Description
Objective: We investigate whether the effect of relative performance information on tournament
performance depends on the tournament’s prize structure. We focus on the effect
of relative performance information on two tournament prize structures: a two-tier structure
in which only the top performer receives a reward and all other contestants receive an
equal payoff that is lower (reward tournament) and a three-tier prize structure in which
the top performer receives a reward, the bottom performer receives a penalty equal to
the amount of that reward, and all remaining contestants receive an equal intermediate
payoff (reward and punish tournament).
Materials Method: We investigate how RPI affects performance in tournaments with different
prize structures via an experiment. In our experiment, each participant competes in
a multi-period tournament against four other participants on a task in which performance
is a function of both individual effort and common uncertainty.
Results: We find that, compared with when relative performance information is not present,
relative performance information has a negative effect on performance in a reward
tournament but a positive effect on performance in a reward and punish tournament.
Relative performance information in tournaments with
different prize structures
Andrew H. Newman
a
, Ivo D. Tafkov
b,?
a
University of South Carolina, Darla Moore School of Business, United States
b
Georgia State University, J. Mack Robinson College of Business, United States
a b s t r a c t
Objective: We investigate whether the effect of relative performance information on tour-
nament performance depends on the tournament’s prize structure. We focus on the effect
of relative performance information on two tournament prize structures: a two-tier struc-
ture in which only the top performer receives a reward and all other contestants receive an
equal payoff that is lower (reward tournament) and a three-tier prize structure in which
the top performer receives a reward, the bottom performer receives a penalty equal to
the amount of that reward, and all remaining contestants receive an equal intermediate
payoff (reward and punish tournament).
Materials Method: We investigate how RPI affects performance in tournaments with dif-
ferent prize structures via an experiment. In our experiment, each participant competes in
a multi-period tournament against four other participants on a task in which performance
is a function of both individual effort and common uncertainty.
Results: We ?nd that, compared with when relative performance information is not pres-
ent, relative performance information has a negative effect on performance in a reward
tournament but a positive effect on performance in a reward and punish tournament. Sup-
plementary analysis reveals that bottom and middle performers drive these differences in
performance, which are due to both differences in effort and in adoption of overly risky
strategies.
Ó 2014 Elsevier Ltd. All rights reserved.
Introduction
A tournament incentive scheme evaluates employees
based on their relative, rather than absolute, performance
level. Tournament compensation is pervasive as more than
half of U.S. corporations use some sort of tournament rank-
ing system that pits employees against colleagues (Berger,
Klassen, Libby, & Webb, 2013; Chen, Williamson, & Zhou,
2012; Hazels & Sasse, 2008; McGregor, 2006). Firms use
tournaments for a variety of employee-types ranging from
production-line workers to salespersons to mutual fund
managers. Employees compete for promotions, bonuses,
and even prizes such as luxurious trips or prime parking
spots (Backes-Gellner & Pull, 2013; Cerdin & Pargneux,
2009; Kempf & Ruenzi, 2008). Tournaments are common
because they can increase productivity by motivating
employees, allow ?rms to avoid paying risk-averse
employees for bearing additional risk associated with com-
mon uncertainty, and help ?rms differentiate the talent of
their workforce (Grote, 2002, 2005; Ng & Lublin, 2010).
Much prior research has focused on understanding when
tournaments produce superior outcomes for the ?rm rela-
tive to other compensation schemes (e.g., Hannan,
Krishnan, & Newman, 2008; Lazear & Rosen, 1981;http://dx.doi.org/10.1016/j.aos.2014.05.004
0361-3682/Ó 2014 Elsevier Ltd. All rights reserved.
?
Corresponding author. Address: 35 Broad Street, Suite 510, Atlanta,
GA 30303, United States. Tel.: +1 404 413 7226; fax: +1 404 413 7203.
E-mail address: [email protected] (I.D. Tafkov).
Accounting, Organizations and Society 39 (2014) 348–361
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
Nalebuff & Stiglitz, 1983). Beyond understanding the set-
tings in which tournaments may be preferable to other
incentive schemes, it is also important for ?rms to under-
stand how the design of the tournament itself can in?u-
ence its effectiveness.
We investigate two key elements of tournament design:
relative performance information (RPI) and prize structure.
We focus on RPI and prize structure because accountants
help ?rms determine the feedback that should be provided
to decision-makers (Bonner & Sprinkle, 2002) as well as
design employee compensation plans (Atkinson, Banker,
Kaplan, & Young, 2001; Indjejikian, 1999). Speci?cally,
we are interested in whether the tournament’s prize struc-
ture in?uences the effect that RPI has on tournament
performance.
In terms of tournament prize structure, many ?rms
incorporate rewards such as monetary bonuses, trips,
and promotions (Gilpatric, 2009; Grote, 2005). Mean-
while, according to a recent survey, 60% of Fortune 500
companies use some form of relative performance-based
compensation scheme that incorporates both reward and
punishment (Cohan, 2012; Kwoh, 2012). Firms such as
General Electric, Metlife, Microsoft, American Express,
AIG, Hewlett Packard, and Yahoo! have used tourna-
ments that incorporate both rewards and punishment,
with punishments coming in such forms as job reassign-
ment, demotion, or even ?ring (Cohan, 2012; Grote,
2005; Kwoh, 2012).
1
For example, General Electric used
relative performance assessments to sort employees into
three groups: a top 20% to whom rewards, promotions
and stock options are showered, a middle 70%, and a bot-
tom 10% who are either ?red or face other disciplinary
actions.
Consistent with the evidence that some ?rms rely
solely on rewards while others rely on both rewards and
punishment, we focus on two tournament prize struc-
tures: a two-tier structure in which only the top performer
receives a reward, and all other contestants receive an
equal payoff that is lower (hereafter, a reward tourna-
ment) and a three-tier structure in which the top per-
former receives a reward, the bottom performer receives
a penalty equal to the amount of that reward, and all
remaining contestants receive an equal intermediate pay-
off (hereafter, a reward and punish tournament). Both
prize structures reward the top performer, and the key dif-
ference across the two structures relates to the treatment
of the nonwinners.
2
In terms of understanding the effect of RPI in tourna-
ments, prior research has focused only on tournaments that
reward the top performers (Casas-Arce & Martinez-Jerez,
2009; Ederer, 2010; Hannan et al., 2008).
3
The empirical evi-
dence shows that RPI informing participants that they are
likely to win the tournament can motivate higher performing
participants to increase or at least maintain their perfor-
mance levels. If only the top performers are rewarded, how-
ever, the majority of participants will receive RPI indicating
that they are unlikely to win the tournament. This leads these
participants either to reduce effort or adopt overly risky
strategies, both of which have a negative effect on their per-
formance. In total, any positive RPI effect on performance for
higher performing participants is outweighed by the
decrease in performance of participants who receive RPI indi-
cating that they are unlikely to win, resulting in an overall
negative effect of RPI on performance.
Prior ?ndings suggest that, because RPI can negatively
affect performance in tournaments, ?rms may reap greater
bene?ts from not providing RPI to tournament partici-
pants. Given the extensive focus on reward tournaments,
our primary focus is to investigate whether this implica-
tion from reward tournaments also applies to another form
of tournament, the reward and punish tournament. Under-
standing the scope of this implication regarding RPI across
alternative prize structures is important because it has
critical rami?cations for how ?rms design their tourna-
ments as well as their information systems.
We investigate how RPI affects performance in tourna-
ments with different prize structures via an experiment. In
our experiment, each participant competes in a multi-per-
iod tournament against four other participants on a task in
which performance is a function of both individual effort
and common uncertainty. We use a multi-period setting
because it allows for a more precise test of our theory
and enhances the generalizability of our results as tourna-
ments are strategic contests (Rankin & Sayre, 2011) and
organizational decisions are typically made in dynamic
multi-period contexts (Hollenbeck, Ilden, Phillips, &
Hedlund, 1994). We manipulate our ?rst factor, tourna-
ment prize structure, between participants at two levels:
reward or reward and punish. Our reward prize structure
provides the winner of the tournament with a monetary
reward while giving all other participants an equal lower
payoff. In contrast, our reward and punish prize structure
provides the winner with a monetary reward, punishes
the loser with a monetary penalty equal to the amount of
the reward, and gives all other participants an equal inter-
mediate payoff. We hold participants expected pay and
prize spread, i.e., the range of payouts between the top
1
These companies use names like ‘‘rank and yank,’’ ‘‘forced ranking,’’
‘‘stack ranking,’’ ‘‘relative performance rating system,’’ ‘‘talent assessment
system,’’ and ‘‘performance procedure’’ to describe such schemes.
2
The term ‘‘prize structure’’ can also be used to refer to any aspect of the
compensation design of a tournament contract. For example, prior research
uses the term to refer to the number of positive rewards present relative to
the number of tournament participants as well as to refer to the prize
spread of any such positive rewards (e.g., Ehrenberg & Bognanno, 1990a,
1990b; Freeman & Gelber, 2010; Knoeber & Thurman, 1994; Lazear &
Rosen, 1981; Lynch, 2005; Orrison, Schotter, & Weigelt, 2004).
3
An exception is Freeman and Gelber (2010) who investigate the effect
of RPI in a six-tier reward prize structure in which all contestants but one
receive a reward and the reward amount per contestant increases gradually
with rank. Freeman and Gelber’s setting differs from ours in an important
way. They use a static, one-period tournament in which contestants
compete after receiving information on their relative abilities. Such a
setting leaves room for effort intensity, but not strategy development, to
in?uence performance. In contrast, we focus on a dynamic, multi-period
setting in which participants must make strategic decisions after being
periodically updated on their chances of winning. We do so because prior
research acknowledges that strategy development is an important dimen-
sion of effort (Bonner & Sprinkle, 2002) and shows that it can affect
tournament performance (Hannan et al., 2008). We discuss the implications
of Freeman and Gelber’s study as they pertain to our study in the
conclusion.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 349
performer and bottom performer, constant across these
two prize structures. We manipulate our second factor,
RPI, between participants at two levels: present or absent.
When RPI is present, we provide RPI in the form of perfor-
mance rank at speci?ed intervals throughout the experi-
ment. When it is absent, we never provide participants
with information about their performance rank during
the tournament.
Consistent with our predictions, when RPI is absent
participants perform better in a reward and punish tourna-
ment than in a reward tournament. We also ?nd a
disordinal interaction between RPI and tournament prize
structure. Speci?cally, RPI has a negative effect on perfor-
mance in a reward tournament but a positive effect on
performance in a reward and punish tournament. As a
result, participants performance is greatest when RPI is
present in a reward and punish tournament. Supplemental
analysis reveals that differences across conditions in
bottom and middle performers’ effort levels and adoption
of overly risky strategies drive the differential effect of
RPI on performance across the two prize structure
conditions.
Our study highlights the inter-related nature of com-
pensation and information system design choices and has
important implications for the design of effective compen-
sation and information systems in a tournament setting.
We contribute to prior research investigating tournaments
in two ways. First, we provide insights on the effect of a
prize structure that includes reward and punishment on
tournament performance. This matters because relative
performance-based compensation contracts that incorpo-
rate reward and punishment are often found in practice
(Cohan, 2012; Grote, 2002, 2005), but little is known about
their effectiveness (Frederickson & Waller, 2005;
Moldovanu, Sela, & Shi, 2012).
