ vt = 1·(I
]
-1)
P
]
(I) = exp_-z
]
|u
c
]
- (u -1)
c
]
]
t-1
u=1
c
[
]
X
]u
_
where X
jt
is a time-varying covariate or seasonality indicator and ?
j
is its associated parameter.
Consequently, the timing of ticket sales for each performance changes with its parameters ?
j
, c
j
,
and ?
j
.
37
Heterogeneity in Sales Patterns
To capture and explain the heterogeneity in sales pattern, we further specify the sales rate
of each performance as a function of its geographic and temporal density measures. We also
include two control covariates to ensure unbiased effects of these scheduling characteristics. The
first covariate is the length of advance-selling period because Moe and Fader (2002, 2009) found
that sales tend to arrive more slowly under a longer advance-selling period. In addition, because
consumers may purchase tickets much earlier when they expect performance to be more
attractive, we incorporate the expected performance attractiveness as the second control
covariate. Consequently, we specify the Weibull parameters (?
j
and c
j
) and the covariate effect
(?
j
) to follow the multivariate normal distribution. We take the log transformation for the
Weibull parameters to ensure positive values:
(11)
_
log(z
]
)
log(c
]
)
[
]
_ ~HIN(µ
j
, L
1
)
where
µ
j
= y
û
+y
1
0E0
]
+y
2
IHP
]
+y
3
I
]
+y
4
E|I
]
]
where ?
j
is the vector of expected Weibull parameters and covariate effect, GEO
j
and TMP
j
are
the geographic and temporal density measures in equations (1) and (2), T
j
is the number of
advance selling weeks for performance j, E[V
j
] is the expected performance attractiveness in
equation (6), and ?
0
, ?
1
, ?
2
, ?
3
, and ?
4
are the vectors of parameters for these covariates.
Therefore, using the parameter results in equation (11), event marketers can understand why
some performances experience earlier timing of ticket sales while other experience later timing
of ticket sales.
38
2.3.5 Endogeneity in Performance Scheduling
Finally, we consider a possible endogeneity in scheduling decisions. Since the live
performance industry has been established and event marketers have scheduled for a variety of
events, event marketers might have incorporated their experiences into a scheduling process. In
other words, how densely event marketers allocate performances across venues may be
dependent on the effect of geographic scheduling, and how densely event marketers allocate
performances across dates may be dependent on the effect of temporal scheduling. Moreover, it
is also likely that event marketers increase the total number of performances and schedule those
on weekend to increase the performance attractiveness. If so, the geographic and temporal
density values vary with the scheduling impact and the expected performance attractiveness.
To control for this type of endogeneity, we refer to a modeling approach proposed by
Manchanda et al (2004). In their research of pharmaceutical detailing, they mentioned that sales
representatives visit various doctors with different frequencies, and sales representatives
determine the frequencies based on how many prescriptions a doctor writes without any detailing
and how strong the effect of detailing is if they visit this doctor. They pointed out the
endogeneity between the decision of detailing and the effect of detailing, and further proposed a
model to correct this endogeneity.
In other words, they specified the expected prescription volume from a doctor as a
function of its baseline volume, the magnitude of detailing, and the detailing effect:
(12) ln(p
ì
-
) ÷ _
[
0ì
(1 -[
2ì
)
_ +_
[
1ì
(1 -[
2ì
)
_ Ðct
where ln(p
ì
-
) is the expected volume of prescription, j
[
0i
(1-[
2i
)
[ is the baseline prescription volume,
Det is the frequency of office visits, and j
[
1i
(1-[
2i
)
[ is the effect of detailing.
39
To model the endogenous detailing behavior, they specify the expected value of detailing
(ln(p
ì
)) as a function of the baseline prescription volume and the effect of detailing:
(13) ln(p
ì
) = y
0
+y
1
_
[
0ì
(1 -[
2ì
)
_ +y
2
_
[
1ì
(1 -[
2ì
)
_
In this way, if detailing is indeed endogenous, the parameter ?
1
or ?
2
will be significantly
different from zero, and the endogeneity between detailing and its effect is under control.
To control for the possible endogenous scheduling behaviors, we take the same modeling
approach as Manchanda et al (2004). We specify geographic and temporal density measures
(GEO
j
and TMP
j
) as a function of expected performance attractiveness (E[V
j
] in equation 6), and
the effects of scheduling characteristics (?
1
and ?
2
in equation 5). Because GEO
j
and TMP
j
in
equation (1) and (2) are between 0 and 1, we take the logit transformation for these density
measures and specify them following the multivariate normal distribution:
(14)
_
logit(0E0
]
)
logit(IHP
]
)
_ ~HIN(m
j
, L
2
)
where
j
m
1]
m
2]
[ = ç
û
+ç
1
E |I
]
] +ç
2
_
0
1
0
2
_
As such, if scheduling decisions are indeed endogenous and reflect on the density
measures, the parameters ?
1
or ?
2
will be significantly different from zero and this endogeneity
will be taken into account.
2.4 Data Description
We contact a national ticket seller to obtain a dataset of live performance events.
Because of the confidentiality agreement with our data provider, we cannot disclose our data
40
provider or the names of events. However, we will describe the nature of the events and the
behaviors observed in the dataset.
In this dataset, there were two events touring across several cities in the U.S. Each event
had a different number of performances in a city and lasted for a different period. For each
performance, we observe its venue location and performance date. In addition, we also have
detailed information regarding when tickets were purchased, for how much money, at which
price levels, and through which channels. Moreover, we are also able to observe the pattern of
ticket sales because transactions were recorded at a daily level. Table 2-1 provides a detailed
description of each field in our data set, which contains abundant information about the live
performance event and has many research opportunities for marketing researchers.
Table 2-1: Description of Variables in the Dataset
2.4.1 Description of Events
The events we have in the dataset are two popular family events. They are live
entertainments that targets on families with young children. In general, there are several types of
family events such as children’s music and theater (e.g., The Wiggles and Dora the Explorer
Live!), circus (e.g., Ringling Brothers and Barnum & Bailey and UniverSoul Circus), ice shows
(e.g., Disney on Ice series), magic shows (e.g., Xtreme Magic and Steve Wyrick ), and so on.
They usually travel across the U.S. or stay in a local market such as Las Vegas. The family
events we have are within the abovementioned categories.
For the two events we have, one sold 2.2 millions of tickets between January and June
2004 and travelled across 50 cities on 245 dates for 449 performances. The other sold 0.8
millions of tickets between January and May 2004 and travelled across 17 cities on 85 dates for
157 performances. The reason for the sales discrepancy is that one event had multiple
41
performing groups touring across cities simultaneously while the other had only one group
performing in our observed time span. Although these two events travelled to numerous cities
during their tours and had three cities in common (i.e., Jacksonville, FL; Miami, FL; Phoenix,
AZ), these events did not perform in those cities at the same time but at least 2.5 months apart.
Therefore, we assume there was no direct competition between these events to affect ticket sales.
Although both events are representative in terms of their ticket sales and the number of
performances, we take only one event in this dissertation to keep the event attractiveness
constant and examine merely the scheduling effect on ticket sales. In this way, once we confirm
a significant scheduling effect, we can further incorporate the renown of different events as an
extension. As such, we choose the event that had more performances. Among 50 cities that this
event toured, we also find some cities were within the same metropolitan markets (e.g., New
York and other metropolitan markets). This observation suggests that the event had multiple
stops in some markets and had higher variations in both geographic and temporal schedules.
Hence, we further select performances in the New York metropolitan market
1
and examine the
impact of geographic and temporal schedules on ticket sales in essay one. In summary, this
event had performances in Continental Arena in East Rutherford, NJ, Nassau Coliseum in
Uniondale, NY, Madison Square Garden in New York, NY, and Sovereign Bank Arena in
Trenton, NJ, respectively between March and June 2004 for 70 performances.
2.4.2 Description of Ticket Sales
Because a dataset of event tickets is not commonly available in marketing, we first
examine ticket sales by price levels and channel types to describe how much money people
usually pay and through which channel. Then, we examine the distribution of ticket sales across
1
We follow the Census Bureau data to define the boundary of a metropolitan market.
42
performances to understand how many tickets each performance sells and when ticket sales
arrive.
Ticket Sales by Price Levels
We first aggregate ticket sales by price levels and performances to examine any different
sales distributions across price levels. On average, the admission fees to a performance
(including face value, facility fees, and service charges) are $30.44 and there are about six price
tiers for consumers to choose. Although seating quality in a venue determines the price levels,
83% of ticket sales are contributed by mid-priced levels (i.e., price levels 2, 3 and 4) with
average price ranges from $20 to $50.
Specifically, price level 3 (mean price=$25.31, std= 3.66) represents 50% of ticket sales,
and price level 4 (mean price= $20.02; std= 3.17) and price level 2 (mean price= $50.16; std=
6.89) contribute 21% and 12% of ticket sales, respectively. In addition, we find the average
admission fees are relatively equal across venues and days of week. In other words, price
variations are within a performance (via price levels) but not across venues or days of week.
Ticket Sales by Channel Types
Next, we aggregate ticket sales by channel types and performances to examine ticket
sales across channels. Although consumers can purchase tickets through any of the six available
channels (i.e., primary box office, secondary box office, Hermes (automatic phone), Internet,
outlet, and phone), we find majority of ticket sales are made through the primary box office
(62% of ticket sales), following by the Internet (22% of ticket sales) and a ticket outlet (11% of
ticket sales). A possible reason for a primary box office to be a dominating channel choice is
that consumers do not have to pay for the convenience charges when they buy tickets in a box
office.
43
Ticket Sales by Performances
We also aggregate ticket sales across performances. As Figure 2-4 shows, a performance
on average sells 8,316 tickets but has its standard deviation being 3,525. Upon a closer look of
the sales distribution by days of week (based on performance dates), we find that weekend
performances tend to have more ticket sales than weekdays. Yet, the variation of ticket sales on
the same day of week is still prominent. The boxplot in Figure 2-5 summarizes the sales
distributions by days of week and indicate a clear variation even on the same day of week. For
example, a Friday performance sells 8,112 tickets on average but has a big standard deviation of
2,509, and a Saturday performance has average ticket sales of 9,552 but has the standard
deviation being 3,808. According to Figure 2-5 and the observation that each performance has
similar price levels, it is convincing that there must be additional factors to explain the variations
in ticket sales. Although one can argue that ticket sales are due to venue capacities, we find the
sizes of capacity in the four venues are similar and there are no sold out for any performance.
Hence, we do not consider the impact of venue capacity on ticket sales in this essay.
Figure 2-4: Sales Distribution by Performances
Figure 2-5: Sales Distribution by Days of Week
Finally, we aggregate daily ticket sales into weekly sales to examine the sales pattern for
each performance. On average, ticket sales arrive up to 15 weeks prior with the range between
11 and 19 weeks. Table 2-2 presents the ticket sales across performances and the breakdown of
weekly sales throughout an advance-selling period in Table 2-2. On average, a performance sells
8,316 tickets with 24% of sales arrived one month prior, 33% of sales arrived 2 to 4 weeks prior,
and 42% of sales arrived in the week of performances. Figure 2-6 shows a sales pattern for a
randomly chosen performance and it is a common pattern in our dataset. However, given this
44
similar pattern across performances, some performances experience ticket sales much earlier
than other performances (e.g., 59% vs. 3% of total sales arrived in the early stage) yet other
performances do not have as many ticket sales arrived in the last week (e.g., 75% vs. 11% of
total sales arrived in the spot stage). The boxplot in Figure 2-7 demonstrates the heterogeneity in
the timing of ticket sales across performances throughout an advance-selling period.
Table 2-2: Summary of Ticket Sales
Figure 2-6: Weekly Sales Pattern of a Performance
Figure 2-7: Heterogeneity in Timing of Ticket Sales across Performances
2.4.3 Covariate Specifications
Before we estimate the proposed model, there are several covariates not directly provided
in the dataset that require our attention. They are the geographic and temporal density measures
(GEO
j
and TMP
j
), the days of week indicators (FRIDAY
j
, SATURDAY
j
, SUNDAY
j
), the
estimated population size in the target market (POP), the length of advance-selling period (T
j
),
and the time-varying covariate (X
jt
). We discuss and specify these covariates in turn.
To compute the density measures, we first refer to venue locations and use the Google
TM
maps to find the driving distance (in miles) between venues. The numbers in Figure 2-8 indicate
the venue locations in the New York metropolitan market and represent the travel sequence
across venues. In addition, the numeric values between venues represent the mileage between
venues (i.e., mile
jj’
). We summarize the geographic distance between venues and the number of
performances in each venue in Figure 2-9 and follow equation (1) to calculate the geographic
density for each performance. Next, we refer to performance dates to calculate the temporal
distance (in absolute values) between performances (day
jj’
). According to the performance dates
45
and their temporal distance to others in Figure 2-10, we apply equation (2) to compute the
temporal density for every performance.
Figure 2-8: Venue Locations and Driving Distances
Figure 2-9: Summary of Geographic Distance
Figure 2-10: Summary of Temporal Distance
To compute the days of week indicators, we refer to performance dates to identify on
which days of week performances take place (i.e., FRIDAY, SATURDAY and SUNDAY).
Although some performances are scheduled on the same date, unfortunately our data does not
indicate time of day for the performances.
Next, we compute the population size in the target market (POP). Given that the event
targets families with young children, we define the target market as the population of families
with children under 10 years old. To compute the market size, we refer to the U.S. Census
Bureau for the 2000 data to find the total number of families with children under 18 years old
and the percentage of all children who are under 10 years old. We multiply these two numbers
to get the family population with children under 10 years old and then multiply the average
family size to get the population size in the target market. According to these calculations, there
are 4,082,615 potential consumers in the New York metropolitan market. We use this market
size and ticket sales across all performances to find the market penetration rate to be 14%, which
means the non-buyer segment represents 86% of the target market.
To measure the number of advance selling weeks (T
j
in equation 11) as a control
covariate, we compute the difference between the first sale date and the performance date for
each performance. Then, we divide this number by seven to convert the advance-selling period
46
to weeks. Although tickets may be available for sales prior to the first sale date, we think our
approach a good proxy given very few sales arrived in the early selling period.
Finally, because we observe Christmas within the advance-selling period for some
performances, we incorporate a time-varying indicator (X
jt
) in equation (10) to control for a
possible pre-Christmas shopping and resulting sales bumps. For each performance j, we code
X
jt
=1 if the advance selling week t is consistent with the pre-Christmas shopping week (i.e., 7
days prior to Christmas). Otherwise, X
jt
=0. Table 2-3 summarizes the descriptive statistics of
covariates.
Table 2-3: Descriptive Statistics of Covariates
2.5 Model Estimation and Benchmark Comparison
2.5.1 Estimation
We choose the Bayesian statistics approach to estimate the number of ticket sales, the
timing of ticket sales, and the endogeneity in a performance schedule simultaneously. We
specify appropriate and diffuse priors for our parameters in the WinBUGS program and estimate
the model over 40,000 iterations. After checking the convergence criteria, we check the
autocorrelation plots for all covariates, discarded 30,000 iterations for burn-in, and use the
remaining iterations as the posterior distribution. We specified the prior distribution of
parameters below:
Priors for modeling the number of ticket sales:
o
0
~N(-6,1u) For the baseline performance attractiveness
o
ì
~N(u,1uu) For the effects of days of week (where i=1, 2,3)
0
ì
~N(u,1uu) For the effect of scheduling characteristics (where i=1, 2)
o
2
~I0(u.1,u.1) For the variance of the performance attractiveness
47
Priors for modeling the timing of ticket sales:
y
ìk
~N(u, 1uu) For the Weibull parameters and the covariate effect
(where i=0, 1, 2, 3, 4 and k= 0, 1, 2, 3)
L
1
-1
~Weibull(I
3
, S)
For the variance-covariance of the Weibull parameters and the covariate
effect
Priors for modeling the endogeneity in performance schedule:
¢
ìk
~N(u,1uu) For the expected geographic and temporal density values
(where i=0, 1, 2 and k= 1, 2)
L
2
-1
~Weibull(I
2
, 2)
2.5.2 Benchmark Comparison
Before presenting our model results, we specify benchmark models to compare with our
proposed model to rule out alternative explanations for our proposed scheduling effects. First,
because the central focus of our modeling efforts is to examine the impact of scheduling
characteristics, one ideal benchmark model is to exclude any scheduling effect but only
incorporate performance attractiveness (i.e., Benchmark 1). Second, some venue locations are
more popular than others. For example, a venue in the New York city might be more attractive
than another venue in Uniondale. Therefore, we consider the second benchmark that
incorporates the venue-specific indicator variables. Finally, Population density around the venue
locations could also explain ticket sales. In other words, event marketers may schedule more
performances in a specific venue because the population density in this venue is high and the
scheduling decision is simply to meet potential market demand in this venue. To rule out this
alternative explanation, we extend our proposed model by including the population density
around each venue location of performances as another explanatory variable (i.e., Benchmark 3).
We estimate our proposed and benchmark models to compare the model fit using the
deviance information criteria (i.e., DIC, Spiegelhalter et al 2002):
48
(15) ÐIC = Ð(ç
) +2pÐ
where Ð(ç
) is the deviance evaluated at the posterior means ç
and pD is the effective number of
parameters in a model, calculated as the difference between the posterior deviance and the
deviance of the posterior mean.
Table 2-4 summarizes the model fit across the benchmark and proposed models.
According to the DIC reported for every model, we find that incorporating the scheduling effect
is definitely superior. Although adding city effects improves the DIC from 157,134 (Benchmark
1) to 156,459 (Benchmark 2), city effects cannot explain the ticket sales as well as does the
proposed model (DIC= 156,221). Moreover, we find our proposed model has a similar fit to
Benchmark 3 (DIC=156,220). Although Benchmark 3 has a smaller DIC value by one unit,
Ntzoufras (2009, p.220) suggests that a model performs better than another does if the DIC
difference is greater than 2. Therefore, we conclude that our proposed model is as good as
Benchmark 3.
Finally, we compare the parameter results between these two models and find significant
and consistent effects of scheduling characteristics on ticket sales. Although the third benchmark
model also shows that the population density and travel distance from consumers’ residences to
venues is relevant, results in this benchmark model still indicate a significant geographic effect.
In other words, although consumers may prefer a venue nearest to them, it is still very likely that
consumers are willing to travel to a farther venue that has a stronger association with a leisure
activity (Okada 2005).
Therefore, we are confident that there are scheduling effects to influence consumer
decisions and ticket sales across performances. We discuss the parameter results and their
implications in the next section.
49
Table 2-4: Benchmark Models and Model Fit
2.6 Results
2.6.1 Number of Ticket Sales
Table 2-5 summarizes our parameter results for the number of ticket sales. First, we find
that performances on Saturday and Sunday have higher attractiveness to increase ticket sales
than those on other days of week (?
2
= 0.35; ?
3
= 0.34). This result is consistent with prior
literature (Corning and Levy 2002; Putler and Lele 2003) and shows the importance of
controlling for performance attractiveness when examining the impact of scheduling
characteristics.
Second, we find that scheduling indeed influences how many tickets each performance
can sell. When performances are scheduled closely in venues and have a shorter geographic
distance to other performances, they attract more consumers and sell more tickets (?
1
= 0.32).
Additionally, when performances are scheduled sparsely along a time span and have a longer
temporal distance to others, they attract more consumers and sell more tickets (?
2
= -0.14).
Because the geographic and temporal density measures have different effects on ticket sales,
these results suggest that consumers evaluate geographic and temporal schedules separately and
have different responses.
Table 2-5: Results for the Number of Ticket Sales
As we mentioned earlier, event marketers’ scheduling and consumers’ decision making
are both unobserved processes to researchers. Hence, we can only speculate possible underlying
mechanisms based on prior literature and our results. One way to interpret different consumer
responses to geographic and temporal schedules is that event marketers have several objectives
when they schedule performances. These objectives influence their scheduling decisions and
50
consumers’ reactions. For example, their objectives may be to signal venue popularity and
enhance assurance of product delivery at the desired time. If so, they will schedule performances
closely in venues of interest yet sparsely across performance dates. From consumers’
perspective, after they see such geographic and temporal schedules, they receive the signal of
venue popularity and find the flexibility in attendance timing. Then, they shape their preferences
to favor performances of such scheduling characteristics and further influence their purchase
decisions. As a result, these performances sell more tickets than other performances.
Alternatively, event marketers may not have a predetermined scheduling strategy in mind.
The geographic and temporal schedules are the consequences of constrained optimization. If so,
consumers will make their own inferences about these performances. For consumers who do not
have specific preferences to venue locations and performance dates, they might evaluate
performances differently based on different contexts. For instance, they may find an event
highly associated with some venues because these venues are close to each other and offer more
performances. Due to the similar venue locations and shorter geographic distance between
venues, these venues may catch consumers’ attention better and become consumers’ preferred
venues when consumers attend an event. Performances in these venues hence share this
common advantage to attract more consumers and sell more tickets.
On the other hand, consumers may perceive an event highly associated with some days of
week because many performances are scheduled around those days. Therefore, these days of
week would catch more attention and become more salient when consumers consider when to
attend. However, consumers usually have uncertainty for the future and prefer a wider range of
dates for selections. Closely scheduled performances at any time could merely substitute one
another and suffer from sales cannibalization.
51
According to these two interpretations, one implication for event marketers is that they
should keep their scheduling strategy (if they indeed have such a strategy) to schedule
performances densely in venue locations but sparsely across performance dates. Even if they do
not have such a strategy but only practice constrained optimization, our results provide them
another useful scheduling implication. That is, they should incorporate the scheduling effects
(i.e., a positive geographic effect but a negative temporal effect) into their decision process as
new constraints to find the most optimal solution.
2.6.2 Timing of Ticket Sales
Table 2-6 describes the parameter estimates and indicates that geographic density and the
number of advance selling weeks have significant effects on the Weibull parameters (?
11
= -7.56;
?
12
= 0.93). Because the objective of this paper is to examine the scheduling effect while
controlling for the number of advance selling weeks, we discuss the geographic effect on the
timing of ticket sales more in details.
