Description
Past accounting experiments have demonstrated signi®cant eects of absorption vs variable costing systems on pri-
cing decisions, but in individual settings that suppressed market features. The main ®nding of the current study is that a
cost-based pricing bias did not persist in laboratory product markets. Given the opportunity to learn from pro®t and
market feedback, sellers revised their price oers toward optimum in a manner that compensated for absorption vs
variable cost signals. The eects of demand conditions, as revealed through actual trades, dominated the eects of
alternative costing systems.
Do cost-based pricing biases persist in laboratory markets?
William S. Waller
a,
*, Brian Shapiro
b
, Galen Sevcik
c
a
Department of Accounting, University of Arizona, McClelland Hall, 301X, Tuscon, AZ 85721, USA
b
University of Minnesota, USA
c
Georgia State University, USA
Abstract
Past accounting experiments have demonstrated signi®cant e?ects of absorption vs variable costing systems on pri-
cing decisions, but in individual settings that suppressed market features. The main ®nding of the current study is that a
cost-based pricing bias did not persist in laboratory product markets. Given the opportunity to learn from pro®t and
market feedback, sellers revised their price o?ers toward optimum in a manner that compensated for absorption vs
variable cost signals. The e?ects of demand conditions, as revealed through actual trades, dominated the e?ects of
alternative costing systems. # 1999 Elsevier Science Ltd. All rights reserved.
1. Introduction
A basic issue in accounting is whether alter-
native information systems a?ect economic deci-
sions. When examining this issue, it is important
to consider the organizational or institutional set-
ting of the decision-maker, e.g. type of ®rm or
market (Hopwood, 1978; Libby, 1990). In man-
agement accounting, there is much interest in how
alternative costing systems a?ect pricing decisions
(Kaplan & Atkinson, 1998). Because pricing deci-
sions naturally occur in markets, research on cost-
based pricing should consider the essential fea-
tures of the market setting in which such decisions
are made. The setting is important for three rea-
sons. First, markets typically give sellers access to
information other than accounting signals, e.g.
past prices and o?ers, which may moderate or
overwhelm the e?ects of costing systems. Second,
pricing decisions require sellers to set goals as well
as process accounting signals, e.g. a 20% pro®t
markup on unit cost. Under market pressure, sell-
ers with alternative costing systems may compen-
sate for unit-cost di?erences when setting the
pro®t goal. Third, markets often give sellers the
opportunity to learn from feedback, with the con-
sequence that initially observed e?ects of costing
systems do not persist.
Experimental research strives for internal valid-
ity through the manipulation of independent vari-
ables, e.g. alternative costing systems, and careful
observation of behavioral e?ects, e.g. pricing
decisions, controlling for other factors. But, such
concern for internal validity does not immunize
experimentalists from the need to consider the
organizational or institutional setting of the deci-
sion-maker (Swieringa & Weick, 1982). Regarding
cost-based pricing, there has been a long line of
accounting experiments providing evidence about
the e?ects of absorption vs variable costing sys-
tems on pricing decisions in individual settings
(Ashton, 1976, 1981; Barnes & Webb, 1986;
0361-3682/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PI I : S0361- 3682( 99) 00009- 4
Accounting, Organizations and Society 24 (1999) 717±739
www.elsevier.com/locate/aos
* Corresponding author. Fax: +1-520-621-3742.
E-mail address: [email protected] (W.S. Waller)
Dyckman, Hoskin & Swieringa, 1982; Hilton,
Swieringa & Turner, 1988; Swieringa, Dyckman &
Hoskin, 1979; Turner & Hilton, 1989). The evi-
dence suggests that these alternative costing sys-
tems cause pricing biases, e.g. systematically
higher price o?ers under absorption costing. As
reviewed below, however, past experiments did
not incorporate essential market features such as
actual seller competition, price revision in light of
feedback, and the threat of bankruptcy. The
question remains as to whether cost-based pricing
biases persist in market settings.
This study contributes evidence about the e?ects
of absorption vs variable costing on pricing deci-
sions in laboratory product markets. In each of
eight markets with a posted-price institution, ten
sellers were randomly assigned absorption or
variable costing systems, and competed by making
o?ers to sell units at a price equal to the reported
unit cost plus a pro®t markup. Although their
costing systems di?ered, all sellers faced the same
increasing marginal cost function with an una-
voidable ®xed cost and a variable cost that
depended on unit sales. Sellers realized a pro®t or
loss from their trades, observed other trades, and
revised their o?ers, for each of 48 periods (six sets
of eight periods), and ultimately survived or went
bankrupt. Predictions of optimal prices and
quantities at competitive equilibrium were derived
independently of costing systems. Demand was
manipulated within and between markets. The
within-market manipulation involved a random
shift for each set of periods. This permitted repe-
ated observations of seller learning (i.e. price revi-
sion) from the starting point of ignorance about
demand. The between-market manipulation
involved the threat of bankruptcy; demand condi-
tions implied that only ®ve (ten) optimizing sellers
could break even on average in the harsh (lenient)
markets. This permitted observation of whether
seller learning accelerated when the threat of
bankruptcy was higher.
Comparable to many laboratory-market studies
(Davis & Holt, 1993), sellers' price o?ers con-
verged toward optimum in each set of periods.
Overall, the median price error (a seller's price
o?er vs the optimal price) decreased by more than
80% from period 1 to 8 of a set. Typically, sellers
who failed to make a sale lowered their price
o?ers, resulting in more units to be sold and mar-
ket eciency (actual pro®t as a percentage of
maximum pro®t) to approach 100%. Absorption
vs variable costing systems caused a bias in price
o?ers only in the ®rst period of the ®rst set, when
sellers had not yet observed actual trades and
were ignorant about demand. The e?ects of
demand conditions, as revealed through trading,
quickly overwhelmed this bias, and the costing
systems had no long-run e?ect on pro®tability or
survival. Price revisions from one period to the
next were strongly associated with both pro®t
variances (a seller's actual vs target pro®t) and
price variances (a seller's price o?er vs the average
market price). The threat of bankruptcy acceler-
ated seller learning, in that relative price revision
from period 1 to 2 was larger in the harsh vs leni-
ent markets, although this e?ect reversed in the
next few periods.
Regarding cost-based pricing biases, this study's
results resembled those of past experiments only in
the initial period. Given pro®t and market feed-
back, the bias did not persist as sellers revised their
price o?ers toward optimum without regard to
absorption vs variable cost signals. Further, the
bias did not re-emerge in the ®rst period of sub-
sequent sets, when sellers again were ignorant
about demand. The e?ects of alternative costing
systems on pricing decisions apparently depend on
individual vs market settings. As discussed below,
there are crucial di?erences between individual
and market settings, including trading institutions,
incentives, and information for learning. The issue
of whether decision behavior observed in indivi-
dual settings persists in markets has attracted
considerable attention from economists and psy-
chologists (Hogarth & Reder, 1986; Lopes, 1994;
Smith, 1991). This study adds to the growing lit-
erature on the issue (Camerer, 1987, 1992;
Camerer, Loewenstein & Weber, 1989; Cox &
Grether, 1996; Ganguly, Kagel & Moser, 1994;
Kachelmeier, 1996). Speci®cally, the strong asso-
ciation between sellers' price revisions and feed-
back variances supports the adaptive learning
model underlying the behavioral theory of the ®rm
(Cyert & March, 1992; Levitt & March, 1988).
Seller learning is an important market mechanism
718 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
(Tversky & Kahneman, 1986), which helps to
reconcile the view of many economists that mar-
kets induce optimal decisions at equilibrium and
the view of many psychologists that individual
decisions are subject to bias. At least in this
study's market setting, simple learning processes
were e?ective in moving biased decisions toward
optimal decisions. Whether learning is e?ective in
more complex settings, e.g. when sellers compete
in multiple product markets, is a question for
future research.
The rest of the paper is organized as follows.
Section 2 discusses di?erences between individual
and market settings, reviews past accounting
experiments on absorption vs variable costing,
and states the hypotheses. Sections 3 and 4
describe the experiment's method and results,
respectively. The last section provides concluding
remarks.
2. Literature review and hypotheses
2.1. Individual and market settings
Psychology and economics both focus on indi-
vidual decision behavior, but from fundamentally
di?erent perspectives (Lopes, 1994). Psychologists
view individuals as cognitive information proces-
sing systems that translate environmental stimuli,
joined with prior knowledge, into behavioral
responses. Such processing involves heuristics, or
simpli®ed procedures, that economize on a limited
capacity for encoding, retrieving, and manipulat-
ing information (Simon, 1978, 1986). Although
generally e?ective, heuristics sometimes produce
biased judgments and decisions, relative to prob-
ability and utility theory (Tversky & Kahneman,
1974, 1986). Many psychological experiments have
demonstrated such biases in individual settings
(Arkes & Hammond, 1986; Kahneman, Slovic &
Tversky, 1982). In contrast, economists build the-
ories of aggregate outcomes, e.g. equilibrium pri-
ces and quantities in competitive markets, based
on the assumption that individuals make rational
choices in terms of utility maximization (Milgrom
& Roberts, 1992). The di?erence in perspectives
has led to considerable debate on the behavioral
foundations of microeconomics and on whether
biases observed in individual settings persist in
markets (Hogarth & Reder, 1986).
A contentious view of the debate sees a contest
between psychology and economics regarding the
empirical validity of the rational-choice model. A
more productive and integrative view seeks ways
in which the limitations of psychology are alle-
viated by economics, and vice versa (Smith, 1991).
Drawing from experimental economics, Smith
stated two broad conclusions from hundreds of
laboratory-market studies: economic theory gen-
erally provides a correct ®rst approximation of
equilibrium outcomes, but is weak in describing
the processes of convergence and economizing on
decision cost. Economic theory derives predictions
using the principle of methodological individual-
ism (Blaug, 1992); i.e. ecient equilibrium out-
comes are implied by the assumedly rational
choices of individuals, along with other condi-
tions. Such derivation is open as to the actual
causal processes that produce aggregate outcomes
and, speci®cally, does not require a disequilibrium
process involving rational choices by actual indi-
viduals (Nelson & Winter, 1982). These con-
siderations suggest limitations in both economics
and psychology. Although generally successful in
predicting equilibria, economic theory provides
limited insight into causal processes. Although
validly describing cognitive processes and con-
straints, psychological experiments are limited by
usage of settings that isolate the individual from
market forces. From an integrative perspective, a
key question is:
Why is it that human subjects in the labora-
tory frequently violate the canons of rational
choice when tested as isolated individuals, but
in the social context of exchange institutions
serve up decisions that are consistent (as
though by magic) with predictive models
based on individual rationality (Smith, 1991,
p. 894)?
To answer this question requires examination of
the market processes by which equilibriumoutcomes
emerge. In this regard, experimental economics
emphasizes the role of trading institutions (i.e. the
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 719
rules governing trades between sellers and buyers),
incentives, and information for learning.
Besides testing equilibrium predictions, a major
contribution of experimental economics has been
extensive empirical evidence regarding the e?ects
of trading institutions on the rate and pattern of
market convergence. For example, double auc-
tions in which each subject may o?er or accept
either bids or asks induce rapid convergence,
whereas posted-price institutions in which sellers
make take-it-or-leave-it o?ers induce slower con-
vergence. Perhaps the most striking demonstra-
tions of institutional e?ects have been market
simulations with zero-intelligence traders pro-
grammed to generate random bids and asks, sub-
ject to only a budget constraint and an
endogenous choice set (Gode & Sunder, 1993; see
also Jamal & Sunder, 1996). Using double auc-
tions, these simulations achieved almost 100%
eciency, despite the absence of arbitrage, bank-
ruptcy, incentives, learning, or even usage of
heuristics. Ecient market outcomes emerged as a
result of the institution and environment, not from
rational choice by individuals (cf. Simon, 1982,
1986; Smith, 1991). Institutional e?ects are a pri-
mary di?erence between individual and market
settings.
Although not necessary for market eciency,
incentives and learning can strongly in¯uence the
convergence process. Standard procedures in
experimental economics include subject payments
under performance-based incentives and repeated
trading periods with feedback to allow for learn-
ing (Davis & Holt, 1993; Friedman & Sunder,
1994). Regarding incentives, in contrast with the
mixed evidence in psychology (Bonner, Young &
Hastie, 1996; Hogarth & Reder, 1986), the pre-
ponderance of evidence in experimental economics
indicates that incentives matter in markets, by
reducing inconsistencies between actual and
rational choice (Smith & Walker, 1993). The
e?ects of incentives in markets may be enhanced
by the presence of multiple, self-interested partici-
pants who a?ect each other's payo? through the
institution (Frey & Eichenberger, 1994). At a
minimum, incentives reduce noise when greater
cognitive e?ort can improve performance.
Regarding learning, the typical pattern of con-
vergence in laboratory markets is for eciency to
be relatively low in the ®rst trading period,
increase signi®cantly in the next few periods, and
increase more gradually in subsequent periods.
While this pattern presumably re¯ects subjects'
revisions in bids and asks in light of performance
feedback, experimental economics contains few
systematic attempts to describe individual learning
processes and their relation to market con-
vergence.
1
This de®ciency suggests an opportunity
for psychology to inform economics about adap-
tation by cognitive information processing sys-
tems. As with incentives, however, generalization
from individual to market settings is problematic.
A primary di?erence is that learning is an inter-
active, social phenomenon in markets where each
subject's performance feedback and observations
of market activity depend on the actions of others
through the institution.
A growing number of experiments have exam-
ined whether biases observed in individual settings
persist in market settings (Camerer, 1987; Camerer
et al., 1989; Cox & Grether, 1996; Duh & Sunder,
1986; Ganguly et al., 1994; Kachelmeier, 1996).
For example, psychological experiments have
shown that subjects in individual settings make
probability judgments using a representativeness
heuristic which causes biases such as ignoring base
rates (Kahneman & Tversky, 1972). Camerer
(1987) ran a series of markets in which subjects
traded assets that paid a state-dependent dividend.
Demand for the assets depended on subjects' pos-
terior probabilities given sample information, and
parameters were speci®ed such that usage of
representativeness vs Bayesian revision implied
di?erent equilibrium prices. Camerer (1987) found
that prices tended toward Bayesian predictions,
with only a small degree of bias attributable to
representativeness. Duh and Sunder (1986) had
similar results. In contrast, Ganguly et al. (1994)
reported that prices in their asset markets were
persistently closer to representativeness-based pre-
dictions than Bayesian predictions, especially in
markets where representativeness implied higher
1
To date, game experiments have emphasized subject learn-
ing far more than market experiments have (see Erev & Roth,
1998; Roth & Erev, 1995).
720 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
prices. Relatively few subjects made unbiased, pre-
trading probability judgments, even after 16 peri-
ods, and these subjects were not suciently active
to drive prices to Bayesian predictions.
Preference reversal is another frequent ®nding in
individual settings (Lichtenstein & Slovic, 1971;
Slovic & Lichtenstein, 1983). Suppose subjects are
asked to perform two tasks: (1) choose lottery A
with a high probability of a moderate payo? or
lottery B with a lower probability of a higher
payo?, and (2) value A and B on a monetary scale.
Many subjects choose A but place a higher value
on B. Cox and Grether (1996) examined pre-
ference reversals by manipulating the response
mode (choice vs valuation), monetary incentives
(strong vs moderate vs none), and setting (indivi-
dual vs market). There were ®ve task repetitions to
allow for learning. The results showed high rates
of preference reversal in the ®rst repetition, but
much lower rates by the ®fth repetition, especially
in market settings. As subjects in markets repeated
the task, they incorporated past prices into their
o?ers.
A ®nal example concerns the sunk-cost fallacy
whereby an individual's decision is a?ected by a
normatively irrelevant historical cost (Arkes &
Blumer, 1985). Kachelmeier (1996) ran a series of
markets to examine one aspect of sunk cost. Sub-
jects were either sellers or buyers of a security.
Prior to trading, each seller was assigned one unit
of the security, an unavoidable sunk cost, and a
redemption value (opportunity cost) in case of no
sale. A between-market manipulation was cost of
sale (sunk cost vs opportunity cost) in a pro®t
feedback report, although the formula for subject
payments did not depend on this variable. The
results showed that bids and asks were persistently
higher in the markets with the sunk-cost format.