Second, we contribute to an emerging stream of
research that has only just begun to consider the role that
prize structure plays in determining the effect of RPI on per-
formance in tournaments (Freeman & Gelber, 2010;
Hannan et al., 2008). We showthat the downside of provid-
ing RPI when only top performers are rewarded, as docu-
mented by prior research and our study, is not universal
to all tournament prize structures. Rather, we provide evi-
dence that the appropriateness of providing RPI varies
based on the tournament’s prize structure. In doing so, we
highlight that ?rms should keep in mind the purpose of
their tournaments when determining whether to provide
RPI. For instance, considering the two tournament prize
structures we examine, our results suggest that, if ?rms
value the overall output of their employees in a tourna-
ment, then they should withhold RPI under a reward prize
structure but provide RPI under a reward and punish prize
structure. If, instead, ?rms value only the output of the
highest performing employee in a tournament, our results
imply that ?rms should consider providing RPI to employ-
ees under both prize structures as RPI has a positive effect
on top performers’ performance under both prize struc-
tures. Finally, in some ?rms that use tournaments, RPI
may be readily available via informal sources (Hannan,
McPhee, Newman, & Tafkov, 2013). Our study suggests that
such ?rms do not necessarily need to discontinue their
tournaments but rather should avoid using a prize struc-
ture that only rewards the top performer.
4
Theory and hypotheses
Tournament prize structure
In a multi-period tournament, performance differences
across contestants emerge over time (Casas-Arce &
Martinez-Jerez, 2009; Ederer, 2010; Hannan et al., 2008).
When RPI is absent, however, contestants are not informed
of these differences and thus lack the insights necessary to
infer where they stand relative to others with respect to
probability of winning. Without RPI, contestants are left to
form assumptions about their relative performance, which
they are likely to deem above-average but not extremely
high. Researchshows that most individuals believethat they
areabove-averageperformers but strategicallyavoidsetting
performance expectations that are too high to prevent per-
formance anxiety and disappointment (Brown, 2007,
2012; Hoorens, 1995; Norem & Cantor, 1986).
We develop our ?rst hypothesis in the context of a tourna-
ment setting where contestants must make assumptions
about their relative standing when deciding how much effort
to exert as well as how much risk to take in an attempt to
improve their performance. To do so, we rely on two single-
period tournament models by Gilpatric (2009) and
Moldovanu and Sela (2001). Although, as discussed below,
there are differences between the settings of these models
and the tournament setting we study, we expect that impor-
tant intuitions fromthesemodels will carryover toour setting.
Gilpatric’s (2009) model provides insights into the design
of a tournament prize structure when identical contestants
simultaneously choose both the mean and the variance of
their performance andwhena ?rmcares about the aggregate
output of all performers, not just the top performer. Accord-
ing to the model, for the ?rm to motivate a desired level of
effort among contestants without inducing costly risk seek-
ing, it should implement a three-tier prize structure instead
of a two-tier one. Speci?cally, the ?rmshouldprovide a mon-
etary reward to the top performer and a monetary punish-
ment to the bottom performer and give all other
contestants some equal amount in between these two pay-
ments (withthe probabilities andsize of the rewardandpun-
ishment being the same). The intuition for this is that,
although the reward can motivate tournament contestants
tochoose highlevels of productive effort, it canalso motivate
themto adopt overly risky strategies. Speci?cally, when fac-
ing multiple opponents, a contestant’s probabilityof winning
a tournament that rewards only the top performer is very
small unless her or his performance is substantially above
the average. Achieving such performance, via productive
effort, is unlikely whena contestant is facing multiple identi-
4
The ?rm’s purpose in using a tournament is likely in?uenced by the
setting in which it is implemented. For instance, in production or sales
settings, ?rms typically use tournaments to motivate the performance of all
of their employees, while in research and development settings, they are
often more concerned with motivating employees to generate at least one
truly innovative idea regardless of how many inferior ideas are also
generated.
350 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
cal opponents. Risky strategies increase the variance in per-
formance output, which can potentially lead to the extreme
performance outcomes neededtowinthe tournament. How-
ever, increasing the variance in performance output beyond
some point may come at the cost of decreased expected per-
formance. The reason, according to Gilpatric, is that increas-
ing the variance in performance output beyond some point
is possible only by selecting strategies with lower expected
performance. Alternatively, seeking out overly risky strate-
gies is costly because it requires time and effort that could
otherwise be devoted toward exerting productive effort.
Under a prize structure that rewards only the top per-
former, the absence of punishment for losing the tourna-
ment means there is less for contestants to lose from
attempting to win the tournament by adopting overly risky
strategies that decrease their expectedperformance. Includ-
ing punishment in the prize structure means that adopting
such risky strategies nowalso increases contestants’ proba-
bility of receiving the punishment, which should reduce
contestants’ incentives to adopt overly risky strategies in
an attempt to win. As such, contestants should adopt fewer
overly risky strategies, resulting in greater expected perfor-
mance, when the prize structure includes both reward and
punishment compared to when it includes only a reward.
Gilpatric’s intuition pertains to the tournament setting
we study even though his model is single-period and
assumes identical contestants. The reason is that the
absence of RPI in a multi-period tournament means con-
testants have no basis by which to conclude that they have
a higher likelihood of winning the tournament than any
other contestant. This is likely to lead contestants to con-
clude that they cannot achieve the extreme performance
needed to win the tournament only by exerting additional
productive effort. Accordingly, similar to Gilpatric’s model,
we expect that, in the absence of punishment (i.e., in the
reward prize structure), contestants are likely to adopt
overly risky strategies in an attempt to win the reward.
Moldovanu and Sela’s (2001) model investigates prize
structure design in a tournament in which contestants with
different privately known abilities simultaneously choose
effort levels and the ?rm is concerned with maximizing
the aggregate output of all performers. According to their
model, in which performance is a function only of effort
and ability, a ?rm should provide more than one reward to
its contestants. The intuition is that second-place (or third-
place, etc.) rewards are more motivating to contestants with
lower abilities thanthe ?rst-place reward, whichmany con-
testants believe they are extremely unlikely to win.
This model’s intuitionregardingthe potential value of mul-
tiple tiers of rewards is applicable to the tournament setting
we study even though the model does not allow contestants
to adopt risky strategies, and it assumes that contestants have
private information regarding their abilities.
5
Speci?cally, if
contestants ina multi-periodtournament expect their perfor-
mance to be above the average but not extremely high, they
may not assess a high likelihood of receiving the reward in a
reward tournament. Accordingly, the motivational effect
from a reward for the winner may be low. In cases like this,
a runner-up reward, such as the one in a three-tier reward
and punish tournament that is provided to contestants who
are not a bottom performer, may be more motivating than a
reward provided to the top performer.
In summary, based on Gilpatric’s model, we expect that
adding a punishment element to the tournament prize
structure will have a positive effect on performance
because the punishment will discourage the adoption of
overly risky strategies that have lower expected perfor-
mance. Additionally, based on Modovanu and Sela’s model,
we expect that moving from a two-tier reward prize struc-
ture to a three-tier reward and punish prize structure will
have a positive effect on performance because contestants
who have low expectations of winning the ?rst-place
reward will be more motivated by the chance to receive
the runner-up reward provided to contestants who neither
win nor lose the tournament.
Formally stated, the hypothesis is:
H1. When relative performance information is absent,
performance under a reward and punish tournament is
greater than performance under a reward tournament.
RPI and tournament prize structure
RPI gives contestants in multi-period tournaments
information on the evolving differences in performance
among their peers that arise from the dynamic nature of
a multi-period tournament (Casas-Arce & Martinez-Jerez,
2009; Ederer, 2010; Hannan et al., 2008). In doing so, RPI
helps contestants make more informed effort decisions in
subsequent periods of the tournament. Both economic
and psychology theories predict that the ultimate effect
of RPI on performance depends on how the RPI’s content
affects contestants’ perceptions of their relative standing
in the tournament. We develop theory to predict that the
difference in overall performance predicted in H1 between
a three-tier prize structure that uses rewards and punish-
ment and a two-tier prize structure that provides only a
reward to the winner will be magni?ed by RPI. Speci?cally,
we develop theory to predict that providing RPI will
decrease performance under a reward prize structure and
increase performance under a reward and punish prize
structure.
6
5
Speci?cally, Moldovanu and Sela’s model assumes that contestants
receive a private signal ex ante regarding their ability as well as where their
ability lies on the distribution of contestants’ potential abilities. Taken
together, these two pieces of information provide contestants with insight
on their relative performance potential before the single-period tourna-
ment. In our tournament setting, contestants do not have this information
ex ante.
6
The single-period models by Gilpatric (2009) and Moldovanu and Sela
(2001) that we relied on to develop our ?rst hypothesis are less useful in
predicting contestant behavior in a multi-period tournament setting when
contestants receive RPI. The reason is that RPI allows contestants to infer
their likelihood of winning the tournament and thus to update their
strategy during the multi-period tournament. Prior research shows that
individual behavior can change substantially when moving from a single-
period setting to a multi-period setting in which interim feedback that
allows for strategy updating, is provided (Hollenbeck et al., 1994; Thaler &
Johnson, 1990). Accordingly, we no longer rely on these models when
developing our remaining hypotheses.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 351
We ?rst present theory to predict that RPI will decrease
performance under a reward prize structure. While this
prediction replicates Hannan et al.’s (2008) result, outlin-
ing the rationale for this prediction helps highlight the
key differences that emerge when RPI is provided under
a reward and punish prize structure. Hannan et al. (2008)
predict and ?nd that, in a reward tournament, RPI moti-
vates the higher performers because it allows them to infer
that they have a high probability of receiving the reward
and thereby have a high marginal bene?t of exerting addi-
tional effort. They thus continue to perform at a high level.
The authors also predict and ?nd, however, that any poten-
tial increase in performance by the higher performers is
outweighed by the decrease in performance from the
lower and middle performers. These performers receive
RPI that indicates that they are unlikely to win the reward
and thereby have a low marginal bene?t of exerting addi-
tional effort. These contestants therefore either decrease
their effort or adopt overly risky strategies that have lower
average expected performance in an attempt to catch up to
their peers. Both approaches result in decrease in perfor-
mance of these contestants.
We expect to replicate the ?ndings of Hannan et al.
(2008). Formally stated, the hypothesis is:
H2a-replication. Relative performance information has a
negative effect on performance in a reward tournament.
We now present theory to predict that RPI will
increase performance under a reward and punish prize
structure. In a reward and punish tournament, the mar-
ginal bene?t of effort increases as one’s performance
moves toward either of the two ends of the performance
spectrum (i.e., very high or very low) because at these
ends the probability of receiving the reward or the pun-
ishment is high. Recall that in the absence of RPI, employ-
ees will expect their performance to be above the average
but not necessarily substantially above it. RPI informs
higher performers that their performance is at the high
end of the performance spectrum and thus that their
probability of receiving the reward is higher than they
expect in the absence of RPI. Such RPI should motivate
these contestants to maintain or even increase their per-
formance levels. From an economic theory perspective,
higher performing contestants have a relatively high mar-
ginal bene?t of exerting additional productive effort
because doing so increases their likelihood of receiving
the reward.
7
From a psychology theory perspective, social
comparison theory suggests that RPI motivates higher per-
forming contestants to sustain high performance levels as
they wish to maintain their relative standing (Beach &
Tesser, 1995; Tesser, 1988). In addition, RPI indicating that
the goal of winning the reward is more attainable than
expected in the absence of RPI boosts contestants’ perfor-
mance expectancy, self-ef?cacy perception and goal com-
mitment (Bandura, 1986; Bandura & Wood, 1989; Locke &
Latham, 1990), all of which also motivate higher performers
to sustain or even increase their performance levels in sub-
sequent periods of the tournament.
Regarding lower performers, both economic and psy-
chology theories predict that RPI has a positive effect on
performance under a reward and punish tournament. In a
reward and punish tournament, contestants worry about
both winning the reward and avoiding the punishment.
In the absence of RPI, if contestants assume that their per-
formance is above average (but neither extremely high nor
extremely low), this means lower performers are over esti-
mating their probability of avoiding punishment. RPI
informs lower performers that their performance is at the
low end of the performance spectrum and thus that their
probability of avoiding punishment is lower than they
expected in the absence of RPI. Therefore, from an eco-
nomic theory perspective, RPI informs lower performers
that their marginal bene?t of additional effort is high
because, although additional effort is unlikely to help them
win the reward, it can help them avoid the punishment.