Table 2-6: Results for Timing of Ticket Sales
However, it is less straightforward to observe the net effect of geographic density on the
timing of ticket sales based on the parameter results. We proceed to simulate performances of
different levels of geographic density to visually show their effects on the timing of ticket sales.
Therefore, Figure 2-11 presents three hypothetical performances of different levels of geographic
density but of the same performance attractiveness. The solid line illustrates the expected timing
of ticket sales resulting from geographic density being the mean value observed in our dataset
(GEO). The two dashed lines show the expected timing of ticket sales for two performances
where their geographic density values are one standard deviation higher or lower than the mean,
respectively. We can observe from Figure 2-11 that once the geographic density increases by
52
one standard deviation from the mean (GEO +1SD), the cumulative ticket sales after 12 weeks of
advance selling increase from 11% to 25% of its expected total amount. In contrast, when the
geographic density decreases by one standard deviation from the mean (GEO -1SD), only 7%
arrived after 12 weeks. Therefore, we conclude that performances in densely scheduled venues
have shorter geographic distance to other performances such that they sell tickets at a faster rate
than those of longer geographic distance.
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales
Our intuition for this result is that venues where these performances are closely scheduled
share the similarity in venue locations and geographic density. These venues can catch
consumers’ attention and lead to an attraction effect. In other words, even after controlling for
performance attractiveness, consumers still think performances in these venues more attractive.
As a result, they are willing to purchase tickets much earlier.
The implication of this result is that event marketers can monitor when and how fast
ticket sales arrive based on the geographic density information across performances. They can
use the expected timing of ticket sales as benchmark measures to compare with realized sales. In
this way, they can be aware of possible sales deviation in an advance-selling period rather than in
the week of performance. Additionally, it is also important for operational and financial
planning because event marketers can adjust their concession and security throughout an
advance-selling period to make sure a performance is not over or under staffed. Moreover, they
can have a better knowledge with the cash flows based on the expected timing of ticket sales.
Regarding the sales bump as a result of pre-Christmas shopping effect (?
j
), we find that
geographic density, number of advance selling weeks, and expected performance attractiveness
all contribute positive effects (?
13
= 4.84; ?
33
= 0.97; ?
43
= 2.45). Yet, temporal density does not
53
have such an effect. In other words, performances of higher attractiveness tend to experience
earlier timing of ticket sales. The attractiveness comes from the expected individual
performance attractiveness (due to the days of week effect) and the geographic density. Because
performances of shorter geographic distance are perceived more attractive even after controlling
for individual performance attractiveness, these performances are more salient to consumers as
Christmas gifts
2
.
2.6.3 Endogeneity in Performance Scheduling
As we have mentioned, the scheduling decision is likely endogenous and dependent on
expected performance attractiveness or effects of geographic and temporal scheduling. As such,
the results discussed above are only managerially meaningful if we accommodate the potential
endogenous scheduling decisions. Table 2-7 indicates some evidence about the endogeneity in
performance scheduling. Specifically, event marketers consider the expected performance
attractiveness when designing a temporal schedule (?
12
= 0.26). When they expect a performance
to be more attractive, they schedule more performances similar to this one. Consequently, there
are more performances scheduled temporally close to each other, resulting in higher temporal
density. However, event marketers neither incorporate performance attractiveness when
designing a geographic schedule, nor do they incorporate the geographic and temporal effects.
This result implies that event marketers may primarily rely on performance attractiveness in their
scheduling decision whether they are aware of the scheduling effects.
Table 2-7: Results for Endogenous Performance Scheduling
2
We considered other covariates, such as the cumulative sales of earlier performances, but found that it had no
significant impact on the timing of ticket sales.
54
2.7 Policy Simulation
To demonstrate the scheduling effects on ticket sales, we conduct a policy simulation by
varying the geographic schedule (Scenario 1) or temporal schedule (Scenario 2) to compare with
the current setting. To make a fair comparison and restrict a new schedule within the same
geographic and temporal range, we only re-allocate one performance, keep the rest unchanged,
and evaluate the differences in ticket sales for the target performance as well as the entire market.
2.7.1 Scenario 1: Geographic Change
According to results of the number of ticket sales, performances generate more sales
volume when they are scheduled in the same or proximate venues and have a higher geographic
density. Hence, in Scenario 1 we reschedule a performance from Venue 2 (Nassau Coliseum) to
Venue 3 (Madison Square Garden) in Figure 2-9, the most densely scheduled venue in our
dataset. As a result, geographic density of this performance increases. We also keep the same
performance date to ensure unchanged performance attractiveness and temporal density.
Figure 2-12 presents the impact of schedule changes on ticket sales. After relocating a
performance from Nassau Coliseum to Madison Square Garden, its ticket sales increase from
5,102 to 7,132, resulting in a difference of 2,031 tickets (which is 24% of average ticket sales per
performance). On the other hand, the overall market sales increase from 532,285 under the
existing schedule to 535,424 tickets in Scenario 1. Note that the 3,140 increases in market sales
are greater than 2,031 increases in a target performance. This increase in market sales provides
the evidence of market expansion. Hence, a geographic schedule after minor changes can attract
more consumers to the rescheduled performance and increase the market penetration.
Figure 2-12: Effect of Schedule Changes on Ticket Sales
55
Relocating the target performance to a different venue also changes the timing of ticket
sales. Under the modified schedule, the cumulative ticket sales of the target performance reach
60% of total sales after 12 weeks, yet the same performance only sells 28% under the original
schedule. Figure 2-13 shows that weekly sales of this performance arrive at a different rate and
results in different patterns. This implies that any monitoring or benchmarking of early ticket
sales needs to incorporate the geographic density in a schedule.
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales
2.7.2 Scenario 2: Temporal Changes
According to our results from studying the number of ticket sales, the second learning is
to disperse performance dates to accommodate more consumers and increase ticket sales.
Therefore, in Scenario 2, we reschedule the same target performance in Scenario 1 to two weeks
earlier. Yet, we keep its venue location and day of week constant. In this way, this target
performance has the same level of performance attractiveness and geographic density, yet with a
lower level of temporal density.
In contrast with Scenario 1 where we see substantial changes in the number and timing of
ticket sales, we observe much smaller changes in Scenario 2. This is consistent with the smaller
parameter values for temporal density compared to those for geographic density (see Table 2-5
and Table 2-6). Specifically, ticket sales for the rescheduled performance increase very slightly
by 385 tickets under the modified temporal schedule and the difference in total sales in the
market is also quite small (772 tickets). Figure 2-12 also shows minimal differences in the
percentage of tickets sold in the first 12 weeks. Hence, these results imply that geographic
scheduling is more important than temporal scheduling. A geographic schedule has a greater
influence on the number and timing of ticket sales than does a temporal schedule.
56
2.8 Conclusions
2.8.1 Summary
In the live entertainment industry, scheduling performances and estimating ticket demand
are two primary tasks facing event marketers. Because these tasks have been treated as two
independent problems by event marketers and marketing researchers, this essay aims to bridge
performance scheduling and demand estimation by examining the potential impact of scheduling
on ticket sales.
According to different streams of literatures, we find that it is possible to see closely
scheduled performances selling more than distantly scheduled performances, yet it is also
possible to see the opposite effect. Therefore, we allow these two possibilities and empirically
test the effect of scheduling characteristics on the number and timing of ticket sales. Specifically,
we characterize performances of the same event by their venue locations and performance dates.
Using their scheduling characteristics, we construct two density measures (i.e., geographic and
temporal density) to capture how close in distance or time performances are scheduled to each
other. Then, we model the number and timing of ticket sales as a function of these density
measures. In addition, we also control for a possible endogeneity in case event marketers
incorporate market responses in their scheduling process.
We contact a national ticket seller to obtain a dataset of a live performance event and use
its ticket sales to examine the impact of its scheduling characteristics on ticket sales. This event
had 70 performances across four venues in the New York metropolitan market and performed
between March and June 2004.
Our results indicate that performances of different scheduling characteristics sell
differently in terms of their number and timing of ticket sales. Specifically, we find that the
57
effect of geographic scheduling differs from the effect of temporal scheduling. Performances
scheduled closely in distance not only sell more tickets but also sell tickets at a faster rate. In
contrast, performances scheduled sparsely in time sell more tickets but do not have an impact on
the timing of sales.
Our explanation for the observed effects is that event marketers may schedule
performances to signal the desirability of venues and accommodate consumers’ uncertain
attendance timing. In this way, performances in the desired venues and along a wider temporal
stretch are more attractive to consumers (even after we have controlled the individual
performance attractiveness).
2.8.2 Limitations and Next Steps
Although this essay shows significant effects of geographic and temporal scheduling on
ticket sales, the results are for one market only. However, event marketers often need to make
scheduling decisions for more than one market. Thus, it is not clear whether the results in essay
one hold in other markets. Hence, the objective of essay two is to use all performance schedules
of the same event to investigate heterogeneous market responses and identify explanatory factors.
We discuss essay two in the next chapter.
58
Table and Figures
Table 2-1: Description of Data Fields
Category in the Data Fields Description
Event Name of event
Identification number Used to differentiate repeat performances of the same event
Performance date Month-Date-Year
Venue location Name of a venue and its location (City and State)
Transaction types Indicate individual purchases, group purchases, school
purchases and so on
Sales date Month-Date-Year
Channel types Six channel types: Primary Box Office, Secondary Box Office,
Hermes (Automatic phone), Internet, Outlet, and Phone.
Price levels Label of price levels
Price paid Indicates the face value, facility charges, service charges, and
the total price paid
Daily tickets Number of tickets sold
59
Table 2-2: Summary of Ticket Sales
Mean Std Dev Min Max
Total Ticket Sales 8,316 3,525 1,827 15,810
Ticket Sales by Stage
Early Sales
(one month prior)
24% 14% 3% 59%
Late Sales
(2-4 weeks prior)
33% 9% 15% 53%
Spot Sales
(performance week)
42% 14% 11% 75%
60
Table 2-3: Descriptive Statistics of Covariates
Description Mean Std Dev Min Max
GEO Geographic Density 0.368 0.185 0.128 0.540
TMP Temporal Density 0.118 0.022 0.063 0.144
FRIDAY Friday performance 0.157 0.367 0 1
SATURDAY Saturday performance 0.300 0.462 0 1
SUNDAY Sunday performance 0.286 0.455 0 1
T
Number of advance-selling weeks 15 2.044 11 19
61
Table 2-4: Benchmark Models and Model Fit
Benchmark Model Proposed Model
1 2 3
Performance attractiveness 9 9 9 9
City effects 9
Population around venues 9
Model Fit
DIC 157,134 156,459 156,220 156,221
62
Table 2-5: Results for the Number of Ticket Sales
Parameter Description Median (STD)
Scheduling Effect
?
1
Effect of geographic density 0.32 (0.05)**
?
2
Effect of temporal density -0.14 (0.05)**
Expected performance attractiveness: E[V
j
]
?
0
Baseline value of event -6.32 (0.12)**
?
1
Friday effect 0.15 (0.13)
?
2
Saturday effect 0.35 (0.12)**
?
3
Sunday effect 0.34 (0.13)**
?
2
Variance of performance attractiveness 5.45 (1.02)
** significant at the 95% highest posterior density
63
Table 2-6: Results for the Timing of Ticket Sales
Parameter Description Median (STD)
Weibull slope parameter: log(?
j
)
?
01
Intercept 3.11 (3.62)
?
11
Effect of geographic density -7.56 (2.20)**
?
21
Effect of temporal density 0.67 (2.99)
?
31
Number of advance-selling weeks -1.08 (0.21)**
?
41
Expected performance attractiveness -0.63 (0.77)
Weibull shape parameter: log(c
j
)
?
02
Intercept -0.64 (1.03)
?
12
Effect of geographic density 0.93 (0.24)**
?
22
Effect of temporal density 0.23 (0.98)
?
32
Number of advance-selling weeks 0.05 (0.02)**
?
42
Expected performance attractiveness -0.16 (0.17)
Time-varying pre-Christmas shopping effect: ?
j
?
03
Intercept 2.09 (3.24)
?
13
Effect of geographic density 4.84 (2.51)*
?
23
Effect of temporal density 1.63 (3.04)
?
33
Number of advance-selling weeks 0.97 (0.28)**
?
43
Expected performance attractiveness 2.45 (0.81)**
Variance-covariance matrix: ?
1
log(?
j
) log(c
j
) ?
j
log(?
j
) 17.38 -1.3 -11.49
log(c
j
) 0.13 0.77
? 12.58
** significant at the 95% highest posterior density
* significant at the 90% highest posterior density
64
Table 2-7: Results for Endogenous Performance Scheduling
Parameter Description Median (STD)
Expected geographic density: m
1
?
01
Intercept 0.18 (0.91)
?
11
Expected performance attractiveness 0.22 (0.15)
?
21
Effect of geographic density 0.04 (0.97)
Expected temporal density: m
2
?
02
Intercept -0.61 (0.75)
?
12
Expected value of performance 0.26 (0.12)**
?
22
Effect of temporal density -0.03 (1.01)
Variance-covariance matrix: ?
2
m
1
m
2
m
1
0.36 0.07
m
2
0.06
** significant at the 95% highest posterior density
65
Figure 2-1: Conceptual Framework of Spatial Decomposition
Geographic
Density
Temporal
Density
Scheduling Characteristics:
Ticket Sales:
(1) Number of Tickets Sold
(2) Timing of Ticket Sales
Attractiveness of
Individual Performances
Geographic
Effect
Temporal
Effect
Endogeneity
66
Figure 2-2: Model Overview
Number of Ticket Sales Timing of Ticket Sales
Scheduling Characteristics
(Geographic and Temporal Density)
67
Figure 2-3: Probability of Ticket Sales over Time
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
r
o
b
a
b
i
l
i
t
y
Advance Selling Week
Case 1
Case 2
Case 3
68
Figure 2-4: Sales Distribution by Performances
sales
20000.00 15000.00 10000.00 5000.00 0.00
F
r
e
q
u
e
n
c
y
12
10
8
6
4
2
0
Mean =8315.70
Std. Dev. =3525.
329
N =70
69
Figure 2-5: Sales Distribution by Days of Week
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
T
i
c
k
e
t
S
a
l
e
s
70
Figure 2-6: Weekly Sales Pattern of a Performance
0
1000
2000
3000
4000
5000
6000
7000
T
i
c
k
e
t
s
a
l
e
s
Advance Selling Week (t)
71
Figure 2-7: Heterogeneity in Sales Pattern Across Performances
0%
10%
20%
30%
40%
50%
60%
70%
80%
1 month prior 2?4 weeks prior last week
%
o
f
t
o
t
a
l
t
i
c
k
e
t
s
a
l
e
s
Advance Selling Period
72
Figure 2-8: Venue Locations and Driving Distances
Note: the number in Figure 2-8 indicates the sequence that the event travelled.
That is, the event went to Venue 1, 2, 3, and 4, respectively.
46
29
90
16
67
67
Note: distance in miles
Source: Google Maps
1
2
3
4
73
Figure 2-9: Summary of Geographic Distance
Continental
Arena
Nassau Coliseum
Madison Square
Garden
Sovereign Bank
Arena
0
5
10
15
20
25
30
35
40
0 10 20 30 40 50 60
N
u
m
b
e
r
o
f
p
e
r
f
o
r
m
a
n
c
e
s
i
n
a
v
e
n
u
e
Average distance to other venues (in miles)
74
Figure 2-10: Summary of Temporal Distance
0
10
20
30
40
50
60
70
2/23/2004 3/14/2004 4/3/2004 4/23/2004 5/13/2004 6/2/2004 6/22/2004
A
v
e
r
a
g
e
d
i
s
t
a
n
c
e
t
o
o
t
h
e
r
p
e
r
f
o
r
m
a
n
c
e
d
a
t
e
s
(
i
n
d
a
y
s
)
Performance date
75
Figure 2-11: Impact of Geographic Density on Timing of Ticket Sales
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
%
o
f
t
o
t
a
l
t
i
c
k
e
t
s
Advance Selling Week
mean GEO +1 SD
mean GEO
mean GEO ?1 SD
Cumulative ticket sales= 25%
Cumulative ticket sales= 11%
Cumulative ticket sales= 7%
76
Figure 2-12: Effect of Schedule Changes on Ticket Sales
Current Schedule:
Target performance=5,102 Market sales=532,285
0
500
1000
1500
2000
2500
3000
3500
Scenario 1 Scenario 2
C
h
a
n
g
e
s
i
n
T
i
c
k
e
t
S
a
l
e
s
target performance market sales
+2031
+385
+772
+3140
77
Figure 2-13: Effect of Schedule Changes on Timing of Ticket Sales
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
1 2 3 4 5 6 7 8 9 10 11 12
%
o
f
T
o
t
a
l
S
a
l
e
s
A
r
r
i
v
e
d
Advance Selling Week
current scenario 1 scenario 2
Cumulative ticket sales= 28% of total
Cumulative ticket sales= 60% of total
78
3 Essay 2: Heterogeneous Market Responses to Performance Schedules and
Their Explanatory Factors
3.1 Introduction
In the first essay, we characterize multiple performances of a single event by their venue
locations and performance dates to understand how their geographic and temporal scheduling
characteristics influence their ticket sales. Using ticket sales of a live performing event in the
New York metropolitan market, we find that the effect of geographic scheduling differs from the
effect of temporal scheduling. Performances scheduled in nearby venues not only sell more
tickets but also sell tickets at a faster rate. In contrast, performances scheduled distantly in time
sell more tickets but do not have an impact on the timing of sales.
However, event marketers often need to make scheduling decisions for more than one
market. Although our finding in essay one has rich implications for event marketers, it is unclear
whether event marketers can apply this finding to all markets. For example, the event analyzed
in essay one had 70 performances in four venues for 32 days in the New York metropolitan
market. However, when it travelled to other markets such as Norfolk, Virginia, it had 19
performances in two venues for 10 days. It also had 21 performances in one venue for 10 days in
the Atlanta area. Thus, performance schedules can vary across markets and these schedules may
not have the same effect on ticket sales across markets. Even if a schedule is the same across
markets, these markets may not respond to their schedule identically. This limitation in essay
one hence motivates our second essay to examine the effect of performance schedules across
markets.
79
Although several studies in the event tickets literature examine descriptive drivers for
event ticket sales, very limited research focuses on examining heterogeneous market responses
(Moore 1966; Weinberg and Shachmut 1978; Havlena and Holak 1988; Reddy et al 1998).
Hence, the objective of essay two is to use all performance schedules of the same event to
investigate heterogeneous market responses and identify explanatory factors. It is important
because events do not always go to the same set of markets when they are on tour. A long
lasting event may go on tour several times and travels to a different set of markets each time.
Hence, once the heterogeneity in market responses and explanatory drivers are known, event
marketers could infer a likely response in a new market or select markets for touring based on
expected market responses.
To accomplish our research objective, we first conduct a preliminary analysis to analyze
all performance schedules and examine their impact on the number of ticket sales. After
applying the model developed in essay one to the entire dataset and estimating market responses
iteratively across markets, we find that the effect of geographic scheduling differs from the effect
of temporal scheduling and this difference is consistent across markets. In terms of the
geographic scheduling, performances scheduled in close venues sell more ticket sales than
sparsely scheduled ones, but this result only hold for markets that use multiple venues.
Regarding the temporal scheduling, in contrast, performances scheduled distantly in time sell
more than the densely scheduled ones.
Although we find consistent scheduling effects from the preliminary results, we also
observe the market responses are of different magnitudes. In other words, some markets are
more responsive to geographic (temporal) schedules than other markets. To identify the factors
that explain these differences across markets, we extend the model developed in essay one to not
80
only examine market-specific response parameters but also investigate observed and unobserved
heterogeneity via the hierarchical Bayesian approach. In addition, we also control for the
possible endogeneity in the performance scheduling process. Among several marketing
characteristics, we choose the size of market population as the first explanatory factor. We also
examine characteristics of a touring event to understand whether participating markets that are
adjacent to each other and the order that an event travels across affect the magnitude of
scheduling effects.
We use the same family event mentioned in essay one and analyze all of its performance
schedules in the dataset to test our model. Because this event sequentially performed 449 times
in 50 cities in the U.S. domestic market between January and June 2004, we aggregate these 50
cities to 42 designated market areas (which will be discussed in detail in the data section). As a
result, there are six markets where the event performed in multiple venues yet all 42 markets
have some variations in their temporal schedules.
Our results show that market and additional scheduling characteristics can both explain
the differences across market responses. First, when a market has a bigger population, the
effects of days of week and baseline attractiveness are attenuated. Our explanation is that
usually there are more events offered to a bigger market than to a smaller one. Consumers in a
bigger market may be used to seeing several competing events offered simultaneously and
having a variety of events to choose from. Hence, they are less responsive to an event no matter
on which day of week it may be on as compared to consumers in a smaller market.
Second, we find the additional scheduling characteristics can partly explain
heterogeneous market responses. With respect to a current market, after an event travels to more
of its geographically adjacent markets, the current market is less responsive to its baseline
81
attractiveness and temporal schedule. A possible reason is that after an event has gone to more
geographically adjacent markets, its newness wears out but its reputation might accumulate over
time. As a result, consumers may refer to other measures such as word of mouth to make their
purchase decisions rather than refer to the baseline attractiveness and its temporal schedule.
Moreover, after an event perform in several markets (whether these markets are adjacent
or not), a market in which an event performs later tends to respond more favorably to a Sunday
performance. Our explanation is that after an event has lasted longer and gone to more markets,
its reputation, or word of mouth, accumulates over time (Reddy et al 1981) even though its
newness may wear out. Because a Sunday performance tends to be the last performance in a
market (at least in the case of our dataset), consumers might think Sunday as their “last
opportunity” to enjoy this event before it leaves for another market. As a result, a market in a
later temporal sequence has a stronger Sunday effect.