Sellers' aversion to a paper loss produced an
upward bias in asks, and buyers reacted by raising
bids. However, the bias did not a?ect market pri-
ces or eciency, because of shifts in the percen-
tages of seller- vs buyer-initiated trades.
In sum, experimental economics provides evi-
dence from hundreds of market studies (not
focused on individual biases) that generally con-
®rms the equilibrium predictions, and indirectly
the rationality assumption, of economic theory.
The much smaller group of studies focused on
individual biases in markets provides more equi-
vocal evidence. Some studies found that individual
biases were reduced, if not eliminated, in market
settings. Other studies found that individual biases
had persistent e?ects on market o?ers and prices.
Before drawing conclusions, more research is nee-
ded on the mechanisms that drive markets and the
conditions under which these mechanisms induce
or fail to induce rational decisions.
2.2. Cost-based pricing
Surveys indicate that ®rms predominantly use
pricing policies that set initial prices equal to unit
cost plus target pro®t (see Dorward, 1987, for a
review). For example, Govindarajan and Anthony
(1983) surveyed over 500 industrial ®rms of the
Fortune 1000, asking respondents to specify ``the
method that comes closest to the one you usually
use in arriving at the normal selling price for your
typical product.'' Variations of absorption (vari-
able) cost-based pricing were speci®ed by 83%
(17%) of the respondents. Shim and Sudit (1995)
similarly surveyed 141 ®rms; 70% (12%) used
absorption (variable) cost-based pricing, and 18%
used ``market-based or competitive'' pricing.
Economists have long criticized cost-based pri-
cing, for several reasons (Oxenfeldt & Baxter,
1961). This procedure uses historical or budgeted
cost, rather than opportunity cost, and average
variable cost, rather than marginal cost. Absorption
costing includes average ®xed cost which is norma-
tively irrelevant to short-run pricing. The pro®t
markup on cost does not explicitly incorporate
information about demand.
Countering such criticism, justi®cations of cost-
based pricing invoke a decision-cost argument.
Given limited knowledge about demand and
opportunity cost, ®rms employ simpli®ed proce-
dures that economize on decision cost (Cyert &
March, 1992). When initial o?ers are subject to
revision depending on the reactions of custo-
mers and competitors, cost-based pricing may
be procedurally rational, i.e. economizing on
decision cost, though not substantively rational,
i.e. optimal without regard to decision cost
(Simon, 1976). Accordingly, costing systems may
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 721
play a role in explaining the disequilibrium process
of price formation involving procedurally rational
sellers, but no role in predicting equilibrium prices
assuming substantively rational choice (Waller,
1995). In this regard, an important issue is whe-
ther alternative costing systems bias price o?ers in
the market convergence process. Using models
and methods from psychology, past accounting
experiments have demonstrated cost-based pricing
biases in individual settings, but none has exam-
ined whether such biases persist in markets.
Early experiments used the lens model from
psychology to examine the sensitivity of sellers'
pricing decisions to changes in costing systems
(Ashton, 1976, 1981; Dyckman et al., 1982;
Swieringa et al., 1979). Ashton (1976) had each
seller set prices for 60 products using three cues,
i.e. unit cost, demand elasticity, and competitor
responsiveness. Absorption vs variable costing
was manipulated on a within-subject basis after 30
prices had been set, with the order of systems var-
ied over subjects. Sellers starting with variable
(absorption) costing also were told that the change
in systems resulted in less (more) useful informa-
tion due to the inclusion (exclusion) of ®xed cost.
Two control groups used either absorption or
variable costing for all 60 products. Ashton con-
structed a regression model of each seller's pricing
policy over the ®rst 30 products, and computed
the mean absolute di?erence between actual and
predicted prices for the last 30 products. This dif-
ference measured sensitivity to the change in cost-
ing systems, which was signi®cantly higher for the
experimental vs control groups. In a critique,
Libby (1976) expressed concerns about the con-
founded manipulation of costing-system change
and stated usefulness, and about di?erences in
cost data for the experimental and control groups.
Responding to these concerns, Swieringa et al.
(1979) performed a replication, isolating the
manipulation of costing-system change and hold-
ing constant the cost data over groups. Swieringa
et al. (1979) found that sellers with a costing-sys-
tem change adjusted their information processing
more than the control groups did. In another
replication using older subjects with more expo-
sure to accounting, Dyckman et al. (1982) repor-
ted similar results.
Two experiments tested hypotheses from Lere
(1986) about the e?ects of absorption vs variable
costing on the correspondence between pro®t-
maximizing decisions and cost-based o?ers made
with a speci®c heuristic (see also Dickhaut & Lere,
1983). The heuristic consisted of the following
steps: (1) the seller suggests a price, p, to his
accountant, (2) the accountant determines expec-
ted demand, E[q(p)], and reports the unit cost, c,
(3) the seller evaluates c against c
+
= E
[pq
/
(p) ÷ q(p)[=E[q
/
(p)[, and (4) the process iterates
until c = c
+
. The degree to which the heuristic
approximates pro®t maximization depends on
absorption vs variable costing.
2
In one experiment
(Hilton et al., 1988), each subject chose between
absorption and variable costing, and made a price
o?er after iterating with a simulated accountant
for up to ten repetitions. Each seller knew the
demand function (or probability distribution of
demand functions) and type of cost function (linear
vs nonlinear and stochastic vs deterministic). Sell-
ers had monetary incentives and pro®t feedback
after each trial. The results did not support Lere's
hypotheses. Sellers' price o?ers di?ered sig-
ni®cantly from prices under the assumed heuristic
as well as from optimal prices, and most sellers
chose absorption over variable costing under all
conditions. Also, mean price o?ers were higher
under absorption vs. variable costing under all
conditions. In an experiment involving quantity
decisions, Turner and Hilton (1989) similarly
found signi®cant divergences from optimum and a
general preference for absorption costing.
The above experiments provided limited evi-
dence regarding the role of absorption vs variable
costing for pricing decisions in markets. The
stream of lens-model studies beginning with Ashton
(1976) examined pricing decisions in individual
settings that lacked most of the distinctive features
2
Lere's (1986) model made the following predictions. Given
a linear, deterministic cost function, variable costing induces
better prices, i.e. prices set by the heuristic are closer to optimal
prices, than absorption costing does. Given a linear, stochastic
cost function and deterministic demand, variable (absorption)
costing induces better prices under risk neutrality (aversion).
Given a nonlinear cost function, absorption costing always
induces better prices.
722 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
of markets discussed earlier. Despite many task
repetitions, sellers had neither monetary incentives
nor pro®t feedback. Although available informa-
tion included cues about market conditions, there
was no actual trade between sellers and buyers.
The later studies (Hilton et al., 1988; Turner &
Hilton, 1989) added improvements such as formal
models that recognized decision cost, monetary
incentives, task repetitions (under changing con-
ditions), and pro®t feedback. However, these stu-
dies assumed decision heuristics involving
interaction between the seller and accountant,
rather than interaction among sellers and buyers
in markets. The studies also assumed that demand
was known, at least as a probability distribution.
All of the above studies involved one-shot o?ers
with no chance of price revision or convergence,
and none included actual seller competition or the
threat of bankruptcy.
Although not concerned with absorption vs
variable costing, Gupta and King (1996) examined
cost-based decisions in a setting that built upon
Hilton et al. (1988). Gupta and King (1996)
manipulated cost-report accuracy and production
complexity in a multiproduct ®rm. Each subject
acted as a monopolist (i.e. no seller competition)
who made cost forecasts for three products given
imperfect cost reports. There was no explicit pri-
cing decision. Instead, each subject's cost forecast
together with a simulated demand function deter-
mined price, quantity, and pro®t. Also, there was
no ®xed or joint cost. Total variable cost depen-
ded on each product's requirements for material
and resources in three conversion processes. The
more accurate cost report was based on two con-
version cost pools, while the less accurate report
was based on only one. Each product in the more
complex ®rm used resources nonproportionately
in the conversion processes, while each product in
the simpler ®rm used resources proportionately.
Subjects went through 20 task repetitions under
stable conditions, e.g. constant demand, with
monetary incentives and feedback. As expected,
the results showed higher pro®t given more accu-
rate cost reports and a simpler ®rm. The results
also revealed systematic learning whereby subjects
revised their forecasts toward optimum, despite
inaccurate cost reports. Learning from experience
may be a procedurally rational substitute for cost-
report accuracy. As elaborated below, this study's
experiment also involved the procedural ration-
ality of seller learning given pro®t and market
feedback.
2.3. Hypotheses
The experiment incorporated many features that,
taken together, distinguish individual and market
settings, e.g. monetary incentives, task repetitions,
pro®t feedback, trading institution, seller competi-
tion, endogenous information about demand, and
the threat of bankruptcy. The experiment produced
evidence about four hypotheses.
The ®rst hypothesis concerns market con-
vergence. It was expected that sellers' price o?ers
would converge toward optimum over each set of
trading periods. Con®rming this expectation was
necessary to establish a general correspondence
between this study's market setting and other
laboratory markets. Competitive equilibrium pre-
dictions were based on market demand and sup-
ply, without regard to costing systems. There were
four lenient markets in which demand tended to be
relatively high (Fig. 1), and four harsh markets in
which demand tended to be relatively low (Fig. 2).
The reason for the between-market manipulation
of demand was to vary the threat of bankruptcy
faced by sellers. A seller went bankrupt when his
cash balance (initial endowment?pro®t or loss)
was negative, and suboptimal decisions were more
likely to cause bankruptcy in the harsh markets.
Each market used six demand functions corre-
sponding to six sets of eight periods. The reason
for the within-market manipulation of demand
was to create recurrent disequilibrium states in
which sellers were ignorant about demand. This
allowed repeated observations of the convergence
process. The supply function (see Figs. 1 and 2)
was constructed by aggregating the sellers' mar-
ginal cost functions. For each set of a market, the
optimal price and quantity (Table 1) were deter-
mined by the intersection of the supply function
and relevant demand function. Seller performance
was measured by price error, i.e. the absolute dif-
ference between a seller's price o?er and the opti-
mal price, where the latter depended on the set and
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 723
market (Table 1).
3
Consistent with convergence
toward optimum, the ®rst hypothesis is:
H1. Sellers' price errors approach zero over periods
1±8 of a set.
The second and main hypothesis concerns the
persistence of a cost-based pricing bias in the
market convergence process. Because of the mar-
ket features discussed earlier, it was expected that
a bias in price o?ers due to absorption vs variable
costing would not persist. Assuming higher price
o?ers under absorption costing in the initial period
(comparable to the ®ndings in individual settings), it
was expected that sellers would respond to pro®t
and market feedback by revising their price o?ers
toward optimumwithout regard to unit-cost signals.
Accordingly, the results should reveal an interactive
e?ect for costing systems and periods, whereby price
o?ers di?er under absorption vs. variable costing in
the early, but not later, periods of a set.
4
H2. A bias in price o?ers due to absorption vs variable
costing does not persist over periods 1±8 of a set.
The third hypothesis concerns seller learning.
Consistent with the behavioral theory of the ®rm
(Cyert & March, 1992; Levitt & March, 1988),
seller learning may be viewed as the adaptation of
current decisions to experiential feedback on past
decisions. Sellers received two kinds of feedback,
pro®t or loss from their own trades and observa-
tions of other trades. Pro®t feedback was measured
by pro®t variance, i.e. the di?erence between a
seller's actual and target pro®t last period. Target
pro®t equaled a seller's markup times o?ered
Fig. 1. Demand and supply for lenient markets.
3
Absolute rather than signed di?erences were used, because
analyses that aggregate over positive and negative di?erences
would understate the average magnitude of deviations from
optimum (cf. Bloom®eld, 1997). Also, price o?ers converged
toward optimum from above (positive di?erences) in some sets,
but from below (negative di?erences) in other sets. Analyses
that aggregate over sets would involve the same understatement
problem.
4
Although sellers returned to a similar state of ignorance
about demand in period 1 of each set, it is an open issue whe-
ther the same pattern of a cost-based pricing bias (i.e. its
emergence and persistence or elimination) unfolds in each set.
Task experience may a?ect the pattern in later sets (see
Results).
724 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
quantity under absorption costing, or, markup
times o?ered quantity minus ®xed cost under
variable costing. When a seller made a sale at the
o?ered price and quantity, the pro®t variance was
zero; otherwise, the pro®t variance was unfavor-
able. Market feedback was measured by price var-
iance, i.e. the di?erence between a seller's price
o?er and the average price per unit sold last per-
iod. It was expected that pro®t and price variances
would be correlated, but not perfectly. High price
o?ers relative to competitors' were likely to result
in both pro®t and price variances, whereas low
price o?ers relative to competitors' were likely to
result in a price variance but no pro®t variance.
Seller learning was measured by price revision, i.e.
the increase or decrease in a seller's price o?er
from t to t ÷ 1(t = 1; :::; 7). It was expected that
larger pro®t and price variances would induce lar-
ger price revisions.
H3. Price revisions depend on feedback regarding
pro®t and price variances.
The last hypothesis concerns the threat of
bankruptcy. Using lenient vs harsh markets as a
proxy for the threat of bankruptcy, it was expected
that sellers' price revisions would accelerate as the
threat of bankruptcy increased. When comparing
price revisions between the lenient vs harsh markets,
however, a possible confound was that di?erences in
optimal prices may have led to di?erences in total
price revisions, regardless of the rate of price revi-
sions. Accordingly, a second measure of learning,
which used each seller as his own control, was rela-
tive price revision, i.e. the ratio of a seller's price
revision from period t to t ÷ 1(t = 1; :::; 7) over his
total price revision from period 1 to 8 of a set. If a
greater threat of bankruptcy causes faster price
revision, then the results should reveal an interactive
e?ect for markets and periods, whereby relative
price revision is higher in the harsh vs. lenient mar-
kets in the early periods of a set.
5
H4. A greater threat of bankruptcy increases rela-
tive price revision in the early periods of a set.
Fig. 2. Demand and supply for harsh markets.
5
By construction, each subject's relative price revision sums
to one over a set of periods. If relative price revision is higher in
the harsh markets for early periods (H4), then the measure
must be higher in the lenient markets for later periods.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 725
3. Method
The procedure for the lenient markets is descri-
bed ®rst, followed by di?erences for the harsh
markets. In each trading period, ten sellers
(undergraduate business students) made o?ers to
sell a product. As a bu?er against possible loss,
each seller received an initial endowment of 50,000
francs, the experimental currency. Making an o?er
consisted of stating the maximum quantity for sale
between one and ®ve units, referring to a schedule
showing the unit cost at each quantity, and
adjusting the unit cost for target pro®t. The sche-
dule was prepared under absorption costing for
®ve sellers and under variable costing for ®ve sell-
ers, holding constant the true cost function.
6
The
unit cost under absorption costing was 1890, 1025,
737, 715 and 702 francs, for one to ®ve units,
respectively; the corresponding numbers under
variable costing were 30, 95, 117, 250 and 330
francs. There were 48 periods, split into six sets of
eight. Buyers were simulated with a computer
program executing the demand functions in Figs.
1 and 2.
7
Demand was constant for each set, but
varied over sets. Speci®cally, D1 in Fig. 1 was
employed for set 1, D2 for set 2, and so on. Peri-
odic pro®t or loss equalled the o?ered price times
units sold minus total variable and ®xed cost.
Variable cost depended on units sold, but ®xed
cost was unavoidable. A bankruptcy occurred
when a seller's franc balance was negative. At the
end of the experiment, each seller's franc balance
was translated into the probability of winning $20
in a lottery, at a rate of 1/100,000. This payo?
structure provided a control for risk preference, in
that a seller who maximized the expected prob-
ability of winning the cash prize was risk-neutral
as to francs when making sales o?ers (Davis &
Holt, 1993). Cost and demand parameters were
speci®ed such that optimal prices and quantities
implied zero pro®t on average (Table 1). The
expected probability of winning the prize was 0.50
(50,000/100,000), and expected pay was $25
(0.50×$20+$15 for participating).
To ensure consistency over markets, the
instructions were presented to sellers by playing a
pre-recorded tape. The speaker on the tape read
the instructions aloud, while sellers read along in a
booklet. The appendix contains the instructions
and o?er sheet for sellers with absorption costing.