8
Additionally, the presence of punishment imposes a cost
on attempting to increase performance via the adoption of
overly risky strategies if such an approach ends up actually
decreasing performance. Accordingly, we expect lower per-
formers in the reward and punish tournament to respond
to RPI by exerting greater productive effort instead of adopt-
ing overly risky strategies. Thus RPI should have a positive
effect on their performance.
Psychology theory leads to similar predictions for lower
performers. While social comparison theory (Brown, Ferris,
Heller, & Keeping, 2007; Buunk & Gibbons, 2007) predicts
that lower performing contestants are motivated to
improve their performance to enhance their relative stand-
ing, expectancy theory, social cognitive theory, and goal-
setting theory (Bandura, 1986; Locke & Latham, 1990)
impose a boundary condition on this prediction. According
to these theories, the size of the goal–performance discrep-
ancy (i.e., the difference between a performance goal and
actual performance) in?uences the effect that RPI has on
performance. A smaller goal–performance discrepancy
motivates individuals because it increases one’s perfor-
mance expectancy and goal commitment. This has a posi-
tive effect on performance. A larger discrepancy
demotivates individuals because it lowers one’s perfor-
mance expectancy and goal commitment. This has a nega-
tive effect on performance. In addition, a larger
7
Higher performing contestants include the top performers, who will
win the reward if they keep their current performance rank, and near the
top performers, who will win the reward only if they improve their rank.
The extent to which the marginal bene?t of additional effort is high may
vary across these two types of performers. While near the top performers
will be motivated to increase their performance to become top performers
and win the reward, the top performers will be motivated to either
maintain their performance if they believe they can win or work harder if
they deem their current performance to be insuf?cient to win due to an
expected increase in performance of near the top performers. We inves-
tigate this issue in our results section.
8
Lower performers include the bottom performers, who will avoid
punishment only if they improve their current performance rank, and near
the bottom performers, who will avoid punishment if they keep their
current performance rank. While the bottom performers are motivated to
increase their performance level to avoid punishment, the near the bottom
performers are motivated to either maintain their performance, if they
deem it suf?cient to avoid the punishment, or increase it, if they deem
maintaining will be insuf?cient to avoid the punishment due to an
expected increase in performance of the bottom performers. We investigate
this issue in our results section.
352 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
goal–performance discrepancy can lead to low strategy
ef?ciency (i.e., ineffective strategy searches or excessive
experimentation), which also hurts performance (Kluger
& DeNisi, 1996).
In a reward and punish tournament, contestants have
two goals: to win by having the highest performance
and to avoid losing by not having the lowest performance.
With regard to the goal of winning the tournament, RPI
indicating that one is a low performer creates a larger
goal–performance discrepancy compared to when contes-
tants receive no RPI and assume that their performance is
above the average. With regard to the goal of not losing
the tournament, RPI creates a goal–performance discrep-
ancy where a discrepancy is absent when contestants do
not have RPI and assume that they are unlikely to be
the one with extremely low performance. Lower perform-
ers should view this latter discrepancy as more manage-
able to overcome than the one associated with the goal
of winning the tournament. The reason is that they now
know that winning the tournament requires outperform-
ing multiple higher-performing contestants whereas not
losing it and, thus, winning the runner-up reward requires
outperforming only one other low performing contestant.
Prior research shows that individuals respond to large
goal–performance discrepancies by lowering their goals
to reduce these discrepancies (Donovan & Williams,
2003; Locke & Latham, 2006). Lower performers in a
reward and punish tournament can do so by focusing on
the goal of not losing rather than the goal of winning
the tournament. Because lower goal–performance dis-
crepancies are more motivating than larger discrepancies
(Locke & Latham, 1990), this shift should motivate lower
performing contestants.
Regarding the middle performers, RPI informs them
that they are slightly less likely to win a reward and
slightly more likely to incur a punishment than they per-
ceive in the absence of RPI. Overall, however, RPI does
not create signi?cant belief revision for these contestants
who expected that their performance would be neither
extremely high nor extremely low. From an economic per-
spective, the slightly decreased marginal bene?t of exert-
ing additional effort to win the reward is
counterbalanced by the slightly increased marginal bene?t
of exerting additional effort to avoid the punishment and
receive the runner up reward. Accordingly, we do not
expect middle performers in the reward and punish tour-
nament to behave differently in the presence of RPI than
in the absence of RPI. Psychology theory leads to similar
predictions for these performers because RPI does not sub-
stantially change their perceived goal–performance
discrepancies.
In summary, relative to when RPI is absent, we predict
that in a reward and punish tournament the presence of
RPI leads to an increase in overall performance. The reason
is that RPI has positive effect on the performance of lower
performers, positive or no effect on the higher performers,
and no effect on the middle performers. Formally stated,
the hypothesis is:
H2b. Relative performance information has a positive
effect on performance in a reward and punish tournament.
Method
Experimental task
We adapt our experimental task from Sprinkle (2000)
and Hannan et al. (2008). We assign each participant to
the role of a production manager responsible for making
production quantity decisions. The task consists of 60 deci-
sion periods broken into 12 trials with ?ve decision periods
per trial. In each period, the objective is to select the pro-
duction quantity (from 1 to 20) that maximizes the com-
pany’s total pro?t (measured in terms of total pro?t
points earned). Each participant competes in a tournament
against four other participants, where participants’ ?nal
performance rank is determined by the total pro?t points
earned over the 60 decision periods.
9
As shown in Fig. 1, the pro?t points earned for a partic-
ular production decision are determined jointly by the par-
ticipant’s production quantity decision and the economic
condition. The economic condition represents a state of
nature and serves as a source of common uncertainty for
participants because they are not informed of the eco-
nomic condition for each trial ex ante. Participants are only
aware that the economic condition ranges from 1 to 20
with equal probability and that, although it remains con-
stant for each of the ?ve periods within a trial, it may vary
across the 12 trials. Each participant faces the same set of
economic conditions, in the same order.
10
Participants have 180 s to complete the ?ve periods of a
trial. A clock is displayed on the computer screen. At the
end of each period within a trial, participants choose
whether they want feedback on their performance. If they
choose to receive feedback, they receive a summary of
their production quantity choices made for each period
in the current trial as well as the number of pro?t points
earned for each of those choices. This feedback remains
on the screen for a minimum of 10 s. Viewing feedback
at the end of a period may help participants make more
pro?table production quantity decisions (i.e., earn more
pro?t points) in the remaining periods of that trial. The
pro?t points earned for a production decision depend on
the economic condition for that trial. Because the eco-
nomic condition remains constant for all ?ve periods,
viewing feedback during a trial can help participants infer
the economic condition for the current trial. To assist with
this inference, we provide participants a table similar to
Fig. 1.
9
We use ?ve-participant tournaments because this allows us to test our
theory in a most ef?cient way. Recall that we argue that RPI can have
different effects on the performance of higher, lower, and middle perform-
ers. Also recall that we differentiate between top performers and near-the-
top performers when discussing the effect of RPI on higher performing
contestants, and between bottom performers and near-the-bottom per-
formers when discussing the effect of RPI on lower performing contestants.
This suggests that we need to have at least ?ve participants in a
tournament to test our predictions. Increasing the size of a tournament
beyond ?ve participants increases the overall number of participants
needed and thereby the cost of our experiment without providing any clear
bene?t with respect to testing our theory.
10
We randomly selected the economic condition for each trial. To
improve control, we made these random choices before the experiment
sessions.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 353
In addition to any compensation based on relative per-
formance, participants also earn $0.01 for every ?ve sec-
onds saved out of the 180 s available per trial. Any
potential gains from inferring the economic condition by
viewing feedback therefore come at a cost because every
time a participant chooses to view feedback for 10 s he is
reducing the compensation he can earn for early comple-
tion of the task. This cost of viewing feedback serves as
our proxy for participants’ disutility of effort. The use of
such a proxy for a cost of effort is a common practice in
experimental accounting research (Hecht, Tafkov, &
Towry, 2012; Sprinkle, 2000) because the number of feed-
back requests meets the three criteria for a measure of task
effort suggested by Baiman (1982).
At the end of each trial, participants in all conditions
automatically receive individual performance feedback.
Speci?cally, they learn the number of pro?t points earned
in the current trial, the cumulative number of pro?t points
earned in all trials completed so far, the number of avail-
able seconds not used in the current trial, and the cumula-
tive number of available seconds not used in all trials
completed so far.
Experimental design
To investigate our hypotheses, we use a multi-period
tournament in which participants make strategic decisions
after being periodically updated on their chances of win-
ning. We use a 2 Â 2 Â 12 mixed experimental design.
The two primary independent variables are the tourna-
ment prize structure (reward or reward and punish) and
RPI (present or absent). We manipulate both variables
between-participants as described in more detail below.
We manipulate trial within-participants as participants
make output-quantity decisions in 60 periods divided into
12 trials of ?ve decision periods each. Our primary depen-
dent variable is performance, measured as the total pro?t
points earned.
Prize structure manipulation
In the reward condition, all participants earn $16 for
completing the 60 decision periods. The participant who
earns the most pro?t points among the ?ve participants
in his respective session receives a reward of $20. In the
reward and punish condition, all participants earn $20 for
completing the 60 decision periods. The participant who
earns the most pro?t points receives a reward of $10, while
the participant who earns the fewest pro?t points is penal-
ized $10. This experimental design ensures that the
expected value of participants’ earnings equals $20 in both
tournament compensation contracts.
11
In addition to holding the expected value of earnings
constant, we also hold the overall prize spread, i.e., range
of earnings from the top performer to the bottom per-
former, constant across the two incentive contract condi-
tions. Speci?cally, the prize spread always equals $20
(reward equals +$20 in reward condition; reward equals
+$10 and penalty equals À$10 in reward and punish
Production Quantity
Economic
Condition 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 5 5 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 5 5 10 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 5 5 10 20 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 5 5 10 20 20 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 5 5 10 20 20 30 30 0 0 0 0 0 0 0 0 0 0 0 0 0
8 5 5 10 20 20 30 30 30 0 0 0 0 0 0 0 0 0 0 0 0
9 5 5 10 20 20 30 30 30 45 0 0 0 0 0 0 0 0 0 0 0
10 5 5 10 20 20 30 30 30 45 45 0 0 0 0 0 0 0 0 0 0
11 5 5 10 20 20 30 30 30 45 45 60 0 0 0 0 0 0 0 0 0
12 5 5 10 20 20 30 30 30 45 45 60 60 0 0 0 0 0 0 0 0
13 5 5 10 20 20 30 30 30 45 45 60 60 60 0 0 0 0 0 0 0
14 5 5 10 20 20 30 30 30 45 45 60 60 60 80 0 0 0 0 0 0
15 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 0 0 0 0 0
16 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 0 0 0 0
17 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 0 0 0
18 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 0 0
19 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 100 0
20 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 100 100
Note: For example, suppose a participant chose a production quantity of 10. If the economic condition was a number from 1 to 9, the participant would earn 0
profit points. If the economic condition was a number from 10 to 20, the participant would earn 45 profit points.
Fig. 1. Pro?t points table.
11
Although we exogenously impose these prize structures, we acknowl-
edge that prize structures can also arise endogenously as a function of the
task characteristics. For instance, according to Aron and Olivella’s (1994)
analytical model, penalty-based contracts emerge in equilibrium in jobs in
which performance is easily measured, whereas reward-based contracts
emerge in equilibrium in jobs in which performance needs to be evaluated
subjectively. Given that the focus of our study is on understanding how the
presence of RPI affects participant behavior across two particular prize
structures, we determined that exogenously imposing the prize structures
was more appropriate. We further discuss this issue in our conclusion.
354 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
condition). We hold the prize spread constant because
prior research shows that it can affect the effort level
exerted in the tournament (Ehrenberg & Bognanno,
1990a, 1990b; Lazear & Rosen, 1981). Without holding
the prize spread constant, we could not disentangle the
prize spread’s effect on performance from the prize struc-
ture’s effect on performance.