We also find that the nature of performance schedules is endogenous. Different from
essay one that endogeneity is found only in a temporal schedule, we find in essay two that
geographic and temporal schedules are both done endogenously after we pool all performances
across markets for analysis. When event marketers expect high performance attractiveness, they
schedule more performances in all venues and tend to allocate those performances around
weekends or along a limited time span. As a result, on average, performances have a shorter
geographic and temporal distance to others and have higher density values. Moreover, when
event marketers understand that consumers prefer performances on dispersed dates because of
uncertain timing of attendance, they decrease the number of weekend performances and/or
disperse performance dates. Consequently, performances in a temporal schedule have lower
82
temporal density values. Since we have accounted for this endogeneity in the estimation process,
the results we present here are unbiased.
The rest of essay two is organized as follows. First, we conduct a preliminary analysis
using all performances observed across markets to show heterogeneous market responses to
performance schedules. Second, we review extant literature to find possible reasons for
heterogeneity and propose our conceptual framework. In the next section, we present the
modeling structure extended from the first essay and discuss the dataset in details. After the
model estimation and benchmark comparison, we present our results and conclude this essay
with limitations and next steps.
3.2 Preliminary Analysis
As described in essay one, a family event went to 50 cities on 245 dates for 449
performances between January and June 2004 (see §2.4.1 Description of Events for more details).
Among which, several cities are within the same metropolitan area and show richer variations in
geographic and temporal schedules. To test whether markets have heterogeneous responses to
their performance schedules, we conduct a preliminary analysis to run the model in equations (5),
(6), and (14) for one market at a time and examine their market specific parameters.
More specifically, we use the designated market area (DMA) to aggregate 50 cities into
42 markets (see the data section for full descriptions) and summarize the market information in
Table 3-1. As Table 3-1 shows, we sort markets by their first performance date and assign a
unique market identification number. Every market has its information listed regarding its first
and last performance dates, number of performances, length of performing period, and venue
usage. Hence, for performances within a market, we characterize these performances by their
83
venue locations and performance dates. Then, we follow the equations (1) and (2) in essay one
to compute their geographic and temporal density and understand how densely or sparsely these
performances are scheduled. Consequently, for each market, we model ticket sales of its
performances as a function of their density measures and performance attractiveness while
controlling for a possible endogeneity in performance scheduling.
Table 3-1: Summary of Performance Schedules across Markets
After iteratively estimating the model across 42 markets, we find that the days of week
represent performance attractiveness and increase ticket sales (?
0
ranges from -3.82 to -8.23; ?
1
ranges from 0.10 to 0.95; ?
2
ranges from 0.24 to 1.48; ?
3
ranges from 0.13 to 1.09). Moreover,
the effect of geographic scheduling differs from the effect of temporal scheduling. Performances
scheduled in nearby venues sell more tickets than those scheduled in distant venues (?
1
ranges
from 0.92 to 4.29). However, this result only holds for markets that use multiple venues. On the
other hand, performances scheduled dispersed across dates sell more than the those scheduled
close in time (?
2
ranges from -0.04 to -2.92). We summarize these market-specific parameters in
Table 3-2.
Table 3-2: Summary of Market Responses to Performance Schedules
According to our preliminary analysis, we confirm consistent yet heterogeneous market
responses to performance schedules. To understand why market responses are different, we
review extant literature that suggests potential explanatory factors for this heterogeneity.
84
3.3 Literature Review and Conceptual Framework
3.3.1 Literature Review
Consumers are different individuals and their preferences and decisions often differ from
one another. However, as the proverb says, “birds of a feather flock together.” It is very
common to observe consumers who have similar tastes living in similar areas and making similar
purchase decisions. Accordingly, when marketers offer a variety of products to consumers, they
often expect heterogeneous consumer responses across zip codes, counties, states, or
metropolitan markets.
Heterogeneity in the unit of analysis is commonly studied in marketing and the entities
include individual consumers, products, firms, markets, and countries. For example, consumers
who have different demographic characteristics or live in different zip codes make different
choices regarding automobiles (Yang and Allenby 2002), book formats (i.e., Print vs. PDF; Jank
and Kannak 2005), or adoptions of online grocer (Choi, Hui, and Bell 2009). Their shopping
behaviors also differ across product types (e.g., motels of high, medium, or low quality, Mazzeo
2002; department stores of upscale, midscale, or discount, Vitorino 2007), store formats (e.g.,
supermarkets, hypermarkets, and discount stores, González-Benito 2005; discounted or regular
retailers, Zhu et al 2007), and brand names (e.g., Wal-mart, K-mart, Target, Zhu and Singh 2009;
Zhu et al 2009). Besides examining the heterogeneity at an individual level, researchers can also
summarize consumer responses across zip codes (Yang and Allenby 2002; Jank and Kannan
2005), metropolitan statistical areas (Zhu and Singh 2009), or countries (Elberse and Eliashberg
2003) to study heterogeneity at an aggregate level.
One way to account for heterogeneity in empirical analyses is to specify individual-
specific parameters (Corning and Levy 2002; González-Benito 2005; Mazzeo 2002; Vitorino
85
2007; Moe and Fader 2009; Zhu and Singh 2009; Zhu et al 2009). For example, Corning and
Levy (2002) specified venue-specific parameters when examining ticket sales across venues to
understand whether consumers of those venues have different responses to product offerings. In
the retail locations, Mazzeo (2002) specified type-specific effects of spatial competition to study
whether motels of low, medium, or high quality types have different spatial effects on
profitability. In addition, Zhu and Singh (2009) used brand-specific parameters to examine
asymmetric competition effects among Wal-mart, K-mart, and Target.
Although heterogeneity has been extensively studied in many contexts, most of prior
studies in the event tickets literature have not yet examined heterogeneous market responses. In
other words, researchers assume the effects of days of week, prices, and promotions are
homogeneous across venues, events, or performances (Moore 1966; Weinberg and Shachmut
1978; Havlena and Holak 1988; Reddy et al 1998). Although Corning and Levy (2002) and Moe
and Fader (2009) are the two exceptions where Corning and Levy (2002) allowed parameters to
be venue specific and Moe and Fader (2009) specified parameters varied with events and price
tiers, they did not identify explanatory factors for their proposed heterogeneity. Given that we
have found the heterogeneous market responses in the preliminary analysis, the objective of
essay two is to identify explanatory factors to explain the differences across markets.
3.3.2 Conceptual Framework
In our conceptual framework, we first discuss possible market characteristics that may
explain differences across markets. Then, we discuss characteristics of a touring event that may
provide context dependent reasons for response heterogeneity.
86
Market Characteristics
When the analysis is made at a market level rather than at an individual level, the first
issue is to define what a market is. Depending on the context of interest, a market can be a
metropolitan statistical area (MSA; Zhu et al 2007), a designated market area defined by A.C.
Nielsen (DMA; Carlyle, Slater, and Chakroff 2008), or a retail trade area (Bronnenberg and Mela
2004). Then, researchers try to find the market characteristics that may explain the difference to
some extent.
In general, metropolitan areas are assumed more similar to other MSAs than to rural
areas, and the similarity or differences may be due to the population size, population density,
income, education, household size, household values, commute time to work, etc. (Mazzeo 2002;
Vitorino 2007; Zhu et al 2007; Zhu and Singh 2009; Zhu et al 2009). For this reason, we
propose that market characteristics can explain the heterogeneous market responses in our
preliminary analysis.
Additional Scheduling Characteristics
We refer to additional scheduling characteristics as characteristics of a touring event. For
example, one characteristic is that its performing group travels from one market to another.
Because this distribution mechanism follows the sequential distribution approach (Elberse and
Eliashberg 2003; Bronnenberg and Mela 2004), we think sequential distribution literature may
provide context dependent characteristics to explain why consumers in different markets react to
performance schedules differently.
Extant works in sequential distribution have focused on the effect of geographic
adjacency on market adoption (Bronnenberg and Mela 2004) and the effect of release timing on
box-office revenues (Elberse and Eliashberg 2003). When Bronnenberg and Mela (2004)
87
studied the spatial evolution of a new product adoption across markets, they found manufacturers
tend to enter markets that are geographically adjacent to a current market. In other words, the
initial market serves as a lead market and its lead market effect rolls out sequentially to
geographically adjacent markets.
On the other hand, the release timing in the distribution also influences how well a
product sells. For example, Elberse and Eliashberg (2003) studied motion pictures to investigate
the issue of release timing between the U.S. market and foreign markets. Although they only
examined the effect of release timing between the initial market and following foreign markets
rather than the effect of release timing along the entire sequence, they still found that shortening
the time lag between two markets increases the revenues of a later market.
Therefore, the sequential distribution literature has traditionally discussed the roles of
geographic adjacency and release timing as important covariates. Whether the geographic
adjacency and release timing in the distribution explains different market responses still remains
unknown and deserves further investigation. Similarly, a live performance event follows a
temporal sequence to travel across markets. Each market along the sequence has different
release timing and some of these markets are geographically adjacent to one another. It is
important to evaluate its geographic adjacency and temporal sequence to understand whether
these additional scheduling characteristics explain different markets responses to performance
schedules. Therefore, we incorporate market characteristics and additional scheduling
characteristics to explain different market responses across markets.
In short, we summarize our conceptual framework in Figure 3-1. Similar to essay one,
we examine the impact of geographic and temporal scheduling on the number of ticket sales and
control for the performance attractiveness through its days of week. In addition, we also control
88
for the possible endogeneity in performance scheduling. Finally, we examine whether and how
market and additional scheduling characteristics explain response heterogeneity across markets.
Figure 3-1: Conceptual Framework
3.4 Model Development
3.4.1 Model Overview
Our model development consists of four steps. First, we use the geographic and temporal
density measures to capture the scheduling characteristics of performances in their associated
markets and understand how densely or sparsely these performances are scheduled. Second,
within each market, we specify market-specific parameters and model ticket sales at a
performance level as a function of these density measures and performance attractiveness. Third,
we control for a possible endogeneity between performance schedules and expected market
responses. Finally, we employ a hierarchical Bayesian (HB) approach to incorporate the
heterogeneity in market responses. Among these four steps, the first three steps are adapted from
essay one, yet the fourth step is the model extension in essay two.
Although the HB approach is not the only method to study heterogeneous market
responses and the latent class analysis (Kamakura and Russell 1989) may be another appropriate
alternative, we choose the HB approach because it can accommodate unobserved heterogeneity
across markets (Rossi and Allenby 2003) in addition to the heterogeneity explained by market
characteristics and additional scheduling characteristics.
3.4.2 Scheduling Characteristics and the Number of Ticket Sales
To begin with, we refer to equations (1) and (2) in essay one to capture the scheduling
characteristics by their geographic and temporal density for all performances in their markets.
89
Then, we refer to equation (5) to rewrite the sales share of performances (and the share of non-
buyers) within a market with market-specific parameters:
(16)
P
m
(]) =
exp(I
]m
) · 0E0
]m
0
1m
IHP
]m
0
2m
1 + ? exp(I
]mi
) · 0E0
]m?
0
1m
IHP
]m?
0
2m
]
m
j
|
=1
where
I
]m
= ?
0m
+?
1m
FRIBAY
jm
+?
2m
SAT0RBAY
jm
+?
3m
S0NBAY
jm
+ ?
jm
; ?
jm
~N(u, ?
s
2
)
where P
m
(j) is the sales share of performance j in market m, V
jm
is its performance attractiveness
(which is a function of days of week), GEO
jm
and TMP
jm
represent the geographic and temporal
density measures, and ?
0m
, ?
1m
, ?
2m
, ?
3m
, ?
1m
, and ?
2m
are market specific parameters. Therefore,
among J
m
performances in market m, their parameters are homogeneous within a market but
heterogeneous across markets. Using the population size in a target market (POP
m
) and the sales
share of a performance, we can calculate the expected ticket sales of a performance (i.e.,
Solcs
m
(]) = P0P
m
× P
m
(])), the market penetration rate of this event (i.e., ? P
m
(])
]
m
]=1
), and the
size of non-buyer segment (i.e., 1 - ? P
m
(])
]
m
]=1
).
3.4.3 Endogeneity in Performance Scheduling
Next, we account for a possible endogeneity in performance scheduling by revising
equation (14) as follows:
(17)
_
logit(0E0
]m
)
logit(IHP
]m
)
_ = ç
û
+ç
1
E |I
]m
] +ç
2
_
0
1m
0
2m
_ +q where p
ì
~N(u, o
q
i
2
I), i = 1,2
where E[V
jm
] is the expected performance attractiveness of performance j in market m, ?
1m
, and
?
2m
represent the effects of geographic and temporal scheduling, and ¢
0
, ¢
1
, and ¢
2
are the
parameters of interest. When ¢
1
or ¢
2
is significant, scheduling is endogenous but has been
taken into account. Note that the specification of equation (17) is the same as equation (14)
90
except that we specify an independent relationship between geographic and temporal density. In
this way, when an event performs in a single venue within a market and only has variation in its
temporal schedule, we can directly drop GEO
jm
in equations (16) and (17) and just investigate a
possible endogeneity in temporal scheduling.
3.4.4 Use of the HB Approach for Response Heterogeneity
Finally, we take the HB approach to explain differences of response parameters across
markets and specify these market-specific parameters as a function of their unobserved
heterogeneity and observed market characteristics and additional scheduling characteristics:
(18)
l
l
l
l
l
l
o
0m
o
1m
o
2m
o
3m
0
1m
0
2m
1
1
1
1
1
1
= A
m
= µX
m
+e where e~N(u, o
c
i
2
I), i = 1,2,S,4,S,6
where ?
m
is a 6-by-1 vector that contains the market-specific parameters (?
0m
, ?
1m
, ?
2m
, ?
3m
, ?
1m
,
and ?
2m
) in equation (16), X
m
is a k-by-1 vector that includes market characteristics and
additional scheduling characteristics, ? is a 6-by-k matrix that represents the effects for these
characteristics, and e is a vector of random errors for the unobserved heterogeneity.
3.5 Data
We contact a national ticket seller to obtain a dataset of two family events and analyze
ticket sales for one event in this dissertation (see §2.4.1 for more information about this event).
In short, this event sequentially performed 449 times in 50 cities in the U.S. domestic market
between January and June 2004. In essay one, we only used ticket sales in the New York
metropolitan market. In essay two, we analyze all performances in the dataset to investigate
heterogeneity in market responses.
91
3.5.1 Definition of Markets
Among 50 venues in the dataset, some venues are in the same MSAs, some are the only
venues in their MSAs, and others are in rural areas (i.e., non-MSAs). To avoid information
losses after aggregating venues to MSAs, we aggregate venues by their designated market areas
(DMAs) for urban and rural cities. The definition and classification of DMAs are proposed by
Nielsen Media Research (Weiner 2000) where each DMA consists of several counties and
consumers in the same DMA receive the same TV broadcasting and media messages (Carlyle,
Slater, and Chakroff 2008). The advantage of aggregating data by DMAs is that, suppose
marketing activities for an event are planned at a DMA level, consumers within the same DMA
are potentially aware of this event even though they may live in a rural area far away from a
venue.
Hence, we refer to a website by the Truck Ads
®
(www.truckads.com) that disaggregates
the U.S. market into several DMAs and lists all individual counties within each DMA. Figure 3-
2 provides an example of the Orlando DMA in Florida. As the map shows, there are nine
counties within this DMA. According to the venue locations in our dataset (i.e., names of venues,
cities, and states), we can identify in which DMA a venue is and which counties are in this DMA.
As a result, we aggregate 50 venues into 42 DMAs in Table 3-3. Figure 3-3 also shows the
locations of these DMAs. Each shaded area represents a DMA and a number in a box is its
market identification number assigned in Table 3-1.
Figure 3-2: Example of a DMA and its county information
Table 3-3: Venue Locations and their Associated DMAs
Figure 3-3: DMA locations
92
3.5.2 Description of Performance Schedules across Markets
According to the DMAs and performance information in our dataset, we summarize the
temporal schedule in each market into Table 3-1 by its first and last dates, length of performing
period, total number of performances, and venue usage in a market. On average, there were
10.69 performances in a market, lasted for 5.79 days, and used 1.19 venues.
Temporal Sequence of the Event Distribution
In Table 3-1, we sort the temporal schedules by their first performance date and observe
the temporal sequence of the event distribution. In general, this event first performed in
Tallahassee in January (market ID= 1), Atlanta in February (market ID= 12), New York in
March (market ID= 16), St. Paul in May (market ID= 33), and Tucson in June (market ID = 42)
3
.
Table 3-1 also indicates a few incidences where two or three markets started performances on the
same date or one to two days apart. For instance, the event had performances in Providence
(market ID= 26) and Worcester (market ID= 27) between May 1 and May 9 while having
performances in La Crosse (market ID= 25) between May 4 and May 5. Therefore, it is likely
that there were up to three performing groups touring in the same period. Because we cannot
identify which groups performed in which markets, we assume that the performing quality is
constant across performing groups and do not affect how markets respond to their performance
schedules. Hence, we analyze all 42 markets together regardless of their performing groups.
Venue Usage of the Event across Markets
In terms of the venue usage across markets, among 42 DMAs, only six markets had more
than one venue in use. These markets were the Greenville-Asheville, Raleigh-Fayetteville,
Norfolk-Hampton, Champaign-Springfield, New York, and Philadelphia DMAs. Moreover,
3
This event kept on travelling after performing in Tucson. However, the performance schedules available in the
dataset were truncated up to June 2004. Thus, we examine the performance schedules of these 42 markets only.
93
except for the New York market that scheduled performances across four venues, the rest had
performances scheduled in two venues. We also find that the event did not always perform in
these venues consecutively within a market. Sometimes the event offered all performances in
one market and then left for a new market, but sometimes it offered a few performances in one
venue and then provided more after a period. Specifically, this event completed all scheduled
performances consecutively in the markets of Norfolk-Hampton, Champaign-Springfield, and
Philadelphia. However, in the market of Greenville-Asheville and Raleigh-Fayetteville, after the
event performed in one venue, it left for other markets and then came back three to four months
later. In the New York market, it performed in three venues consecutively, left for other markets,
and then returned seven weeks later.
Geographic Adjacency of Markets
Upon a closer look of these markets and their locations in Figure 3-3, we also find that
this event had an extensive tour in the East Coast and some of the Mid-West markets. Moreover,
this event sometimes went to near or adjacent markets but occasionally traveled to an isolated
market. For example, this event went to Orlando (market ID=2) and Tampa (market ID=3) in a
consecutive order but went to an isolated market in Tucson (market ID=42).
Descriptions of Ticket Sales
Table 3-4 summarizes the average ticket sales at a performance level and its total ticket
sales at a market level. Across all markets, a performance can sell an average of 3,825 tickets
with a standard deviation being 1,162 tickets. However, depending on in which market a
performance is, it can sell as many as 8,316 tickets in New York (market ID= 16) or as little as
585 tickets in Madison (market ID= 28). In terms of ticket sales at a market level, an event can
sell an average of 51,905 tickets with a standard deviation being 90,992. Although it seems that
94
markets that use multiple venues experience more ticket sales (e.g., New York and Philadelphia
are ranked as the top two best selling markets), markets of a single venue usage do not
necessarily sell less. For example, the event performed in only one venue in Atlanta (market
ID=12) had market sales ranked at the third place. Moreover, Miami is also a single venue DMA
but its market sales exceeded the DMAs of Greenville-Ashville (market ID=9) and Champaign-
Springfield (market ID=20). Because the focus of essay two is on sales at performance level and
response heterogeneity across markets, we will address the issue of market sales in essay three.
Table 3-4: Summary of Ticket Sales across Markets
3.5.3 Covariate Specifications
Before we estimate our model, we still have to measure geographic and temporal density
as scheduling characteristics, create the days of week indicators for performance attractiveness,
and compute the population size of the target markets. Moreover, we need to select market
characteristics and measure the geographic adjacency and temporal sequence for the additional
scheduling characteristics. Hence, we discuss each covariate in turn.
Geographic and Temporal Density Measures
First, we calculate the geographic density for performances in markets (GEO
jm
) where
more than one venue is used. These markets are Greenville-Asheville, Raleigh-Fayetteville,
Norfolk-Hampton, Champaign-Springfield, New York, and Philadelphia. We start with
identifying venue locations on the Google Maps to compute the geographic distance (in miles)
between venues in the same market. Then, we apply the equation (1) in essay one to compute
the geographic density for each performance. As a result, depending on in which market a
performance is, the average geographic density of a performance ranges between 0.368 and
0.556 with the average across markets being 0.476 and the standard deviation being 0.09.
95
Next, we use the equation (2) in essay one to calculate the temporal density for
performances in their associated markets (TMP
jm
). On average, a performance in a market has
its temporal density ranging from 0.118 to 0.667 with the average across markets being 0.476
and the standard deviation being 0.09. Table 3-5 provides the descriptive statistics of these
covariates.
Table 3-5: Descriptive Statistics of Covariates across Markets
Days of Week Indicators and Market Population
According to observed performance dates in the dataset, we further create the days of
week indicators (FRIDAY
jm
, SATURDAY
jm
, and SUNDAY
jm
) to measure performance
attractiveness. On average, a market has 17% of performances on Friday, 38% on Saturday, and
25% on Sunday. However, there are markets without any Friday, Saturday, or Sunday
performances, as shown in those blank cells in Table 3-5. We also summarize the descriptive
statistics of days of week covariates in Table 3-5.
To compute a population size (POP
m
) in a target market (i.e., family population with
children under 10 years of age) across all DMAs, we refer to the Census Bureau to collect
relevant population information at the county level and then aggregate the population size by
DMAs, the same approach used in essay one. As a result, the average population size in a target
market is 439,662 with a standard deviation being 647,091. Table 3-6 presents the summary
information of the population size in each DMA.
Table 3-6: Descriptive Statistics of Market Characteristics
Market Characteristics
In addition to population size in a target market, we consider population density, family
income, and average family size as other potential market characteristics to explain response
96
heterogeneity. Different from the extant literature, we choose the information at a family level
rather than at a household level because our event targets at families with young children. Hence,
we first download the U.S. 2000 Census data at the county level to match the counties of interest
in our 42 DMAs. Because the population size in each county differs, we weight the market
characteristics in each county by its population size to calculate the average value for each DMA.