The instructions and o?er sheet for sellers with
variable costing were identical, except for the cost
schedule. The instructions described the market
setting in detail: the trading institution, task of
making o?ers, sequence of events in a period,
initial endowment and payo? structure, general
nature of demand and timing of random shifts,
limits on unit sales, pro®t computation, periodic
®xed cost and variable cost, cost schedule, price
limits (0±2000 francs), and bankruptcy rule.
Table 1
Optimal prices, quantities, and pro®t (per seller)
Demand Price Quantity Profit
A. Lenient markets
D1 755 5 265
D2 315 3 ÷1,265
D3 1,155 5 2,265
D4 955 5 1,265
D5 415 3 ÷965
D6 215 3 ÷1,565
Average 635 4 0
B. Harsh markets
D1 395 3 ÷1.025
D2 195 3 ÷1,625
D3 855 5 765
D4 705 5 15
D5 295 3 ÷1,325
D6 65 1 ÷1,825
Average 418.33 3.33 ÷836.66
6
The true cost function was a ®xed cost of 1860 francs, and
a marginal cost of 30 francs for the ®rst unit sold, 160 francs
for the second and third units sold, and 650 francs for the
fourth and ®fth units sold. Relatively large steps in the cost
function were used to ensure nontrivial di?erences in optimal
prices among alternative demand levels in each market (Figs. 1
and 2 and Table 1).
7
To illustrate, D1 in Fig. 1 implied that buyers were willing
to acquire one unit at a price of 1245 francs, two units at 1235
francs, three units at 1225 francs, four units at 1215 francs, . . .,
and 50 units at 755 francs. Suppose that two sellers each o?ered
to sell two units, but at di?erent prices, i.e. 1210 francs (seller
A) and 1220 francs (seller B). The trades would be two units at
1210 francs for seller A and one unit at 1220 francs for seller B.
726 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
After reading the instructions, sellers were
shown four examples covering various possible
sales o?ers and demand conditions. For simpli-
city, each example included only two sellers, but
gave detailed explanations for completing o?er
sheets and interpreting feedback. Example 1 was a
case in which the sellers o?ered di?erent prices
and quantities, and buyers accepted both o?ers.
Example 2 repeated the o?ers of Example 1, but
demand was lower such that only one seller made
a sale. Example 3 was a case in which the sellers
o?ered the same price, but demand was not su-
cient for each to sell the o?ered quantities. When
sellers o?ered the same price, the number of units
sold were divided as evenly as possible. When an
even distribution was not possible, the remaining
unit(s) was allocated randomly. Example 4 was a
case in which a seller went bankrupt. The exam-
ples were prepared under absorption costing for
®ve subjects and variable costing for ®ve subjects.
The speaker on the tape did not refer to speci®c
values of cost or markups. All subjects heard the
same description of the market procedure.
In each period, sellers had 3 min to complete the
o?er sheet, after which the sheets were collected
and the o?ers were entered into a computer. Peri-
odic feedback consisted of two parts. One part
was market activity (i.e. price and quantity for
each sale) which was displayed publicly on large
screens. Another screen displayed publicly the
number of surviving sellers, which was updated
when a bankruptcy occurred. The other part was
each seller's results (i.e. the price and quantity for
a sale, revenue, total cost, pro®t or loss, and
beginning and ending franc balances) which were
displayed privately on a terminal. These displays
continued until the o?ers for the next period
were entered. To ensure that sellers paid atten-
tion to both types of feedback, there was a 20-s
lag between initiation of the public display of
market results and private display of each seller's
results.
This procedure was repeated for 48 periods. The
entire experiment, including the instructions, las-
ted approximately 2.5 h. At the end, the lottery
was played separately for each seller with a posi-
tive franc balance, who was paid according to the
above description.
The harsh markets followed the same proce-
dure, with one exception.
8
Because of the di?er-
ence in pro®t opportunities, the cash prize was
raised to $100 in order to equate expected pay for
all markets. Given optimal behavior in the harsh
markets (Table 1), each seller's ending franc balance
would be 9840 francs (50,000÷48×836.66). The
expected probability of winning the prize was.10
(9,840/100,000), and expected pay was $25
(0.10×$100+$15). This procedure held expected pay
constant at $25 for the lenient and harsh markets.
4. Results
4.1. Convergence toward optimum
H1 predicted that sellers' price errors approach
zero over periods 1±8 of a set. Table 2 reports
descriptive statistics regarding seller and market
performance by set and period, separately for the
lenient and harsh markets.
9
The ®rst column
shows medians for price error over sellers; e.g., in
period 1 of set 1, the median price error was 245
(628) francs, relative to the optimal price of 755
(395) francs, for the 40 sellers in the lenient (harsh)
markets. Changes in median price errors from
period 1±8 show a clear pattern of convergence
toward optimum. Averaging over sets for the
8
In two of the harsh markets, the number of periods was
allowed to exceed 48 (to at most 51), in order to achieve a
minimum bankruptcy rate of 4/10. The intent was to observe
whether bankruptcies were related to costing systems. There
were 22 survivers in the harsh markets after all periods; 13 were
assigned absorption costing and 9 were assigned variable cost-
ing (
2
= 1:62; p > 0:10). To facilitate comparisons between
markets, the Results section analyzes the data for periods 1±48
only.
9
All laboratory-market studies with multiple trading peri-
ods share the problem of serial dependence among observations
over time. Indeed, H1 predicts a particular pattern of serial
dependence. The approach taken here is to present the results
on a disaggregated basis and, as appropriate, present additional
aggregate tests. The disaggregated analyses were not intended
as independent hypothesis tests. Repeated-measures analysis of
variance was used in the aggregate tests, with set and period as
within-subject factors. These tests accommodated serial depen-
dence by assessing the e?ects of each within-subject factor, and
between-subjects factors, with respect to distinct error terms
(Neter, Kutner, Nachtsheim & Wasserman, 1996).
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 727
Table 2
Seller and market performance
a
Seller performance Market performance
Set-period Price error Quantity error Profit error No. of trades/ no. of sellers Efficiency
1÷1 245 (628) 2 (1) 755 (835) 24/40 (4/40) 69% (10%)
1÷2 153 (348) .5 (1) 725 (835) 28/40 (11/40) 81 (45)
1÷3 145 (127) 0 (1) 475 (655) 32/40 (21/40) 89 (72)
1÷4 117 (105) 0 (0) 475 (217) 34/40 (25/40) 93 (77)
1÷5 105 (72) 0 (0) 507 (360) 31/40 (23/40) 93 (80)
1÷6 98 (75) 0 (0) 475 (235) 34/40 (26/40) 94 (82)
1÷7 96 (72) 0 (0) 445 (225) 34/40 (27/40) 94 (84)
1÷8 89 (62) 0 (.5) 392 (216) 35/40 (28/40) 96 (86)
2÷1 486 (455) 2 (0) 595 (235) 1/40 (7/40) 0 (26)
2÷2 395 (248) 2 (0) 595 (235) 6/40 (12/40) 27 (26)
2÷3 207 (202) 2 (0) 595 (235) 14/40 (15/40) 58 (65)
2÷4 160 (155) 1 (0) 595 (235) 17/40 (19/40) 68 (76)
2÷5 110 (135) 1 (0) 545 (235 20/40 (23/40) 80 (81)
2÷6 83 (123) .5 (0) 241 (291) 28/40 (25/40) 84 (83)
2÷7 59 (110) 0 (0) 257 (236) 31/40 (29/40) 88 (86)
2÷8 46 (100) 0 (0) 193 (235) 31/40 (31/40) 87 (88)
3÷1 411 (405) 0 (2) 2,180 (1,827) 35/40 (37/40) 89 (86)
3÷2 255 (108) 0 (0) 1,275 (649) 38/40 (36/40) 96 (92)
3÷3 194 (55) 0 (0) 1,025 (280) 38/40 (36/40) 95 (95)
3÷4 145 (44) 0 (0) 775 (216) 38/40 (38/40) 98 (96)
3÷5 104 (44) 0 (0) 537 (210) 37/40 (35/40) 95 (96)
3÷6 94 (54) 0 (0) 475 (270) 40/40 (36/40) 97 (95)
3÷7 63 (75) 0 (0) 360 (370) 38/40 (35/40) 99 (94)
3÷8 55 (73) 0 (0) 302 (322) 36/40 (36/40) 94 (94)
4÷1 205 (99) 0 (0) 1,025 (497) 37/40 (31/40) 91 (91)
4÷2 86 (5) 0 (0) 462 (415) 38/40 (33/40) 98 (95)
4÷3 50 (110) 0 (0) 275 (537) 37/40 (34/40) 96 (96)
4÷4 45 (114) 0 (0) 225 (550) 37/40 (37/40) 95 (96)
4÷5 43 (96) 0 (0) 222 (47) 38/40 (35/40) 96 (96)
4÷6 42 (93) 0 (0) 210 (517) 38/40 (34/40) 96 (97)
4÷7 42 (81) 0 (0) 212 (462) 38/40 (35/40) 96 (97)
4÷8 35 (57) 0 (0) 165 (285) 39/40 (35/40) 97 (98)
5÷1 335 (425) 2 (2) 895 (235) 4/40 (4/40) 10 (17)
5÷2 185 (155) 1 (0) 725 (235) 20/40 (22/40) 54 (66)
5÷3 79 (105) 0 (0) 375 (366) 25/40 (26/40) 81 (80)
5÷4 35 (77) 0 (0) 157 (390) 29/40 (29/40) 92 (86)
5÷5 16 (54) 0 (0) 228 (415) 33/40 (32/40) 95 (90)
5÷6 15 (44) 0 (0) 91 (414) 35/40 (34/40) 97 (92)
5÷7 13 (35) 0 (0) 43 (373) 36/40 (35/40) 97 (93)
5÷8 12 (27) 0 (0) 43 (358) 39/40 (36/40) 97 (94)
6÷1 586 (645) 2 (4) 295 (35) 1/40 (0/40) 0 (0)
6÷2 235 (335) 0 (2) 295 (35) 15/40 (1/40) 50 (0)
6÷3 160 (165) 0 (2) 295 (35) 20/40 (6/39) 71 (29)
6÷4 112 (57) 0 (1) 295 (35) 26/40 (15/38) 84 (73)
6÷5 88 (30) 0 (0) 268 (35) 28/40 (23/38) 87 (84)
6÷6 81 (20) 0 (0) 244 (23) 31/40 (27/38) 90 (87)
6÷7 65 (15) 0 (0) 193 (15) 32/40 (29/33) 91 (91)
6÷8 59 (15) 0 (0) 165 (15) 34/40 (23/31) 93 (90)
a
The ®rst (second) entry in each cell pertains to the lenient (harsh) markets. Entries in the ®rst through third columns are medians
over sellers for price error, quantity error, and pro®t error, respectively. Entries in the fourth column are frequencies of sellers with
trades. Entries in the ®fth column are medians over markets for eciency.
728 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
lenient (harsh) markets, the median price error
decreased by 85% (81%) from periods 1 to 8. For
each set, a Wilcoxon sign test was used to compare
the price errors in period 1 and 8, separately for
the lenient and harsh markets; in each case, the
decrease was highly signi®cant (p < 0:001). As an
aggregate test, a repeated-measures analysis of
variance (ANOVA) was performed using LN
(price error) as the dependent measure, with sets
(1±5) and periods (1±8) as within-subjects factors,
and lenient vs harsh markets and absorption vs
variable costing as between-subjects factors. Exclud-
ing set 6 avoided the problemof missing values due to
bankruptcies.
10
Using logged price o?ers as the
dependent variable reduced the inordinate e?ects of
unusually high or low o?ers. The ®rst column of
Table 3 shows the main and interactive e?ects (results
in other columns are discussed later). The relevant
result for H1 was the signi®cant e?ect of periods
(F = 254:96; p < 0:001). Consistent with the con-
vergence pattern in the ®rst column of Table 2, the
mean of LN (price error) decreased monotonically
from period 1 to 8. These results support H1.
Table 3 shows other signi®cant e?ects for LN(price
error), including sets (F = 86:43; p < 0:001), markets
(F = 9:60; p < 0:003), periods×markets (F = 7:39;
p < 0:001), sets×markets (F = 29:10; p < 0:001),
periods×sets (F = 13:16; p < 0:001), and periods
×markets×sets (F = 11:52; p < 0:001). The mean of
LN(price error) decreased fromset 1 to 5, which may
be attributed to task experience and changing
demand conditions. The mean generally was higher
for harsh vs lenient markets, and the di?erence was
larger in later periods. The market e?ect also varied
over sets. The mean was higher for harsh (lenient)
markets in sets 2 and 4 (3 and 5), and about the same
in set 1. In all sets and markets, the mean decreased
over periods, but at somewhat di?erent rates.
11
In addition to price errors, Table 2 provides
descriptive statistics on other measures of seller
and market performance. The second and third
columns show the median quantity errors and
pro®t errors, respectively. A quantity error was
de®ned as the absolute di?erence between a seller's
quantity o?er and the optimal quantity, and a
pro®t error as the absolute di?erence between a
seller's actual and optimal pro®t, where the opti-
mal values depended on the set and market (Table 1).
The fourth column shows the number of trades
divided by the number of participating sellers. The
last column shows the median eciency over
markets. Eciency increased toward 100% as the
total quantity sold increased toward optimum, but
at a decreasing rate because buyers' marginal
value was decreasing and sellers' marginal cost
was increasing.
12
Generally, these measures were
consistent with the decrease in price errors over
periods. In most sets and markets, median quantity
errors quickly decreased to zero, median pro®t
errors decreased over periods, and both the per-
centage of sellers with trades and market eciency
approached 100%.
13
These convergence results
are comparable to many laboratory-market stu-
dies.
10
As expected, some sellers in the harsh markets went
bankrupt, beginning in period 3 of set 6; nine sellers went
bankrupt by period 8 of set 6.
11
As to the interactive e?ect of periods × markets × sets,
the mean of LN (price error) was higher for the harsh (lenient)
markets in later periods of sets 2 and 4 (1); the mean was higher
for the harsh (lenient) markets in most periods of set 3 (5).
12
To illustrate, the demand function for set 1 of the lenient
markets implied that buyers valued the ®rst unit at 1245 francs,
second unit at 1235 francs, third unit at 1225 francs, . . ., and 50th
unit at 755 francs. Given the optimal quantity of 50 units, the
aggregate value to buyers was 50,000 francs, and the aggregate
variable cost to sellers was 16,500 francs (10×30+20×160+
20×650), so that maximum pro®t was 33,500 francs. Suppose that
three units were sold by di?erent sellers (at any prices between 30
and 1225 francs), resulting in aggregate pro®t of 3615 francs
(1245+1235+1225÷30×3). In this case, eciency would be 11%
(3,615/33,500).
13
There were some exceptions to the general convergence
pattern. Regarding pro®t errors, there was minimal con-
vergence in sets 2, 5, and 6 of the harsh markets, which may be
attributed to the relatively small change in pro®t when a sale
was made vs no sale. In set 6 of the harsh markets, for example,
no sale implied a pro®t of -1,860 francs, compared to optimal
pro®t of ÷1825 francs assuming the sale of one unit at 65
francs. Regarding the percentage of sellers with trades, the
pattern was di?erent for sets 3 and 4, compared to the other
sets, in that the percentage started and stayed high. Figs. 1 and
2 show a large upward shift in demand from set 2 to 3, and a
small downward shift from set 3 to 4. Although sellers knew
that shifts in demand were random, prices in period 8 of a set
apparently a?ected o?ers in period 1 of the next set. Such car-
ryover e?ects explain why there were more trades from the start
in sets 3 and 4.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 729
4.2. E?ects of costing systems
H2 predicted that a bias in price o?ers due to
absorption vs variable costing does not persist
over periods 1 to 8 of a set. Table 4 reports an
ANOVA for each period using LN (price o?er) as
the dependent variable, with absorption vs vari-
able costing and lenient vs harsh markets as
between-subjects factors. The ®rst and second
columns indicate whether the mean of LN (price
o?er) was higher for absorption vs variable cost-
ing, and for lenient vs harsh markets, respectively.