Holding the overall prize spread constant while chang-
ing from a two-tier prize structure in the reward condition
to a three-tier prize structure in the reward and punish
condition creates underlying differences in the two con-
tracts. Speci?cally, the prize spread of $20 between the
winner and the nonwinners in the reward condition (com-
puted as $36–$16) is greater than the prize spread of $10
between the winner and those that neither win nor lose
(i.e., those in the middle) in the reward and punish condi-
tion (computed as $30–$20). Although the participants in
the reward and punish condition have a lower cost of not
winning than those in the reward condition ($10 vs. $20),
the participants in the reward and punish condition also
have a greater cost of losing than those in the reward con-
dition ($10 vs. $0). Consistent with our hypothesis devel-
opment, we expect that the combination of a $10 cost of
losing and a $10 cost of not winning in the reward and
punish condition will motivate more productive effort than
a $20 cost of not winning in the reward condition.
It is important to highlight that our reward and punish
prize structure differs from our reward prize structure in
two ways. First, the reward and punish prize structure
includes a penalty element that is lacking from the reward
prize structure. Second, the reward and punish prize struc-
ture is a three-level prize structure that sorts participants
as winners, nonwinners, and losers, whereas the reward
prize structure is a two-level prize structure that sorts peo-
ple as winners and nonwinners. The purpose of our exper-
iment design is not to distinguish the incremental effect
each of these two differences has on tournament perfor-
mance but to compare the effect of a reward and punish
prize structure to that of a reward prize structure. Put dif-
ferently, we cannot comment on whether any differences
in performance across our prize structure conditions are
driven by the presence of punishment, the addition of a
third-tier to the pay structure, or both.
RPI manipulation
We manipulate RPI between-participants at two levels:
absent or present. In the absent condition, we do not pro-
vide RPI during the tournament. In the present condition,
we provide participants with their cumulative perfor-
mance rank, from #1 to #5, on a quarterly basis at the
end of the 3rd, 6th, 9th, and 12th trials. Importantly, in
addition to knowing that everyone faces the same eco-
nomic conditions over the course of the 12 trials, partici-
pants in all conditions are also aware that all participants
have the same three states of nature in each quarter. In
other words, participants know that the quarterly RPI is
informative because the common error is the same for all
participants. This also means that the total amount of pos-
sible pro?t points is the same for all participants in each
quarter. RPI appears on the computer screen following
the individual performance feedback and remains on the
screen for 15 s.
Participants and procedures
In total, 80 students from a large public university par-
ticipated in 16 experimental sessions (four per experimen-
tal condition). Participants were randomly assigned to one
of the four experimental conditions based on which ses-
sion they attended. The mean age of the participants was
22.2 years, and 55% of them were female. There were no
signi?cant differences across conditions for age and gender
(all p > 0.18, two-tailed). Each session was conducted in an
experimental computer lab using z-tree software
(Fischbacher, 2007) and lasted approximately 75 min.
Upon arrival, we assigned participants a participation
number and provided them with a written description of
the experiment, examples of the task they would be com-
pleting, and details regarding their compensation contract.
We then required them to complete a short quiz to ensure
that they understood the experiment. Upon completion of
the quiz, the ?rst trial began. After the end of the 12th trial,
participants completed a post-experimental questionnaire
that collected process-related and demographic informa-
tion. Finally, to assure anonymity, we paid participants
via sealed envelopes based on their previously selected
participation number. Including a $5 show-up fee, partici-
pants earned an average of $28.13 on the task.
Results
Table 1 reports descriptive statistics related to our main
dependent variable, performance. Fig. 2 provides a graphi-
cal summary of the results.
Hypotheses tests
H1 predicts that, when RPI is absent, performance will
be greater under the reward and punish prize structure
than under the reward prize structure. Our test results,
presented in Table 2, show that, when RPI is absent, partic-
ipants in the reward and punish condition earn signi?-
cantly more pro?t points than participants in the reward
condition (2239.00 vs. 1990.75 from Table 1, t = 3.18,
p = 0.01, one-tailed).
12
This supports H1.
H2a and H2b predict an interaction between RPI and
prize structure, such that the effect of RPI on performance
is negative under a reward prize structure and positive
under a reward and punish prize structure. Panel A of
Table 3 presents an ANOVA for which the dependent vari-
able is pro?t points earned and the two independent vari-
ables are RPI and prize structure. As expected, we ?nd a
signi?cant RPI Â prize structure interaction (F = 3.86,
p = 0.05, two-tailed). Fig. 2 shows that the interaction is dis-
ordinal, that is, of the form predicted in H2a and H2b.
12
We also run this and subsequent tests using OLS regressions, in which
we cluster the data by tournament group (i.e., by each group of ?ve
participants competing against each other). Results of these analyses are
inferentially identical to those reported in the paper.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 355
Simple effects analysis, presented in Panel B of Table 3,
indicates that under a reward prize structure participants
earn fewer pro?t points when RPI is present than when
RPI is absent (1776.00 vs. 1990.75 from Table 1, t = 1.87,
p = 0.03, one-tailed). This result replicates Hannan et al.
(2008) and supports H2a. However, under a reward and
punish prize structure, participants earn more pro?t points
when RPI is present than when RPI is absent (2409.75 vs.
2239.00 from Table 1, t = 2.19, p = 0.02, one-tailed). Thus
H2b is also supported. Overall, the results support our pre-
diction that the effect of RPI on tournament performance
depends on the tournament’s prize structure.
Following from H1, H2a, and H2b performance should
be greatest when RPI is provided under a reward and
punish prize structure. Our tests’ results (untabulated)
show that participants earn signi?cantly more pro?t points
Table 1
Descriptive statistics: performance
a
(mean, (standard deviation)).
Prize structure
c
Reward Reward and punish
Relative performance information
b
No RPI 1990.75 (453.75) 2239.00 (319.80)
RPI 1776.00 (688.81) 2409.75 (215.15)
a
Performance is the total pro?t points earned during the 12 trials (60 periods). The pro?t points earned in each period are based on the economic
condition in that trial as well as participants’ production quantity choice. The pro?t points earned in a period represent the performance outcome for that
period. As such, total pro?t points earned over all 60 periods (12 trials) measures total performance.
b
RPI was manipulated at two levels. Participants in the No RPI condition received no relative performance information during the course of the
tournament or at the end. Participants in the RPI condition received their relative performance rank, from #1 to #5, at the end of trials 3, 6, 9, and 12.
c
Prize structure was manipulated at two levels: a reward was paid to only the winner of the tournament (reward condition) or a reward was paid to the
winner and a penalty was imposed on the loser (reward and punish condition). The expected pay and the prize spread were held constant across these two
conditions.
Reward
Reward and
Punish
Fig. 2. Effect of RPI
a
and Prize Structure
b
on Performance
c
.
a
RPI was manipulated at two levels. Participants in the No RPI condition received no relative
performance information on their performance during the course of the tournament or at the end. Participants in the RPI condition received their relative
performance rank, from #1 to #5, at the end of trials 3, 6, 9, and 12.
b
Prize structure was manipulated at two levels: a reward was paid only to the winner of
the tournament (reward condition), or a reward was paid to the winner and a penalty was imposed on loser (reward and punish condition). The expected
pay and the prize spread were held constant across these two conditions.
c
Performance is the total pro?t points earned during the 12 trials (60 periods). The
pro?t points earned in each period are based on the economic condition in that trial as well as participants’ production quantity choice. The pro?t points
earned in a period represent the performance outcome for that period. As such, total pro?t points earned over all 60 periods (12 trials) measures total
performance.
Table 2
Test of H1.
Hypothesis Mean difference t-Statistics p-Value
Dependent variable – performance
a
No RPI
b
/R&P
c
> No RPI/reward
c
248.25 3.18 0.01
*
*
All p-values are reported on a one-tailed basis due to our directional predictions for these effects.
a
Performance is the total pro?t points earned during the 12 trials (60 periods). The pro?t points earned in each period are based on the economic
condition in that trial as well as participants’ production quantity choice. The pro?t points earned in a period represent the performance outcome for that
period. As such, total pro?t points earned over all 60 periods (12 trials) measures total performance.
b
RPI was manipulated at two levels. Participants in the No RPI condition received no relative performance information on their performance during the
course of the tournament or at the end. Participants in the RPI condition received their relative performance rank, from #1 to #5, at the end of trials 3, 6, 9,
and 12.
c
Prize structure was manipulated at two levels: a reward was paid only to the winner of the tournament (reward condition) or a reward was paid to the
winner and a penalty was imposed on the loser (reward and punish condition). The expected pay and the prize spread were held constant across these two
conditions.
356 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
in this condition compared to each of the other three condi-
tions (all p < 0.05, two-tailed). This demonstrates the bene?t
of using rewards and punishment as complements in a
tournament prize structure, especially when RPI is
provided.
Supplemental analysis
RPI content
Our main analysis documents that RPI has a differential
effect on participants’ overall performance across the two
prize structure conditions. This prior analysis focuses on
the presence of RPI. As stated in our hypotheses develop-
ment for H2a and H2b, we expect that differences in how
participants respond to the content of the RPI play a critical
role in creating any performance differences across our
prize structure conditions. To test this expectation, we
examine the effect of RPI content on participants’ change
in performance during the tournament.
To do so, we de?ne change in performance as pro?t
points earned in the last three trials (4th quarter) minus
pro?t points earned in the ?rst three trials (1st quarter).
We use performance in these two quarters to calculate
change in performance because, by experimental design,
the set of economic conditions (states of nature) is the
same for the ?rst and fourth quarters, although the order
is randomized within each quarter. As such, the common
error term does not in?uence this comparison as the total
pro?t potential is the same in the ?rst and fourth quarters.
We classify participants based on participants’ perfor-
mance rank at the midpoint of the tournament. We use
the midpoint RPI because it allows us to directly investi-
gate how the content of the feedback in?uences changes
in participants’ behavior during the tournament.
13
We code
participants with a rank of 1 or 2 as top performers, partic-
ipants with a rank of 3 as middle performers, and those with
a rank of 4 or 5 as bottom performers. However, because
within a given prize structure we ?nd a similar RPI effect
for the middle and the bottom performers, we combine
these groups and refer to them as nontop performers to
avoid redundant discussions.
14
We ?nd a signi?cant three-way RPI Â prize struc-
ture  rank interaction (F = 2.71, p = 0.10, two-tailed). To
gain insight on the nature of the interaction, we conduct
two independent ANOVAs, one each for the top and nontop
performers. For the top performers, we ?nd a positive RPI
effect on change in performance (t = 2.35, p = 0.05, two-
tailed) but no signi?cant RPI Â prize structure interaction
(F = 1.56, p = 0.22, two-tailed). The performance of top per-
formers who received RPI increased from the ?rst quarter
to the fourth quarter more than the performance of top
performers who did not receive RPI in both the reward
condition (51.88 vs. 18.75) and the reward and punish con-
dition (116.25 vs. 58.75). This reveals that, consistent with
our expectations outlined during our hypotheses develop-
ment, providing RPI bene?ts top performers’ performance
over time, regardless of the prize structure.
For the nontop performers, we ?nd an RPI Â prize struc-
ture interaction (F = 8.63, p < 0.01, two-tailed). Simple effect
analysis ?nds a negative RPI effect on change in perfor-
mance in the reward condition. The performance of those
who received RPI decreased from the ?rst quarter to the
fourth quarter, while the performance of those who did
not receive RPI increased (À62.08 vs. +24.17, t = 1.93,
Table 3
Tests of H2a and H2b.