Table 3-6 presents the summary statistics across these DMAs.
Additional Scheduling Characteristics
We consider two additional characteristics of a performance schedule to explain the
heterogeneity across markets. The first characteristic we examine reflects the order of markets
that appear in a schedule. The second is the number of geographically adjacent markets with
respect to a focal market. We refer to the previous covariate as the temporal sequence and the
later as geographic adjacency.
To measure the temporal sequence, we refer to Table 3-1 that sorts markets by their first
day of performance to check which market is in the earliest distribution timeline (Order=1).
Then, we go down the list to assign an increasing number to markets in a later distribution
timeline. For example, according to Table 3-1, Tallahassee is the first market and Orlando is the
second market. We assign Order
1
=1 and Order
2
=2. For markets that had the first performance
on the same date (e.g., Columbia and Greenville-Asheville), we assigned an equal rank to these
markets (i.e., Order
8
= Order
9
= 8). Table 3-7 summarizes the descriptive statistics of temporal
sequence across markets.
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics
According to the lead market effect (Bronnenberg and Mela 2004), markets that are
adjacent and adopt a new product first have an impact on adjacent markets that have not yet been
97
adopted. We follow this logic to measure the geographic adjacency among our participating
DMAs. Hence, we first refer to Figure 3-3 to locate these 42 DMAs and check which markets
are adjacent to one another. Then, we refer to the order of each market to count how many
adjacent markets an event went to before it arrives to a focal market. Finally, we use this number
to represent the extent of geographic adjacency of a participating market. As a result, an event
went to an average of 0.95 spatially adjacent markets before it goes to a focal market. Table 3-7
summarizes the descriptive statistics of geographic adjacency across markets
4
.
The purpose of essay two is to explain heterogeneous market responses and propose
explanatory factors. Thus, we use the same set of covariates in the HB approach in equation (18)
and choose the population size in a target market, the order of markets that an event travels, and
the number of geographically adjacent markets for a focal market. We also examined other
market characteristics as shown in Table 3-6. Although population density is another significant
explanatory factor, it has the same effect as the population size. On the other hand, we find the
average family income and family size cannot explain any response heterogeneity across markets.
3.6 Model Estimation and Benchmark Comparison
3.6.1 Estimation
We choose the hierarchical Bayesian approach to estimate the number of ticket sales,
heterogeneous market responses, and endogeneity in performance scheduling simultaneously.
We specify appropriate and diffuse priors for our parameters in the WinBUGS program and
estimate the model over 10,000 iterations. After checking the convergence criteria, we examine
the autocorrelation plots for all covariates, discard the first 5,000 iterations for burn-in, and use
4
We also count how many adjacent markets a focal market has regardless of when the event performed in these
markets. However, this alternative measure did not explain any of the response heterogeneity.
98
the remaining iterations as the posterior distribution. We specify the prior distributions of
parameters below:
Priors for modeling performance attractiveness:
?
s
2
~I0(u.1,u.1)
Priors for modeling the heterogeneity in market responses:
[
kì
~N(u,1u) where k= 1, 2, 3, 4 and i=1, 2, 3, 4, 5, 6
o
c
i
2
~I0(u.1,u.1) where i=1, 2, 3, 4, 5, 6
Priors for modeling the endogeneity in performance scheduling:
¢
kì
~N(u,1u) where k= 0, 1, 2 and i=1, 2
p
ì
~N(u,1u) where k=1, 2
3.6.2 Benchmark Comparison
Before presenting our model results, we compare the model fit between the proposed and
benchmark models. Because our proposed model aims to explain the heterogeneous market
responses to performance schedules, we think a homogeneous model that does not account for
any heterogeneity to be an appropriate benchmark (i.e., Benchmark 1). In addition, we also
compare the proposed model with the preliminary analysis (i.e., Benchmark 2). After comparing
the model fit using the DIC measure (equation 15), we find that the DIC of the proposed model
is 5590.63. This fit measure is lower than the heterogeneous model without explanatory factors
(DIC=5613.94) and the homogeneous model (DIC=6927.14). Hence, our proposed model has
incorporated the market heterogeneity and explained the differences across markets.
99
3.7 Results
3.7.1 Heterogeneous Market Responses to Performance Schedules
According to the market-specific parameters, we find that markets respond to
performance schedules consistently. Although the effect sizes vary from markets to markets, all
expected values of parameters (i.e., effects of baseline performance attractiveness, days of week,
geographic density, and temporal density) are consistently positive or negative. Figure 3-4
visually presents these heterogeneous parameter values. Specifically, the parameter values of the
baseline performance attractiveness is consistently negative across markets (E[?
0m
] ranges from
-4.43 to -7.10) and performances on Friday, Saturday, or Sunday have incremental attractiveness
to generate more ticket sales (E[?
1m
] ranges from 0.08 to 0.39; E[?
2m
] ranges from 0.16 to 0.66;
E[?
3m
] rangers from 0.18 to 0.44).
Figure 3-4: Heterogeneous Market Responses
Moreover, markets have consistent responses to their geographic and temporal schedules.
For markets where multiple venues are in use, densely scheduled performances across venues
sell more tickets than sparsely scheduled performances (E[?
1m
] ranges from 0.81 to 2.13). In
contrast, sparsely scheduled performances along a time span sell more than densely scheduled
performances (E[?
2m
] ranges from -0.10 to -0.59). These results are consistent with results in
essay one and our preliminary analysis. Hence, we confirm the consistent (yet heterogeneous)
market responses to performance schedules.
3.7.2 Explanatory Factors for Market Heterogeneity
Our results in Table 3-8 report explanatory factors for the heterogeneity in market
responses. First, when a market has a bigger population, the effects of days of week and baseline
attractiveness are attenuated (?
11
= -0.25; ?
12
= -0.03; ?
13
= -0.05; ?
14
= -0.02). Our explanation is that
100
usually there are more events offered to a bigger market than to a smaller one. Consumers in a
bigger market may be used to seeing several competing events offered simultaneously and have a
variety of events to choose. Hence, they are less responsive to an event (no matter on which
days of week it is) than are consumers in a smaller market.
Table 3-8: Explanatory Factors for Market Heterogeneity
Second, we find that additional scheduling characteristics can partly explain
heterogeneous market responses. With respect to a current market, after an event travels to more
of its geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule (?
21
= -0.14; ?
26
= 0.10). A possible reason is that after an
event has gone to more spatially adjacent markets, its newness wears out but its reputation might
accumulate over time. As a result, consumers may refer to other measures such as word of
mouth to make their purchase decisions rather than refer to the baseline attractiveness and its
temporal schedule.
Moreover, after an event follows its temporal sequence to perform in several markets
(whether these markets are adjacent or not), a current market in a late distribution sequence tends
to respond more favorably to a Sunday performance (?
34
=0.10). This result is also graphically
shown in Figure 3-4 (d). Our explanation is that after an event has lasted longer and gone to
more markets, its reputation or word of mouth accumulates over time (Reddy et al 1981) even
though its newness may wear out. Because a Sunday performance tends to be the last
performance in a market (at least it is the case in our dataset), consumers might think Sunday as
their “last opportunity” to enjoy this event before it leaves for another market. As a result, a
market in a later temporal sequence has a stronger Sunday effect.
101
3.7.3 Endogenous Scheduling Decision
We also find an endogeneity in performance schedules. Different from essay one where
only endogeneity is found in a temporal schedule, in essay two we find that geographic and
temporal schedules are both done endogenously after we pool all performances across markets
for analysis. When event marketers expect high performance attractiveness, they schedule more
performances in all venues (?
11
= 0.18) and tend to allocate those performances around weekends
or along a limited time span (?
12
= 0.84). As a result, on average, performances have a shorter
geographic and temporal distance to others and have higher density values. Moreover, when
event marketers understand that consumers prefer performances on dispersed dates because of
uncertain timing of attendance, event marketers decrease the number of weekend performances
and/or disperse performance dates (?
22
= 1.30). Consequently, performances in a temporal
schedule have lower temporal density values. Since we have accounted for this endogeneity in
the estimation process, the results we present here are unbiased. Table 3-9 summarizes the
results of the endogeneity in performance scheduling.
Table 3-9: Endogenous Performance Schedules
3.8 Conclusions
3.8.1 Summary
In the first essay, we observe multiple performances of a single event and examine them
by their venue locations and performance dates to understand how their geographic and temporal
scheduling characteristics influence their ticket sales. However, event marketers often need to
make scheduling decisions for more than one market. Although our finding in essay one has rich
implications for event marketers, it is unclear whether event marketers can apply this finding to
102
all markets. Hence, the objective of essay two is to use all performance schedules of the same
event to investigate heterogeneous market responses and identify explanatory factors.
To accomplish our research objective, we first conduct a preliminary analysis and find
consistent scheduling effects. However, we also observe the market responses are of different
magnitudes. To identify the factors that explain these differences across markets, we extend the
model developed in essay one to not only examine market-specific response parameters but also
investigate observed and unobserved heterogeneity via the hierarchical Bayesian approach.
Among several marketing characteristics, we choose the size of market population as the first
explanatory factor. We also use additional scheduling characteristics along the distribution of
this event to examine whether geographic adjacency between markets and temporal sequence
along the distribution affect the magnitude of scheduling effects.
We use the same family event mentioned in essay one and aggregate the 50 cities it
travelled to 42 designated market areas. Our results show that market characteristics and
additional scheduling characteristics can both explain the differences across market responses.
First, when a market has a bigger population, the effects of days of week and baseline
attractiveness are attenuated. Second, with respect to a current market, after an event travels to
more of its geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule. Moreover, after an event follows its temporal sequence to
perform in several markets, a current market in a late distribution sequence tends to respond
more favorably to a Sunday performance. We also find an endogeneity in performance
schedules. Since we have accounted for this endogeneity in the estimation process, the results
we present here are unbiased.
103
3.8.2 Limitations and Next Steps
This essay examines heterogeneous market responses to performance schedules and
contributes to the event tickets literature by investigating explanatory factors. Results of this
essay provide a more generalizable scheduling guideline for event marketers and assist event
marketers in anticipating potential market response based on market characteristics and
additional scheduling characteristics.
However, the limitation of this essay is that we allow the additional scheduling
characteristics to explain the heterogeneity in market responses (i.e., parameter effects) but have
not yet directly examined whether additional scheduling characteristics in an event distribution
affect ticket sales across markets. As the sequential distribution literature suggests, preceding
markets tend to have some effects on later markets through their spatial adjacency (Bronnenberg
and Mela 2004) or time lag between release timing (Elberse and Eliashberg 2003). It is
important to examine whether sales in different markets are independent or not. We continue
discussing this issue in essay three.
104
Tables and Figures
Table 3-1: Summary of DMA markets
Market
ID
Market
First
Date
Last
Date
Number of
Performances
Number of
Show Dates
Number of
Venues
1 Tallahassee 1/1 1/4 6 4 1
2 Orlando 1/2 1/4 6 3 1
3 Tampa 1/7 1/11 8 5 1
4 Miami 1/8 1/18 16 10 1
5 Jacksonville 1/14 1/18 9 5 1
6 Birmingham 1/21 1/25 10 5 1
7 Nashville 1/22 1/25 8 4 1
8 Columbia 1/28 2/1 9 5 1
9 Greenville-Asheville 1/28 6/13 19 10 2
10 Raleigh-Fayetteville 2/5 5/23 17 9 2
11 Richmond 2/11 2/16 11 6 1
12 Atlanta 2/12 2/22 21 10 1
13 Norfolk-Hampton 2/18 2/29 19 10 2
14 Cincinnati 2/25 2/29 9 5 1
15 Charlotte 3/3 3/7 11 5 1
16 New York 3/3 6/6 70 32 4
17 Wheeling 3/17 3/20 7 4 1
18 Louisville 3/24 3/28 7 5 1
19 Terre Haute 4/6 4/7 4 2 1
20 Champaign-Springfield 4/9 4/25 11 6 2
21 Carbondale 4/13 4/14 3 2 1
22 Philadelphia 4/14 5/2 31 16 2
23 Charleston 4/15 4/18 8 4 1
24 Albany 4/29 5/2 8 4 1
25 La Crosse 5/4 5/5 4 2 1
26 Providence 5/5 5/9 9 5 1
27 Worcester 5/5 5/9 8 5 1
28 Madison 5/7 5/9 6 3 1
29 Hartford 5/12 5/16 9 5 1
30 Dayton 5/13 5/16 7 4 1
31 Mankato 5/18 5/19 3 2 1
32 Rochester 5/19 5/23 8 5 1
33 St. Paul 5/21 5/23 6 3 1
34 Cedar Rapids 5/25 5/26 4 2 1
35 Hershey 5/26 5/31 9 6 1
36 Sedalia 5/29 5/31 5 3 1
37 Memphis 6/2 6/5 6 4 1
38 Evansville 6/3 6/6 6 4 1
39 Wilkes-Barre 6/9 6/13 8 5 1
40 North Little Rock 6/16 6/20 8 5 1
41 Macon 6/23 6/27 8 5 1
42 Tucson 6/24 6/27 7 4 1
Average (STD) 10.69 (10.81) 5.79 (4.97) 1.19 (0.55)
105
Table 3-2: Summary of Market Responses to Performance Schedules
Market ID ?
0
?
1
?
2
?
3
?
1
?
2
1 -5.5 ** 0.3 1.32 ** 0.15 -1.8 **
2 -5.76 ** 1.19 ** 0.58 ** -0.79 **
3 -6.32 ** 0.75 ** 1.48 ** 0.88 ** -1.33 **
4 -6.97 ** 0.3 0.79 ** 0.61 ** -1.24 **
5 -5.96 ** 0.9 ** 1.33 ** 0.7 ** -1.69 **
6 -5.42 ** 0.62 ** 0.89 ** 0.22 -2.04 **
7 -5.99 ** 0.86 ** 1.42 ** 0.73 ** -1.4 **
8 -5.92 ** 0.95 ** 1.36 ** 0.59 ** -1.81 **
9 -7.17 ** 0.66 ** 0.94 ** 0.6 ** 0.96 ** -1.8 **
10 -6.16 ** 0.5 ** 0.81 ** 0.41 ** 4.29 ** -2.92 **
11 -5.66 ** 0.9 ** 1.3 ** 1.09 ** -2.38 **
12 -6.72 ** 0.24 0.5 ** 0.34 ** -1.2 **
13 -7.23 ** 0.54 ** 0.75 ** 0.48 ** 0.92 ** -0.84 **
14 -5.68 ** 0.4 * 0.91 ** 0.41 ** -1.74 **
15 -5.78 ** 0.69 ** 0.96 ** 0.53 ** -1.67 **
16 -8.23 ** 0.1 0.24 ** 0.2 ** 0.51 ** -1.16 **
17 -5.4 ** 0.94 ** 1.19 ** -2.22 **
18 -6.17 ** 0.56 * 1.21 ** 0.22 -1.18 **
19 -3.82 ** -2.22 **
20 -6.03 ** 0.40 ** 0.35 ** 2.09 ** -2.04 **
21 -4.43 ** -1.17 **
22 -7.18 ** 0.28 ** 0.37 ** 0.24 ** 0.99 ** -1.3 **
23 -5.68 ** 0.83 ** 1.27 ** 0.61 ** -1.8 **
24 -5.62 ** 0.89 ** 1.4 ** 0.79 ** -2.1 **
25 -4.79 ** -0.66 **
26 -5.63 ** 0.34 * 0.85 ** 0.27 * -1.71 **
27 -6.82 ** 0.53 ** 1.27 ** 0.71 ** -0.37 **
28 -6.1 ** 0.98 ** 0.29 -0.04
29 -5.86 ** 0.35 0.84 ** 0.29 * -1.32 **
30 -6.2 ** 0.7 ** 1.46 ** 0.9 ** -1.25 **
31 -3.83 ** -2.12 **
32 -6.19 ** 0.66 ** 1.34 ** 0.87 ** -1.41 **
33 -6.44 ** 1.05 ** 0.5 ** -0.59 *
34 -4.59 ** -0.98 **
35 -6.83 ** 0.45 ** 1.11 ** 0.71 ** -0.34 **
36 -6.35 ** 0.65 ** 0.77 ** 0.08
37 -6.13 ** 0.59 * 1.24 ** -1.08 **
38 -6.04 ** 0.35 1.16 ** 0.13 -0.88 **
39 -6 ** 0.51 ** 1.13 ** 0.6 ** -1.47 **
40 -6.06 ** 0.58 ** 1.19 ** 0.6 ** -1.51 **
41 -6.14 ** 0.48 * 1.22 ** 0.57 ** -1.34 **
42 -5.79 ** 0.54 ** 1.36 ** 0.82 ** -1.7 **
**: significant at the 95% highest posterior density
* : significant at the 90% highest posterior density
106
Table 3-3: Venue Locations and their Associated DMAs
Venue City State DMA Market
Pepsi Arena Albany NY Albany
Philips Arena Atlanta GA Atlanta
Bjcc Arena Birmingham AL Birmingham
Southern Illinois University Carbondale IL Carbondale
Us Cellular Center Cedar Rapids IA Cedar Rapids
Assembly Hall Champaign IL Champaign-Springfield
Prairie Capital Convention Center Springfield IL Champaign-Springfield
Charleston Civic Center Charleston WV Charleston
Charlotte Coliseum Charlotte NC Charlotte
Us Bank Arena Cincinnati OH Cincinnati
Colonial Center Columbia SC Columbia
Ej Nutter Center Dayton OH Dayton
Roberts Municipal Stadium Evansville IN Evansville
Asheville Civic Center Asheville NC Greenville-Asheville
Bi-Lo Center Greenville SC Greenville-Asheville
Hartford Civic Center Hartford CT Hartford
Giant Center Hershey PA Hershey
Jacksonville Veterans Memorial Arena Jacksonville FL Jacksonville
La Crosse Center Lacrosse WI Lacrosse
Louisville Gardens Louisville KY Louisville
Macon Centreplex Macon GA Macon
Alliant Energy Center Coliseum Madison WI Madison
Midwest Wireless Civic Center Mankato MN Mankato
Pyramid Arena Memphis TN Memphis
American Airlines Arena Miami FL Miami
Gaylord Entertainment Ctr Nashville TN Nashville
Continental Arena East Rutherford NJ New York
Madison Square Garden New York NY New York
Sovereign Bank Arena Trenton NJ New York
Nassau Coliseum Uniondale NY New York
Hampton Coliseum Hampton VA Norfolk-Hampton
Norfolk Scope Arena Norfolk VA Norfolk-Hampton
Alltel Arena North Little Rock AK North Little Rock
Td Waterhouse Centre Orlando FL Orlando
Boardwalk Hall Atlantic City NJ Philadelphia
Wachovia Spectrum Philadelphia PA Philadelphia
Dunkin' Donuts Center Providence RI Providence
Crown Center Of Cumberland Fayetteville NC Raleigh-Fayetteville
Rbc Center Raleigh NC Raleigh-Fayetteville
Richmond Coliseum Richmond VA Richmond
Blue Cross Arena Rochester NY Rochester
Mathewson Center Sedalia MO Sedalia
Xcel Energy Center St Paul MN St Paul
Tallahassee Leon County Civic Center Tallahassee FL Tallahassee
St Pete Times Forum Tampa FL Tampa
Hulman Center Terre Haute IN Terre Haute
Tcc Arena Tucson AZ Tucson
Wesbanco Arena Wheeling WV Wheeling
Wachovia Arena Wilkes-Barre PA Wilkes-Barre
Worcester Centrum Centre Worcester MA Worcester
107
Table 3-4: Summary of Ticket Sales
Market ID Market
Ticket Sales per Performance
Market Sales
Mean STD Min Max
1 Tallahassee 2362 1236 1659 4858 14174
2 Orlando 4348 396 3893 4952 26088
3 Tampa 6660 2056 2644 9294 53277
4 Miami 5192 2499 2150 10635 83075
5 Jacksonville 5038 1148 3357 6766 45341
6 Birmingham 8072 2048 5237 11214 80720
7 Nashville 6359 1944 3897 9274 50869
8 Columbia 3571 1147 2511 5707 32136
9 Greenville-Asheville 3475 2306 800 7716 66024
10 Raleigh-Fayetteville 5481 3480 1376 12747 93171
11 Richmond 6355 1240 3863 7703 69901
12 Atlanta 6611 1908 2499 9969 138827
13 Norfolk-Hampton 5209 1433 3095 8093 98977
14 Cincinnati 7788 2501 2638 10228 70092
15 Charlotte 7457 2647 2935 11459 82030
16 New York 8316 3525 1827 15810 582099
17 Wheeling 1848 450 993 2469 12939
18 Louisville 2734 317 2326 3173 19137
19 Terre Haute 3031 422 2545 3563 12124
20 Champaign-Springfield 1045 527 491 1910 11500
21 Carbondale 3182 1321 1838 4478 9546
22 Philadelphia 4497 2182 1548 8915 139421
23 Charleston 4407 651 3434 5442 35258
24 Albany 6489 1489 3456 8084 51914
25 La Crosse 907 219 603 1096 3629
26 Providence 5088 1065 2803 6216 45795
27 Worcester 3233 995 1521 4838 25864
28 Madison 585 210 278 914 3508
29 Hartford 5137 1258 2768 6925 46237
30 Dayton 2704 614 2011 3669 18928
31 Mankato 869 321 646 1237 2608
32 Rochester 2314 588 1472 3322 18513
33 St. Paul 1855 337 1516 2324 11130
34 Cedar Rapids 2037 287 1689 2356 8149
35 Hershey 758 147 574 982 6825
36 Sedalia 1088 203 855 1304 5442
37 Memphis 3135 321 2624 3539 18811
38 Evansville 972 429 507 1780 5834
39 Wilkes-Barre 2921 933 2018 4583 23369
40 North Little Rock 2879 827 2044 4281 23029
41 Macon 1183 369 735 1691 9461
42 Tucson 3465 824 2437 4379 24257
Market average 3825 1162 2098 5712
51905
(STD= 90992)
108
Table 3-5: Descriptive Statistics of Covariates across Markets
Market
ID
FRIDAY SATURDAY SUNDAY GEO TMP Population of
Target Market Mean STD Mean STD Mean STD Mean STD Mean STD
1 0.167 0.41 0.500 0.55 0.167 0.41 0.539 0.14 104326
2 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 477402
3 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 578505
4 0.125 0.34 0.375 0.50 0.250 0.45 0.299 0.04 741879
5 0.222 0.44 0.333 0.50 0.222 0.44 0.476 0.10 282479
6 0.200 0.42 0.300 0.48 0.200 0.42 0.470 0.07 321845
7 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 430888
8 0.222 0.44 0.333 0.50 0.222 0.44 0.476 0.10 169819
9 0.211 0.42 0.316 0.48 0.263 0.45 0.482 0.03 0.232 0.05 344151
10 0.235 0.44 0.353 0.49 0.176 0.39 0.492 0.09 0.245 0.05 463117
11 0.182 0.40 0.273 0.47 0.273 0.47 0.457 0.10 183972
12 0.190 0.40 0.286 0.46 0.286 0.46 0.307 0.04 1074597
13 0.211 0.42 0.316 0.48 0.263 0.45 0.497 0.03 0.296 0.04 345541
14 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 435717
15 0.182 0.40 0.273 0.47 0.273 0.47 0.473 0.07 466459
16 0.157 0.37 0.300 0.46 0.286 0.46 0.368 0.18 0.118 0.02 4100325
17 0.286 0.49 0.429 0.53 0.520 0.11 53729
18 0.143 0.38 0.429 0.53 0.143 0.38 0.470 0.13 298428
19 0.667 0.00† 66486
20 0.273 0.47 0.545 0.52 0.182 0.40 0.461 0.05 0.318 0.06 156173
21 0.667 0.14 155322
22 0.129 0.34 0.290 0.46 0.258 0.44 0.556 0.17 0.221 0.03 1321626
23 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 211135
24 0.250 0.46 0.375 0.52 0.250 0.46 0.530 0.09 225057
25 0.667 0.00† 86726
26 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 287211
27 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 1126579
28 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 147178
29 0.111 0.33 0.333 0.50 0.222 0.44 0.463 0.08 463979
30 0.143 0.38 0.429 0.53 0.286 0.49 0.532 0.12 240286
31 0.667 0.14 17074
32 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 186215
33 0.167 0.41 0.500 0.55 0.333 0.52 0.611 0.11 802661
34 0.667 0.00† 140908
35 0.111 0.33 0.333 0.50 0.222 0.44 0.438 0.11 316530
36 0.400 0.55 0.400 0.55 0.567 0.09 423202
37 0.167 0.41 0.500 0.55 0.506 0.13 347835
38 0.167 0.41 0.500 0.55 0.167 0.41 0.539 0.14 128297
39 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 227157
40 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 230179
41 0.125 0.35 0.375 0.52 0.250 0.46 0.470 0.11 109799
42 0.143 0.38 0.429 0.53 0.286 0.49 0.532 0.12 174998
Market
Average
0.172 0.40 0.382 0.51 0.250 0.46 0.476 0.09 0.476 0.09 439662
Note:
1. Markets without any observations in days of week are shown in blank cells.
2. Markets without any variation in their geographic schedules are shown in blank cells.
† Performances in market 19, 25, and 34 had equal temporal distance to other performances in the same market.
109
Table 3-6: Descriptive Statistics of Market Characteristics
Market Characteristics N Mean STD Min Max
POP
m
Population of target market 42 439662 647091 17074 4100325
POP_Density
m
Population density (in the log term) 42 5.717 0.85 3.787 7.615
Income
m
Family income (in the log term) 42 10.594 0.15 10.260 10.883
Family_Size
m
Average family size 42 3.028 0.09 2.857 3.276
110
Table 3-7: Descriptive Statistics of Additional Scheduling Characteristics
Additional Scheduling Characteristics N Mean STD Min Max
ADJ
m
Number of geographically adjacent
markets performing prior to a focal
market
42 0.95 1.00 0 4
ORDER
m
Distribution order in the temporal
sequence
42 21.43 12.3 1 42
111
Table 3-8: Sources of Heterogeneous Market Responses
Parameter Description Median (Std Dev)
Effect of Baseline performance attractiveness: E[?
0m
]
?
01
Intercept -4.30 (0.21)**
?
11
Population size in a target market -0.25 (0.06)**
?
21
Num. of geographically contiguous markets -0.14 (0.09)*
?
31
Num. of preceding markets in temporal sequence -0.12 (0.16)
Effect of Friday performances: E[?
1m
]
?
02
Intercept 0.31 (0.73)**
?
12
Population size in a target market -0.03 (0.01)**
?
22
Num. of geographically contiguous markets 0.00 (0.03)
?
32
Num. of preceding markets in temporal sequence 0.03 (0.06)
Effect of Saturday performances: E[?
2m
]
?
03
Intercept 0.51 (0.07)**
?
13
Population size in a target market -0.05 (0.01)**
?
23
Num. of geographically contiguous markets 0.01 (0.02)
?
33
Num. of preceding markets in temporal sequence 0.06 (0.05)
Effect of Sunday performances: E[?
3m
]
?
04
Intercept 0.21 (0.07)**
?
14
Population size in a target market -0.02 (0.01)**
?
24
Num. of geographically contiguous markets 0.00 (0.03)
?
34
Num. of preceding markets in temporal sequence 0.10 (0.05)**
Effect of Geographic density: E[?
1m
]
?
05
Intercept 1.22 (0.85)*
?
15
Population size in a target market -0.18 (0.32)
?
25
Num. of geographically contiguous markets -0.25 (1.19)
?
35
Num. of preceding markets in temporal sequence 0.86 (1.53)
Effect of Temporal density: E[?
2m
]
?
06
Intercept -0.61 (0.15)**
?
16
Population size in a target market 0.04 (0.04)
?
26
Num. of geographically contiguous markets 0.10 (0.06)**
?
36
Num. of preceding markets in temporal sequence 0.09 (0.11)
* significant at the 90% highest posterior density
** significant at the 95% highest posterior density
112
Table 3-9: Results of Performance Schedule Model
Parameter Description Median (Std Dev)
Expected geographic density: m
1
?
01
Intercept -0.06 (0.47)
?
11
Expected performance attractiveness 0.18 (0.06)**
?
21
Effect of geographic density 0.09 (0.14)
Expected temporal density: m
2
?
02
Intercept 3.35 (0.27)**
?
12
Expected value of performance 0.84 (0.05)**
?
22
Effect of temporal density 1.30 (0.18)**
** significant at the 95% highest posterior density
113
Figure 3-1: Conceptual Framework of Heterogeneous Market Responses
Attractiveness of
Performances
(1) Geographic
Density
At a performance level:
number of ticket sales
Temporal
Effect
Geographic
Effect
Scheduling Characteristics:
(2) Temporal
Density
Days of Week
Effects
Touring
Characteristics
Market
Characteristics
114
Figure 3-2: Example of a DMA and its county information
115
Figure 3-3: Locations of DMA Markets and Performing Sequence
116
Figure 3-4: Heterogeneous Market Responses
(a) Expected baseline effect: E[?
0
]
(b) Expected Friday effect: E[?
1
]
(c) Expected Saturday effect: E[?
2
]
(d) Expected Sunday effect: E[?
3
]
(e) Expected geographic effect: E[?
1
]
(f) Expected temporal effect: E[?
2
]
117
4 Essay 3: Sequential Distribution of a Live Performance Event
4.1 Introduction
When and where to schedule performances are two of the most important decisions
facing event marketers in the live entertainment industry. When event marketers schedule a tour
for an event, they have to design a performance schedule within each participating market and
determine an overall travel sequence across markets. Therefore, their scheduling decisions are
within and across markets and may have different effects on ticket sales.
In the first two essays, we have shown the effect of within-market scheduling and
identified explanatory factors for heterogeneous market responses. Specifically, we find that
venue locations in a geographic schedule influences ticket sales differently from do performance
dates in a temporal schedule. Densely scheduled performances across venues sell more tickets,
yet densely scheduled performances across times sell fewer tickets. Moreover, the population
size, geographic adjacency between markets, and temporal sequence in an event distribution can
explain heterogeneous market responses to some extent.
Because essays one and two have studied the effect of within-market scheduling and left
the impact of across-market scheduling unknown, essay three examines an event distribution
across markets and its impact on market sales. Specifically, an event distribution involves
scheduling across markets. Event marketers first decide a touring sequence at one time. Then, a
performing group follows this sequence to travel from one market to another. This group
performs in one market at a time, provides a few shows within a venue, and then leaves for
another market. Therefore, the mechanism of an event distribution is the same as the sequential
distribution.
118
Sequential distribution has been studied in marketing literature where researchers study
the market roll out of a new product (Bronnenberg and Mela 2004) and movie releases across
markets or channels (Lehmann and Weinberg 2000; Elberse and Eliashberg 2003; Chintagunta,
Gopinath, and Venkataraman 2009). These works show a dependent relationship between
preceding and following markets and indicate the effect of sequential distribution on sales or
profitability. In addition, they suggest the underlying reasons for the effect of sequential
distribution to be the lead market effect from geographically adjacent markets (Bronnenberg and
Mela 2004), word of mouth effect from previous markets (Elberse and Eliashberg 2003;
Chintagunta et al 2009), or effect of release timing between channels (Lehmann and Weinberg
2000). Hence, it is common to observe how well a new product sells in previous markets to
influence whether other markets adopt this product, when following markets launch this product,
and how well this product sells.
Similarly, when an event is distributed across markets, it is likely to see preceding
markets influencing following markets. This influence may come from geographic adjacency,
word of mouth, or release timing. Although essay two uses geographical adjacency between
markets and temporal sequence in a distribution to explain heterogeneous market responses, it
has not yet explored the possibility that markets may have a more direct dependent relationship.
Consequently, the objective of essay three is to examine whether an event distributed
sequentially across markets has an effect such that ticket sales in preceding markets can
influence sales in those following markets. We refer to such an effect as the carryover effect in
this essay.
To achieve this objective, we model ticket sales of each market as a function of its
performance schedule within a market and potential carryovers from an event distribution.
119
However, one modeling challenge is that scheduling decisions are correlated and endogenous
with the demand. For instance, the number of performances, the number of venues booked, and
the number of days scheduled may be highly correlated with one another. If we simply use these
scheduling decisions to explain market demand, these scheduling variables will be highly
correlated and suffer from the issue of collinearity.
To solve this issue, in addition to modeling supply and demand simultaneously to account
for the endogeneity, we have to use variables that are independent of one another yet still
represent the scheduling influences. Thus, in the demand model for event tickets, we use the size
of trading areas of booked venue(s) in a market as one instrumental variable for the scheduling
influence. We also use the distribution of performance dates in a schedule as the second
instrumental variable for the supply of an event on different dates.
On the other hand, we capture carryovers from an event distribution from the beginning
of its tour. Because an event travels to markets at different times and each market along the
distribution has different release timing, we employ the spatially weighted approach to account
for ticket sales of preceding markets as well as their release timing.
To specify the supply decisions simultaneously with the demand, we assume that event
scheduling within a market consists of three related decisions. First, event marketers consider
how many seats they have to provide in order to sell an expected number of tickets. We call this
decision an overall supply in a market. Next, they decide how many venues they need to reserve
given the number of seats needed in a market. This is the decision of venue usage. In the
meantime, they have to determine how many days they need to book given the number of seats
needed and the size of a venue capacity. We refer to this decision as the day usage decision. In
120
this way, we take into account the influence of expected demand on the supply decisions and use
variables to incorporate the scheduling influences on demand.
We contact a national ticket seller to obtain a dataset of live performance events and use
ticket sales for one event to test our model. This particular event sequentially performed 449
times in 50 cities in the U.S. domestic market between January and June 2004. Because we
aggregated these cities into 42 DMAs in essay two, we proceed to analyze ticket sales at a
market level in essay three.
The first finding of our demand model is that an event experiences more market sales
when more consumers are within the primary trading area of its venue(s). The intuition behind is
that when an event is more accessible to consumers and has more consumers within its primary
trading area, it provides more convenience to consumers due to a shorter travel distance. As a
result, an event with a larger trading area because of using more venues in a market can
accommodate more consumers and increase ticket sales.
Second, we show that an event sells more tickets when it has performances on various
dates in a market. In other words, when an event has more performances available to a market
and has a dispersed temporal schedule, it provides more flexibility to consumers especially for
those who have higher uncertainty about whether they could attend at a particular time.
Therefore, an event with a bigger variance in the distribution of performance dates sells more
tickets in a market.
Third, we show that an event distribution has a carryover effect on ticket sales. However,
the influence is across multiple venues within the same market but not across different markets.
In other words, when an event performs from one market to another, its ticket sales in preceding
markets do not affect sales in following markets. Yet, when this event performs in more than
121
one venue, its ticket sales in a preceding venue carry over to a later venue and influence its
overall market sales.
We think the nature of the family event analyzed in this essay is the underlying reason for
a carryover effect significant within a market but not across markets. Because this family event
targets young children and their parents, it is easier to observe children and parents discussing
event information within a market than across markets. Moreover, this family event travels
within the U.S. and often goes to a similar set of markets after a year or longer. Thus, these
markets do not have to depend on other participating markets but can rely on their own historical
experience to determine the quality of this event.
Finally, we show the endogeneity in the supply of an event. Event marketers use the
expected market demand to determine how many seats they need to provide, and this overall
supply further influences the number of venues and days they schedule the event.
The rest of essay three is organized as follows. We start with reviewing relevant
literature and constructing the conceptual framework. Next, we introduce our model
development and describe our data. After the model estimation and benchmark comparisons, we
discuss our results and conclude this essay.
4.2 Literature Review and Conceptual Framework
Essay three centers on literature in sequential distribution. Although we have reviewed
some relevant works in essay two, we discuss this literature in depth to show its mechanism and
possible effects in turn.
4.2.1 Sequential Distribution
Sequential distribution has been studied in movie and retailing contexts. The concept of
sequential distribution is that a new product starts in one channel or market and then gradually
122
distributes to another. Hence, as time passes by, the product availability increases and reaches
more consumers (Lehmann and Weinberg 2000). Based on where sequential distribution takes
place, we categorize extant works into two mechanisms: (1) sequential distribution across
channels and (2) sequential distribution across markets.
Sequential Distribution across Channels
Sequential distribution across channels refers to a new product released from one
channel to another, and it is a very common mechanism in the movie industry. One key
objective of research in this stream is to understand the impact of release timing of a new movie
title on its box-office revenues (Lehmann and Weinberg 2000; Hennig-Thurau et al 2006;
Hennig-Thurau, Houston, and Walsh 2007). In this way, researchers can suggest the optimal
release timing of a movie to another channel.
Specifically, Lehmann and Weinberg (2000) examined the optimal release timing from
movie theaters to video rental stores, and they found shortening the release timing (compared
with current practice) leads to increases in profits. On the other hand, Hennig-Thurau et al (2006)
studied revenue drivers in different movie channels (i.e., theater and video). They found that
release timing has a stronger influence on short-term box-office revenues than on long-term
revenues. Yet, release timing does not affect how well a movie sells on the video channel.
Another work by Hennig-Thurau et al (2007) examined the optimal release timing across
four distribution channels (i.e., theater, video purchase, video rental, and video on demand) and
further indicated that changing the order of distribution or shortening the release timing increases
profits for movie studios. To sum up, studies in this category emphasize the effect of time lags
between channels and show higher profitability due to a shortened time lag between channels.
123
Sequential Distribution across Markets
Sequential distribution across markets refers to a new product distributed from one
market to another, and it is observed in the movie and retailing industries. A primary objective
of this research stream is to investigate whether and how previous market success carries over to
later markets and influences market entry decisions or product sales in later markets (Elberse and
Eliashberg 2003; Bronnenberg and Mela 2004; Chintagunta et al 2009).
In terms of the movie industry, no matter whether movies are of limited or wide release
within a market, a common approach is to distribute movies sequentially across markets. For
instance, Elberse and Eliashberg (2003) investigated the relationship between the U.S. domestic
and a foreign market. Within a market, they concluded that box-office revenues of a movie title
and the number of screens in a theater in a preceding week affect the supply and demand in a
following week. They also found that total box-office revenues in the U.S market tend to carry
over to a foreign market and influence the supply and demand for the same movie in the opening
week. However, when they tested an interaction effect between the U.S. box-office revenues and
the time lag on a foreign market, they found this carryover effect only significant for the screen
management decision but insignificant for market demand.
Similarly, Chintagunta et al (2009) studied a sequentially released movie across the U.S.
local markets and examined the effect of online word of mouth (measured by the valance of
online reviews) from previous markets on the box-office revenues on the opening day in a new
market. They found that a time lag between an initial market and a current market negatively
affects sales, yet the average user rating of online reviews positively influences sales. However,
reviews are accumulated from the opening of a movie up to a current market. Researchers did
124
not incorporate the joint effect between the release timing and user rating generated in different
markets.
On the other hand, in the retailing context, Bronnenberg and Mela (2004) examined
manufacturers’ market entry decisions and retailers’ chain adoption decisions for newly
introduced frozen pizzas. They suggested that past market entry decisions influence whether a
manufacturer enters a new market, and past chain adoption decisions affect whether a chain in a
new market adopts this product. However, because the focus of this paper was on the supply
side, researchers did not investigate the effect of previous market entry or retail adoption on
demand in following markets.
Rationale for the Effect of Sequential Distribution
One primary reason for retailers or movie studios to practice sequential distribution is to
prevent financial losses of a failing launch (Lehmann and Weinberg 2000). Moreover, there are
several underlying reasons to explain why sequential distribution would be effective and
influence sales in later markets or channels.
For example, the success-breed-success effect (Elberse and Eliashberg 2003; Hennig-
Thuran et al 2006) shows a previous success in preceding markets or channels may be replicated
more easily in later entities. The word-of-mouth effect (Elberse and Eliashberg 2003; Hennig-
Thuran et al 2007; Chintagunta et al 2009) suggests that people exchange opinions and their
experiences influence how other people think. Moreover, the lead market effect (Bronnenberg
and Mela 2004) posits that similar behaviors tend to take place in spatially adjacent markets.
Thus, it is easier for adjacent markets to observe a focal market and imitate behaviors in this
focal market.
125
Summary
To sum up, studies in sequential distribution show a dependent relationship between
preceding and following markets and indicate the effect of sequential distribution on sales or
profitability. However, one limitation is that some, if not all, of these papers assume the decision
of sequential distribution is made one at a time rather than simultaneously (Lehmann and
Weinberg 2000; Elberse and Eliashberg 2003; Bronnenberg and Mela 2004; Hennig-Thuran et al
2006; Hennig-Thuran et al 2007). In other words, decisions of release timing, market entry,
chain adoption, and screen management are made sequentially after managers observe outcomes
(i.e., adoption decisions, box-office revenues, or profitability) from previous adoptions.
When managers have to plan a new product launch simultaneously for all participating
markets, it is not clear whether these markets still have a dependent relationship such that sales
in preceding markets carry over to following markets and influence their sales. Accordingly,
essay three contributes to this literature by studying an event distribution and its impact while a
touring sequence has to be planned at one time for an event to travel sequentially across markets.
4.2.2 Conceptual Framework
The objective of essay three is to examine whether markets along an event distribution
have a dependent relationship such that ticket sales of preceding markets have a carryover effect
to influence ticket sales in following markets. Hence, in our conceptual framework, we first
discuss how a performance schedule in a market influences its ticket sales. Then, we discuss
why preceding markets along an event distribution could influence following markets and what
the possible impact might be. Finally, we discuss the endogeneity between supply and demand
for an event.
126
Effects of Performance Schedule within a Market
When an event provides many performances in a market, its performance schedule is of a
relatively larger scale, compared with an event providing fewer performances. Among these
performances, if event marketers choose to book multiple venues and each venue is surrounded
by densely populated consumers, this event will be able to reach more consumers and have a
bigger primary trading area (Huff 1964) due to its enhanced spatial accessibility (Betancourt
2004). On the other hand, if these performances are at different times of day across various days
of week, this event will provide more flexibility to consumers and can deliver the performing
contents at consumers’ desired times (Betancourt 2004). In this way, the supply of a
performance schedule within a market influences how well an event sells in this market.
Effects of Carryovers from an Event Distribution
Moreover, as an event travels across markets and incurs a varying number of ticket sales,
it is possible to see preceding markets influencing following markets due to the effect of event
distribution. In other words, how well an event sells in a market may influence its sales in
adjacent markets via the lead market effect (Bronnenberg and Mela 2004). Addition, it is
possible that when an event sells well in one market, event marketers try to replicate this success
in another market because of the success-breed-success effect (Elberse and Eliashberg 2003;
Hennig-Thuran et al 2006).
On the other hand, it is also likely that consumers who have attended an event talk about
this event online or offline thus influencing people who have not yet attended. In this way, the
word of mouth of this event may travel across markets and influence people in different areas
(Elberse and Eliashberg 2003; Chingatunga et al 2009). Hence, the more ticket sales an event
127
experiences from previous markets, the higher is its volume of word of mouth, and the more
likely an event will sell well.
One special case occurs when an event travels to multiple venues of the same market. It
is possible that people who have gone to an event in a preceding venue express their opinions
about this event and influence other people in the same market. If so, an event distribution will
not only influence ticket sales across markets but also affect ticket sales across venues within the
same market. Consequently, the carryover effect from an event distribution may be across
markets as well as within a market (but across multiple venues).
Finally, although the population size of a market may influence the baseline market
demand, once we control for this market characteristic, a sequentially distributed event may still
influence its market sales through its performance schedule within a market and carryovers from
an event distribution.