The number of observations was 80 (10 sellers×8
markets) in each ANOVA, except for periods 3±8
of set 6, which omitted bankrupt sellers. Focusing
on set 1, costing systems had a signi®cant e?ect in
period 1 (F = 8:88; p < 0:004); the mean (median)
price o?er was 1162 (1048) francs under absorp-
tion costing vs 933 (968) francs under variable
costing. The higher unit cost under absorption
costing apparently caused higher price o?ers,
when sellers had not yet observed actual trades
and were ignorant about demand. In period 2, the
mean (median) price o?er was 908 (877) francs
under absorption costing vs 836 (772) francs under
variable costing. Although price o?ers were still
higher under absorption costing, the e?ect of
costing systems was no longer signi®cant. In peri-
ods 3±8, price o?ers under absorption vs. variable
costing were about the same. In contrast, harsh vs
lenient markets had an insigni®cant e?ect in per-
iod 1, but highly signi®cant e?ects in the remain-
ing periods. R
2
increased from 0.11 in period 1 to
0.97 in period 8, due to the increasingly strong
e?ects of markets and decreasing variation in price
o?ers. The results for set 1 support H2. The initi-
ally observed bias did not persist, but instead was
overwhelmed by demand conditions as revealed
through actual trades.
The results for sets 2±6 showed a similar pattern
of weak e?ects due to costing systems, strong
e?ects due to markets, and increasing R
2
over
periods. Unlike set 1, however, there was only a
small cost-based pricing bias in period 1 of the
subsequent sets (Table 4). Although the mean
value of LN(price o?er) was higher under absorp-
tion costing in period 1 of each set, costing sys-
tems had no signi®cant e?ect. This result may be
attributed to the combination of three factors.
First, variation in price o?ers was relatively high in
period 1 of each set, which lowered the likelihood
of detecting of a signi®cant bias. Second, despite
the random shifts in demand over sets, prices in
period 8 of a set often a?ected o?ers in period 1 of
the next set (see note 13). Such carryover e?ects
would mitigate a cost-based pricing bias. Third,
some sellers learned during set 1 that their costing
system was of limited usefulness, and placed less
reliance on unit-cost signals in subsequent sets.
As an aggregate test, an ANOVA was per-
formed using LN(price o?er) as the dependent
variable, with periods (1±8) and sets (1±5) as
within-subjects factors, and absorption vs variable
costing and lenient vs harsh markets as between-
subjects factors. The second column of Table 3
shows the main and interactive e?ects. Consistent
with the above results, there was an insigni®cant
e?ect for costing systems (F = 1:20; p > 0:28), and a
signi®cant e?ect for markets (F = 409:46;
p < 0:001). The signi®cant e?ects for sets
Table 3
Analyses of variance for price errors, o?ers, and relative price
revision
a
LN
(price error)
LN
(price o?er)
Relative
price revision
Within-subjects e?ects:
P 254.96 (.001) 53.69 (.001) 374.99 (.001)
P × M 7.39 (.001) 1.43 (.19) 12.85 (.001)
P × CS 0.54 (.80) 2.03 (.05) 0.22 (.97)
P × M × CS 0.94 (.47) 0.69 (.68) 0.73 (.63)
S 86.43 (.001) 1,154.22 (.001) ±
S × M 29.10 (.001) 29.51 (.001) ±
S × CS 1.02 (.39) 1.06 (.38) ±
S × M × CS 1.39 (.24) 1.17 (.33) ±
P × S 13.16 (.001) 67.68 (.001) 3.39 (.001)
P × S × M 11.52 (.001) 8.35 (.001) 2.85 (.001)
P × S × CS 0.74 (.84) 1.14 (.28) 1.31 (.14)
P × S × M × CS 0.78 (.78) 1.02 (.44) 0.65 (.90)
Between-subjects e?ects:
M 9.60 (.003) 409.46 (.001) ±
CS 0.02 (.87) 1.20 (.28) ±
M × CS 1.13 (.29) 1.66 (.20) ±
a
P-stands for periods; S-for sets; M-for harsh vs lenient
markets, and CS for absorption vs variable costing systems.
Entries are F statistics with signi®cance levels in parentheses.
730 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
Table 4
Analyses of variance for LN (price o?er)
a
Group with higher LN (price o?er) Main and interactive e?ects
Set-period Absorption (A)
vs. variable (V)
Lenient (L)
vs harsh (H)
Costing
systems
Markets Costing systems
× markets
R
2
1±1 A L 8.88 (0.004) 0.31 (.58) 1.63 (0.21) 0.11
1±2 A L 1.63 (0.21) 11.68 (0.001) 0.01 (.91) 0.15
1±3 V L 0.17 (0.68) 74.99 (0.001) 0.45 (0.51) 0.49
1±4 V L 0.74 (0.39) 107.61 (0.001) 0.03 (0.86) 0.59
1±5 V L 0.85 (0.36) 140.77 (0.001) 4.52 (0.037) 0.64
1±6 A L 0.12 (0.73) 418.03 (0.001) 2.15 (0.15) 0.84
1±7 V L 0.17 (0.68) 309.30 (0.001) 0.20 (0.66) 0.80
1±8 A L 0.21 (0.65) 2,395.89 (0.001) 0.30 (0.59) 0.97
2±1 A L 1.83 (0.18) 14.46 (0.001) 2.03 (0.16) 0.17
2±2 V L 0.02 (0.89) 41.74 (0.001) 0.04 (0.84) 0.35
2±3 V L 0.09 (0.77) 1.40 (0.24) 1.49 (0.23) 0.02
2±4 A L 0.95 (0.33) 15.70 (0.001) 0.09 (0.77) 0.18
2±5 A L 2.70 (0.10) 48.14 (0.001) 0.01 (0.91) 0.40
2±6 A L 0.03 (0.86) 45.19 (0.001) 0.01 (0.95) 0.37
2±7 A L 0.31 (0.58) 105.12 (0.001) 0.31 (0.58) 0.58
2±8 A L 0.05 (0.83) 133.88 (0.001) 3.69 (0.06) 0.63
3±1 A L 0.01 (0.93) 34.15 (0.001) 0.05 (0.83) 0.31
3±2 A L 0.02 (0.90) 20.22 (0.001) 0.06 (0.81) 0.21
3±3 V L 4.03 (0.048) 58.25 (0.001) 0.53 (0.47) 0.45
3±4 A L 0.19 (0.66) 84.74 (0.001) 0.24 (0.62) 0.53
3±5 V L 0.43 (0.51) 33.20 (0.001) 0.36 (0.55) 0.31
3±6 V L 0.02 (0.89) 127.80 (0.001) 0.26 (0.61) 0.63
3±7 A L 0.05 (0.83) 182.37 (0.001) 0.10 (0.76) 0.71
3±8 V L 0.26 (0.61) 71.55 (0.001) 1.92 (0.17) 0.48
4±1 A L 0.01 (0.97) 1.00 (0.32) 0.51 (0.48) 0.01
4±2 A L 0.37 (0.55) 23.96 (0.001) 0.15 (0.70) 0.24
4±3 A L 0.01 (0.97) 133.22 (0.001) 0.07 (0.79) 0.64
4±4 V L 2.67 (0.11) 99.35 (0.001) 2.99 (0.088) 0.56
4±5 A L 0.13 (0.72) 554.63 (0.001) 0.49 (0.48) 0.88
4±6 A L 0.01 (0.91) 879.31 (0.001) 0.01 (0.95) 0.92
4±7 A L 0.08 (0.78) 1,153.29 (0.001) 0.06 (0.81) 0.94
4±8 V L 0.18 (0.67) 1,408.03 (0.001) 0.16 (0.69) 0.95
5±1 A L 1.10 (0.30) 0.15 (0.70) 0.66 (0.42) 0.02
5±2 A L 0.12 (0.73) 21.24 (0.001) 0.12 (0.73) 0.22
5±3 A L 1.50 (0.22) 48.25 (0.001) 0.03 (0.87) 0.40
5±4 A L 1.13 (0.29) 152.46 (0.001) 1.81 (0.18) 0.66
5±5 V L 0.42 (0.52) 273.79 (0.001) 0.70 (0.41) 0.78
5±6 V L 0.43 (0.51) 348.34 (0.001) 0.19 (0.66) 0.82
5±7 V L 1.20 (0.28) 402.96 (0.001) 0.17 (0.69) 0.84
5±8 V L 0.65 (0.42) 592.76 (0.001) 0.14 (0.71) 0.89
6±1 A L 1.14 (0.29) 11.11 (0.001) 2.38 (0.13) 0.14
6±2 A H 0.79 (0.38) 0.43 (0.51) 2.03 (0.16) 0.02
6±3 V L 1.42 (0.24) 8.61 (0.004) 2.17 (0.15) 0.11
6±4 V L 3.70 (0.058) 23.85 (0.001) 3.75 (0.057) 0.26
6±5 A L 0.04 (0.84) 62.98 (0.001) 0.11 (0.74) 0.46
6±6 V L 1.91 (0.17) 37.43 (0.001) 2.03 (0.16) 0.34
6±7 V L 0.89 (0.35) 51.52 (0.001) 1.12 (0.29) 0.43
6±8 V L 0.94 (0.34) 4.51 (0.037) 0.02 (0.90) 0.07
a
Each row shows the results of a separate analysis of variance. The ®rst and second columns indicate the groups with the higher means
of LN (price o?er). The third through ®fth columns present the F statistic, with its signi®cance level in parentheses, for each of the main
and interactive e?ects. The last column reports the coecient of determination.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 731
Table 5
Analyses of variance for price revisions
a
E?ects of feedback variances and markets
Set-period Price
revision
Profit
variance
Price
variance
Variances
(combined)
Pro®t var.
(incremental)
Price var.
(incremental)
Markets R
2
1±1 ÷195 2649 450 41.76 (0.001) 5.98 (0.02) 11.34 (0.001) 19.37 (0.001) 0.58
1±2 ÷21 1366 170 36.79 (0.001) 1.46 (0.23) 12.39 (0.001) 12.94 (0.001) 0.53
1±3 ÷12 0 43 39.19 (0.001) 0.63 (0.43) 27.12 (0.001) 0.64 (0.43) 0.51
1±4 0 0 19 34.34 (0.001) 1.30 (0.26) 20.65 (0.001) 0.84 (0.36) 0.48
1±5 0 0 21 46.70 (0.001) 3.65 (0.06) 29.49 (0.001) 0.25 (0.62) 0.55
1±6 ÷3 0 13 51.19 (0.001) 0.87 (0.36) 55.92 (0.001) 0.08 (0.78) 0.57
1±7 0 0 9 73.36 (0.001) 1.36 (0.25) 67.00 (0.001) 0.10 (0.76) 0.66
2±1 ÷142 2005 588 24.71 (0.001) 15.60 (0.001) 0.43 (0.51) 0.11 (0.74) 0.40
2±2 ÷59 1006 78 28.00 (0.001) 6.52 (0.01) 3.32 (0.07) 6.41 (0.01) 0.45
2±3 ÷37 811 37 21.02 (0.001) 0.58 (0.45) 20.96 (0.001) 3.35 (0.07) 0.37
2±4 ÷23 375 19 53.27 (0.001) 2.45 (0.12) 38.65 (0.001) 0.04 (0.84) 0.58
2±5 ÷15 5 9 60.62 (0.001) 0.20 (0.66) 58.66 (0.001) 6.95 (0.01) 0.63
2±6 ÷10 0 5 39.17 (0.001) 0.73 (0.39) 54.58 (0.001) 0.64 (0.43) 0.51
2±7 ÷7 0 4 23.27 (0.001) 1.21 (0.28) 31.27 (0.001) 1.61 (0.21) 0.39
3±1 175 0 ÷26 35.70 (0.001) 3.95 (0.05) 44.14 (0.001) 8.48 (0.01) 0.51
3±2 72 0 1 36.93 (0.001) 14.83 (0.001) 33.36 (0.001) 1.53 (0.22) 0.50
3±3 37 0 5 28.57 (0.001) 10.74 (0.002) 21.83 (0.001) 1.60 (0.21) 0.44
3±4 19 0 16 67.03 (0.001) 21.13 (0.001) 44.26 (0.001) 0.18 (0.67) 0.64
3±5 15 0 4 47.21 (0.001) 18.52 (0.001) 23.03 (0.001) 0.03 (0.88) 0.55
3±6 6 0 4 34.71 (0.001) 31.09 (0.001) 5.19 (0.03) 0.70 (0.40) 0.48
3±7 4 0 4 18.49 (0.001) 27.14 (0.001) 0.19 (0.66) 5.78 (0.02) 0.36
4±1 50 0 13 96.70 (0.001) 19.36 (0.001) 54.30 (0.001) 7.78 (0.01) 0.73
4±2 25 0 13 79.71 (0.001) 63.37 (0.001) 14.94 (0.001) 24.15 (0.001) 0.71
4±3 14 0 4 42.60 (0.001) 28.32 (0.001) 8.35 (0.01) 50.81 (0.001) 0.64
4±4 0 0 5 65.71 (0.001) 30.73 (0.001) 26.51 (0.001) 36.86 (0.001) 0.69
4±5 0 0 2 26.22 (0.001) 18.04 (0.001) 7.70 (0.01) 22.13 (0.001) 0.50
4±6 0 0 1 20.48 (0.001) 7.42 (0.01) 13.25 (0.001) 28.68 (0.001) 0.48
4±7 0 0 2 24.80 (0.001) 3.52 (0.07) 26.01 (0.001) 43.52 (0.001) 0.55
5±1 ÷250 1965 203 19.18 (0.001) 3.21 (0.08) 7.58 (0.01) 8.41 (0.01) 0.38
5±2 ÷50 599 59 76.35 (0.001) 0.02 (0.96) 39.72 (0.001) 0.61 (0.44) 0.67
5±3 ÷27 0 21 64.45 (0.001) 0.45 (0.51) 45.76 (0.001) 2.99 (0.09) 0.63
5±4 ÷12 0 9 65.54 (0.001) 0.27 (0.61) 68.00 (0.001) 1.47 (0.23) 0.64
5±5 ÷5 0 8 122.92 (0.001) 4.96 (0.03) 196.61 (0.001) 0.12 (0.73) 0.76
5±6 ÷4 0 4 22.50 (0.001) 0.29 (0.59) 31.36 (0.001) 0.47 (0.49) 0.37
5±7 ÷3 0 2 22.50 (0.001) 0.29 (0.59) 31.36 (0.001) 0.47 (0.49) 0.37
6±1 ÷303 1995 720 3.12 (0.05) 1.63 (0.21) 0.09 (0.76) 6.06 (0.02) 0.14
6±2 ÷100 849 208 6.57 (0.002) 0.90 (0.35) 1.40 (0.24) 1.79 (0.19) 0.17
6±3 ÷50 249 69 7.77 (0.001) 0.34 (0.57) 4.21 (0.04) 0.46 (0.50) 0.18
6±4 ÷20 6 23 44.25 (0.001) 0.77 (0.38) 40.28 (0.001) 0.14 (0.71) 0.55
6±5 ÷10 0 8 8.27 (0.001) 1.77 (0.19) 15.49 (0.001) 1.20 (0.28) 0.19
6±6 ÷6 0 9 4.37 (0.02) 0.32 (0.57) 4.41 (0.04) 4.68 (0.03) 0.16
6±7 ÷2 0 2 8.01 (0.001) 1.79 (0.19) 5.49 (0.02) 36.73 (0.001) 0.45
a
Each row shows the results of a separate analysis of variance. The ®rst through third columns report the medians for price
revision, pro®t variance, and price variance, respectively. The fourth through seventh columns present the F statistic, with its
signi®cance level in parentheses, for the combined e?ects of pro®t and price variances, the incremental e?ect of pro®t var-
iances after adjusting for price variances and markets, the incremental e?ect of price variances after adjusting for pro®t var-
iances and markets, and the main e?ect for markets, respectively. The last column reports the coecient of determination.