Source of variation df Mean square F p-Value
Panel A: Omnibus ANOVA on performance
a
Between subjects
RPI
b
1 9680.00 0.05 0.83
*
Prize structure
c
1 3889620.00 18.77
Objective: We investigate whether the effect of relative performance information on tournament
performance depends on the tournament’s prize structure. We focus on the effect
of relative performance information on two tournament prize structures: a two-tier structure
in which only the top performer receives a reward and all other contestants receive an
equal payoff that is lower (reward tournament) and a three-tier prize structure in which
the top performer receives a reward, the bottom performer receives a penalty equal to
the amount of that reward, and all remaining contestants receive an equal intermediate
payoff (reward and punish tournament).
Materials Method: We investigate how RPI affects performance in tournaments with different
prize structures via an experiment. In our experiment, each participant competes in
a multi-period tournament against four other participants on a task in which performance
is a function of both individual effort and common uncertainty.
Results: We find that, compared with when relative performance information is not present,
relative performance information has a negative effect on performance in a reward
tournament but a positive effect on performance in a reward and punish tournament.
Relative performance information in tournaments with
different prize structures
Andrew H. Newman
a
, Ivo D. Tafkov
b,?
a
University of South Carolina, Darla Moore School of Business, United States
b
Georgia State University, J. Mack Robinson College of Business, United States
a b s t r a c t
Objective: We investigate whether the effect of relative performance information on tour-
nament performance depends on the tournament’s prize structure. We focus on the effect
of relative performance information on two tournament prize structures: a two-tier struc-
ture in which only the top performer receives a reward and all other contestants receive an
equal payoff that is lower (reward tournament) and a three-tier prize structure in which
the top performer receives a reward, the bottom performer receives a penalty equal to
the amount of that reward, and all remaining contestants receive an equal intermediate
payoff (reward and punish tournament).
Materials Method: We investigate how RPI affects performance in tournaments with dif-
ferent prize structures via an experiment. In our experiment, each participant competes in
a multi-period tournament against four other participants on a task in which performance
is a function of both individual effort and common uncertainty.
Results: We ?nd that, compared with when relative performance information is not pres-
ent, relative performance information has a negative effect on performance in a reward
tournament but a positive effect on performance in a reward and punish tournament. Sup-
plementary analysis reveals that bottom and middle performers drive these differences in
performance, which are due to both differences in effort and in adoption of overly risky
strategies.
Ó 2014 Elsevier Ltd. All rights reserved.
Introduction
A tournament incentive scheme evaluates employees
based on their relative, rather than absolute, performance
level. Tournament compensation is pervasive as more than
half of U.S. corporations use some sort of tournament rank-
ing system that pits employees against colleagues (Berger,
Klassen, Libby, & Webb, 2013; Chen, Williamson, & Zhou,
2012; Hazels & Sasse, 2008; McGregor, 2006). Firms use
tournaments for a variety of employee-types ranging from
production-line workers to salespersons to mutual fund
managers. Employees compete for promotions, bonuses,
and even prizes such as luxurious trips or prime parking
spots (Backes-Gellner & Pull, 2013; Cerdin & Pargneux,
2009; Kempf & Ruenzi, 2008). Tournaments are common
because they can increase productivity by motivating
employees, allow ?rms to avoid paying risk-averse
employees for bearing additional risk associated with com-
mon uncertainty, and help ?rms differentiate the talent of
their workforce (Grote, 2002, 2005; Ng & Lublin, 2010).
Much prior research has focused on understanding when
tournaments produce superior outcomes for the ?rm rela-
tive to other compensation schemes (e.g., Hannan,
Krishnan, & Newman, 2008; Lazear & Rosen, 1981;http://dx.doi.org/10.1016/j.aos.2014.05.004
0361-3682/Ó 2014 Elsevier Ltd. All rights reserved.
?
Corresponding author. Address: 35 Broad Street, Suite 510, Atlanta,
GA 30303, United States. Tel.: +1 404 413 7226; fax: +1 404 413 7203.
E-mail address: [email protected] (I.D. Tafkov).
Accounting, Organizations and Society 39 (2014) 348–361
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
Nalebuff & Stiglitz, 1983). Beyond understanding the set-
tings in which tournaments may be preferable to other
incentive schemes, it is also important for ?rms to under-
stand how the design of the tournament itself can in?u-
ence its effectiveness.
We investigate two key elements of tournament design:
relative performance information (RPI) and prize structure.
We focus on RPI and prize structure because accountants
help ?rms determine the feedback that should be provided
to decision-makers (Bonner & Sprinkle, 2002) as well as
design employee compensation plans (Atkinson, Banker,
Kaplan, & Young, 2001; Indjejikian, 1999). Speci?cally,
we are interested in whether the tournament’s prize struc-
ture in?uences the effect that RPI has on tournament
performance.
In terms of tournament prize structure, many ?rms
incorporate rewards such as monetary bonuses, trips,
and promotions (Gilpatric, 2009; Grote, 2005). Mean-
while, according to a recent survey, 60% of Fortune 500
companies use some form of relative performance-based
compensation scheme that incorporates both reward and
punishment (Cohan, 2012; Kwoh, 2012). Firms such as
General Electric, Metlife, Microsoft, American Express,
AIG, Hewlett Packard, and Yahoo! have used tourna-
ments that incorporate both rewards and punishment,
with punishments coming in such forms as job reassign-
ment, demotion, or even ?ring (Cohan, 2012; Grote,
2005; Kwoh, 2012).
1
For example, General Electric used
relative performance assessments to sort employees into
three groups: a top 20% to whom rewards, promotions
and stock options are showered, a middle 70%, and a bot-
tom 10% who are either ?red or face other disciplinary
actions.
Consistent with the evidence that some ?rms rely
solely on rewards while others rely on both rewards and
punishment, we focus on two tournament prize struc-
tures: a two-tier structure in which only the top performer
receives a reward, and all other contestants receive an
equal payoff that is lower (hereafter, a reward tourna-
ment) and a three-tier structure in which the top per-
former receives a reward, the bottom performer receives
a penalty equal to the amount of that reward, and all
remaining contestants receive an equal intermediate pay-
off (hereafter, a reward and punish tournament). Both
prize structures reward the top performer, and the key dif-
ference across the two structures relates to the treatment
of the nonwinners.
2
In terms of understanding the effect of RPI in tourna-
ments, prior research has focused only on tournaments that
reward the top performers (Casas-Arce & Martinez-Jerez,
2009; Ederer, 2010; Hannan et al., 2008).
3
The empirical evi-
dence shows that RPI informing participants that they are
likely to win the tournament can motivate higher performing
participants to increase or at least maintain their perfor-
mance levels. If only the top performers are rewarded, how-
ever, the majority of participants will receive RPI indicating
that they are unlikely to win the tournament. This leads these
participants either to reduce effort or adopt overly risky
strategies, both of which have a negative effect on their per-
formance. In total, any positive RPI effect on performance for
higher performing participants is outweighed by the
decrease in performance of participants who receive RPI indi-
cating that they are unlikely to win, resulting in an overall
negative effect of RPI on performance.
Prior ?ndings suggest that, because RPI can negatively
affect performance in tournaments, ?rms may reap greater
bene?ts from not providing RPI to tournament partici-
pants. Given the extensive focus on reward tournaments,
our primary focus is to investigate whether this implica-
tion from reward tournaments also applies to another form
of tournament, the reward and punish tournament. Under-
standing the scope of this implication regarding RPI across
alternative prize structures is important because it has
critical rami?cations for how ?rms design their tourna-
ments as well as their information systems.
We investigate how RPI affects performance in tourna-
ments with different prize structures via an experiment. In
our experiment, each participant competes in a multi-per-
iod tournament against four other participants on a task in
which performance is a function of both individual effort
and common uncertainty. We use a multi-period setting
because it allows for a more precise test of our theory
and enhances the generalizability of our results as tourna-
ments are strategic contests (Rankin & Sayre, 2011) and
organizational decisions are typically made in dynamic
multi-period contexts (Hollenbeck, Ilden, Phillips, &
Hedlund, 1994). We manipulate our ?rst factor, tourna-
ment prize structure, between participants at two levels:
reward or reward and punish. Our reward prize structure
provides the winner of the tournament with a monetary
reward while giving all other participants an equal lower
payoff. In contrast, our reward and punish prize structure
provides the winner with a monetary reward, punishes
the loser with a monetary penalty equal to the amount of
the reward, and gives all other participants an equal inter-
mediate payoff. We hold participants expected pay and
prize spread, i.e., the range of payouts between the top
1
These companies use names like ‘‘rank and yank,’’ ‘‘forced ranking,’’
‘‘stack ranking,’’ ‘‘relative performance rating system,’’ ‘‘talent assessment
system,’’ and ‘‘performance procedure’’ to describe such schemes.
2
The term ‘‘prize structure’’ can also be used to refer to any aspect of the
compensation design of a tournament contract. For example, prior research
uses the term to refer to the number of positive rewards present relative to
the number of tournament participants as well as to refer to the prize
spread of any such positive rewards (e.g., Ehrenberg & Bognanno, 1990a,
1990b; Freeman & Gelber, 2010; Knoeber & Thurman, 1994; Lazear &
Rosen, 1981; Lynch, 2005; Orrison, Schotter, & Weigelt, 2004).
3
An exception is Freeman and Gelber (2010) who investigate the effect
of RPI in a six-tier reward prize structure in which all contestants but one
receive a reward and the reward amount per contestant increases gradually
with rank. Freeman and Gelber’s setting differs from ours in an important
way. They use a static, one-period tournament in which contestants
compete after receiving information on their relative abilities. Such a
setting leaves room for effort intensity, but not strategy development, to
in?uence performance. In contrast, we focus on a dynamic, multi-period
setting in which participants must make strategic decisions after being
periodically updated on their chances of winning. We do so because prior
research acknowledges that strategy development is an important dimen-
sion of effort (Bonner & Sprinkle, 2002) and shows that it can affect
tournament performance (Hannan et al., 2008). We discuss the implications
of Freeman and Gelber’s study as they pertain to our study in the
conclusion.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 349
performer and bottom performer, constant across these
two prize structures. We manipulate our second factor,
RPI, between participants at two levels: present or absent.
When RPI is present, we provide RPI in the form of perfor-
mance rank at speci?ed intervals throughout the experi-
ment. When it is absent, we never provide participants
with information about their performance rank during
the tournament.
Consistent with our predictions, when RPI is absent
participants perform better in a reward and punish tourna-
ment than in a reward tournament. We also ?nd a
disordinal interaction between RPI and tournament prize
structure. Speci?cally, RPI has a negative effect on perfor-
mance in a reward tournament but a positive effect on
performance in a reward and punish tournament. As a
result, participants performance is greatest when RPI is
present in a reward and punish tournament. Supplemental
analysis reveals that differences across conditions in
bottom and middle performers’ effort levels and adoption
of overly risky strategies drive the differential effect of
RPI on performance across the two prize structure
conditions.
Our study highlights the inter-related nature of com-
pensation and information system design choices and has
important implications for the design of effective compen-
sation and information systems in a tournament setting.
We contribute to prior research investigating tournaments
in two ways. First, we provide insights on the effect of a
prize structure that includes reward and punishment on
tournament performance. This matters because relative
performance-based compensation contracts that incorpo-
rate reward and punishment are often found in practice
(Cohan, 2012; Grote, 2002, 2005), but little is known about
their effectiveness (Frederickson & Waller, 2005;
Moldovanu, Sela, & Shi, 2012).
Second, we contribute to an emerging stream of
research that has only just begun to consider the role that
prize structure plays in determining the effect of RPI on per-
formance in tournaments (Freeman & Gelber, 2010;
Hannan et al., 2008). We showthat the downside of provid-
ing RPI when only top performers are rewarded, as docu-
mented by prior research and our study, is not universal
to all tournament prize structures. Rather, we provide evi-
dence that the appropriateness of providing RPI varies
based on the tournament’s prize structure. In doing so, we
highlight that ?rms should keep in mind the purpose of
their tournaments when determining whether to provide
RPI. For instance, considering the two tournament prize
structures we examine, our results suggest that, if ?rms
value the overall output of their employees in a tourna-
ment, then they should withhold RPI under a reward prize
structure but provide RPI under a reward and punish prize
structure. If, instead, ?rms value only the output of the
highest performing employee in a tournament, our results
imply that ?rms should consider providing RPI to employ-
ees under both prize structures as RPI has a positive effect
on top performers’ performance under both prize struc-
tures. Finally, in some ?rms that use tournaments, RPI
may be readily available via informal sources (Hannan,
McPhee, Newman, & Tafkov, 2013). Our study suggests that
such ?rms do not necessarily need to discontinue their
tournaments but rather should avoid using a prize struc-
ture that only rewards the top performer.