Endogenous Supply and Demand of an Event Distribution
Meanwhile, because supply is often endogenous with demand, it is likely that the supply
of an event in a market influences its market sales, and the expected demand in this market
affects the supply of the same market. Therefore, our conceptual framework for an event
distribution must consider supply and demand simultaneously and allow the expected demand
and other scheduling constraints to influence the supply decision. Figure 4-1 below indicates the
endogenous relationship and summarizes the effects of performance schedule and event
distribution on market sales.
Figure 4-1: Conceptual Framework of Essay Three
Specifically, we assume event marketers make several scheduling decisions for a
touring event. Similar to the screen management decision in the movie industry (Elberse and
128
Eliashberg 2003; Basuroy, Desai, and Talukdar 2006; Eliashberg et al 2007), the first scheduling
decision is to determine the total number of seats an event needs to provide in individual markets.
Because the total number of seats is the maximum possible seats that an event can sell within a
market, we call this an overall supply decision. To endogenize the overall supply with market
sales, we assume event marketers rely on a size of market population and expected market
demand to set a desirable amount of supply. Hence, they may increase the overall supply when
they expect higher demand in a bigger market. In addition, after setting an overall supply of an
event, other scheduling issues are deciding how many venues to use and how many days to book
for this event. We call these decisions as venue usage and day usage and assume these decisions
are as a result of the market characteristics (e.g., population size or population density) and
scheduling constraints (e.g., venue capacity, venue availability, or facility rental fees).
4.3 Model Development
4.3.1 Overview
To test our conceptual framework, we model the supply and demand for an event
simultaneously. At the supply side, we model the total number of seats, venues, and days needed.
At the demand side, we specify ticket sales of each market as a function of its performance
schedule within a market and potential carryovers from an event distribution. However, one
modeling challenge is that scheduling decisions are correlated and endogenous with market
demand. For instance, the number of performances, the number of venues booked, and the
number of days scheduled in a market may be highly correlated with one another. If we simply
use these scheduling decisions to explain market demand, these covariates will be highly
correlated and suffer from the issue of collinearity.
129
To solve this issue, in addition to modeling supply and demand simultaneously to account
for the endogeneity, we have to use variables that are independent of one another yet still
represent the scheduling influences. Thus, in the demand model for event tickets, we use the size
of trading areas of booked venue(s) in a market as one variable for the scheduling influence. We
also use the distribution of performance dates in a schedule as the second variable for the supply
of an event on different dates.
Although our modeling approach is similar to that of Elberse and Eliashberg (2003) who
specified the number of screens and box-office revenues simultaneously for a sequentially
distributed movie, our approach differs from theirs because we model several elements in the
scheduling decisions in additional to just the capacity decision.
As such, we model the linerized supply and demand to follow the multivariate normal
distribution.
(19)
Y =
l
l
l
l
ln(Solcs
m
)
ln(Scots
m
)
ln(Icnucs
m
)
ln(Ðoys
m
) 1
1
1
1
~HIN__
y
1m
y
2m
y
3m
y
4m
_ , L_
where Sales
m
is the number of ticket sales in market m (i.e., market demand), Seats
m
is the total
number of seats supplied in market m (i.e., overall supply), Venues
m
is the number of venues
used in a market (i.e., venue usage), and Days
m
is the number of days available in a schedule (i.e.,
day usage). These dependent variables have expected values y
m
and a variance-covariance
matrix ?. In this way, the correlations between supply decisions and the correlation between
supply and demand are controlled in the variance-covariance matrix.
4.3.2 Demand Equation: Market Sales
According to equation (19), we model expected market sales as a function of its
performance schedule within a market and potential carryovers from an event distribution:
130
(20) y
1m
= o
0
+ X
1m
u
1
+X
2m
u
2
+Z
m
u
3
where y
1m
is the expected ticket sales (in the log term) in market m, X
1
is a vector of variables to
represent the scheduling influences, X
2
represents potential carryovers from an event distribution,
Z
m
is a vector of market characteristics used as control covariates, and ? is the vector of
associated parameter effects. We discuss the operationalization of these covariates in variable
specification.
4.3.3 Supply Equations: Overall Supply, Venue Usage, and Day Usage
Although some studies assume an exogenous supply decision (Swami et al 1999;
Eliashberg et al 2005; Eliashberg et al 2007; Chintagunta et al 2009), we propose an endogenous
and positive relationship between overall supply and market demand (Elberse and Eliashberg
2003; Basuroy et al 2006).
We assume that event scheduling within a market consists of several related decisions.
First, event marketers consider how many seats they have to provide in order to sell an expected
number of tickets. Hence, we specify the expected number of seats as a function of its expected
market demand and market characteristics:
(21) y
2m
= [
0
+?
1
y
1m
+Z
m
µ
2
where y
2m
is the expected number of seats (in the log term) provided, y
1m
is the expected market
sales (in the log term), Z
m
is a vector of market characteristics used as control covariates, and ?
is the vector of associated parameter effects.
Second, we assume the following decisions are to decide how many venues to reserve
and how many days to book. Specifically, given the amount of supply event marketers have to
provide, they can also refer to market characteristics such as market land area and population
density in a market to determine how many venues to book. They can also refer to the size of
131
venue capacity and the cost of venue rental to decide how many days are needed in order to be
cost effective. For example, if a market is densely populated and its land area is big, event
marketers may consider booking more venues to increase the spatial accessibility of this event.
On the other hand, if the average venue capacity is small and the daily cost of renting a venue is
high, event marketers may consider booking fewer days but scheduling more performances in a
day. We specify the expected venue and day usage as follows:
(22) y
3m
= y
0
+?
1
y
2m
+Z
m
y
2
y
4m
= 0
0
+ ?
1
y
2m
+Z
m
6
2
where y
2m
is the expected number of seats (in the log term) provided, y
2m
is the expected number
of venues (in the log term) needed, y
3m
is the expected number of days (in the log term) needed,
Z
m
is a vector of market characteristics, and ? and ? are the vectors of associated parameter
effects.
Similar to Elberse and Eliashberg (2003), the advantage of specifying supply and demand
equations in the way above is that parameters in equations (20), (21), and (22) represent
elasticity of covariates. These parameter values suggest how changes in their covariates result in
changes in demand and supply. We can also compare ? in equation (20) to rank the importance
of these covariates on market sales.
4.3.4 Variable Specifications
Because of the endogenous and correlated scheduling decisions, we aim to find
representative variables that are correlated with a performance schedule but do not have a severe
collinearity in the demand equation. Among several possible measures, we find the size of
trading area of venues and the flexibility in a temporal schedule may serve the modeling purpose.
The rationale is that when an event uses multiple venues in its performance schedule, it has a
132
bigger primary trading area and is more accessible for consumers within this area. In addition,
when this event has more performances at different times of day and on various days of week, it
provides consumers with higher flexibility to attend at their own convenience. Since these
benefits are relevant with a performance schedule but are not highly collinear, we propose three
variables and discuss how we operationalize these variables in turn.
Size of Trading Area
We follow the retail trading area in the retail location literature (Huff 1964; Applebaum
1966; Cliquet 1998) to compute the size of trading area of booked venue(s) in a market.
Specifically, we measure the size of trading area by referring to the population density around
each venue of a performance schedule. Because population density can determine the size of a
potential market (Huff 1964) yet it is not uniformly distributed (Donthu and Rust 1989), we
assume the zip code of each venue as its primary trading area and consumers living in the same
zip code to be the potential consumers. Therefore, the proxy of total consumers within the
primary trading areas is as follows:
(23)
Sizc
m
= Ðcnsity
¡
vcnuc
m
¡=1
where Size
m
is the size of trading area of selected venues, x
1m
is the total number of consumers
within the primary trading areas and Density
v
is the population density around the zip code of
venue v.
However, the level of people’s willingness to travel can expand or shrink the trading area
of a venue (Huff 1964; Applebaum 1966). In other words, when people have higher tolerance to
travel, a venue is able to reach more people and has a bigger trading area. To consider the factor
of travel tolerance, we further include an adjustment term and rewrite equation (23) as follows:
133
(24)
Sizc
m
= Jcnsity
¡
· z
¡
vcnucs
m
¡=1
where ?
v
is an adjustment term to represent consumers’ willingness to travel.
To approximate consumers’ willingness to travel, we use the average commute time as a
proxy. Our assumption is that consumers who spend more time commuting on a daily basis are
more willing to travel and have higher travel tolerance. Hence, we compute ?
v
as an index
relative to the average (Mazzeo 2002):
(25) z
¡
=
commutc
¡
mcon(commutc
¡
)
; v : e {m = 1,2, …, H]
where commute
v
is the average commute time (in minutes) for people living in zip code v and
mean(commute
v
) is the average. In this way, if ?
v
> 1, people are willing to travel farther, and
venue v has an expended trading area to reach more people. In contrast, if ?
v
< 1, the trading
area of venue v shrinks. Finally, if ?
v
= 1, the trading area of a venue is as it is.
Flexibility in a Temporal Schedule
Because the flexibility in a temporal schedule represents how easily consumers can attend
an event at different times of day and across various days of week, we use the distribution of
performance dates in a schedule as other variables. Specifically, we compute the average
number of performances per day for the flexibility during a day. In addition, among all
performances available to a market, we use its distribution and calculate its variance to represent
the flexibility during a week. Therefore, if an event has more performances per day and has a
bigger variance, consumers will have more flexibility to attend this event at their own
convenience.
134
Carryovers from an Event Distribution
As we mentioned in the conceptual framework, carryovers from an event distribution
may affect sales in later markets. Moreover, when an event performs in multiple venues within a
market, the carryover effect may also exist across venues but within this market. Therefore, we
measure across-market carryovers and within-market carryovers in this section.
To measure across-market carryovers since the beginning of an event distribution, we use
ticket sales from preceding markets as a proxy (Elberse and Eliashberg 2003; Bronnenberg and
Mela 2004). In this way, various amounts of ticket sales from preceding markets represent
different magnitudes of carryovers from these markets. However, because an event travels to
markets at a different time and each market along the distribution has different release timing, we
employ the spatially weighted approach to account for ticket sales of preceding markets as well
as their release timing (Yang and Allenby 2003; Bronnenberg and Mela 2004; Choi, Hui, and
Bell 2009).
(26)
AC
m
= W
1
Sales
m
= w
1mmi
Solcs
mi
M
m
|
=1
mi=m
where
w
1mmi
=
1
exp(J
1
(m, m'))
where AC
m
represents the amount of across-market carryovers up to market m, Sales
m
is a vector
of ticket sales that market m’ has occurred up to the beginning of market m (m’?m), and W
1
is a
spatial weight matrix where each element (W
1mm’
) is an exponentially weighted distance measure
between a preceding market m’ and a current market m (d
1
(m,m’)).
Although it is arguable that our across-market measure only considers the temporal
sequence of participating markets but not considers the geographic adjacency between markets,
we think our measure is better because this essay focuses on understanding the effect of an entire
135
distribution rather than the effect of spatially adjacent markets only. However, it is possible to
include the geographic adjacency between markets as an extension and study the effect of spatial
adjacency.
Therefore, before measuring d
1
in equation (26), we observe markets are temporally
adjacent in several ways. Figure 4-2 below illustrates various possibilities of market
connectedness, where a darker color represents a preceding market m’, a lighter color indicates a
current market m, and the width of a box is the length of a performing period. In our dataset, we
observe several ways of market connectedness. We classify them into three cases and discuss
each one in turn.
Figure 4-2: Illustration of Market Connectedness
Specifically, in the first case, Apart, performances in two markets are apart from each
other in a few days or are tightly connected. Case 1.a and Case 1.b in Figure 4-2 illustrates these
situations, respectively. In the second case, Overlap, performances in two markets have an
overlapped performing period. This overlapped period may be for a few days or for a longer
period. Case 2.a and Case 2.b in Figure 4-2 illustrates these situations, respectively. Note that
in the first two cases, an event only employs one venue to provide performances. If there are
multiple performing groups touring in the U.S. at the same time, it is possible to observe markets
without any time lag in between or markets have performances at the same time. Case 1.b, Case
2.a, and Case 2.b in Figure 4-2 represent these situations.
On the other hand, when an event performs in two markets and uses multiple venues in
one of the markets, it is likely that performances in a current market are in between two venues
of another market or overlaps with one of the venues. Case 3.a, In between a break, and Case
3.b, Overlap with one venue, in Figure 4-2 are examples of these situations. However, when an
136
event performs in two markets and both employ multiple venues, it is possible to see performing
periods overlapping in some venues yet apart in others (i.e., Figure 4-2, Case 3.c: between and
overlap). Therefore, Case 3.b and Case 3.c in Figure 4-2 exist if an event has more than one
performing group touring at the same time.
According to these different situations, we measure d
1
(in weeks) differently. In Case 1
(Apart), we measure d
1
by calculating the time lag (in days) between two markets and then
converting this value to week:
(27) J
1
(m, m
i
) = #oport 7 ?
where #apart indicates the time lag between two markets in days.
Case 2 represents a situation that markets have an overlapped performing period.
Because there are performances showing concurrently in two markets, we assume the carryovers
from these markets should have no decay but a stronger effect. In other words, markets of some
overlapped performing periods should have stronger influence on each other than markets that
are apart from each other. Therefore, overlapped markets should have higher spatial weights.
To do this, we allow d
1
(m,m’) in equation (26) to be negative such that its associated spatial
weight (W
1mm’
) is larger:
(28) J
1
(m, m
i
) = -1 × #o:crlop 7 ?
where #overlap indicates the number of days that two markets have overlapped performances.
This specification ensures overlapped markets in Case 2 have higher weights than markets in
Case 1.
Finally, in Case 3, venues are either in between a break (Case 3.a), overlap with one
venue (Case 3.b), or between and overlap (Case 3.c). We consider each condition separately and
measure d
1
accordingly:
137
• Case 3.a In between a break
Case 3.a is when performances in market m are in between the first and second venue of
market m’. Because by the time of market m, sales in the second venue of market m’ have not
yet occurred. Hence, d
1
(m,m’) is the time lag between performances in the first venue of market
m’ and those in the venue of market m. That is, Case 3.a is similar to Case 1.a. We measure d
1
using equation (27). However, the associated market sales (Sales
m’
) in equation (26) are up to
the end of first venue of market m’ only.
• Case 3.b Overlap with one venue
Case 3.b is when performances in market m overlap with the first venue of market m’.
This is a similar case to Case 2. Therefore, we measure d
1
using equation (28) yet sales for
market m’ are up to the end of first venue only.
• Case 3.c Between and overlap
Because Case 3.c is a combination of Case 1 and Case 2, we need to consider the number
of days markets are apart and overlapped at the same time. Hence, we allow #apart and #overlap
to cancel each other and measure d
1
accordingly:
(29) J
1
(m, m
i
) = |#oport + (-1 × #o:crlop)] 7 ?
To measure within-market carryovers from preceding venues in a market, we take the
same spatially weighted approach and assign a different weight based on the time lag to a current
venue v:
(30)
wC
m
= W
2
Sales
v
= w
2¡¡i
Solcs
¡i
¡,¡iem
where
w
2¡¡i
=
1
exp(J
2
, :'))138
where WC
m
is the within-market carryovers for market m (if it is a market that uses multiple
venues), Sales
v’
is a vector of ticket sales that a preceding venue v’ has occurred up to the
beginning of venue v, and W
2
is a spatial weight matrix to measure the temporal difference (d
2
)
between a current venue v and its preceding venue v’ in market m.
In a market where an event employs multiple venues, performances tend to take place
sequentially with a number of days apart from one another. Hence, we measure d
2
in a similar
way to Case 1 in equation (27):
(31) J
2
, :i
) = #oport 7 ?
4.3.5 Model Summary
To sum up, we model the supply and demand simultaneously. At the demand side, we
use three variables and two spatially weighted measures to represent the scheduling influences
from a performance schedule and carryovers from an event distribution. We also use the
population size as a market characteristic to control for the baseline effect on market demand. At
the supply side, we model the total number of seats, venues, and days an event needs to provide
for each market. We allow the expected market demand to influence the total number of seats
and assume event marketers book venues and days based on their overall supply. Meanwhile,
we take into account the potential effects of market characteristics and scheduling constraints in
the supply decision.
Finally, in case there are omitted yet correlated variables to affect supply and demand
jointly, we allow correlated error terms and estimate supply and demand simultaneously. We
rewrite our model specification as follow and present the relationships among these covariates in
Figure 4-3:
139
(32) Y =
l
l
l
l
ln(Solcs
m
)
ln(Scots
m
)
ln(Icnucs
m
)
ln(Ðoys
m
) 1
1
1
1
~HIN__
y
1m
y
2m
y
3m
y
4m
_ , L_
y
1m
= o
0
+o
1
ln(Sizc
m
) +o
2
ln(Intcnsity
m
) +o
3
ln(Iorioncc
m
)
+o
4
ln(AC
m
) +o
5
ln(wC
m
) +o
6
ln(P0P
m
)
y
2m
= [
0
+[
1
y
1m
+[
2
ln(P0P
m
)
y
3m
= y
0
+y
1
y
2m
+y
2
ln(Arco
m
) +y
3
ln(Ðcnsity
m
)
y
4m
= 0
0
+0
1
y
2m
+0
2
ln(Copocity
m
) +0
3
ln(Fccs
m
)
where
M = market 1,2,…,M market
Sales
m
= Number of ticket sales (market demand)
Seats
m
= Number of seats supplied (overall supply)
Venues
m
= Number of venues booked (venue usage)
Days
m
= Number of days scheduled (day usage)
Size
m
= Size of trading area of venues
Intenstiy
m
= Average number of performances per show day
Variance
m
= Variance in the distribution of performance dates
AC
m
= Across-market carryovers up to market m
WC
m
= Within-market carryovers in market m
POP
m
= Size of market population
Area
m
= Size of market land area (in square miles)
Density
m
= Population density in market m
Capacity
m
= Avg. capacity of selected venues in market m
Rental
m
= Avg. rental fees in market m
Figure 4-3: Overview of Model Development
4.4 Data
We contact a national ticket seller to obtain a dataset of live performance events and use
ticket sales for one event to test our model. This event sequentially performed 449 times in 50
140
cities in the U.S. domestic market between January and June 2004. We have aggregated the data
into 42 DMAs in essay two and analyze this aggregated dataset in essay three
5
.
4.4.1 Touring Sequence
We have discussed how the event travels across markets in essay two. Table 3-1 lists the
first and last dates of performances and the venue usage for each DMA and Figure 3-3 shows the
touring sequence graphically. Among these 42 DMAs, the event performed in multiple venues
in six DMAs where we observe two ways of within-market touring.
One approach is that the event first performed in one venue in a focal DMA, left for other
DMAs, and then returned to the focal DMA again but performed in a different venue.
Specifically, the event took this approach in the Greenville-Asheville DMA (market ID= 9), the
Raleigh-Fayetteville DMA (market ID= 10), and the Champaign-Springfield DMA (market
ID=20). In the Greenville-Asheville DMA, the event first performed in Greenville between
January 28 and February 1, left for other DMAs, and then returned to Asheville on June 9, 2004.
In the Raleigh-Fayetteville DMA, it performed in Raleigh between February 5 and February 9,
left for other DMAs, and then returned to Fayetteville on May 20, 2004. Finally, in the
Champaign-Springfield DMA, it performed in Springfield between April 9 and April 11 and then
in Champaign on April 23.
The other approach is that the event performed in venues within a market consecutively
and then left for other DMAs. For example, when it performed in the Norfolk-Hampton DMA
(market ID= 13), it first stayed in Norfolk between February 18 and February 22 and then went
to Hampton on February 25. In the Philadelphia DMA (market ID= 22), it first performed in
Philadelphia between April 14 and April 25 and then went to Atlantic City on April 28.
5
Essay one provides detailed discussion about the event and essay two describes how we aggregate sales data into
42 DMAs.
141
However, the tour for the New York DMA (market ID= 16) is a combination of these two
approaches. The event first performed in East Rutherford, Uniondale, and New York
consecutively between March 3 and April 11. Then, it left for other DMAs and finally returned
to Trenton on June 3.
4.4.2 Covariates in the Demand Model
Venue Usage and Size of Trading Area
In terms of the venue usage within a market, we find that a selected venue may not be in
the center of a DMA but it is often located in a densely populated area. For example, Figure 4-4
illustrates the venue locations of the six DMAs discussed above and shows the population
density around each venue. Therefore, it is reasonable to assume that the zip code of a venue is
the primary trading area for this venue and consumers living in this zip code are the potential
consumers in the trading area. As such, it is meaningful to use the population density to measure
the size of potential market.
Figure 4-4: Venue Locations and Surrounding Population Density
Specifically, to calculate the size of trading area for selected venues in a market, we refer
to the data collected from the U.S. Census Bureau for population, land area (in square miles), and
consumers’ travel time to work (in minutes) at a zip code level for all venues observed in the
dataset. We further divide the population by the size of land area for the population density of
each venue (Venue
v
). On average, there are 3,289 consumers in the same zip code of a venue
with the standard deviation being 4,239.
In addition, the average travel time to work across all observed venues is 20.70 minutes
with the standard deviation being 5.71. We divide travel time to work for each venue
(Commute
v
) by the sample average to get the index of travel tolerance (?
v
). Hence, the mean
142
travel tolerance across venues is 1 with its standard deviation being 0.28. Using the new
information above, we compute the size of trading area at a venue level. Then, for DMAs that
have multiple venues, we aggregate across venues to conclude that an average size of trading
area in a market is about 4,291 consumers with the standard deviation being 9,968. Table 4-1
indicates the descriptive statistics of the size of trading area across markets (in the log term).
Table 4-1: Descriptive Statistics of Covariates (in the log term)
Day Usage and Flexibility Measures
We summarize the touring dates across markets in Figure 4-5. On average, it took an
event 5.79 days performing in a market with the number of show days ranging from two to 32.