732 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
(F = 1; 154:22; p < 0:001) and sets × markets
(F = 29:51; p < 0:001) further indicate the large
impact of demand conditions. The signi®cant e?ects
for periods (F = 53:69; p < 0:001), periods × sets
(F = 67:68; p < 0:001), and periods × sets × mar-
kets (F=8.35, p
Past accounting experiments have demonstrated signi®cant eects of absorption vs variable costing systems on pri-
cing decisions, but in individual settings that suppressed market features. The main ®nding of the current study is that a
cost-based pricing bias did not persist in laboratory product markets. Given the opportunity to learn from pro®t and
market feedback, sellers revised their price oers toward optimum in a manner that compensated for absorption vs
variable cost signals. The eects of demand conditions, as revealed through actual trades, dominated the eects of
alternative costing systems.
Do cost-based pricing biases persist in laboratory markets?
William S. Waller
a,
*, Brian Shapiro
b
, Galen Sevcik
c
a
Department of Accounting, University of Arizona, McClelland Hall, 301X, Tuscon, AZ 85721, USA
b
University of Minnesota, USA
c
Georgia State University, USA
Abstract
Past accounting experiments have demonstrated signi®cant e?ects of absorption vs variable costing systems on pri-
cing decisions, but in individual settings that suppressed market features. The main ®nding of the current study is that a
cost-based pricing bias did not persist in laboratory product markets. Given the opportunity to learn from pro®t and
market feedback, sellers revised their price o?ers toward optimum in a manner that compensated for absorption vs
variable cost signals. The e?ects of demand conditions, as revealed through actual trades, dominated the e?ects of
alternative costing systems. # 1999 Elsevier Science Ltd. All rights reserved.
1. Introduction
A basic issue in accounting is whether alter-
native information systems a?ect economic deci-
sions. When examining this issue, it is important
to consider the organizational or institutional set-
ting of the decision-maker, e.g. type of ®rm or
market (Hopwood, 1978; Libby, 1990). In man-
agement accounting, there is much interest in how
alternative costing systems a?ect pricing decisions
(Kaplan & Atkinson, 1998). Because pricing deci-
sions naturally occur in markets, research on cost-
based pricing should consider the essential fea-
tures of the market setting in which such decisions
are made. The setting is important for three rea-
sons. First, markets typically give sellers access to
information other than accounting signals, e.g.
past prices and o?ers, which may moderate or
overwhelm the e?ects of costing systems. Second,
pricing decisions require sellers to set goals as well
as process accounting signals, e.g. a 20% pro®t
markup on unit cost. Under market pressure, sell-
ers with alternative costing systems may compen-
sate for unit-cost di?erences when setting the
pro®t goal. Third, markets often give sellers the
opportunity to learn from feedback, with the con-
sequence that initially observed e?ects of costing
systems do not persist.
Experimental research strives for internal valid-
ity through the manipulation of independent vari-
ables, e.g. alternative costing systems, and careful
observation of behavioral e?ects, e.g. pricing
decisions, controlling for other factors. But, such
concern for internal validity does not immunize
experimentalists from the need to consider the
organizational or institutional setting of the deci-
sion-maker (Swieringa & Weick, 1982). Regarding
cost-based pricing, there has been a long line of
accounting experiments providing evidence about
the e?ects of absorption vs variable costing sys-
tems on pricing decisions in individual settings
(Ashton, 1976, 1981; Barnes & Webb, 1986;
0361-3682/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PI I : S0361- 3682( 99) 00009- 4
Accounting, Organizations and Society 24 (1999) 717±739
www.elsevier.com/locate/aos
* Corresponding author. Fax: +1-520-621-3742.
E-mail address: [email protected] (W.S. Waller)
Dyckman, Hoskin & Swieringa, 1982; Hilton,
Swieringa & Turner, 1988; Swieringa, Dyckman &
Hoskin, 1979; Turner & Hilton, 1989). The evi-
dence suggests that these alternative costing sys-
tems cause pricing biases, e.g. systematically
higher price o?ers under absorption costing. As
reviewed below, however, past experiments did
not incorporate essential market features such as
actual seller competition, price revision in light of
feedback, and the threat of bankruptcy. The
question remains as to whether cost-based pricing
biases persist in market settings.
This study contributes evidence about the e?ects
of absorption vs variable costing on pricing deci-
sions in laboratory product markets. In each of
eight markets with a posted-price institution, ten
sellers were randomly assigned absorption or
variable costing systems, and competed by making
o?ers to sell units at a price equal to the reported
unit cost plus a pro®t markup. Although their
costing systems di?ered, all sellers faced the same
increasing marginal cost function with an una-
voidable ®xed cost and a variable cost that
depended on unit sales. Sellers realized a pro®t or
loss from their trades, observed other trades, and
revised their o?ers, for each of 48 periods (six sets
of eight periods), and ultimately survived or went
bankrupt. Predictions of optimal prices and
quantities at competitive equilibrium were derived
independently of costing systems. Demand was
manipulated within and between markets. The
within-market manipulation involved a random
shift for each set of periods. This permitted repe-
ated observations of seller learning (i.e. price revi-
sion) from the starting point of ignorance about
demand. The between-market manipulation
involved the threat of bankruptcy; demand condi-
tions implied that only ®ve (ten) optimizing sellers
could break even on average in the harsh (lenient)
markets. This permitted observation of whether
seller learning accelerated when the threat of
bankruptcy was higher.
Comparable to many laboratory-market studies
(Davis & Holt, 1993), sellers' price o?ers con-
verged toward optimum in each set of periods.
Overall, the median price error (a seller's price
o?er vs the optimal price) decreased by more than
80% from period 1 to 8 of a set. Typically, sellers
who failed to make a sale lowered their price
o?ers, resulting in more units to be sold and mar-
ket eciency (actual pro®t as a percentage of
maximum pro®t) to approach 100%. Absorption
vs variable costing systems caused a bias in price
o?ers only in the ®rst period of the ®rst set, when
sellers had not yet observed actual trades and
were ignorant about demand. The e?ects of
demand conditions, as revealed through trading,
quickly overwhelmed this bias, and the costing
systems had no long-run e?ect on pro®tability or
survival. Price revisions from one period to the
next were strongly associated with both pro®t
variances (a seller's actual vs target pro®t) and
price variances (a seller's price o?er vs the average
market price). The threat of bankruptcy acceler-
ated seller learning, in that relative price revision
from period 1 to 2 was larger in the harsh vs leni-
ent markets, although this e?ect reversed in the
next few periods.
Regarding cost-based pricing biases, this study's
results resembled those of past experiments only in
the initial period. Given pro®t and market feed-
back, the bias did not persist as sellers revised their
price o?ers toward optimum without regard to
absorption vs variable cost signals. Further, the
bias did not re-emerge in the ®rst period of sub-
sequent sets, when sellers again were ignorant
about demand. The e?ects of alternative costing
systems on pricing decisions apparently depend on
individual vs market settings. As discussed below,
there are crucial di?erences between individual
and market settings, including trading institutions,
incentives, and information for learning. The issue
of whether decision behavior observed in indivi-
dual settings persists in markets has attracted
considerable attention from economists and psy-
chologists (Hogarth & Reder, 1986; Lopes, 1994;
Smith, 1991). This study adds to the growing lit-
erature on the issue (Camerer, 1987, 1992;
Camerer, Loewenstein & Weber, 1989; Cox &
Grether, 1996; Ganguly, Kagel & Moser, 1994;
Kachelmeier, 1996). Speci®cally, the strong asso-
ciation between sellers' price revisions and feed-
back variances supports the adaptive learning
model underlying the behavioral theory of the ®rm
(Cyert & March, 1992; Levitt & March, 1988).
Seller learning is an important market mechanism
718 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
(Tversky & Kahneman, 1986), which helps to
reconcile the view of many economists that mar-
kets induce optimal decisions at equilibrium and
the view of many psychologists that individual
decisions are subject to bias. At least in this
study's market setting, simple learning processes
were e?ective in moving biased decisions toward
optimal decisions. Whether learning is e?ective in
more complex settings, e.g. when sellers compete
in multiple product markets, is a question for
future research.
The rest of the paper is organized as follows.
Section 2 discusses di?erences between individual
and market settings, reviews past accounting
experiments on absorption vs variable costing,
and states the hypotheses. Sections 3 and 4
describe the experiment's method and results,
respectively. The last section provides concluding
remarks.
2. Literature review and hypotheses
2.1. Individual and market settings
Psychology and economics both focus on indi-
vidual decision behavior, but from fundamentally
di?erent perspectives (Lopes, 1994). Psychologists
view individuals as cognitive information proces-
sing systems that translate environmental stimuli,
joined with prior knowledge, into behavioral
responses. Such processing involves heuristics, or
simpli®ed procedures, that economize on a limited
capacity for encoding, retrieving, and manipulat-
ing information (Simon, 1978, 1986). Although
generally e?ective, heuristics sometimes produce
biased judgments and decisions, relative to prob-
ability and utility theory (Tversky & Kahneman,
1974, 1986). Many psychological experiments have
demonstrated such biases in individual settings
(Arkes & Hammond, 1986; Kahneman, Slovic &
Tversky, 1982). In contrast, economists build the-
ories of aggregate outcomes, e.g. equilibrium pri-
ces and quantities in competitive markets, based
on the assumption that individuals make rational
choices in terms of utility maximization (Milgrom
& Roberts, 1992). The di?erence in perspectives
has led to considerable debate on the behavioral
foundations of microeconomics and on whether
biases observed in individual settings persist in
markets (Hogarth & Reder, 1986).
A contentious view of the debate sees a contest
between psychology and economics regarding the
empirical validity of the rational-choice model. A
more productive and integrative view seeks ways
in which the limitations of psychology are alle-
viated by economics, and vice versa (Smith, 1991).
Drawing from experimental economics, Smith
stated two broad conclusions from hundreds of
laboratory-market studies: economic theory gen-
erally provides a correct ®rst approximation of
equilibrium outcomes, but is weak in describing
the processes of convergence and economizing on
decision cost. Economic theory derives predictions
using the principle of methodological individual-
ism (Blaug, 1992); i.e. ecient equilibrium out-
comes are implied by the assumedly rational
choices of individuals, along with other condi-
tions. Such derivation is open as to the actual
causal processes that produce aggregate outcomes
and, speci®cally, does not require a disequilibrium
process involving rational choices by actual indi-
viduals (Nelson & Winter, 1982). These con-
siderations suggest limitations in both economics
and psychology. Although generally successful in
predicting equilibria, economic theory provides
limited insight into causal processes. Although
validly describing cognitive processes and con-
straints, psychological experiments are limited by
usage of settings that isolate the individual from
market forces. From an integrative perspective, a
key question is:
Why is it that human subjects in the labora-
tory frequently violate the canons of rational
choice when tested as isolated individuals, but
in the social context of exchange institutions
serve up decisions that are consistent (as
though by magic) with predictive models
based on individual rationality (Smith, 1991,
p. 894)?
To answer this question requires examination of
the market processes by which equilibriumoutcomes
emerge. In this regard, experimental economics
emphasizes the role of trading institutions (i.e. the
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 719
rules governing trades between sellers and buyers),
incentives, and information for learning.
Besides testing equilibrium predictions, a major
contribution of experimental economics has been
extensive empirical evidence regarding the e?ects
of trading institutions on the rate and pattern of
market convergence. For example, double auc-
tions in which each subject may o?er or accept
either bids or asks induce rapid convergence,
whereas posted-price institutions in which sellers
make take-it-or-leave-it o?ers induce slower con-
vergence. Perhaps the most striking demonstra-
tions of institutional e?ects have been market
simulations with zero-intelligence traders pro-
grammed to generate random bids and asks, sub-
ject to only a budget constraint and an
endogenous choice set (Gode & Sunder, 1993; see
also Jamal & Sunder, 1996). Using double auc-
tions, these simulations achieved almost 100%
eciency, despite the absence of arbitrage, bank-
ruptcy, incentives, learning, or even usage of
heuristics. Ecient market outcomes emerged as a
result of the institution and environment, not from
rational choice by individuals (cf. Simon, 1982,
1986; Smith, 1991). Institutional e?ects are a pri-
mary di?erence between individual and market
settings.
Although not necessary for market eciency,
incentives and learning can strongly in¯uence the
convergence process. Standard procedures in
experimental economics include subject payments
under performance-based incentives and repeated
trading periods with feedback to allow for learn-
ing (Davis & Holt, 1993; Friedman & Sunder,
1994). Regarding incentives, in contrast with the
mixed evidence in psychology (Bonner, Young &
Hastie, 1996; Hogarth & Reder, 1986), the pre-
ponderance of evidence in experimental economics
indicates that incentives matter in markets, by
reducing inconsistencies between actual and
rational choice (Smith & Walker, 1993). The
e?ects of incentives in markets may be enhanced
by the presence of multiple, self-interested partici-
pants who a?ect each other's payo? through the
institution (Frey & Eichenberger, 1994). At a
minimum, incentives reduce noise when greater
cognitive e?ort can improve performance.
Regarding learning, the typical pattern of con-
vergence in laboratory markets is for eciency to
be relatively low in the ®rst trading period,
increase signi®cantly in the next few periods, and
increase more gradually in subsequent periods.
While this pattern presumably re¯ects subjects'
revisions in bids and asks in light of performance
feedback, experimental economics contains few
systematic attempts to describe individual learning
processes and their relation to market con-
vergence.
1
This de®ciency suggests an opportunity
for psychology to inform economics about adap-
tation by cognitive information processing sys-
tems. As with incentives, however, generalization
from individual to market settings is problematic.
A primary di?erence is that learning is an inter-
active, social phenomenon in markets where each
subject's performance feedback and observations
of market activity depend on the actions of others
through the institution.
A growing number of experiments have exam-
ined whether biases observed in individual settings
persist in market settings (Camerer, 1987; Camerer
et al., 1989; Cox & Grether, 1996; Duh & Sunder,
1986; Ganguly et al., 1994; Kachelmeier, 1996).
For example, psychological experiments have
shown that subjects in individual settings make
probability judgments using a representativeness
heuristic which causes biases such as ignoring base
rates (Kahneman & Tversky, 1972). Camerer
(1987) ran a series of markets in which subjects
traded assets that paid a state-dependent dividend.
Demand for the assets depended on subjects' pos-
terior probabilities given sample information, and
parameters were speci®ed such that usage of
representativeness vs Bayesian revision implied
di?erent equilibrium prices. Camerer (1987) found
that prices tended toward Bayesian predictions,
with only a small degree of bias attributable to
representativeness. Duh and Sunder (1986) had
similar results. In contrast, Ganguly et al. (1994)
reported that prices in their asset markets were
persistently closer to representativeness-based pre-
dictions than Bayesian predictions, especially in
markets where representativeness implied higher
1
To date, game experiments have emphasized subject learn-
ing far more than market experiments have (see Erev & Roth,
1998; Roth & Erev, 1995).
720 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
prices. Relatively few subjects made unbiased, pre-
trading probability judgments, even after 16 peri-
ods, and these subjects were not suciently active
to drive prices to Bayesian predictions.
Preference reversal is another frequent ®nding in
individual settings (Lichtenstein & Slovic, 1971;
Slovic & Lichtenstein, 1983). Suppose subjects are
asked to perform two tasks: (1) choose lottery A
with a high probability of a moderate payo? or
lottery B with a lower probability of a higher
payo?, and (2) value A and B on a monetary scale.
Many subjects choose A but place a higher value
on B. Cox and Grether (1996) examined pre-
ference reversals by manipulating the response
mode (choice vs valuation), monetary incentives
(strong vs moderate vs none), and setting (indivi-
dual vs market). There were ®ve task repetitions to
allow for learning. The results showed high rates
of preference reversal in the ®rst repetition, but
much lower rates by the ®fth repetition, especially
in market settings. As subjects in markets repeated
the task, they incorporated past prices into their
o?ers.
A ®nal example concerns the sunk-cost fallacy
whereby an individual's decision is a?ected by a
normatively irrelevant historical cost (Arkes &
Blumer, 1985). Kachelmeier (1996) ran a series of
markets to examine one aspect of sunk cost. Sub-
jects were either sellers or buyers of a security.
Prior to trading, each seller was assigned one unit
of the security, an unavoidable sunk cost, and a
redemption value (opportunity cost) in case of no
sale. A between-market manipulation was cost of
sale (sunk cost vs opportunity cost) in a pro®t
feedback report, although the formula for subject
payments did not depend on this variable. The
results showed that bids and asks were persistently
higher in the markets with the sunk-cost format.