4
Theory and hypotheses
Tournament prize structure
In a multi-period tournament, performance differences
across contestants emerge over time (Casas-Arce &
Martinez-Jerez, 2009; Ederer, 2010; Hannan et al., 2008).
When RPI is absent, however, contestants are not informed
of these differences and thus lack the insights necessary to
infer where they stand relative to others with respect to
probability of winning. Without RPI, contestants are left to
form assumptions about their relative performance, which
they are likely to deem above-average but not extremely
high. Researchshows that most individuals believethat they
areabove-averageperformers but strategicallyavoidsetting
performance expectations that are too high to prevent per-
formance anxiety and disappointment (Brown, 2007,
2012; Hoorens, 1995; Norem & Cantor, 1986).
We develop our ?rst hypothesis in the context of a tourna-
ment setting where contestants must make assumptions
about their relative standing when deciding how much effort
to exert as well as how much risk to take in an attempt to
improve their performance. To do so, we rely on two single-
period tournament models by Gilpatric (2009) and
Moldovanu and Sela (2001). Although, as discussed below,
there are differences between the settings of these models
and the tournament setting we study, we expect that impor-
tant intuitions fromthesemodels will carryover toour setting.
Gilpatric’s (2009) model provides insights into the design
of a tournament prize structure when identical contestants
simultaneously choose both the mean and the variance of
their performance andwhena ?rmcares about the aggregate
output of all performers, not just the top performer. Accord-
ing to the model, for the ?rm to motivate a desired level of
effort among contestants without inducing costly risk seek-
ing, it should implement a three-tier prize structure instead
of a two-tier one. Speci?cally, the ?rmshouldprovide a mon-
etary reward to the top performer and a monetary punish-
ment to the bottom performer and give all other
contestants some equal amount in between these two pay-
ments (withthe probabilities andsize of the rewardandpun-
ishment being the same). The intuition for this is that,
although the reward can motivate tournament contestants
tochoose highlevels of productive effort, it canalso motivate
themto adopt overly risky strategies. Speci?cally, when fac-
ing multiple opponents, a contestant’s probabilityof winning
a tournament that rewards only the top performer is very
small unless her or his performance is substantially above
the average. Achieving such performance, via productive
effort, is unlikely whena contestant is facing multiple identi-
4
The ?rm’s purpose in using a tournament is likely in?uenced by the
setting in which it is implemented. For instance, in production or sales
settings, ?rms typically use tournaments to motivate the performance of all
of their employees, while in research and development settings, they are
often more concerned with motivating employees to generate at least one
truly innovative idea regardless of how many inferior ideas are also
generated.
350 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
cal opponents. Risky strategies increase the variance in per-
formance output, which can potentially lead to the extreme
performance outcomes neededtowinthe tournament. How-
ever, increasing the variance in performance output beyond
some point may come at the cost of decreased expected per-
formance. The reason, according to Gilpatric, is that increas-
ing the variance in performance output beyond some point
is possible only by selecting strategies with lower expected
performance. Alternatively, seeking out overly risky strate-
gies is costly because it requires time and effort that could
otherwise be devoted toward exerting productive effort.
Under a prize structure that rewards only the top per-
former, the absence of punishment for losing the tourna-
ment means there is less for contestants to lose from
attempting to win the tournament by adopting overly risky
strategies that decrease their expectedperformance. Includ-
ing punishment in the prize structure means that adopting
such risky strategies nowalso increases contestants’ proba-
bility of receiving the punishment, which should reduce
contestants’ incentives to adopt overly risky strategies in
an attempt to win. As such, contestants should adopt fewer
overly risky strategies, resulting in greater expected perfor-
mance, when the prize structure includes both reward and
punishment compared to when it includes only a reward.
Gilpatric’s intuition pertains to the tournament setting
we study even though his model is single-period and
assumes identical contestants. The reason is that the
absence of RPI in a multi-period tournament means con-
testants have no basis by which to conclude that they have
a higher likelihood of winning the tournament than any
other contestant. This is likely to lead contestants to con-
clude that they cannot achieve the extreme performance
needed to win the tournament only by exerting additional
productive effort. Accordingly, similar to Gilpatric’s model,
we expect that, in the absence of punishment (i.e., in the
reward prize structure), contestants are likely to adopt
overly risky strategies in an attempt to win the reward.
Moldovanu and Sela’s (2001) model investigates prize
structure design in a tournament in which contestants with
different privately known abilities simultaneously choose
effort levels and the ?rm is concerned with maximizing
the aggregate output of all performers. According to their
model, in which performance is a function only of effort
and ability, a ?rm should provide more than one reward to
its contestants. The intuition is that second-place (or third-
place, etc.) rewards are more motivating to contestants with
lower abilities thanthe ?rst-place reward, whichmany con-
testants believe they are extremely unlikely to win.
This model’s intuitionregardingthe potential value of mul-
tiple tiers of rewards is applicable to the tournament setting
we study even though the model does not allow contestants
to adopt risky strategies, and it assumes that contestants have
private information regarding their abilities.
5
Speci?cally, if
contestants ina multi-periodtournament expect their perfor-
mance to be above the average but not extremely high, they
may not assess a high likelihood of receiving the reward in a
reward tournament. Accordingly, the motivational effect
from a reward for the winner may be low. In cases like this,
a runner-up reward, such as the one in a three-tier reward
and punish tournament that is provided to contestants who
are not a bottom performer, may be more motivating than a
reward provided to the top performer.
In summary, based on Gilpatric’s model, we expect that
adding a punishment element to the tournament prize
structure will have a positive effect on performance
because the punishment will discourage the adoption of
overly risky strategies that have lower expected perfor-
mance. Additionally, based on Modovanu and Sela’s model,
we expect that moving from a two-tier reward prize struc-
ture to a three-tier reward and punish prize structure will
have a positive effect on performance because contestants
who have low expectations of winning the ?rst-place
reward will be more motivated by the chance to receive
the runner-up reward provided to contestants who neither
win nor lose the tournament.
Formally stated, the hypothesis is:
H1. When relative performance information is absent,
performance under a reward and punish tournament is
greater than performance under a reward tournament.
RPI and tournament prize structure
RPI gives contestants in multi-period tournaments
information on the evolving differences in performance
among their peers that arise from the dynamic nature of
a multi-period tournament (Casas-Arce & Martinez-Jerez,
2009; Ederer, 2010; Hannan et al., 2008). In doing so, RPI
helps contestants make more informed effort decisions in
subsequent periods of the tournament. Both economic
and psychology theories predict that the ultimate effect
of RPI on performance depends on how the RPI’s content
affects contestants’ perceptions of their relative standing
in the tournament. We develop theory to predict that the
difference in overall performance predicted in H1 between
a three-tier prize structure that uses rewards and punish-
ment and a two-tier prize structure that provides only a
reward to the winner will be magni?ed by RPI. Speci?cally,
we develop theory to predict that providing RPI will
decrease performance under a reward prize structure and
increase performance under a reward and punish prize
structure.
6
5
Speci?cally, Moldovanu and Sela’s model assumes that contestants
receive a private signal ex ante regarding their ability as well as where their
ability lies on the distribution of contestants’ potential abilities. Taken
together, these two pieces of information provide contestants with insight
on their relative performance potential before the single-period tourna-
ment. In our tournament setting, contestants do not have this information
ex ante.
6
The single-period models by Gilpatric (2009) and Moldovanu and Sela
(2001) that we relied on to develop our ?rst hypothesis are less useful in
predicting contestant behavior in a multi-period tournament setting when
contestants receive RPI. The reason is that RPI allows contestants to infer
their likelihood of winning the tournament and thus to update their
strategy during the multi-period tournament. Prior research shows that
individual behavior can change substantially when moving from a single-
period setting to a multi-period setting in which interim feedback that
allows for strategy updating, is provided (Hollenbeck et al., 1994; Thaler &
Johnson, 1990). Accordingly, we no longer rely on these models when
developing our remaining hypotheses.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 351
We ?rst present theory to predict that RPI will decrease
performance under a reward prize structure. While this
prediction replicates Hannan et al.’s (2008) result, outlin-
ing the rationale for this prediction helps highlight the
key differences that emerge when RPI is provided under
a reward and punish prize structure. Hannan et al. (2008)
predict and ?nd that, in a reward tournament, RPI moti-
vates the higher performers because it allows them to infer
that they have a high probability of receiving the reward
and thereby have a high marginal bene?t of exerting addi-
tional effort. They thus continue to perform at a high level.
The authors also predict and ?nd, however, that any poten-
tial increase in performance by the higher performers is
outweighed by the decrease in performance from the
lower and middle performers. These performers receive
RPI that indicates that they are unlikely to win the reward
and thereby have a low marginal bene?t of exerting addi-
tional effort. These contestants therefore either decrease
their effort or adopt overly risky strategies that have lower
average expected performance in an attempt to catch up to
their peers. Both approaches result in decrease in perfor-
mance of these contestants.
We expect to replicate the ?ndings of Hannan et al.
(2008). Formally stated, the hypothesis is:
H2a-replication. Relative performance information has a
negative effect on performance in a reward tournament.
We now present theory to predict that RPI will
increase performance under a reward and punish prize
structure. In a reward and punish tournament, the mar-
ginal bene?t of effort increases as one’s performance
moves toward either of the two ends of the performance
spectrum (i.e., very high or very low) because at these
ends the probability of receiving the reward or the pun-
ishment is high. Recall that in the absence of RPI, employ-
ees will expect their performance to be above the average
but not necessarily substantially above it. RPI informs
higher performers that their performance is at the high
end of the performance spectrum and thus that their
probability of receiving the reward is higher than they
expect in the absence of RPI. Such RPI should motivate
these contestants to maintain or even increase their per-
formance levels. From an economic theory perspective,
higher performing contestants have a relatively high mar-
ginal bene?t of exerting additional productive effort
because doing so increases their likelihood of receiving
the reward.
7
From a psychology theory perspective, social
comparison theory suggests that RPI motivates higher per-
forming contestants to sustain high performance levels as
they wish to maintain their relative standing (Beach &
Tesser, 1995; Tesser, 1988). In addition, RPI indicating that
the goal of winning the reward is more attainable than
expected in the absence of RPI boosts contestants’ perfor-
mance expectancy, self-ef?cacy perception and goal com-
mitment (Bandura, 1986; Bandura & Wood, 1989; Locke &
Latham, 1990), all of which also motivate higher performers
to sustain or even increase their performance levels in sub-
sequent periods of the tournament.
Regarding lower performers, both economic and psy-
chology theories predict that RPI has a positive effect on
performance under a reward and punish tournament. In a
reward and punish tournament, contestants worry about
both winning the reward and avoiding the punishment.
In the absence of RPI, if contestants assume that their per-
formance is above average (but neither extremely high nor
extremely low), this means lower performers are over esti-
mating their probability of avoiding punishment. RPI
informs lower performers that their performance is at the
low end of the performance spectrum and thus that their
probability of avoiding punishment is lower than they
expected in the absence of RPI. Therefore, from an eco-
nomic theory perspective, RPI informs lower performers
that their marginal bene?t of additional effort is high
because, although additional effort is unlikely to help them
win the reward, it can help them avoid the punishment.