In terms of days of week for performances, we find that most markets (32 out of 42 DMAs) tend
to offer the last performance on Sunday. However, depending on the number of performances
offered in a market, some markets started the first performance on Wednesday (22 out of 42
DMAs), some on Thursday (10 out of 42 DMAs), and the rest on Tuesday or Friday. In other
words, the flexibility that an event provides with consumers to attend on various days of week is
heterogeneous across markets. Descriptively, nine DMAs have performances for two to three
days in a week, nine DMAs have performances across four days, 14 DMAs have performances
across five days, and 10 DMAs have performances more than six days.
Figure 4-5: Touring Dates across Markets
After examining the distribution of performance dates for each market, we find an
average market has its variance in the distribution of performance dates to be 199.71 days with
the standard deviation being 831.68. This skewed distribution is due to four DMAs that have
performances in multiple venues and have a long lag between venues. Therefore, if we exclude
143
these four DMAs, the average value of variance is 2.325 days with its standard deviation being
3.37.
In addition, on average, there are 1.79 performances available during a show day with the
standard deviation being 0.21. Although we do not know the specific times for day of our
performances, we still find the flexibility that an event provides with consumers to attend at
different times of day to be heterogeneous across markets. Table 4-1 indicates the descriptive
statistics of the flexibility in days of week and for times of day (in the log term).
Across-Market and Within-Market Carryovers
Figure 4-5 also illustrates how markets are temporally connected with one another.
According to venue usage and the first and last dates of performances, we classify 42 DMAs into
markets that are completely apart from one another (Case 1 in Figure 4-2), overlapped (Case 2),
or are in between multiple venues (Case 3). We further apply equations (27), (28), or (29) to
compute the time lag between markets to ensure all preceding markets have different spatial
weights in equation (26) and calculate across-market carryovers accordingly. Hence, the average
across-market carryovers are 73,963.88 with the standard deviation being 65,688.67.
Figure 4-5: Touring Dates across Markets
Similarly, we use equation (31) to calculate the time lag between venues within a market
for those six DMAs that have multiple venues. As a result, we have an average within-market
carryovers being 51,564.62 and its standard deviation being 73,673.07. Table 4-1 indicates the
descriptive statistics of the across-market and within-market carryovers (in the log term).
Population Size
The last covariate in the demand equation is the population size in each market. Because
we have computed the target market population in essay two, we include the descriptive statistics
144
in Table 4-1. The average population size in a market is 439,662 and the standard deviation is
647091.
4.4.3 Covariates in the Supply Model
Venue Capacity and Market Capacity
Before we study the overall supply in each market (Seat
m
) as the first supply decision, we
need to know the number of performances and the seating capacity for each venue. However, we
do not have information regarding venue capacity in our dataset, so we refer to venue websites
and the Wikipedia to collect seating capacity data.
According to venue websites, venues have various configurations for different events
(e.g., basketball games, hockey games, concerts, etc.). Hence, we choose the format that is the
closest to the setting of a family event and record the associated capacity as the venue capacity
(Capacity
v
). Using all venue capacity in the same market, we further compute the average venue
capacity in a market (Capacity
m
) for equation (32) and observe heterogeneity in venue capacity.
On average, a venue has 13,612 seats with standard deviation being 4,682. Fifty percent of
venues have capacity between 10,423 seats (quartile 1) and 17,315 seats (quartile 3).
Next, we multiply the venue capacity by the number of performances in this venue to
know the total number of seats supplied by this venue and then sum across all venues in the same
market to get the market capacity (Seats
m
). On average, an event offers 156,747 seats in a
market with its standard deviation being 184,537. Some markets only supplied a few seats (e.g.,
Mankato DMA offered 14,496 seats in a total) but some markets provided as many as 1,159,059
seats (i.e., New York DMA).
According to the market sales and market capacity, we find that at most 50% of market
capacity was filled. On average, only 24% of total capacity was filled in a market. Figure 4-6
145
presents the rate of filled capacity across markets. Table 4-1 also indicates the descriptive
statistics of the overall market supply (in the log term).
Figure 4-6: Capacity-Filled Rate across Markets
Market Land Area and Population Density
To study the number of venues a market needs to book in equation (32), we still need
information about the market land area and population density at a market level. Hence, we use
the U.S. Census Bureau statistics to obtain the market land area (in square miles) and the size of
population at a county level.
After we divide the county population by its land area, we understand the population
density at a county level. Then, we sum across all counties within the same DMA to get the
population density (Density
m
) at a market level. We also compute the land area in a market
(Area
m
) by summing the land area across counties in the same DMA. Therefore, an average
market has 3,846 square miles of land and its population density is 422 people per square mile.
Table 4-1 indicates the descriptive statistics of the market land area and population density (in
the log term).
Facility Rental Fees
We also use the facility rental fees (Fees
m
) to study the number of days event marketers
need to book in a market. However, because this information is not publicly available, we
assume rental fees of a venue are positively correlated with housing values in the same zip code.
In other words, when a median value of a single-family house is high, it is very likely that rental
fees of a venue in this zip code are also high. Hence, we collect housing market data from the
U.S. Census Bureau based on this assumption. The distribution of the housing value across
markets is skewed. The average value (in thousands) is 2055 and the standard deviation is 4855.
146
The median of the housing value is 94.2. Table 4-1 also summarizes the descriptive statistics of
this covariate (in the log term).
Correlation
Finally, before we estimate the proposed model, we also check the correlation among our
covariates and the correlation among all dependent variables. Table 4-2 reports the correlation
between covariates in the demand model and Table 4-3 indicates that dependent variables are
moderate or highly correlated. Hence, our proposed model that correlates all error terms among
dependent variables has accounted for this issue.
Table 4-2: Correlation Coefficient of Demand Covariates
Table 4-3: Correlation Coefficient of Dependent Variables
4.5 Estimation and Results
4.5.1 Model Estimation
We estimate demand and supply equations simultaneously using the Bayesian approach
where parameters are specified to follow diffuse prior distribution as follows:
Priors for model of market sales:
o
0
~N(1u,1u) and o
ì
~N(u,1uu) where i=1, 2, 3, 4, 5, 6
Priors for model of overall supply:
[
0
~N(11,1u) and [
ì
~N(u,1uu) where i=1, 2
Priors for model of venue usage:
y
0
~N(u.11,1u) and y
ì
~N(u,1uu) where i=1, 2, 3
Priors for model of day usage:
0
0
~N(1.S,1u) and 0
ì
~N(u,1uu) where i=1, 2, 3
Priors for the variance-covariance of Supply and Demand:
L
-1
~wcibull(I
4
, 4)
147
We run 30,000 iterations in WinBUGs. After checking the convergence criteria, we
check the autocorrelation plots for all covariates, discard the first 20,000 iterations for burn-in,
and use the remaining iterations as the posterior distribution.
4.5.2 Results of Demand Equation
Scheduling Influence in a Performance Schedule
First, we show that the size of trading area of selected venues in a market has a positive
effect on market sales. When an event performs in multiple venues in a market or when its
venues are located in densely populated areas, this event sells more tickets in this market
(?
1
=0.160). Our explanation is that scheduling performances in multiple venues or selecting
venues in densely populated area can improve the spatial accessibility of this event (Huff 1964;
Donthu and Rust 1989; Betancourt 2004). In this way, consumers living within the trading area
do not have to travel a longer distance to attend an event. When there are more consumers
within the trading area of a market, it is more likely for this market to sell more tickets (Huff
1964; Applebaum 1966; Cliquet 1988).
Second, we find the flexibility of performance dates in a schedule has a positive effect on
ticket sales. When the variance in the distribution of performance dates increases, an event
provides consumers with higher flexibility to attend on various days of week (Betancourt 2004)
and sell more tickets in a market (?
3
=0.235). However, when there are more performances in a
show day, the flexibility for times of day does not contribute to ticket sales.
Carryovers in an Event Distribution
Moreover, we show that a sequentially distributed event has an effect on ticket sales.
However, the influence is across multiple venues within the same market but not across different
markets. In other words, when an event performs from one market to another, its ticket sales in
148
preceding markets do not affect sales in following markets. Yet, when this event performs in
more than one venue within a market, its ticket sales in a preceding venue can carry over to a
later venue and influence its overall market sales (?
5
=0.083).
We think the nature of the family event analyzed in this essay is the underlying reason for
a carryover effect significant within a market but not across markets. Because this family event
targets young children and their parents, it is easier to observe children and parents discussing
event information within a market than across markets. Moreover, this family event travels
within the U.S. and often goes to a similar set of markets after a year or longer. Thus, these
markets do not have to depend on other participating markets but can rely on their own historical
experience to determine the quality of this event.
Our results are consistent with the results in Elberse and Eliashberg (2003). Specifically,
they used sales in a preceding week as the volume of word of mouth and found a significant
effect on box-office revenues in a following week within the same market. This is similar to our
carryover effect within a market except that our carryover effect is across venues but not over
time. On the other hand, when Elberse and Eliashberg (2003) measured the interaction effect
between market sales in a domestic market and the time lag between a domestic and foreign
market, they found this interaction, or weighted word of mouth, effect insignificant across
markets. Although our spatially weighted measure for the carryovers across markets is similar
to their measure except that we take into account all of the participating markets instead of just
the initial market, our results are consistent with their work and support their results.
Elasticity
Another advantage of our model specification is that parameter estimates in our demand
model suggest the elasticity for all covariates. After comparing elasticity across significant
149
covariates, we conclude that the variance in the distribution of performance dates is more
important than the size of trading area of selected venues and the carryovers of an event
distribution within a market, respectively in this order. Although market population has a higher
elasticity (?
6
= 0.264), this is a market characteristic that event marketers cannot manipulate in
their scheduling decisions. Furthermore, these elasticity values suggest event marketers which
factors to strengthen. If trade-offs among these factors have to be made, they can make rational
decisions based on the elasticity. We summarize the parameter estimates in our demand model
in Table 4-4.
Table 4-4: Results of the Demand Model
4.5.3 Results of Supply Equations
From the results of the supply model, we confirm the endogeneity between the supply
and demand for an event. Event marketers use the expected market demand to determine how
many seats they need to provide (?
1
=0.771) but not the size of market population. Moreover, in
terms of the venue usage, when event marketers need to increase their overall supply, they tend
to schedule performances in more venues (?
1
=0.473) but do not consider the size of market land
area or population density in this market.
On the other hand, when event marketers evaluate how many days to book for an event,
they consider how many seats they need to supply (?
1
= 0.929) and the average venue capacity in
a market (?
2
= -0.703). In other words, when they need to provide more seats to a market, they
book more days for this event and have a longer performing duration. However, the number of
days needed decreases with the venue capacity. Although we do not find average housing value
influences the decision of day usage, it is possible that this variable is not an appropriate proxy
150
for the rental facility fees in a market. Table 4-5 below summarizes the parameter values in the
supply equations.
Table 4-5: Results of the Supply Model
4.5.4 Correlated Demand and Supply
Finally, we find correlated error terms in the supply and demand models. In other words,
there are unspecified covariates affecting both supply and demand at the same time. The
correlation between the market sales and overall supply is 0.549 and the correlation between the
market sales and day usage is 0.499. On the other hand, the decisions of overall supply and day
usage are also correlated (corr(?
Seats
, ?
Days
)=0.823)). Hence, it is necessary to assume correlated
error terms in our model to avoid biased parameter estimates. Table 4-6 below indicates the
correlation coefficients among our supply and demand models.
Table 4-6: Correlation between Supply and Demand Models
4.6 Conclusions
4.6.1 Summary
When an event travels across markets and has its distribution sequence planned at one
time, it is not clear whether and how the sequential distribution of this event influences ticket
sales in each participating market. The objective of essay three is to examine whether markets
along an event distribution have a dependent relationship and whether ticket sales of preceding
markets have a carryover effect to influence ticket sales in following markets.
To achieve this objective, we model ticket sales of each market as a function of its
performance schedule within a market and potential carryovers from an event distribution.
Specifically, we employ three variables to represent the scheduling influences from various
venues and performance dates, and take the spatially weighted approach to capture carryovers of
151
participating markets that have different ticket sales and release timing. We also specify the
supply and demand for this event simultaneously to account for a likely endogeneity. At the
supply side, we model separate but correlated decisions of overall supply, venue usage, and day
usage. In this way, our proposed model provides better understanding of scheduling effects on
demand and control for the endogenous supply and demand.
We contact a national ticket seller to obtain a dataset of a live performance event and
analyze ticket sales at a market level. The first result indicates that an event experiences more
market sales when it plays in several venues and has a bigger trading area to accommodate more
potential consumers. Second, an event sells more tickets when its performances are dispersed
across days of week but not during many times a day. In other words, the flexibility in a
temporal schedule is along the days of week to attract more consumer attendance.
Third, we find a significant effect of carryovers from an event distribution. When an
event performs in multiple venues within a market, ticket sales in a preceding market carry over
to later venues and influence its market sales although this carryover effect does not exist across
participating markets. Finally, we find supply and demand for an event to be endogenous. Event
marketers use the expected market demand to determine how many seats they need to provide,
and this overall supply further influences how many venues they reserve and how many days
they book in a schedule.
4.6.2 Conclusion
This essay contributes to the sequential distribution literature by studying an event
distribution where its touring sequence is set at one time rather than sequentially. We show that
the impact of sequential distribution exists locally but not across markets. In other words,
although markets along a tour do not have a dependent relationship, venues of the same market
152
have a dependent relationship and preceding venues can influence ticket sales in later venues and
ticket sales in this market.
The methodological contribution of this essay is to employ variables for an endogenous
and correlated performance schedule. By modeling the supply and demand simultaneously and
having the variables in the demand equation, we ensure unbiased scheduling effects and provide
actionable implications for event marketers.
To sum up, the first implication of essay three lies in the decision of venue usage. After
event marketers decide how many venues to book, they can refer to the size of trading area of
each alternative venue and select among these venues accordingly. Moreover, they can add or
drop venues based on the size of trading area if the desirable number of venues is not feasible in
a market. The second implication lies in the decision of day usage. Event marketers should
consider not to allocate multiple performances in a day but to disperse performances across
various days of week. However, event marketers should still evaluate the overall costs for such
scheduling changes in any venue or day usage.
Finally, when a touring sequence has to be planned simultaneously, it is not necessary for
event markets to consider any dependent relationship across markets. However, if this event
performs in multiple venues in a specific market, it is preferred that event marketers schedule
this event in a more important venue prior to other venues. In other words, the scheduling
objective for a touring event should be to minimize the travel distance across markets but
maximize the within-market carryovers for venues in the same market.
153
Tables and Figures
Table 4-1: Descriptive Statistics of Covariates (in the log term)
Variable Label N Mean Std Dev Min Max
Covariates in the demand equation
Sales Number of ticket sales
42 10.18 1.15 7.87 13.27
Size Size of trading area
42 7.54 1.26 4.78 11.08
Intensity Avg. number of performances in a
show day
42 0.58 0.12 0.34 0.79
Variance Variance in the distribution of
performance dates
42 1.02 2.20 -1.10 8.44
AC Across-market carryovers
42 10.93 0.74 9.52 12.75
WC Within-market carryovers
6 5.16 8.33 -7.45 12.13
POP Population size
42 12.46 0.92 9.75 15.23
Covariates in the supply equations
Seats Number of seats supplied
42 11.59 0.85 9.58 13.96
Venues Number of venues booked
42 0.12 0.30 0 1.39
Days Number of days supplied
42 1.57 0.56 0.69 3.47
Area Size of market land area (in square
miles)
42 7.90 0.80 6.55 9.81
Density Population density of a market
42 5.72 0.85 3.79 7.62
Capacity Avg. capacity of selected venues in a
market
42 9.45 0.42 8.06 10.09
Rental Avg. housing value of a market (in
thousands)
42 5.24 1.85 3.60 9.71
154
Table 4-2: Correlation Coefficient of Demand Covariates
ln(Size) ln(Intensity) ln(Variance) ln(AC) ln(WC) ln(POP)
ln(Size) --
ln(Intensity) 0.41 --
ln(Variance) 0.52 0.22 --
ln(AC) -0.02 0.08 -0.00 --
ln(WC) 0.44 0.22 0.20 0.11 --
ln(POP) 0.63 0.28 0.45 -0.23 0.16 --
155
Table 4-3: Correlation Coefficient of Dependent Variables
ln(Sales) ln(Seats) ln(Venues) ln(Days)
ln(Sales) --
ln(Seats) 0.87 --
ln(Venues) 0.51 0.57 --
ln(Days) 0.80 0.86 0.73 --
156
Table 4-4: Results of the Demand Model
Parameter Effect Median (Std Dev)
ln(Sales): Expected ticket sales in a market
?
0
Intercept 5.329 (1.164) **
?
1
Effect of geographic coverage 0.161 (0.073) *
?
2
Effect of intensity 0.128 (0.168)
?
3
Effect of variance 0.235 (0.038) **
?
4
Effect of across-market carryover 0.002 (0.063)
?
5
Effect of within-market carryover 0.083 (0.021) **
?
6
Effect of population size 0.264 (0.086) **
** significant at the 95% highest posterior density
157
Table 4-5: Results of the Supply Model
Parameters Median (Std Dev)
ln(Seats): Expected number of seats supplied in a market
?
0
Intercept 4.766 (0.832) **
?
1
Effect of expected market sales 0.771 (0.086) **
?
2
Effect of population -0.073 (0.050)
ln(Venues): Expected number of venues needed
?
0
Intercept -4.621 (0.709) **
?
1
Effect of planned overall supply 0.473 (0.067) **
?
2
Effect of market land area -0.042 (0.046)
?
3
Effect of market population density -0.087 (0.050)
ln(Days): Expected number of days needed
?
0
Intercept -2.169 (1.097)
**
?
1
Effect of planned overall supply 0.929 (0.070)
**
?
2
Effect avg. venue capacity -0.703 (0.098)
**
?
3
Effect of housing value -0.024 (0.028)
** significant at the 95% highest posterior density
158
Table 4-6: Correlation between Supply and Demand Models
?
?
Sales
?
Seats
?
Venues
?
Sales
--
?
Seats
0.549
**
--
?
Venues
-0.169
-0.280 --
?
Days
0.499
**
0.823
**
-0.273
** significant at the 95% highest posterior density
159
Figure 4-1: Conceptual Framework
Market
Characteristics
Scheduling costs
& constraints
Carryovers across
and within a market
Endogeneity
Sales in Previous
Markets and Venues
Demand for an Event
Ticket sales
at a market level
Supply of an Event
(Scheduling Decisions)
1. Overall supply
2. Venue usage
3. Day usage
Figure 4
Case (1):
(1.a) A fe
Case (2):
(2.a) Littl
Case (3):
(3.a) In be
(3.c) Betw
4-2: Illustrati
Apart
w days apart
Overlap
e overlapped
In between a
etween a brea
ween and over
ion of Marke
a market of m
ak
rlap
et Connected
multiple ven
160
dness
(1.b) Tig
(2.b) Mu
nues
(3.b) Ov
ghtly connect
uch overlappe
verlap with on
t
ed
ne venue
previ
curren
ious market m
nt market m
’
161
Figure 4-3: Overview of Model Development
Size of
Trading Area
Market
population
Venue
capacity
Rental
facility fees
Population
density
Market land
area
Within?market
carryover
Sales
Across?market
carryover
Seats
Venues
Days
Intensity
Variance
?
1
?
2
?
3
?
4
?
5
?
6
?
1
?
2
?
1
?
2
?
3
?
1
?
2
?
3
162
Figure 4-4: Venue Locations
(a) Greenville-Asheville
(b) Raleigh-Fayetteville
(c) Norfolk-Hampton
(d) New York
(e) Champaign-Springfield
(f) Philadelphia
163
Figure 4-5: Touring Dates across Markets
12/30/2003
1/19/2004
2/8/2004
2/28/2004
3/19/2004
4/8/2004
4/28/2004
5/18/2004
6/7/2004
6/27/2004
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41
P
e
r
f
o
r
m
a
n
c
e
D
a
t
e
s
Market ID
164
Figure 4-6: Capacity-Filled Rate across Markets
0%
10%
20%
30%
40%
50%
60%
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41
%
C
a
p
a
c
i
t
y
F
i
l
l
e
d
Market ID
165
5 Conclusion
5.1 Summary
Scheduling is an important decision facing event marketers in the live entertainment
industry. When they schedule a tour for an event, they have to design a performance schedule
within each participating market and determine an overall travel sequence across markets.
Therefore, their scheduling decisions are within and across markets, which may have different
effects on ticket sales.
Although marketing research in the live entertainment industry has focused on
identifying drivers for ticket sales, researchers have not evaluated whether the scheduling
decisions influence how markets respond. In other words, researchers treat the supply and
demand for an event as two separate problems and they have not yet investigated the relationship
between supply and demand.
As such, this dissertation analyzes performance schedules of a live performance event
and examines the effects on ticket sales within and across markets. Specifically, essay one
investigates whether and how performances of similar scheduling characteristics sell differently
in terms of how many tickets each performance sells and when ticket sales arrive. We use the
venue locations and performance dates as the scheduling characteristics for each performance
and measure the similarity in these scheduling characteristics by the geographic distance between
venues and the temporal distance between performance dates.
Methodologically, we use the competing destination model to examine the number of
ticket sales and the Weibull hazard model for the timing of ticket sales. In addition, we also
control for a possible endogeneity between a performance schedule and it demand effect. Using
166
70 performances in the New York market, we show that performances sell differently depending
on how similar their venue locations and performance dates are.
In other words, when performances are in the same or nearby venues, they have higher
similarity in a geographic schedule. From consumers’ perspective, they may perceive nearby
venues to be more attractive due to this similarity in venue locations. As a result, more
consumers purchase tickets for those performances and they are more willing to purchase these
tickets early. On the other hand, when performances are on the same or closer dates, they have
higher similarity in a temporal schedule. Because consumers often have uncertainty about their
consumption state and prefer various attendance timing for choices, shorter temporal distance
between performances result in higher competition and sales cannibalization.