Sellers' aversion to a paper loss produced an
upward bias in asks, and buyers reacted by raising
bids. However, the bias did not a?ect market pri-
ces or eciency, because of shifts in the percen-
tages of seller- vs buyer-initiated trades.
In sum, experimental economics provides evi-
dence from hundreds of market studies (not
focused on individual biases) that generally con-
®rms the equilibrium predictions, and indirectly
the rationality assumption, of economic theory.
The much smaller group of studies focused on
individual biases in markets provides more equi-
vocal evidence. Some studies found that individual
biases were reduced, if not eliminated, in market
settings. Other studies found that individual biases
had persistent e?ects on market o?ers and prices.
Before drawing conclusions, more research is nee-
ded on the mechanisms that drive markets and the
conditions under which these mechanisms induce
or fail to induce rational decisions.
2.2. Cost-based pricing
Surveys indicate that ®rms predominantly use
pricing policies that set initial prices equal to unit
cost plus target pro®t (see Dorward, 1987, for a
review). For example, Govindarajan and Anthony
(1983) surveyed over 500 industrial ®rms of the
Fortune 1000, asking respondents to specify ``the
method that comes closest to the one you usually
use in arriving at the normal selling price for your
typical product.'' Variations of absorption (vari-
able) cost-based pricing were speci®ed by 83%
(17%) of the respondents. Shim and Sudit (1995)
similarly surveyed 141 ®rms; 70% (12%) used
absorption (variable) cost-based pricing, and 18%
used ``market-based or competitive'' pricing.
Economists have long criticized cost-based pri-
cing, for several reasons (Oxenfeldt & Baxter,
1961). This procedure uses historical or budgeted
cost, rather than opportunity cost, and average
variable cost, rather than marginal cost. Absorption
costing includes average ®xed cost which is norma-
tively irrelevant to short-run pricing. The pro®t
markup on cost does not explicitly incorporate
information about demand.
Countering such criticism, justi®cations of cost-
based pricing invoke a decision-cost argument.
Given limited knowledge about demand and
opportunity cost, ®rms employ simpli®ed proce-
dures that economize on decision cost (Cyert &
March, 1992). When initial o?ers are subject to
revision depending on the reactions of custo-
mers and competitors, cost-based pricing may
be procedurally rational, i.e. economizing on
decision cost, though not substantively rational,
i.e. optimal without regard to decision cost
(Simon, 1976). Accordingly, costing systems may
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 721
play a role in explaining the disequilibrium process
of price formation involving procedurally rational
sellers, but no role in predicting equilibrium prices
assuming substantively rational choice (Waller,
1995). In this regard, an important issue is whe-
ther alternative costing systems bias price o?ers in
the market convergence process. Using models
and methods from psychology, past accounting
experiments have demonstrated cost-based pricing
biases in individual settings, but none has exam-
ined whether such biases persist in markets.
Early experiments used the lens model from
psychology to examine the sensitivity of sellers'
pricing decisions to changes in costing systems
(Ashton, 1976, 1981; Dyckman et al., 1982;
Swieringa et al., 1979). Ashton (1976) had each
seller set prices for 60 products using three cues,
i.e. unit cost, demand elasticity, and competitor
responsiveness. Absorption vs variable costing
was manipulated on a within-subject basis after 30
prices had been set, with the order of systems var-
ied over subjects. Sellers starting with variable
(absorption) costing also were told that the change
in systems resulted in less (more) useful informa-
tion due to the inclusion (exclusion) of ®xed cost.
Two control groups used either absorption or
variable costing for all 60 products. Ashton con-
structed a regression model of each seller's pricing
policy over the ®rst 30 products, and computed
the mean absolute di?erence between actual and
predicted prices for the last 30 products. This dif-
ference measured sensitivity to the change in cost-
ing systems, which was signi®cantly higher for the
experimental vs control groups. In a critique,
Libby (1976) expressed concerns about the con-
founded manipulation of costing-system change
and stated usefulness, and about di?erences in
cost data for the experimental and control groups.
Responding to these concerns, Swieringa et al.
(1979) performed a replication, isolating the
manipulation of costing-system change and hold-
ing constant the cost data over groups. Swieringa
et al. (1979) found that sellers with a costing-sys-
tem change adjusted their information processing
more than the control groups did. In another
replication using older subjects with more expo-
sure to accounting, Dyckman et al. (1982) repor-
ted similar results.
Two experiments tested hypotheses from Lere
(1986) about the e?ects of absorption vs variable
costing on the correspondence between pro®t-
maximizing decisions and cost-based o?ers made
with a speci®c heuristic (see also Dickhaut & Lere,
1983). The heuristic consisted of the following
steps: (1) the seller suggests a price, p, to his
accountant, (2) the accountant determines expec-
ted demand, E[q(p)], and reports the unit cost, c,
(3) the seller evaluates c against c
+
= E
[pq
/
(p) ÷ q(p)[=E[q
/
(p)[, and (4) the process iterates
until c = c
+
. The degree to which the heuristic
approximates pro®t maximization depends on
absorption vs variable costing.
2
In one experiment
(Hilton et al., 1988), each subject chose between
absorption and variable costing, and made a price
o?er after iterating with a simulated accountant
for up to ten repetitions. Each seller knew the
demand function (or probability distribution of
demand functions) and type of cost function (linear
vs nonlinear and stochastic vs deterministic). Sell-
ers had monetary incentives and pro®t feedback
after each trial. The results did not support Lere's
hypotheses. Sellers' price o?ers di?ered sig-
ni®cantly from prices under the assumed heuristic
as well as from optimal prices, and most sellers
chose absorption over variable costing under all
conditions. Also, mean price o?ers were higher
under absorption vs. variable costing under all
conditions. In an experiment involving quantity
decisions, Turner and Hilton (1989) similarly
found signi®cant divergences from optimum and a
general preference for absorption costing.
The above experiments provided limited evi-
dence regarding the role of absorption vs variable
costing for pricing decisions in markets. The
stream of lens-model studies beginning with Ashton
(1976) examined pricing decisions in individual
settings that lacked most of the distinctive features
2
Lere's (1986) model made the following predictions. Given
a linear, deterministic cost function, variable costing induces
better prices, i.e. prices set by the heuristic are closer to optimal
prices, than absorption costing does. Given a linear, stochastic
cost function and deterministic demand, variable (absorption)
costing induces better prices under risk neutrality (aversion).
Given a nonlinear cost function, absorption costing always
induces better prices.
722 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
of markets discussed earlier. Despite many task
repetitions, sellers had neither monetary incentives
nor pro®t feedback. Although available informa-
tion included cues about market conditions, there
was no actual trade between sellers and buyers.
The later studies (Hilton et al., 1988; Turner &
Hilton, 1989) added improvements such as formal
models that recognized decision cost, monetary
incentives, task repetitions (under changing con-
ditions), and pro®t feedback. However, these stu-
dies assumed decision heuristics involving
interaction between the seller and accountant,
rather than interaction among sellers and buyers
in markets. The studies also assumed that demand
was known, at least as a probability distribution.
All of the above studies involved one-shot o?ers
with no chance of price revision or convergence,
and none included actual seller competition or the
threat of bankruptcy.
Although not concerned with absorption vs
variable costing, Gupta and King (1996) examined
cost-based decisions in a setting that built upon
Hilton et al. (1988). Gupta and King (1996)
manipulated cost-report accuracy and production
complexity in a multiproduct ®rm. Each subject
acted as a monopolist (i.e. no seller competition)
who made cost forecasts for three products given
imperfect cost reports. There was no explicit pri-
cing decision. Instead, each subject's cost forecast
together with a simulated demand function deter-
mined price, quantity, and pro®t. Also, there was
no ®xed or joint cost. Total variable cost depen-
ded on each product's requirements for material
and resources in three conversion processes. The
more accurate cost report was based on two con-
version cost pools, while the less accurate report
was based on only one. Each product in the more
complex ®rm used resources nonproportionately
in the conversion processes, while each product in
the simpler ®rm used resources proportionately.
Subjects went through 20 task repetitions under
stable conditions, e.g. constant demand, with
monetary incentives and feedback. As expected,
the results showed higher pro®t given more accu-
rate cost reports and a simpler ®rm. The results
also revealed systematic learning whereby subjects
revised their forecasts toward optimum, despite
inaccurate cost reports. Learning from experience
may be a procedurally rational substitute for cost-
report accuracy. As elaborated below, this study's
experiment also involved the procedural ration-
ality of seller learning given pro®t and market
feedback.
2.3. Hypotheses
The experiment incorporated many features that,
taken together, distinguish individual and market
settings, e.g. monetary incentives, task repetitions,
pro®t feedback, trading institution, seller competi-
tion, endogenous information about demand, and
the threat of bankruptcy. The experiment produced
evidence about four hypotheses.
The ®rst hypothesis concerns market con-
vergence. It was expected that sellers' price o?ers
would converge toward optimum over each set of
trading periods. Con®rming this expectation was
necessary to establish a general correspondence
between this study's market setting and other
laboratory markets. Competitive equilibrium pre-
dictions were based on market demand and sup-
ply, without regard to costing systems. There were
four lenient markets in which demand tended to be
relatively high (Fig. 1), and four harsh markets in
which demand tended to be relatively low (Fig. 2).
The reason for the between-market manipulation
of demand was to vary the threat of bankruptcy
faced by sellers. A seller went bankrupt when his
cash balance (initial endowment?pro®t or loss)
was negative, and suboptimal decisions were more
likely to cause bankruptcy in the harsh markets.
Each market used six demand functions corre-
sponding to six sets of eight periods. The reason
for the within-market manipulation of demand
was to create recurrent disequilibrium states in
which sellers were ignorant about demand. This
allowed repeated observations of the convergence
process. The supply function (see Figs. 1 and 2)
was constructed by aggregating the sellers' mar-
ginal cost functions. For each set of a market, the
optimal price and quantity (Table 1) were deter-
mined by the intersection of the supply function
and relevant demand function. Seller performance
was measured by price error, i.e. the absolute dif-
ference between a seller's price o?er and the opti-
mal price, where the latter depended on the set and
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 723
market (Table 1).
3
Consistent with convergence
toward optimum, the ®rst hypothesis is:
H1. Sellers' price errors approach zero over periods
1±8 of a set.
The second and main hypothesis concerns the
persistence of a cost-based pricing bias in the
market convergence process. Because of the mar-
ket features discussed earlier, it was expected that
a bias in price o?ers due to absorption vs variable
costing would not persist. Assuming higher price
o?ers under absorption costing in the initial period
(comparable to the ®ndings in individual settings), it
was expected that sellers would respond to pro®t
and market feedback by revising their price o?ers
toward optimumwithout regard to unit-cost signals.
Accordingly, the results should reveal an interactive
e?ect for costing systems and periods, whereby price
o?ers di?er under absorption vs. variable costing in
the early, but not later, periods of a set.
4
H2. A bias in price o?ers due to absorption vs variable
costing does not persist over periods 1±8 of a set.
The third hypothesis concerns seller learning.
Consistent with the behavioral theory of the ®rm
(Cyert & March, 1992; Levitt & March, 1988),
seller learning may be viewed as the adaptation of
current decisions to experiential feedback on past
decisions. Sellers received two kinds of feedback,
pro®t or loss from their own trades and observa-
tions of other trades. Pro®t feedback was measured
by pro®t variance, i.e. the di?erence between a
seller's actual and target pro®t last period. Target
pro®t equaled a seller's markup times o?ered
Fig. 1. Demand and supply for lenient markets.
3
Absolute rather than signed di?erences were used, because
analyses that aggregate over positive and negative di?erences
would understate the average magnitude of deviations from
optimum (cf. Bloom®eld, 1997). Also, price o?ers converged
toward optimum from above (positive di?erences) in some sets,
but from below (negative di?erences) in other sets. Analyses
that aggregate over sets would involve the same understatement
problem.
4
Although sellers returned to a similar state of ignorance
about demand in period 1 of each set, it is an open issue whe-
ther the same pattern of a cost-based pricing bias (i.e. its
emergence and persistence or elimination) unfolds in each set.
Task experience may a?ect the pattern in later sets (see
Results).
724 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
quantity under absorption costing, or, markup
times o?ered quantity minus ®xed cost under
variable costing. When a seller made a sale at the
o?ered price and quantity, the pro®t variance was
zero; otherwise, the pro®t variance was unfavor-
able. Market feedback was measured by price var-
iance, i.e. the di?erence between a seller's price
o?er and the average price per unit sold last per-
iod. It was expected that pro®t and price variances
would be correlated, but not perfectly. High price
o?ers relative to competitors' were likely to result
in both pro®t and price variances, whereas low
price o?ers relative to competitors' were likely to
result in a price variance but no pro®t variance.
Seller learning was measured by price revision, i.e.
the increase or decrease in a seller's price o?er
from t to t ÷ 1(t = 1; :::; 7). It was expected that
larger pro®t and price variances would induce lar-
ger price revisions.
H3. Price revisions depend on feedback regarding
pro®t and price variances.
The last hypothesis concerns the threat of
bankruptcy. Using lenient vs harsh markets as a
proxy for the threat of bankruptcy, it was expected
that sellers' price revisions would accelerate as the
threat of bankruptcy increased. When comparing
price revisions between the lenient vs harsh markets,
however, a possible confound was that di?erences in
optimal prices may have led to di?erences in total
price revisions, regardless of the rate of price revi-
sions. Accordingly, a second measure of learning,
which used each seller as his own control, was rela-
tive price revision, i.e. the ratio of a seller's price
revision from period t to t ÷ 1(t = 1; :::; 7) over his
total price revision from period 1 to 8 of a set. If a
greater threat of bankruptcy causes faster price
revision, then the results should reveal an interactive
e?ect for markets and periods, whereby relative
price revision is higher in the harsh vs. lenient mar-
kets in the early periods of a set.
5
H4. A greater threat of bankruptcy increases rela-
tive price revision in the early periods of a set.
Fig. 2. Demand and supply for harsh markets.
5
By construction, each subject's relative price revision sums
to one over a set of periods. If relative price revision is higher in
the harsh markets for early periods (H4), then the measure
must be higher in the lenient markets for later periods.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 725
3. Method
The procedure for the lenient markets is descri-
bed ®rst, followed by di?erences for the harsh
markets. In each trading period, ten sellers
(undergraduate business students) made o?ers to
sell a product. As a bu?er against possible loss,
each seller received an initial endowment of 50,000
francs, the experimental currency. Making an o?er
consisted of stating the maximum quantity for sale
between one and ®ve units, referring to a schedule
showing the unit cost at each quantity, and
adjusting the unit cost for target pro®t. The sche-
dule was prepared under absorption costing for
®ve sellers and under variable costing for ®ve sell-
ers, holding constant the true cost function.
6
The
unit cost under absorption costing was 1890, 1025,
737, 715 and 702 francs, for one to ®ve units,
respectively; the corresponding numbers under
variable costing were 30, 95, 117, 250 and 330
francs. There were 48 periods, split into six sets of
eight. Buyers were simulated with a computer
program executing the demand functions in Figs.
1 and 2.
7
Demand was constant for each set, but
varied over sets. Speci®cally, D1 in Fig. 1 was
employed for set 1, D2 for set 2, and so on. Peri-
odic pro®t or loss equalled the o?ered price times
units sold minus total variable and ®xed cost.
Variable cost depended on units sold, but ®xed
cost was unavoidable. A bankruptcy occurred
when a seller's franc balance was negative. At the
end of the experiment, each seller's franc balance
was translated into the probability of winning $20
in a lottery, at a rate of 1/100,000. This payo?
structure provided a control for risk preference, in
that a seller who maximized the expected prob-
ability of winning the cash prize was risk-neutral
as to francs when making sales o?ers (Davis &
Holt, 1993). Cost and demand parameters were
speci®ed such that optimal prices and quantities
implied zero pro®t on average (Table 1). The
expected probability of winning the prize was 0.50
(50,000/100,000), and expected pay was $25
(0.50×$20+$15 for participating).
To ensure consistency over markets, the
instructions were presented to sellers by playing a
pre-recorded tape. The speaker on the tape read
the instructions aloud, while sellers read along in a
booklet. The appendix contains the instructions
and o?er sheet for sellers with absorption costing.