8
Additionally, the presence of punishment imposes a cost
on attempting to increase performance via the adoption of
overly risky strategies if such an approach ends up actually
decreasing performance. Accordingly, we expect lower per-
formers in the reward and punish tournament to respond
to RPI by exerting greater productive effort instead of adopt-
ing overly risky strategies. Thus RPI should have a positive
effect on their performance.
Psychology theory leads to similar predictions for lower
performers. While social comparison theory (Brown, Ferris,
Heller, & Keeping, 2007; Buunk & Gibbons, 2007) predicts
that lower performing contestants are motivated to
improve their performance to enhance their relative stand-
ing, expectancy theory, social cognitive theory, and goal-
setting theory (Bandura, 1986; Locke & Latham, 1990)
impose a boundary condition on this prediction. According
to these theories, the size of the goal–performance discrep-
ancy (i.e., the difference between a performance goal and
actual performance) in?uences the effect that RPI has on
performance. A smaller goal–performance discrepancy
motivates individuals because it increases one’s perfor-
mance expectancy and goal commitment. This has a posi-
tive effect on performance. A larger discrepancy
demotivates individuals because it lowers one’s perfor-
mance expectancy and goal commitment. This has a nega-
tive effect on performance. In addition, a larger
7
Higher performing contestants include the top performers, who will
win the reward if they keep their current performance rank, and near the
top performers, who will win the reward only if they improve their rank.
The extent to which the marginal bene?t of additional effort is high may
vary across these two types of performers. While near the top performers
will be motivated to increase their performance to become top performers
and win the reward, the top performers will be motivated to either
maintain their performance if they believe they can win or work harder if
they deem their current performance to be insuf?cient to win due to an
expected increase in performance of near the top performers. We inves-
tigate this issue in our results section.
8
Lower performers include the bottom performers, who will avoid
punishment only if they improve their current performance rank, and near
the bottom performers, who will avoid punishment if they keep their
current performance rank. While the bottom performers are motivated to
increase their performance level to avoid punishment, the near the bottom
performers are motivated to either maintain their performance, if they
deem it suf?cient to avoid the punishment, or increase it, if they deem
maintaining will be insuf?cient to avoid the punishment due to an
expected increase in performance of the bottom performers. We investigate
this issue in our results section.
352 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
goal–performance discrepancy can lead to low strategy
ef?ciency (i.e., ineffective strategy searches or excessive
experimentation), which also hurts performance (Kluger
& DeNisi, 1996).
In a reward and punish tournament, contestants have
two goals: to win by having the highest performance
and to avoid losing by not having the lowest performance.
With regard to the goal of winning the tournament, RPI
indicating that one is a low performer creates a larger
goal–performance discrepancy compared to when contes-
tants receive no RPI and assume that their performance is
above the average. With regard to the goal of not losing
the tournament, RPI creates a goal–performance discrep-
ancy where a discrepancy is absent when contestants do
not have RPI and assume that they are unlikely to be
the one with extremely low performance. Lower perform-
ers should view this latter discrepancy as more manage-
able to overcome than the one associated with the goal
of winning the tournament. The reason is that they now
know that winning the tournament requires outperform-
ing multiple higher-performing contestants whereas not
losing it and, thus, winning the runner-up reward requires
outperforming only one other low performing contestant.
Prior research shows that individuals respond to large
goal–performance discrepancies by lowering their goals
to reduce these discrepancies (Donovan & Williams,
2003; Locke & Latham, 2006). Lower performers in a
reward and punish tournament can do so by focusing on
the goal of not losing rather than the goal of winning
the tournament. Because lower goal–performance dis-
crepancies are more motivating than larger discrepancies
(Locke & Latham, 1990), this shift should motivate lower
performing contestants.
Regarding the middle performers, RPI informs them
that they are slightly less likely to win a reward and
slightly more likely to incur a punishment than they per-
ceive in the absence of RPI. Overall, however, RPI does
not create signi?cant belief revision for these contestants
who expected that their performance would be neither
extremely high nor extremely low. From an economic per-
spective, the slightly decreased marginal bene?t of exert-
ing additional effort to win the reward is
counterbalanced by the slightly increased marginal bene?t
of exerting additional effort to avoid the punishment and
receive the runner up reward. Accordingly, we do not
expect middle performers in the reward and punish tour-
nament to behave differently in the presence of RPI than
in the absence of RPI. Psychology theory leads to similar
predictions for these performers because RPI does not sub-
stantially change their perceived goal–performance
discrepancies.
In summary, relative to when RPI is absent, we predict
that in a reward and punish tournament the presence of
RPI leads to an increase in overall performance. The reason
is that RPI has positive effect on the performance of lower
performers, positive or no effect on the higher performers,
and no effect on the middle performers. Formally stated,
the hypothesis is:
H2b. Relative performance information has a positive
effect on performance in a reward and punish tournament.
Method
Experimental task
We adapt our experimental task from Sprinkle (2000)
and Hannan et al. (2008). We assign each participant to
the role of a production manager responsible for making
production quantity decisions. The task consists of 60 deci-
sion periods broken into 12 trials with ?ve decision periods
per trial. In each period, the objective is to select the pro-
duction quantity (from 1 to 20) that maximizes the com-
pany’s total pro?t (measured in terms of total pro?t
points earned). Each participant competes in a tournament
against four other participants, where participants’ ?nal
performance rank is determined by the total pro?t points
earned over the 60 decision periods.
9
As shown in Fig. 1, the pro?t points earned for a partic-
ular production decision are determined jointly by the par-
ticipant’s production quantity decision and the economic
condition. The economic condition represents a state of
nature and serves as a source of common uncertainty for
participants because they are not informed of the eco-
nomic condition for each trial ex ante. Participants are only
aware that the economic condition ranges from 1 to 20
with equal probability and that, although it remains con-
stant for each of the ?ve periods within a trial, it may vary
across the 12 trials. Each participant faces the same set of
economic conditions, in the same order.
10
Participants have 180 s to complete the ?ve periods of a
trial. A clock is displayed on the computer screen. At the
end of each period within a trial, participants choose
whether they want feedback on their performance. If they
choose to receive feedback, they receive a summary of
their production quantity choices made for each period
in the current trial as well as the number of pro?t points
earned for each of those choices. This feedback remains
on the screen for a minimum of 10 s. Viewing feedback
at the end of a period may help participants make more
pro?table production quantity decisions (i.e., earn more
pro?t points) in the remaining periods of that trial. The
pro?t points earned for a production decision depend on
the economic condition for that trial. Because the eco-
nomic condition remains constant for all ?ve periods,
viewing feedback during a trial can help participants infer
the economic condition for the current trial. To assist with
this inference, we provide participants a table similar to
Fig. 1.
9
We use ?ve-participant tournaments because this allows us to test our
theory in a most ef?cient way. Recall that we argue that RPI can have
different effects on the performance of higher, lower, and middle perform-
ers. Also recall that we differentiate between top performers and near-the-
top performers when discussing the effect of RPI on higher performing
contestants, and between bottom performers and near-the-bottom per-
formers when discussing the effect of RPI on lower performing contestants.
This suggests that we need to have at least ?ve participants in a
tournament to test our predictions. Increasing the size of a tournament
beyond ?ve participants increases the overall number of participants
needed and thereby the cost of our experiment without providing any clear
bene?t with respect to testing our theory.
10
We randomly selected the economic condition for each trial. To
improve control, we made these random choices before the experiment
sessions.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 353
In addition to any compensation based on relative per-
formance, participants also earn $0.01 for every ?ve sec-
onds saved out of the 180 s available per trial. Any
potential gains from inferring the economic condition by
viewing feedback therefore come at a cost because every
time a participant chooses to view feedback for 10 s he is
reducing the compensation he can earn for early comple-
tion of the task. This cost of viewing feedback serves as
our proxy for participants’ disutility of effort. The use of
such a proxy for a cost of effort is a common practice in
experimental accounting research (Hecht, Tafkov, &
Towry, 2012; Sprinkle, 2000) because the number of feed-
back requests meets the three criteria for a measure of task
effort suggested by Baiman (1982).
At the end of each trial, participants in all conditions
automatically receive individual performance feedback.
Speci?cally, they learn the number of pro?t points earned
in the current trial, the cumulative number of pro?t points
earned in all trials completed so far, the number of avail-
able seconds not used in the current trial, and the cumula-
tive number of available seconds not used in all trials
completed so far.
Experimental design
To investigate our hypotheses, we use a multi-period
tournament in which participants make strategic decisions
after being periodically updated on their chances of win-
ning. We use a 2 Â 2 Â 12 mixed experimental design.
The two primary independent variables are the tourna-
ment prize structure (reward or reward and punish) and
RPI (present or absent). We manipulate both variables
between-participants as described in more detail below.
We manipulate trial within-participants as participants
make output-quantity decisions in 60 periods divided into
12 trials of ?ve decision periods each. Our primary depen-
dent variable is performance, measured as the total pro?t
points earned.
Prize structure manipulation
In the reward condition, all participants earn $16 for
completing the 60 decision periods. The participant who
earns the most pro?t points among the ?ve participants
in his respective session receives a reward of $20. In the
reward and punish condition, all participants earn $20 for
completing the 60 decision periods. The participant who
earns the most pro?t points receives a reward of $10, while
the participant who earns the fewest pro?t points is penal-
ized $10. This experimental design ensures that the
expected value of participants’ earnings equals $20 in both
tournament compensation contracts.
11
In addition to holding the expected value of earnings
constant, we also hold the overall prize spread, i.e., range
of earnings from the top performer to the bottom per-
former, constant across the two incentive contract condi-
tions. Speci?cally, the prize spread always equals $20
(reward equals +$20 in reward condition; reward equals
+$10 and penalty equals À$10 in reward and punish
Production Quantity
Economic
Condition 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 5 5 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 5 5 10 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 5 5 10 20 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 5 5 10 20 20 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 5 5 10 20 20 30 30 0 0 0 0 0 0 0 0 0 0 0 0 0
8 5 5 10 20 20 30 30 30 0 0 0 0 0 0 0 0 0 0 0 0
9 5 5 10 20 20 30 30 30 45 0 0 0 0 0 0 0 0 0 0 0
10 5 5 10 20 20 30 30 30 45 45 0 0 0 0 0 0 0 0 0 0
11 5 5 10 20 20 30 30 30 45 45 60 0 0 0 0 0 0 0 0 0
12 5 5 10 20 20 30 30 30 45 45 60 60 0 0 0 0 0 0 0 0
13 5 5 10 20 20 30 30 30 45 45 60 60 60 0 0 0 0 0 0 0
14 5 5 10 20 20 30 30 30 45 45 60 60 60 80 0 0 0 0 0 0
15 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 0 0 0 0 0
16 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 0 0 0 0
17 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 0 0 0
18 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 0 0
19 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 100 0
20 5 5 10 20 20 30 30 30 45 45 60 60 60 80 80 95 95 95 100 100
Note: For example, suppose a participant chose a production quantity of 10. If the economic condition was a number from 1 to 9, the participant would earn 0
profit points. If the economic condition was a number from 10 to 20, the participant would earn 45 profit points.
Fig. 1. Pro?t points table.
11
Although we exogenously impose these prize structures, we acknowl-
edge that prize structures can also arise endogenously as a function of the
task characteristics. For instance, according to Aron and Olivella’s (1994)
analytical model, penalty-based contracts emerge in equilibrium in jobs in
which performance is easily measured, whereas reward-based contracts
emerge in equilibrium in jobs in which performance needs to be evaluated
subjectively. Given that the focus of our study is on understanding how the
presence of RPI affects participant behavior across two particular prize
structures, we determined that exogenously imposing the prize structures
was more appropriate. We further discuss this issue in our conclusion.
354 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
condition). We hold the prize spread constant because
prior research shows that it can affect the effort level
exerted in the tournament (Ehrenberg & Bognanno,
1990a, 1990b; Lazear & Rosen, 1981). Without holding
the prize spread constant, we could not disentangle the
prize spread’s effect on performance from the prize struc-
ture’s effect on performance.