Since results in essay one are limited in the New York metropolitan market, essay two
analyzes performance schedules across 42 markets and examines whether the results in essay one
are heterogeneous across markets and if there are any explanatory factors to explain differences
across markets. We first conduct a preliminary analysis using the same model developed in
essay one and compare the effects of scheduling characteristics across markets. After confirming
the heterogeneous market responses to performance schedules, we employ the hierarchical
Bayesian approach to identify explanatory factors for differences across markets.
Our results show that market population, geographic adjacency between markets, and
temporal sequence in an event distribution can explain different market responses. First, when a
market has a bigger population, the effects of days of week and baseline attractiveness are
attenuated. Second, with respect to a current market, after an event travels to more of its
geographically adjacent markets, the current market is less responsive to its baseline
attractiveness and temporal schedule. Third, after an event follows its temporal sequence to
167
perform in several markets, a current market in a late distribution sequence tends to respond
more favorably to a Sunday performance.
As such, essay one examines the impact of a performance schedule in a single market,
and essay two analyzes all performance schedules observed in the dataset and uses market and
additional scheduling characteristics to explain different market responses. However, the event
in our dataset is sequentially distributed across its participating markets. It is important to
understand whether the carryover effect exists due to an event distribution and how this
carryover effect influences ticket sales of participating markets.
Therefore, essay three analyzes the distribution of this live performance event and
examines whether ticket sales of preceding markets carry over to following markets and
influence ticket sales in those following markets. Besides controlling for the effect of a
performance schedule within a market, we model ticket sales of each market as a function of its
potential carryovers from an event distribution. We also model the supply and demand for an
event simultaneously to account for a possible endogeneity. Specifically, we use the size of
trading area(s) of scheduled venue(s) in a market and flexibility in the distribution of
performance dates in a schedule to represent the scheduling influences and employ the spatially
weighted approach to incorporated carryovers of preceding markets and their different release
timing.
Our results show that when an event has a larger trading area in a market and/or offers
more performances along a dispersed time span, it tends to sell more tickets in this market.
Moreover, we show that an event distribution has an effect on ticket sales. However, the effect
of carryovers is significant across venues of the same market but not across markets. When an
168
event performs in more than one venue, its ticket sales in a preceding venue carry over to a later
venue and influence its overall market sales.
5.2 General Discussions
In general, this dissertation indicates several scheduling effects. We classify these effects
into effect of within-market scheduling and effect of across-market scheduling. Moreover, we
study the effect of within-market scheduling into two aspects. Essays one and two focus on the
effect of within-market scheduling at a performance level, yet essay three addresses the same
effect at a market level. We sum up these scheduling effects and discuss the differences in turn.
First, for the effect of within-market scheduling at a performance level in essays one and
two, we find that the effect of a geographic schedule differs from the effect of a temporal
schedule. The similarity in venue locations benefit ticket sales at a performance level but the
similarity in performance dates cannibalizes ticket sales. Moreover, market responses to
performance schedules are heterogeneous and can be explained via the market and additional
scheduling characteristics. Second, for the effect of within-market scheduling at a market level
in essay three, we find that an event sells more tickets when it employs more venues in its
geographic schedule and serve a bigger trading area. It also sells more when the distribution of
its performance dates in a temporal schedule has a big variance to offer consumers greater
flexibility in attendance timing. Third, for the effect of across-market scheduling in essay three,
we find that markets are not dependent on one another, but venues within the same market have a
dependent relationship to influence ticket sales in this market.
Although the effect of a geographic schedule at a performance level seems contradictory
to the effect at a market level, these results are in fact complimentary. Although the first two
essays suggest event marketers to decrease the geographic distance between performances and
169
increase the similarity in venue locations, this suggestion is for markets that employing multiple
venues only. When an event performs in just one single venue, event marketers can follow the
learning in essay three to select a venue that has the biggest trading area. In other words, if event
marketers decide to schedule an event in multiple venues, they can apply their learning in essay
three to select venues and then allocate performances to these venues based on the first two
essays.
Consequently, the order of these essays allows us to understand the scheduling effects
from a performance level to a market level. Essay one starts with examining the effect of a
performance schedule at a performance level in a single market and concludes that performances
scheduled closely in distance but distantly in time can experience more ticket sales. In addition,
essay one also suggests performances experiencing earlier timing of ticket sales when these
performances are scheduled in nearby venues. To test the generalizability of these results and
explain the heterogeneity across markets, essay two expands the scope of analysis and confirms
the scheduling impact in all participating markets of a touring event.
Finally, essay three examines whether an event has a carryover effect when it
sequentially distributes across markets. This essay concludes that markets do not influence one
another on their ticket sales yet their venues within the same market have such an effect.
Although one may argue that the third essay does not have to be conducted after the first two
essays, we choose this sequence to investigate the carryover effect after we can understand and
control for the effect of a performance schedule within a market.
Finally, although it is arguable that essay three could have used the density measures
developed in essay one, we choose to use three variables for the following reasons. First, the
scheduling characteristics represent the similarity between performances in a schedule. They do
170
not represent how well an event is able to serve its trading area at a market level or express the
flexibility in attendance timing in a temporal schedule. Second, the measure of geographic
density in essay one is applicable only when an event performs in multiple venues in a market.
For markets where an event performs in one venue only, there is no variation in its geographic
distance yet the trading area of this single venue could still influence ticket sales. Therefore, it is
necessary to use different measures to differentiate the effect of scheduling characteristics at a
performance level from the effect of a performance schedule at a market level.
5.3 Contributions
This dissertation has both empirical and academic contributions to the marketing field.
Empirically, we show that performance schedules do affect ticket sales. Managers can use
performance schedules to estimate ticket sales at a performance or market level. Event
marketers can use these estimates as benchmarks to monitor a pattern of ticket sales and even
allocate marketing resources accordingly.
Academically, the findings in this dissertation enrich literature in event tickets and
sequential distribution. We introduce new drivers of ticket sales to the event tickets literature
such that researchers can use new differential measures to explain variations in the number and
timing of ticket sales. Moreover, we examine a sequential distribution problem in a new context
where simultaneous planning is needed and find the effect of sequential distribution only within
a local market but not across markets.
5.4 Limitation and Future Research
The primary limitation in this dissertation is that we do not have access to consumer
identification data. Although individual transactions are observed, we cannot model a
consumer’s decision process to understand the effect of a performance schedule at a finer level.
171
To resolve this issue, we could apply an agent based modeling approach. Using our model
results as aggregated parameter values, we can further simulate individual consumers in a market
using the U.S. Census data and allow variations in agents’ preferences. This future direction will
better assist event marketers in performance scheduling and allow researchers to study marketing
problems using a complexity system.
Another future research lies in the pricing structure of event tickets. As our data suggest,
the total price that a consumer pays includes the face value, facility fees, and convenience
charges, where face value represents the highest share in the total price paid, followed by
convenience charges and facility fees. When summarizing consumers’ channel usage, we find
that consumers tend to purchase in box offices to avoid paying for convenience fees. However,
convenience fees are the major revenue source for ticket sellers. If ticket sellers and event
promoters could collaborate and re-structure the pricing breakdowns (e.g., the merger between
Ticketmaster, a primary ticket seller, and Live Nation, an event promoter), it is likely that
consumers’ ticket purchases will migrate to the Internet or other channels. This new topic
involves pricing and channel strategies and we leave it for a future direction.
172
Reference
Applebaum, William (1966), “Methods for Determining Store Trade Areas, Market Penetration,
and Potential Sales,” Journal of Marketing Research, 3 (2), pp.127-141.
Basuroy, Suman, Kalpesh Kaushik Desai, and Debabrata Talukdar (2006), “An Empirical
Investigation of Signaling in the Motion Picture Industry,” Journal of Marketing Research, 43
(2), pp.287-295.
Betancourt, Roger R. (2004), Economics of Retailing and Distribution, Edward Elgar Publishing
Ltd.
Bradlow, Eric T., Bart Bronnenberg, Gary J. Russell, Neeraj Arora, Sri Devi Duvvuri, Frankel
Ter Hofstede, Catarina Sismeiro, Raphael Thomadsen, and Sha Yang (2005), “Spatial Models in
Marketing,” Marketing Letters, 16 (3-4) pp.267-278.
Bronnenberg, Bart J. and Carl F. Mela (2004), “Market Roll-Out and Retailer Adoption for New
Brands,” Marketing Science, 23 (4), pp.500-518.
Cachon, Gérard P. (2004), “The Allocation of Inventory Risk in a Supply Chain: Push, Pull, and
Advance-Purchase Discount Contracts,” Management Science, 50 (2), p.222-238.
Carlyle, Kellie E., Michael D. Slater, and Jennifer L. Chakroff (2008), “Newspaper Coverage of
Intimate Partner Violence: Skewing Representations of Risk,” Journal of Communication, 58 (1),
pp168-186.
Chintagunta, Pradeep K., Shyam Gopinath, and Sriram Venkataraman (2009), “Online Word-of-
mouth Effecs on the Offline Sales of Sequentially released new products: An Application to the
Movie Market,” working paper.
Choi, Jeonghye, Sam K. Hui, David R. Bell (2009), “Spatio-Temporal Analysis of Imitation
Behavior in Adoption of an Online Grocery Retailer,” Journal of Marketing Research,
forthcoming.
Chu, Wujin and Woosik Chu (1994), “Signaling Quality by Selling through a Reputable Retailer:
an Example of Renting the Reputation of Another Agent,” Marketing Science, 13 (2), pp.177-
189.
Cliquet, Gerard (1998), “Integration and territory coverage of the hypermarket industry in France:
A relative entropy measure,” International Review of Retail, Distribution and Consumer
Research, 8 (2), pp.205-224.
Corning, Jonathan and Armando Levy (2002), “Demand for Live Theater with Market
Segmentation and Seasonality,” Journal of Cultural Economics, 26 (3), pp.217-235.
173
Currim, Imran S., Charles B. Weinberg, Dick R. Wittink (1981), “Design of Subscription
Programs for a Performing Arts Series,” Journal of Consumer Research, 8 (2), pp.67-75
Desiraju, Ramarao and Steven M. Shugan (1999), “Strategic Service Pricing and Yield
Management,” Journal of Marketing, 63 (1), pp.44-56.
Dobson, Gregory and Phillip J. Lederer Lederer (1993), “Airline Scheduling and Routing in a
Hub-and-Spoke System,” Transportation Science, 27 (3), pp.281-297.
Donthu, Naveen and Roland T. Rust (1989), “Estimating Geographic Customer Densities Using
Kernel Density Estimation,” Marketing Science, 8 (2), pp.191-203.
Dorndorf, Ulrich, Florian Jaehn, Chen Lin, Hui Ma, and Erwin Pesch (2007), “Disruption
Management in Flight Gate Scheduling,” Statistica Neerlandica, 61 (1), pp.92-114.
Elberse, Anita and Jehoshua Eliashberg (2003), “Demand and Supply Dynamics for Sequentially
Released Products in International Markets: The Case of Motion Pictures,” Marketing Science,
22 (3), pp.329-354.
Eliashberg, Jehoshua, Sanjeev Swami, Charles B. Weinberg, and Berend Wierenga (2009),
“Evolutionary Approach o the Development of Decision Support Systems in the Movie Industry,”
Decision Support Systems, 47 (1), p.1-12.
____, Anita Elberse, and Mark A.A.M. Leenders (2006), “The Motion Picture Industry: Critical
Issues in Practice, Current Research, and New Research Directions,” Marketing Science, 25 (6),
pp.638-661.
____, Quintus Hegie, Jason Ho, Dennis Huisman, Steven J. Miller, Sanjeev Swami, Charles B.
Weinberg, and Berend Wierenga (2007), “Demand-Driven Scheduling of Movies in a Multiplex,”
working paper.
Etschmaier, Maximilian M. and Dennis F. X. Mathaisel (1985), “Airline Scheduling: An
Overview,” Transportation Science, 19 (2), pp.127-138.
Fotheringham, A. Stewart (1983), “Some theoretical aspects of destination choice and their
relevance to production-constrained gravity models,” Environment and Planning A, 15 (8),
pp.1121-1132.
____ (1988), “Consumer Store Choice and Choice Set Definition,” Marketing Science, 7 (3),
pp.299-310.
Gallego, Guillerimo and Garrett van Ryzin (1994), “Optimal Dynamic Pricing of Inventories
with Stochastic Demand over Finite Horizons,” Management Science, 40 (8), pp.999-1020.
González-Benito, Óscar (2005), “Spatial Competitive Interaction of Retail Store Formats:
Modeling Proposal and Empirical Results,” Journal of Business Research, 58 (4), pp.457-466.
174
Havlena, William J. and Susan L. Holak (1988), “The Influence of Variety on the Demand for
Bundles of Musical Performances,” Advances in Consumer Research, 15, pp.22-26.
Hennig-Thurau, Thorsten, Mark B. Houston, Gianfranco Walsh (2006), “The Differing Roles of
Success Drivers Across Sequential Channels: An Application to the Motion Picture Industry,”
Journal of the Academy of Marketing Science, 34 (4), pp.559-575.
____, Victor Henning, Henrik Sattler, Felix Eggers, and Mark B. Houston (2007), “The Last
Picture Show? Timing and Order of Movie Distribution Channels,” Journal of Marketing, 71 (4),
pp.63-83.
Huber, Joel, John W. Payne, and Christopher Puto (1982), “Adding Asymmetrically Dominated
Alternatives: Violations of Regularity and the Similarity Hypothesis,” Journal of Consumer
Research, 9 (1), pp.90-98.
____ and Christopher Puto (1983), “Market Boundaries and Product Choice: Illustrating
Attraction and Substitution Effects,” Journal of Consumer Research, 10 (1), pp.31-44.
Huff, David L. (1964), “Defining and Estimating a Trading Area,” Journal of Marketing, 28 (3),
pp.34-38.
Jank, Wolfgang and P.K. Kannan (2005), “Understanding Geographic Markets of Online Firms
Using Spatial Models of Consumer Choice,” Marketing Science, 24 (4), pp.623-634.
Jarrah, Ahmad I., Jon Goodstein, and Ram Narasimhan (2000), “An Efficient Airline Re-
Fleeting Model for the Incremental Modification of Planned Fleet Assignments,” Transportation
Science, 34 (4), pp.349-362.
Kamakura, Wagner A. and Gary J. Russell (1989), “A Probabilistic Choice Model for Market
Segmentation and Elasticity Structure,” Journal of Marketing Research, 26 (4), pp. 379-390.
Lehmann, Donald R. and Charles B. Weinberg (2000), “Sales Through Sequential Distribution
Channels: An Application to Movies and Videos,” Journal of Marketing, 64 (3), pp.18-33.
Leslie, Phillip (2004), “Price Discrimination in Broadway Theater,” RAND Journal of
Economics, 35 (3), pp.520-541.
Lohatepanont, Manoj and Cynthia Barnhart (2004), “Airline Schedule Planning: Integrated
Models and Algorithms for Schedule Design and Fleet Assignment,” Transportation Science, 38
(1), pp.19-32.
Luce, R. Duncan (1959), Individual Choice Behavior, New York: John Wiley & Sons, Inc.
Lutz, Nancy A. and V. Padmanabhan (1995), “Why do We Observe Minimal Warranties?”
Marketing Science, 14 (4), pp.417-441.
175
Manchanda, Puneet, Peter E. Rossi, and Pradeep K. Chintagunta (2004), “Response Modeling
with Nonrandom Marketing-Mix Variables,” Journal of Marketing Research, 41 (4), pp.467-478.
Mazzeo, Michael J. (2002), “Product Choice and Oligopoly Market Structure,” RAND Journal of
Economics, 33 (2), pp.221-242.
McCardle, Kevin, Kumar Rajaram, Christopher S. Tang (2004), “Advance Booking Discount
Programs Under Retail Competition,” Management Science, 50 (5), pp.701-708.
McFadden, Daniel L. (1974), “Conditional Logit Analysis of Qualitative Choice Behavior,” in
P.Zarembka, ed., Frontiers in Econometrics, Academic Press, New York, pp.105-142.
McGill, Jeffrey I. and Garrett J. van Ryzin (1999), “Revenue Management: Research Overview
and Prospects,” Transportation Science, 33 (2), pp.233-256.
Moe, Wendy W. and Peter S. Fader (2002), “Using Advance Purchase Orders to Forecast New
Product Sales,” Marketing Science, 21 (3), pp.347-364.
____ and ____ (2009), “The Role of Price Tiers in Advance Purchasing of Event Tickets,”
Journal of Service Research, forthcoming.
Moore, Thomas Gale (1966), “The Demand for Broadway Theatre Tickets,” The Review of
Economics and Statistics, 48 (1), pp.79-87.
Moorthy, Sridhar and Kannan Srinivasan (1995), “Signaling Quality with a Money-Back
Guarantee: The Role of Transaction Costs,” Marketing Science, 14 (4), pp.442-466.
Ntzoufras, Ioannis (2009), Bayesian Modeling Using WinBUGS, John Wiley & Sons, Inc.,
Hoboken, New Jersey.
Okada, Erica Mina (2005), “Justification Effects on Consumer Choice of Hedonic and Utilitarian
Goods,” Journal of Marketing Research, 42 (1), pp.43-53.
Orhun, A. Yesim (2005), “Spatial Differentiation in the Supermarket Industry,” working paper.
Padmanabhan, V. and Ram C. Rao (1993), “Warranty Policy and Extended Service Contracts:
Theory and an Application to Automobiles,” Marketing Science, 12 (3), pp.230-247.
Prelec, Drazen and George Loewenstein (1998), “The Red and the Black: Mental Accounting of
Savings and Debt,” Marketing Science, 17 (1), pp.4-28.
Purohit, Devavrat and Joydeep Srivastava (2001), “Effect of Manufacturer Reputation, Retailer
Reputation, and Product Warranty on Consumer Judgments of Product Quality: A Cue
Diagnosticity Framework,” Journal of Consumer Psychology, 10 (3), pp.123-134.
176
Putler, Daniel S. and Shilpa Lele (2003), “An Easily Implemented Framework for Forecasting
Ticket Sales to Performing Arts Events,” Marketing Letters, 14 (4), pp.307-320.
Reddy, Srinivas K., Vanitha Swaminathan, and Carol M. Motley (1998), “Exploring the
Determinants of Broadway Show Success,” Journal of Marketing Research, 35 (3), pp.370-383.
Rossi, Peter E. and Greg M. Allenby (2003), “Bayesian Statistics and Marketing,” Marketing
Science, 22 (3), pp.304-328.
Seim, Katja (2006), “An Empirical Model for Firm Entry with Endogenous Product-Type
Choices,” RAND Journal of Economics, 37 (3), pp.619-640.
Shugan, Steven M. and Jinhong Xie (2000), “Advance Pricing of Services and Other
Implications of Separating Purchase and Consumption,” Journal of Service Research, 2 (3),
pp.227-239.
Simonson (1989), “Choice Based on Reasons: The Case of Attraction and Compromise Effects,”
Journal of Consumer Research, 16 (3), pp.158-174.
Soberman, David A. (2003), “Simultaneous Signaling and Screening with Warranties,” Journal
of Marketing Research, 40 (2), pp.176-192.
Spiegelhalter, David J., Nicola G. Best, Bradley P. Carlin, and Angelika van der Linde (2002),
“Bayesian Measures of Model Complexity and Fit,” Journal of the Royal Statistical Society:
Series B (Statistical Methodology) , 64 (4), pp.583-639.
Swami, Sanjeev, Jehoshua Eliashberg, and Charles B. Weinberg (1999), “SliverScreener: a
Modeling Approach to Movie Screens Management,” Marketing Science, 18 (3), pp.352-372.
Talluri, Kalyan and Garrett van Ryzin (2004), “Revenue Management Under a General Discrete
Choice Model of Consumer Behavior,” Management Science, 50 (1), pp.15-33.
Tang, Christopher S., Kumar Rajaram, Aydin Alptekinoglu, and Jihong Ou (2004), “The
Benefits of Advance Booking Discount Program: Model and Analysis,” Management Science,
50 (4), pp.465-478.
Thomadsen, Raphael (2007), “Product Positioning and Competition: The Role of Location in the
Fast Food Industry,” Marketing Science, 26 (6), pp.792-804.
Venkatesh R. and Vijay Mahajan (1993), “A Probabilistic Approach to Pricing a Bundle of
Products or Services,” Journal of Marketing Research, 30 (4), pp.494-508.
Vitorino, Maria Ana (2007), “Empirical Entry Games with complementarities: an Application to
the Shopping Center Industry,” working paper.
Watson, Randal (2005), “Entry and Location Choice in Eyewear Retailing,” working paper.
177
Weinberg, Charles B. and Kenneth M. Shachmut (1978), “ARTS PLAN: a Model Based System
for Use in Planning a Performing Arts Series,” Management Science, 24 (6), pp.654-664.
___ (1986), “ARTS PLAN: Implementation, Evolution, and Usage,” Marketing Science, 5 (2),
pp.143-158.
Weatherford, Lawrence R., Samuel E. Bodily, and Phillip E. Pfeifer (1993), “Modeling the
Customer Arrival Process and Comparing Decision Rules in Perishable Asset Revenue
Management Situations,” Transportation Science, 27 (3), pp.239-251.
Weiner , Richard (2000), “Of Bulldog Editions and Matters,” Public Relations Tactics, 7 (9),
pp.7.
Xie, Jinhong and Steven M. Shugan (2001), “Electronic Tickets, Smart Cards, and Online
Prepayments: When and How to Advance Sell,” Marketing Science, 20 (3), pp.219-243.
Yang, Sha and Greg M. Allenby (2003), “Modeling Interdependent Consumer Preferences,”
Journal of Marketing Research, 40 (3), pp.282-294.
Zhu, Ting and Vishal Singh (2009), “Spatial Competition with Endogenous Location Choices: an
Application to Discount Retailing,” Quantitative Marketing and Economics, 7 (1), pp.1-35.
___, ___, and Mark Manuszak (2009), “Market Structure and Competition in the Retail Discount
Industry,” Journal of Marketing Research, forthcoming.
__, __, and Anthony Dukes (2007), “Local Competition, Entry, and Agglomeration,” working
paper.
doc_855010833.pdf