The instructions and o?er sheet for sellers with
variable costing were identical, except for the cost
schedule. The instructions described the market
setting in detail: the trading institution, task of
making o?ers, sequence of events in a period,
initial endowment and payo? structure, general
nature of demand and timing of random shifts,
limits on unit sales, pro®t computation, periodic
®xed cost and variable cost, cost schedule, price
limits (0±2000 francs), and bankruptcy rule.
Table 1
Optimal prices, quantities, and pro®t (per seller)
Demand Price Quantity Profit
A. Lenient markets
D1 755 5 265
D2 315 3 ÷1,265
D3 1,155 5 2,265
D4 955 5 1,265
D5 415 3 ÷965
D6 215 3 ÷1,565
Average 635 4 0
B. Harsh markets
D1 395 3 ÷1.025
D2 195 3 ÷1,625
D3 855 5 765
D4 705 5 15
D5 295 3 ÷1,325
D6 65 1 ÷1,825
Average 418.33 3.33 ÷836.66
6
The true cost function was a ®xed cost of 1860 francs, and
a marginal cost of 30 francs for the ®rst unit sold, 160 francs
for the second and third units sold, and 650 francs for the
fourth and ®fth units sold. Relatively large steps in the cost
function were used to ensure nontrivial di?erences in optimal
prices among alternative demand levels in each market (Figs. 1
and 2 and Table 1).
7
To illustrate, D1 in Fig. 1 implied that buyers were willing
to acquire one unit at a price of 1245 francs, two units at 1235
francs, three units at 1225 francs, four units at 1215 francs, . . .,
and 50 units at 755 francs. Suppose that two sellers each o?ered
to sell two units, but at di?erent prices, i.e. 1210 francs (seller
A) and 1220 francs (seller B). The trades would be two units at
1210 francs for seller A and one unit at 1220 francs for seller B.
726 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
After reading the instructions, sellers were
shown four examples covering various possible
sales o?ers and demand conditions. For simpli-
city, each example included only two sellers, but
gave detailed explanations for completing o?er
sheets and interpreting feedback. Example 1 was a
case in which the sellers o?ered di?erent prices
and quantities, and buyers accepted both o?ers.
Example 2 repeated the o?ers of Example 1, but
demand was lower such that only one seller made
a sale. Example 3 was a case in which the sellers
o?ered the same price, but demand was not su-
cient for each to sell the o?ered quantities. When
sellers o?ered the same price, the number of units
sold were divided as evenly as possible. When an
even distribution was not possible, the remaining
unit(s) was allocated randomly. Example 4 was a
case in which a seller went bankrupt. The exam-
ples were prepared under absorption costing for
®ve subjects and variable costing for ®ve subjects.
The speaker on the tape did not refer to speci®c
values of cost or markups. All subjects heard the
same description of the market procedure.
In each period, sellers had 3 min to complete the
o?er sheet, after which the sheets were collected
and the o?ers were entered into a computer. Peri-
odic feedback consisted of two parts. One part
was market activity (i.e. price and quantity for
each sale) which was displayed publicly on large
screens. Another screen displayed publicly the
number of surviving sellers, which was updated
when a bankruptcy occurred. The other part was
each seller's results (i.e. the price and quantity for
a sale, revenue, total cost, pro®t or loss, and
beginning and ending franc balances) which were
displayed privately on a terminal. These displays
continued until the o?ers for the next period
were entered. To ensure that sellers paid atten-
tion to both types of feedback, there was a 20-s
lag between initiation of the public display of
market results and private display of each seller's
results.
This procedure was repeated for 48 periods. The
entire experiment, including the instructions, las-
ted approximately 2.5 h. At the end, the lottery
was played separately for each seller with a posi-
tive franc balance, who was paid according to the
above description.
The harsh markets followed the same proce-
dure, with one exception.
8
Because of the di?er-
ence in pro®t opportunities, the cash prize was
raised to $100 in order to equate expected pay for
all markets. Given optimal behavior in the harsh
markets (Table 1), each seller's ending franc balance
would be 9840 francs (50,000÷48×836.66). The
expected probability of winning the prize was.10
(9,840/100,000), and expected pay was $25
(0.10×$100+$15). This procedure held expected pay
constant at $25 for the lenient and harsh markets.
4. Results
4.1. Convergence toward optimum
H1 predicted that sellers' price errors approach
zero over periods 1±8 of a set. Table 2 reports
descriptive statistics regarding seller and market
performance by set and period, separately for the
lenient and harsh markets.
9
The ®rst column
shows medians for price error over sellers; e.g., in
period 1 of set 1, the median price error was 245
(628) francs, relative to the optimal price of 755
(395) francs, for the 40 sellers in the lenient (harsh)
markets. Changes in median price errors from
period 1±8 show a clear pattern of convergence
toward optimum. Averaging over sets for the
8
In two of the harsh markets, the number of periods was
allowed to exceed 48 (to at most 51), in order to achieve a
minimum bankruptcy rate of 4/10. The intent was to observe
whether bankruptcies were related to costing systems. There
were 22 survivers in the harsh markets after all periods; 13 were
assigned absorption costing and 9 were assigned variable cost-
ing (
2
= 1:62; p > 0:10). To facilitate comparisons between
markets, the Results section analyzes the data for periods 1±48
only.
9
All laboratory-market studies with multiple trading peri-
ods share the problem of serial dependence among observations
over time. Indeed, H1 predicts a particular pattern of serial
dependence. The approach taken here is to present the results
on a disaggregated basis and, as appropriate, present additional
aggregate tests. The disaggregated analyses were not intended
as independent hypothesis tests. Repeated-measures analysis of
variance was used in the aggregate tests, with set and period as
within-subject factors. These tests accommodated serial depen-
dence by assessing the e?ects of each within-subject factor, and
between-subjects factors, with respect to distinct error terms
(Neter, Kutner, Nachtsheim & Wasserman, 1996).
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 727
Table 2
Seller and market performance
a
Seller performance Market performance
Set-period Price error Quantity error Profit error No. of trades/ no. of sellers Efficiency
1÷1 245 (628) 2 (1) 755 (835) 24/40 (4/40) 69% (10%)
1÷2 153 (348) .5 (1) 725 (835) 28/40 (11/40) 81 (45)
1÷3 145 (127) 0 (1) 475 (655) 32/40 (21/40) 89 (72)
1÷4 117 (105) 0 (0) 475 (217) 34/40 (25/40) 93 (77)
1÷5 105 (72) 0 (0) 507 (360) 31/40 (23/40) 93 (80)
1÷6 98 (75) 0 (0) 475 (235) 34/40 (26/40) 94 (82)
1÷7 96 (72) 0 (0) 445 (225) 34/40 (27/40) 94 (84)
1÷8 89 (62) 0 (.5) 392 (216) 35/40 (28/40) 96 (86)
2÷1 486 (455) 2 (0) 595 (235) 1/40 (7/40) 0 (26)
2÷2 395 (248) 2 (0) 595 (235) 6/40 (12/40) 27 (26)
2÷3 207 (202) 2 (0) 595 (235) 14/40 (15/40) 58 (65)
2÷4 160 (155) 1 (0) 595 (235) 17/40 (19/40) 68 (76)
2÷5 110 (135) 1 (0) 545 (235 20/40 (23/40) 80 (81)
2÷6 83 (123) .5 (0) 241 (291) 28/40 (25/40) 84 (83)
2÷7 59 (110) 0 (0) 257 (236) 31/40 (29/40) 88 (86)
2÷8 46 (100) 0 (0) 193 (235) 31/40 (31/40) 87 (88)
3÷1 411 (405) 0 (2) 2,180 (1,827) 35/40 (37/40) 89 (86)
3÷2 255 (108) 0 (0) 1,275 (649) 38/40 (36/40) 96 (92)
3÷3 194 (55) 0 (0) 1,025 (280) 38/40 (36/40) 95 (95)
3÷4 145 (44) 0 (0) 775 (216) 38/40 (38/40) 98 (96)
3÷5 104 (44) 0 (0) 537 (210) 37/40 (35/40) 95 (96)
3÷6 94 (54) 0 (0) 475 (270) 40/40 (36/40) 97 (95)
3÷7 63 (75) 0 (0) 360 (370) 38/40 (35/40) 99 (94)
3÷8 55 (73) 0 (0) 302 (322) 36/40 (36/40) 94 (94)
4÷1 205 (99) 0 (0) 1,025 (497) 37/40 (31/40) 91 (91)
4÷2 86 (5) 0 (0) 462 (415) 38/40 (33/40) 98 (95)
4÷3 50 (110) 0 (0) 275 (537) 37/40 (34/40) 96 (96)
4÷4 45 (114) 0 (0) 225 (550) 37/40 (37/40) 95 (96)
4÷5 43 (96) 0 (0) 222 (47) 38/40 (35/40) 96 (96)
4÷6 42 (93) 0 (0) 210 (517) 38/40 (34/40) 96 (97)
4÷7 42 (81) 0 (0) 212 (462) 38/40 (35/40) 96 (97)
4÷8 35 (57) 0 (0) 165 (285) 39/40 (35/40) 97 (98)
5÷1 335 (425) 2 (2) 895 (235) 4/40 (4/40) 10 (17)
5÷2 185 (155) 1 (0) 725 (235) 20/40 (22/40) 54 (66)
5÷3 79 (105) 0 (0) 375 (366) 25/40 (26/40) 81 (80)
5÷4 35 (77) 0 (0) 157 (390) 29/40 (29/40) 92 (86)
5÷5 16 (54) 0 (0) 228 (415) 33/40 (32/40) 95 (90)
5÷6 15 (44) 0 (0) 91 (414) 35/40 (34/40) 97 (92)
5÷7 13 (35) 0 (0) 43 (373) 36/40 (35/40) 97 (93)
5÷8 12 (27) 0 (0) 43 (358) 39/40 (36/40) 97 (94)
6÷1 586 (645) 2 (4) 295 (35) 1/40 (0/40) 0 (0)
6÷2 235 (335) 0 (2) 295 (35) 15/40 (1/40) 50 (0)
6÷3 160 (165) 0 (2) 295 (35) 20/40 (6/39) 71 (29)
6÷4 112 (57) 0 (1) 295 (35) 26/40 (15/38) 84 (73)
6÷5 88 (30) 0 (0) 268 (35) 28/40 (23/38) 87 (84)
6÷6 81 (20) 0 (0) 244 (23) 31/40 (27/38) 90 (87)
6÷7 65 (15) 0 (0) 193 (15) 32/40 (29/33) 91 (91)
6÷8 59 (15) 0 (0) 165 (15) 34/40 (23/31) 93 (90)
a
The ®rst (second) entry in each cell pertains to the lenient (harsh) markets. Entries in the ®rst through third columns are medians
over sellers for price error, quantity error, and pro®t error, respectively. Entries in the fourth column are frequencies of sellers with
trades. Entries in the ®fth column are medians over markets for eciency.
728 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
lenient (harsh) markets, the median price error
decreased by 85% (81%) from periods 1 to 8. For
each set, a Wilcoxon sign test was used to compare
the price errors in period 1 and 8, separately for
the lenient and harsh markets; in each case, the
decrease was highly signi®cant (p < 0:001). As an
aggregate test, a repeated-measures analysis of
variance (ANOVA) was performed using LN
(price error) as the dependent measure, with sets
(1±5) and periods (1±8) as within-subjects factors,
and lenient vs harsh markets and absorption vs
variable costing as between-subjects factors. Exclud-
ing set 6 avoided the problemof missing values due to
bankruptcies.
10
Using logged price o?ers as the
dependent variable reduced the inordinate e?ects of
unusually high or low o?ers. The ®rst column of
Table 3 shows the main and interactive e?ects (results
in other columns are discussed later). The relevant
result for H1 was the signi®cant e?ect of periods
(F = 254:96; p < 0:001). Consistent with the con-
vergence pattern in the ®rst column of Table 2, the
mean of LN (price error) decreased monotonically
from period 1 to 8. These results support H1.
Table 3 shows other signi®cant e?ects for LN(price
error), including sets (F = 86:43; p < 0:001), markets
(F = 9:60; p < 0:003), periods×markets (F = 7:39;
p < 0:001), sets×markets (F = 29:10; p < 0:001),
periods×sets (F = 13:16; p < 0:001), and periods
×markets×sets (F = 11:52; p < 0:001). The mean of
LN(price error) decreased fromset 1 to 5, which may
be attributed to task experience and changing
demand conditions. The mean generally was higher
for harsh vs lenient markets, and the di?erence was
larger in later periods. The market e?ect also varied
over sets. The mean was higher for harsh (lenient)
markets in sets 2 and 4 (3 and 5), and about the same
in set 1. In all sets and markets, the mean decreased
over periods, but at somewhat di?erent rates.
11
In addition to price errors, Table 2 provides
descriptive statistics on other measures of seller
and market performance. The second and third
columns show the median quantity errors and
pro®t errors, respectively. A quantity error was
de®ned as the absolute di?erence between a seller's
quantity o?er and the optimal quantity, and a
pro®t error as the absolute di?erence between a
seller's actual and optimal pro®t, where the opti-
mal values depended on the set and market (Table 1).
The fourth column shows the number of trades
divided by the number of participating sellers. The
last column shows the median eciency over
markets. Eciency increased toward 100% as the
total quantity sold increased toward optimum, but
at a decreasing rate because buyers' marginal
value was decreasing and sellers' marginal cost
was increasing.
12
Generally, these measures were
consistent with the decrease in price errors over
periods. In most sets and markets, median quantity
errors quickly decreased to zero, median pro®t
errors decreased over periods, and both the per-
centage of sellers with trades and market eciency
approached 100%.
13
These convergence results
are comparable to many laboratory-market stu-
dies.
10
As expected, some sellers in the harsh markets went
bankrupt, beginning in period 3 of set 6; nine sellers went
bankrupt by period 8 of set 6.
11
As to the interactive e?ect of periods × markets × sets,
the mean of LN (price error) was higher for the harsh (lenient)
markets in later periods of sets 2 and 4 (1); the mean was higher
for the harsh (lenient) markets in most periods of set 3 (5).
12
To illustrate, the demand function for set 1 of the lenient
markets implied that buyers valued the ®rst unit at 1245 francs,
second unit at 1235 francs, third unit at 1225 francs, . . ., and 50th
unit at 755 francs. Given the optimal quantity of 50 units, the
aggregate value to buyers was 50,000 francs, and the aggregate
variable cost to sellers was 16,500 francs (10×30+20×160+
20×650), so that maximum pro®t was 33,500 francs. Suppose that
three units were sold by di?erent sellers (at any prices between 30
and 1225 francs), resulting in aggregate pro®t of 3615 francs
(1245+1235+1225÷30×3). In this case, eciency would be 11%
(3,615/33,500).
13
There were some exceptions to the general convergence
pattern. Regarding pro®t errors, there was minimal con-
vergence in sets 2, 5, and 6 of the harsh markets, which may be
attributed to the relatively small change in pro®t when a sale
was made vs no sale. In set 6 of the harsh markets, for example,
no sale implied a pro®t of -1,860 francs, compared to optimal
pro®t of ÷1825 francs assuming the sale of one unit at 65
francs. Regarding the percentage of sellers with trades, the
pattern was di?erent for sets 3 and 4, compared to the other
sets, in that the percentage started and stayed high. Figs. 1 and
2 show a large upward shift in demand from set 2 to 3, and a
small downward shift from set 3 to 4. Although sellers knew
that shifts in demand were random, prices in period 8 of a set
apparently a?ected o?ers in period 1 of the next set. Such car-
ryover e?ects explain why there were more trades from the start
in sets 3 and 4.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 729
4.2. E?ects of costing systems
H2 predicted that a bias in price o?ers due to
absorption vs variable costing does not persist
over periods 1 to 8 of a set. Table 4 reports an
ANOVA for each period using LN (price o?er) as
the dependent variable, with absorption vs vari-
able costing and lenient vs harsh markets as
between-subjects factors. The ®rst and second
columns indicate whether the mean of LN (price
o?er) was higher for absorption vs variable cost-
ing, and for lenient vs harsh markets, respectively.