Holding the overall prize spread constant while chang-
ing from a two-tier prize structure in the reward condition
to a three-tier prize structure in the reward and punish
condition creates underlying differences in the two con-
tracts. Speci?cally, the prize spread of $20 between the
winner and the nonwinners in the reward condition (com-
puted as $36–$16) is greater than the prize spread of $10
between the winner and those that neither win nor lose
(i.e., those in the middle) in the reward and punish condi-
tion (computed as $30–$20). Although the participants in
the reward and punish condition have a lower cost of not
winning than those in the reward condition ($10 vs. $20),
the participants in the reward and punish condition also
have a greater cost of losing than those in the reward con-
dition ($10 vs. $0). Consistent with our hypothesis devel-
opment, we expect that the combination of a $10 cost of
losing and a $10 cost of not winning in the reward and
punish condition will motivate more productive effort than
a $20 cost of not winning in the reward condition.
It is important to highlight that our reward and punish
prize structure differs from our reward prize structure in
two ways. First, the reward and punish prize structure
includes a penalty element that is lacking from the reward
prize structure. Second, the reward and punish prize struc-
ture is a three-level prize structure that sorts participants
as winners, nonwinners, and losers, whereas the reward
prize structure is a two-level prize structure that sorts peo-
ple as winners and nonwinners. The purpose of our exper-
iment design is not to distinguish the incremental effect
each of these two differences has on tournament perfor-
mance but to compare the effect of a reward and punish
prize structure to that of a reward prize structure. Put dif-
ferently, we cannot comment on whether any differences
in performance across our prize structure conditions are
driven by the presence of punishment, the addition of a
third-tier to the pay structure, or both.
RPI manipulation
We manipulate RPI between-participants at two levels:
absent or present. In the absent condition, we do not pro-
vide RPI during the tournament. In the present condition,
we provide participants with their cumulative perfor-
mance rank, from #1 to #5, on a quarterly basis at the
end of the 3rd, 6th, 9th, and 12th trials. Importantly, in
addition to knowing that everyone faces the same eco-
nomic conditions over the course of the 12 trials, partici-
pants in all conditions are also aware that all participants
have the same three states of nature in each quarter. In
other words, participants know that the quarterly RPI is
informative because the common error is the same for all
participants. This also means that the total amount of pos-
sible pro?t points is the same for all participants in each
quarter. RPI appears on the computer screen following
the individual performance feedback and remains on the
screen for 15 s.
Participants and procedures
In total, 80 students from a large public university par-
ticipated in 16 experimental sessions (four per experimen-
tal condition). Participants were randomly assigned to one
of the four experimental conditions based on which ses-
sion they attended. The mean age of the participants was
22.2 years, and 55% of them were female. There were no
signi?cant differences across conditions for age and gender
(all p > 0.18, two-tailed). Each session was conducted in an
experimental computer lab using z-tree software
(Fischbacher, 2007) and lasted approximately 75 min.
Upon arrival, we assigned participants a participation
number and provided them with a written description of
the experiment, examples of the task they would be com-
pleting, and details regarding their compensation contract.
We then required them to complete a short quiz to ensure
that they understood the experiment. Upon completion of
the quiz, the ?rst trial began. After the end of the 12th trial,
participants completed a post-experimental questionnaire
that collected process-related and demographic informa-
tion. Finally, to assure anonymity, we paid participants
via sealed envelopes based on their previously selected
participation number. Including a $5 show-up fee, partici-
pants earned an average of $28.13 on the task.
Results
Table 1 reports descriptive statistics related to our main
dependent variable, performance. Fig. 2 provides a graphi-
cal summary of the results.
Hypotheses tests
H1 predicts that, when RPI is absent, performance will
be greater under the reward and punish prize structure
than under the reward prize structure. Our test results,
presented in Table 2, show that, when RPI is absent, partic-
ipants in the reward and punish condition earn signi?-
cantly more pro?t points than participants in the reward
condition (2239.00 vs. 1990.75 from Table 1, t = 3.18,
p = 0.01, one-tailed).
12
This supports H1.
H2a and H2b predict an interaction between RPI and
prize structure, such that the effect of RPI on performance
is negative under a reward prize structure and positive
under a reward and punish prize structure. Panel A of
Table 3 presents an ANOVA for which the dependent vari-
able is pro?t points earned and the two independent vari-
ables are RPI and prize structure. As expected, we ?nd a
signi?cant RPI Â prize structure interaction (F = 3.86,
p = 0.05, two-tailed). Fig. 2 shows that the interaction is dis-
ordinal, that is, of the form predicted in H2a and H2b.
12
We also run this and subsequent tests using OLS regressions, in which
we cluster the data by tournament group (i.e., by each group of ?ve
participants competing against each other). Results of these analyses are
inferentially identical to those reported in the paper.
A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361 355
Simple effects analysis, presented in Panel B of Table 3,
indicates that under a reward prize structure participants
earn fewer pro?t points when RPI is present than when
RPI is absent (1776.00 vs. 1990.75 from Table 1, t = 1.87,
p = 0.03, one-tailed). This result replicates Hannan et al.
(2008) and supports H2a. However, under a reward and
punish prize structure, participants earn more pro?t points
when RPI is present than when RPI is absent (2409.75 vs.
2239.00 from Table 1, t = 2.19, p = 0.02, one-tailed). Thus
H2b is also supported. Overall, the results support our pre-
diction that the effect of RPI on tournament performance
depends on the tournament’s prize structure.
Following from H1, H2a, and H2b performance should
be greatest when RPI is provided under a reward and
punish prize structure. Our tests’ results (untabulated)
show that participants earn signi?cantly more pro?t points
Table 1
Descriptive statistics: performance
a
(mean, (standard deviation)).
Prize structure
c
Reward Reward and punish
Relative performance information
b
No RPI 1990.75 (453.75) 2239.00 (319.80)
RPI 1776.00 (688.81) 2409.75 (215.15)
a
Performance is the total pro?t points earned during the 12 trials (60 periods). The pro?t points earned in each period are based on the economic
condition in that trial as well as participants’ production quantity choice. The pro?t points earned in a period represent the performance outcome for that
period. As such, total pro?t points earned over all 60 periods (12 trials) measures total performance.
b
RPI was manipulated at two levels. Participants in the No RPI condition received no relative performance information during the course of the
tournament or at the end. Participants in the RPI condition received their relative performance rank, from #1 to #5, at the end of trials 3, 6, 9, and 12.
c
Prize structure was manipulated at two levels: a reward was paid to only the winner of the tournament (reward condition) or a reward was paid to the
winner and a penalty was imposed on the loser (reward and punish condition). The expected pay and the prize spread were held constant across these two
conditions.
Reward
Reward and
Punish
Fig. 2. Effect of RPI
a
and Prize Structure
b
on Performance
c
.
a
RPI was manipulated at two levels. Participants in the No RPI condition received no relative
performance information on their performance during the course of the tournament or at the end. Participants in the RPI condition received their relative
performance rank, from #1 to #5, at the end of trials 3, 6, 9, and 12.
b
Prize structure was manipulated at two levels: a reward was paid only to the winner of
the tournament (reward condition), or a reward was paid to the winner and a penalty was imposed on loser (reward and punish condition). The expected
pay and the prize spread were held constant across these two conditions.
c
Performance is the total pro?t points earned during the 12 trials (60 periods). The
pro?t points earned in each period are based on the economic condition in that trial as well as participants’ production quantity choice. The pro?t points
earned in a period represent the performance outcome for that period. As such, total pro?t points earned over all 60 periods (12 trials) measures total
performance.
Table 2
Test of H1.
Hypothesis Mean difference t-Statistics p-Value
Dependent variable – performance
a
No RPI
b
/R&P
c
> No RPI/reward
c
248.25 3.18 0.01
*
*
All p-values are reported on a one-tailed basis due to our directional predictions for these effects.
a
Performance is the total pro?t points earned during the 12 trials (60 periods). The pro?t points earned in each period are based on the economic
condition in that trial as well as participants’ production quantity choice. The pro?t points earned in a period represent the performance outcome for that
period. As such, total pro?t points earned over all 60 periods (12 trials) measures total performance.
b
RPI was manipulated at two levels. Participants in the No RPI condition received no relative performance information on their performance during the
course of the tournament or at the end. Participants in the RPI condition received their relative performance rank, from #1 to #5, at the end of trials 3, 6, 9,
and 12.
c
Prize structure was manipulated at two levels: a reward was paid only to the winner of the tournament (reward condition) or a reward was paid to the
winner and a penalty was imposed on the loser (reward and punish condition). The expected pay and the prize spread were held constant across these two
conditions.
356 A.H. Newman, I.D. Tafkov / Accounting, Organizations and Society 39 (2014) 348–361
in this condition compared to each of the other three condi-
tions (all p < 0.05, two-tailed). This demonstrates the bene?t
of using rewards and punishment as complements in a
tournament prize structure, especially when RPI is
provided.
Supplemental analysis
RPI content
Our main analysis documents that RPI has a differential
effect on participants’ overall performance across the two
prize structure conditions. This prior analysis focuses on
the presence of RPI. As stated in our hypotheses develop-
ment for H2a and H2b, we expect that differences in how
participants respond to the content of the RPI play a critical
role in creating any performance differences across our
prize structure conditions. To test this expectation, we
examine the effect of RPI content on participants’ change
in performance during the tournament.
To do so, we de?ne change in performance as pro?t
points earned in the last three trials (4th quarter) minus
pro?t points earned in the ?rst three trials (1st quarter).
We use performance in these two quarters to calculate
change in performance because, by experimental design,
the set of economic conditions (states of nature) is the
same for the ?rst and fourth quarters, although the order
is randomized within each quarter. As such, the common
error term does not in?uence this comparison as the total
pro?t potential is the same in the ?rst and fourth quarters.
We classify participants based on participants’ perfor-
mance rank at the midpoint of the tournament. We use
the midpoint RPI because it allows us to directly investi-
gate how the content of the feedback in?uences changes
in participants’ behavior during the tournament.
13
We code
participants with a rank of 1 or 2 as top performers, partic-
ipants with a rank of 3 as middle performers, and those with
a rank of 4 or 5 as bottom performers. However, because
within a given prize structure we ?nd a similar RPI effect
for the middle and the bottom performers, we combine
these groups and refer to them as nontop performers to
avoid redundant discussions.
14
We ?nd a signi?cant three-way RPI Â prize struc-
ture  rank interaction (F = 2.71, p = 0.10, two-tailed). To
gain insight on the nature of the interaction, we conduct
two independent ANOVAs, one each for the top and nontop
performers. For the top performers, we ?nd a positive RPI
effect on change in performance (t = 2.35, p = 0.05, two-
tailed) but no signi?cant RPI Â prize structure interaction
(F = 1.56, p = 0.22, two-tailed). The performance of top per-
formers who received RPI increased from the ?rst quarter
to the fourth quarter more than the performance of top
performers who did not receive RPI in both the reward
condition (51.88 vs. 18.75) and the reward and punish con-
dition (116.25 vs. 58.75). This reveals that, consistent with
our expectations outlined during our hypotheses develop-
ment, providing RPI bene?ts top performers’ performance
over time, regardless of the prize structure.
For the nontop performers, we ?nd an RPI Â prize struc-
ture interaction (F = 8.63, p < 0.01, two-tailed). Simple effect
analysis ?nds a negative RPI effect on change in perfor-
mance in the reward condition. The performance of those
who received RPI decreased from the ?rst quarter to the
fourth quarter, while the performance of those who did
not receive RPI increased (À62.08 vs. +24.17, t = 1.93,
Table 3
Tests of H2a and H2b.
Source of variation df Mean square F p-Value
Panel A: Omnibus ANOVA on performance
a
Between subjects
RPI
b
1 9680.00 0.05 0.83
*
Prize structure
c
1 3889620.00 18.77