The number of observations was 80 (10 sellers×8
markets) in each ANOVA, except for periods 3±8
of set 6, which omitted bankrupt sellers. Focusing
on set 1, costing systems had a signi®cant e?ect in
period 1 (F = 8:88; p < 0:004); the mean (median)
price o?er was 1162 (1048) francs under absorp-
tion costing vs 933 (968) francs under variable
costing. The higher unit cost under absorption
costing apparently caused higher price o?ers,
when sellers had not yet observed actual trades
and were ignorant about demand. In period 2, the
mean (median) price o?er was 908 (877) francs
under absorption costing vs 836 (772) francs under
variable costing. Although price o?ers were still
higher under absorption costing, the e?ect of
costing systems was no longer signi®cant. In peri-
ods 3±8, price o?ers under absorption vs. variable
costing were about the same. In contrast, harsh vs
lenient markets had an insigni®cant e?ect in per-
iod 1, but highly signi®cant e?ects in the remain-
ing periods. R
2
increased from 0.11 in period 1 to
0.97 in period 8, due to the increasingly strong
e?ects of markets and decreasing variation in price
o?ers. The results for set 1 support H2. The initi-
ally observed bias did not persist, but instead was
overwhelmed by demand conditions as revealed
through actual trades.
The results for sets 2±6 showed a similar pattern
of weak e?ects due to costing systems, strong
e?ects due to markets, and increasing R
2
over
periods. Unlike set 1, however, there was only a
small cost-based pricing bias in period 1 of the
subsequent sets (Table 4). Although the mean
value of LN(price o?er) was higher under absorp-
tion costing in period 1 of each set, costing sys-
tems had no signi®cant e?ect. This result may be
attributed to the combination of three factors.
First, variation in price o?ers was relatively high in
period 1 of each set, which lowered the likelihood
of detecting of a signi®cant bias. Second, despite
the random shifts in demand over sets, prices in
period 8 of a set often a?ected o?ers in period 1 of
the next set (see note 13). Such carryover e?ects
would mitigate a cost-based pricing bias. Third,
some sellers learned during set 1 that their costing
system was of limited usefulness, and placed less
reliance on unit-cost signals in subsequent sets.
As an aggregate test, an ANOVA was per-
formed using LN(price o?er) as the dependent
variable, with periods (1±8) and sets (1±5) as
within-subjects factors, and absorption vs variable
costing and lenient vs harsh markets as between-
subjects factors. The second column of Table 3
shows the main and interactive e?ects. Consistent
with the above results, there was an insigni®cant
e?ect for costing systems (F = 1:20; p > 0:28), and a
signi®cant e?ect for markets (F = 409:46;
p < 0:001). The signi®cant e?ects for sets
Table 3
Analyses of variance for price errors, o?ers, and relative price
revision
a
LN
(price error)
LN
(price o?er)
Relative
price revision
Within-subjects e?ects:
P 254.96 (.001) 53.69 (.001) 374.99 (.001)
P × M 7.39 (.001) 1.43 (.19) 12.85 (.001)
P × CS 0.54 (.80) 2.03 (.05) 0.22 (.97)
P × M × CS 0.94 (.47) 0.69 (.68) 0.73 (.63)
S 86.43 (.001) 1,154.22 (.001) ±
S × M 29.10 (.001) 29.51 (.001) ±
S × CS 1.02 (.39) 1.06 (.38) ±
S × M × CS 1.39 (.24) 1.17 (.33) ±
P × S 13.16 (.001) 67.68 (.001) 3.39 (.001)
P × S × M 11.52 (.001) 8.35 (.001) 2.85 (.001)
P × S × CS 0.74 (.84) 1.14 (.28) 1.31 (.14)
P × S × M × CS 0.78 (.78) 1.02 (.44) 0.65 (.90)
Between-subjects e?ects:
M 9.60 (.003) 409.46 (.001) ±
CS 0.02 (.87) 1.20 (.28) ±
M × CS 1.13 (.29) 1.66 (.20) ±
a
P-stands for periods; S-for sets; M-for harsh vs lenient
markets, and CS for absorption vs variable costing systems.
Entries are F statistics with signi®cance levels in parentheses.
730 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
Table 4
Analyses of variance for LN (price o?er)
a
Group with higher LN (price o?er) Main and interactive e?ects
Set-period Absorption (A)
vs. variable (V)
Lenient (L)
vs harsh (H)
Costing
systems
Markets Costing systems
× markets
R
2
1±1 A L 8.88 (0.004) 0.31 (.58) 1.63 (0.21) 0.11
1±2 A L 1.63 (0.21) 11.68 (0.001) 0.01 (.91) 0.15
1±3 V L 0.17 (0.68) 74.99 (0.001) 0.45 (0.51) 0.49
1±4 V L 0.74 (0.39) 107.61 (0.001) 0.03 (0.86) 0.59
1±5 V L 0.85 (0.36) 140.77 (0.001) 4.52 (0.037) 0.64
1±6 A L 0.12 (0.73) 418.03 (0.001) 2.15 (0.15) 0.84
1±7 V L 0.17 (0.68) 309.30 (0.001) 0.20 (0.66) 0.80
1±8 A L 0.21 (0.65) 2,395.89 (0.001) 0.30 (0.59) 0.97
2±1 A L 1.83 (0.18) 14.46 (0.001) 2.03 (0.16) 0.17
2±2 V L 0.02 (0.89) 41.74 (0.001) 0.04 (0.84) 0.35
2±3 V L 0.09 (0.77) 1.40 (0.24) 1.49 (0.23) 0.02
2±4 A L 0.95 (0.33) 15.70 (0.001) 0.09 (0.77) 0.18
2±5 A L 2.70 (0.10) 48.14 (0.001) 0.01 (0.91) 0.40
2±6 A L 0.03 (0.86) 45.19 (0.001) 0.01 (0.95) 0.37
2±7 A L 0.31 (0.58) 105.12 (0.001) 0.31 (0.58) 0.58
2±8 A L 0.05 (0.83) 133.88 (0.001) 3.69 (0.06) 0.63
3±1 A L 0.01 (0.93) 34.15 (0.001) 0.05 (0.83) 0.31
3±2 A L 0.02 (0.90) 20.22 (0.001) 0.06 (0.81) 0.21
3±3 V L 4.03 (0.048) 58.25 (0.001) 0.53 (0.47) 0.45
3±4 A L 0.19 (0.66) 84.74 (0.001) 0.24 (0.62) 0.53
3±5 V L 0.43 (0.51) 33.20 (0.001) 0.36 (0.55) 0.31
3±6 V L 0.02 (0.89) 127.80 (0.001) 0.26 (0.61) 0.63
3±7 A L 0.05 (0.83) 182.37 (0.001) 0.10 (0.76) 0.71
3±8 V L 0.26 (0.61) 71.55 (0.001) 1.92 (0.17) 0.48
4±1 A L 0.01 (0.97) 1.00 (0.32) 0.51 (0.48) 0.01
4±2 A L 0.37 (0.55) 23.96 (0.001) 0.15 (0.70) 0.24
4±3 A L 0.01 (0.97) 133.22 (0.001) 0.07 (0.79) 0.64
4±4 V L 2.67 (0.11) 99.35 (0.001) 2.99 (0.088) 0.56
4±5 A L 0.13 (0.72) 554.63 (0.001) 0.49 (0.48) 0.88
4±6 A L 0.01 (0.91) 879.31 (0.001) 0.01 (0.95) 0.92
4±7 A L 0.08 (0.78) 1,153.29 (0.001) 0.06 (0.81) 0.94
4±8 V L 0.18 (0.67) 1,408.03 (0.001) 0.16 (0.69) 0.95
5±1 A L 1.10 (0.30) 0.15 (0.70) 0.66 (0.42) 0.02
5±2 A L 0.12 (0.73) 21.24 (0.001) 0.12 (0.73) 0.22
5±3 A L 1.50 (0.22) 48.25 (0.001) 0.03 (0.87) 0.40
5±4 A L 1.13 (0.29) 152.46 (0.001) 1.81 (0.18) 0.66
5±5 V L 0.42 (0.52) 273.79 (0.001) 0.70 (0.41) 0.78
5±6 V L 0.43 (0.51) 348.34 (0.001) 0.19 (0.66) 0.82
5±7 V L 1.20 (0.28) 402.96 (0.001) 0.17 (0.69) 0.84
5±8 V L 0.65 (0.42) 592.76 (0.001) 0.14 (0.71) 0.89
6±1 A L 1.14 (0.29) 11.11 (0.001) 2.38 (0.13) 0.14
6±2 A H 0.79 (0.38) 0.43 (0.51) 2.03 (0.16) 0.02
6±3 V L 1.42 (0.24) 8.61 (0.004) 2.17 (0.15) 0.11
6±4 V L 3.70 (0.058) 23.85 (0.001) 3.75 (0.057) 0.26
6±5 A L 0.04 (0.84) 62.98 (0.001) 0.11 (0.74) 0.46
6±6 V L 1.91 (0.17) 37.43 (0.001) 2.03 (0.16) 0.34
6±7 V L 0.89 (0.35) 51.52 (0.001) 1.12 (0.29) 0.43
6±8 V L 0.94 (0.34) 4.51 (0.037) 0.02 (0.90) 0.07
a
Each row shows the results of a separate analysis of variance. The ®rst and second columns indicate the groups with the higher means
of LN (price o?er). The third through ®fth columns present the F statistic, with its signi®cance level in parentheses, for each of the main
and interactive e?ects. The last column reports the coecient of determination.
W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739 731
Table 5
Analyses of variance for price revisions
a
E?ects of feedback variances and markets
Set-period Price
revision
Profit
variance
Price
variance
Variances
(combined)
Pro®t var.
(incremental)
Price var.
(incremental)
Markets R
2
1±1 ÷195 2649 450 41.76 (0.001) 5.98 (0.02) 11.34 (0.001) 19.37 (0.001) 0.58
1±2 ÷21 1366 170 36.79 (0.001) 1.46 (0.23) 12.39 (0.001) 12.94 (0.001) 0.53
1±3 ÷12 0 43 39.19 (0.001) 0.63 (0.43) 27.12 (0.001) 0.64 (0.43) 0.51
1±4 0 0 19 34.34 (0.001) 1.30 (0.26) 20.65 (0.001) 0.84 (0.36) 0.48
1±5 0 0 21 46.70 (0.001) 3.65 (0.06) 29.49 (0.001) 0.25 (0.62) 0.55
1±6 ÷3 0 13 51.19 (0.001) 0.87 (0.36) 55.92 (0.001) 0.08 (0.78) 0.57
1±7 0 0 9 73.36 (0.001) 1.36 (0.25) 67.00 (0.001) 0.10 (0.76) 0.66
2±1 ÷142 2005 588 24.71 (0.001) 15.60 (0.001) 0.43 (0.51) 0.11 (0.74) 0.40
2±2 ÷59 1006 78 28.00 (0.001) 6.52 (0.01) 3.32 (0.07) 6.41 (0.01) 0.45
2±3 ÷37 811 37 21.02 (0.001) 0.58 (0.45) 20.96 (0.001) 3.35 (0.07) 0.37
2±4 ÷23 375 19 53.27 (0.001) 2.45 (0.12) 38.65 (0.001) 0.04 (0.84) 0.58
2±5 ÷15 5 9 60.62 (0.001) 0.20 (0.66) 58.66 (0.001) 6.95 (0.01) 0.63
2±6 ÷10 0 5 39.17 (0.001) 0.73 (0.39) 54.58 (0.001) 0.64 (0.43) 0.51
2±7 ÷7 0 4 23.27 (0.001) 1.21 (0.28) 31.27 (0.001) 1.61 (0.21) 0.39
3±1 175 0 ÷26 35.70 (0.001) 3.95 (0.05) 44.14 (0.001) 8.48 (0.01) 0.51
3±2 72 0 1 36.93 (0.001) 14.83 (0.001) 33.36 (0.001) 1.53 (0.22) 0.50
3±3 37 0 5 28.57 (0.001) 10.74 (0.002) 21.83 (0.001) 1.60 (0.21) 0.44
3±4 19 0 16 67.03 (0.001) 21.13 (0.001) 44.26 (0.001) 0.18 (0.67) 0.64
3±5 15 0 4 47.21 (0.001) 18.52 (0.001) 23.03 (0.001) 0.03 (0.88) 0.55
3±6 6 0 4 34.71 (0.001) 31.09 (0.001) 5.19 (0.03) 0.70 (0.40) 0.48
3±7 4 0 4 18.49 (0.001) 27.14 (0.001) 0.19 (0.66) 5.78 (0.02) 0.36
4±1 50 0 13 96.70 (0.001) 19.36 (0.001) 54.30 (0.001) 7.78 (0.01) 0.73
4±2 25 0 13 79.71 (0.001) 63.37 (0.001) 14.94 (0.001) 24.15 (0.001) 0.71
4±3 14 0 4 42.60 (0.001) 28.32 (0.001) 8.35 (0.01) 50.81 (0.001) 0.64
4±4 0 0 5 65.71 (0.001) 30.73 (0.001) 26.51 (0.001) 36.86 (0.001) 0.69
4±5 0 0 2 26.22 (0.001) 18.04 (0.001) 7.70 (0.01) 22.13 (0.001) 0.50
4±6 0 0 1 20.48 (0.001) 7.42 (0.01) 13.25 (0.001) 28.68 (0.001) 0.48
4±7 0 0 2 24.80 (0.001) 3.52 (0.07) 26.01 (0.001) 43.52 (0.001) 0.55
5±1 ÷250 1965 203 19.18 (0.001) 3.21 (0.08) 7.58 (0.01) 8.41 (0.01) 0.38
5±2 ÷50 599 59 76.35 (0.001) 0.02 (0.96) 39.72 (0.001) 0.61 (0.44) 0.67
5±3 ÷27 0 21 64.45 (0.001) 0.45 (0.51) 45.76 (0.001) 2.99 (0.09) 0.63
5±4 ÷12 0 9 65.54 (0.001) 0.27 (0.61) 68.00 (0.001) 1.47 (0.23) 0.64
5±5 ÷5 0 8 122.92 (0.001) 4.96 (0.03) 196.61 (0.001) 0.12 (0.73) 0.76
5±6 ÷4 0 4 22.50 (0.001) 0.29 (0.59) 31.36 (0.001) 0.47 (0.49) 0.37
5±7 ÷3 0 2 22.50 (0.001) 0.29 (0.59) 31.36 (0.001) 0.47 (0.49) 0.37
6±1 ÷303 1995 720 3.12 (0.05) 1.63 (0.21) 0.09 (0.76) 6.06 (0.02) 0.14
6±2 ÷100 849 208 6.57 (0.002) 0.90 (0.35) 1.40 (0.24) 1.79 (0.19) 0.17
6±3 ÷50 249 69 7.77 (0.001) 0.34 (0.57) 4.21 (0.04) 0.46 (0.50) 0.18
6±4 ÷20 6 23 44.25 (0.001) 0.77 (0.38) 40.28 (0.001) 0.14 (0.71) 0.55
6±5 ÷10 0 8 8.27 (0.001) 1.77 (0.19) 15.49 (0.001) 1.20 (0.28) 0.19
6±6 ÷6 0 9 4.37 (0.02) 0.32 (0.57) 4.41 (0.04) 4.68 (0.03) 0.16
6±7 ÷2 0 2 8.01 (0.001) 1.79 (0.19) 5.49 (0.02) 36.73 (0.001) 0.45
a
Each row shows the results of a separate analysis of variance. The ®rst through third columns report the medians for price
revision, pro®t variance, and price variance, respectively. The fourth through seventh columns present the F statistic, with its
signi®cance level in parentheses, for the combined e?ects of pro®t and price variances, the incremental e?ect of pro®t var-
iances after adjusting for price variances and markets, the incremental e?ect of price variances after adjusting for pro®t var-
iances and markets, and the main e?ect for markets, respectively. The last column reports the coecient of determination.
732 W.S. Waller et al. / Accounting, Organizations and Society 24 (1999) 717±739
(F = 1; 154:22; p < 0:001) and sets × markets
(F = 29:51; p < 0:001) further indicate the large
impact of demand conditions. The signi®cant e?ects
for periods (F = 53:69; p < 0:001), periods × sets
(F = 67:68; p < 0:001), and periods × sets × mar-
kets (F=8.35, p