Dr. Mirko Kremer ()
Assistant Professor of Supply Chain Management am Smeal College of Business, Pennsylvania State University.
Seine Forschung untersucht die verhaltenswissenschaftlichen Aspekte menschlichen Entscheidungsverhaltens
im Kontext des Supply Chain Management, 460 Business Building, State College, PA 16803, USA,
Prof. Dr. Stefan Minner ()
Inhaber des Lehrstuhls für Logistik und Supply Chain Management an der Fakultät für Wirtschaftswissenschaf-
ten der Universität Wien. Seine Forschungsschwerpunkte liegen in der Entwicklung und Analyse quantitativer
Optimierungsverfahren in der Logistik unter besonderer Berücksichtigung von Dynamik und Unsicherheit.
Anwendungen sind die Gestaltung logistischer Netzwerke, das servicegradorientierte Bestandsmanagement
und die Planung und Steuerung unternehmensübergreifender Lieferketten, University of Vienna, Brünner
Straße 72, 1210 Vienna, Austria, Email: email@example.com
ZfB-Special Issue 4/2008 83
The Human Element in Inventory Decision Making
under Uncertainty – A Review of Experimental Evidence
in the Newsvendor Model
Mirko Kremer, Stefan Minner
● Empirical evidence for the newsvendor problem shows that the normative solution
has limited predictive power as human decision makers tend to order quantities closer
to mean demand.
● We review the literature on experimental results and discuss potential behavioral
explanations for the observations.
● Based on these explanations we provide an overview on potential debiasing strategies.
Keywords Newsvendor Model · Human Experiments · Debiasing · Information ·
JEL: C91, D81, D83, M11
84 ZfB-Special Issue 4/2008
Model-based research has generated a tremendous body of literature contributing to our
understanding of how firm’s operations should be managed. When it comes to implemen-
tation, the success of theoretically supported operations tools and techniques depends
crucially on the descriptive accuracy of their assumptions on managerial behavior. Since
people are the common factor in real world operations processes, we need a better under-
standing of human behavior in order to improve these processes. Still, many models keep
sticking to the restrictive assumptions that people are 1) not a major factor in the phenom-
ena under study, 2) deterministic in their actions, 3) predictable in their actions, 4) inde-
pendent of others, 5) not part of the product, 6) emotionless and 7) observable (Boudreau,
A growing number of researchers have started revisiting these assumptions in order to
step towards a descriptively more accurate operations theory. Bendoly et al. (2006) pro-
vide a broad overview on the emerging field of Behavioral Operations Management.
Given the early stage of the field, it is yet too soon to recognize a coherent body of behav-
ioral operations theory. But robust behavioral pattern have started to emerge, especially
in two clusters of behavioral studies on concepts central to operations management theo-
ry. First, we note a swiftly increasing number of experimental studies revolving the bull-
whip effect, which is one of the theoretically most intensively studied phenomena in
supply chain management (Sterman, 1989; Croson and Donohue, 2003, 2005, 2006; Wu
and Katok, 2005). Secondly, we observe an increasing number of empirical studies on the
newsvendor model which is widely used to analyze and teach „optimal“ operations man-
agement under uncertainty, and is the focus of this review.
Model-based research on the newsvendor problem is overwhelming (for a review,
Khouja, 1999) and predominantly follows the normative principles of traditional micro-
economics. The applicability of the model is broad in the sense that its logic serves as a
building block for more complex models that seek to provide normative guidance as to
supply chain design, e.g. by contracts (Cachon, 2003), information systems (Chen, 2003),
or electronically enabled excess inventory markets (Lee and Whang, 2002). Clearly, a
thorough understanding of how real people tackle the problem can prove valuable for
improving teaching, guiding managerial practice as well as refining an important part of
In this paper, we review existing empirical evidence on actual newsvendor behavior
gathered through human experiments. The accumulated empirical evidence to date is
largely inconsistent with the normative benchmark commonly taught in the classroom
and used in academic research, and thus indicates that the model‘s simplicity does not
translate into an accurate description of real behavior.
We divide our review into two major building blocks. First, Section B. discusses the
underlying cognitive processes behind four interrelated decision biases which can account
for the mean ordering pattern observed in newsvendor experiments. Secondly, Section C.
explores the antecedents for these decision biases, and reviews debiasing strategies. The
explicit distinction between biases, antecedent sources of biases, and strategies to debias,
provides the reader with a guided access to the current state of empirical research on the
newsvendor problem. Section D. highlights the key insights gathered thus far, and dis-
ZfB-Special Issue 4/2008 85
cusses the merits, as well as limitations, of using laboratory experiments to advance the
behavioral aspects of operations management theory.
B. The Newsvendor – Normative Prescription, Empirical Regularities,
and Competing Explanations
I. The Benchmark
Consider a newsvendor buying q units of a good at a constant and known unit price c prior
to a selling season, earning a revenue p per unit sold. At the time of the ordering decision,
demand D is uncertain with a known distribution function Φ(D), inverse Φ-1, expected
value , and standard deviation . After realization of D=d the statewise profit is
(q,d)=p·min(q,d)-cq. The maximization of total expected profit from an order q
(1) ED[(q, D)] = ∫0
p – c
yields the optimal critical fractile solution q* = Φ–1冢9冣. For the remainder of this
paper, and following Schweitzer and Cachon (2000), products are labeled high profit
p – c 1
(HP) if 9 ≥ 3 and thus q* ≥ for symmetric demand distributions. They are labeled
p – c 1
low profit (LP) if 9 < 3 and thus q* < .
II. The Observation
Schweitzer and Cachon (2000) is the first study to test the newsvendor model’s empirical
validity in a controlled laboratory experiment. In their base experiment, subjects make 30
inventory decisions under a known uniform demand distribution. In 15 rounds the known
price p and unit cost c are set such that q* > (high profit), in the remaining 15 rounds
they entail q* < (low profit). The key observation is an order regression to the mean, i.
e. decision maker’s intuitively select order quantities that are too high for low profit
products and too low for high profit products, relative to the risk neutral benchmark q*.
This mean ordering behavior is a robust finding in a set of follow-up studies which we
review in the following (Bolton and Katok, 2007; Benzion et al., 2005; Lurie and Swami-
nathan, 2007; Kremer et al., 2007; Katok and Wu, 2007; Bostian et al., 2006; Thonemann
et al., 2007).
Schweitzer and Cachon (2000) refute various competing explanations for too-low-too-
high pattern based on the fact that most alternative objective functions imply a unidirec-
tional deviation from the expected profit maximizing benchmark q*. Of particular theo-
retical importance is the notion of risk aversion because it deviates from the risk-neutral
benchmark q* used in standard treatments of the problem, while not (necessarily) leaving
the normative accounts well-established by expected utility theory. Eeckhoudt et al.
86 ZfB-Special Issue 4/2008
(2004) analyze the general case of a utility-maximizing risk-averse newsvendor. Lau
(1980) analyzes the newsvendor problem with a mean variance criterion and Anvari
(1987) discusses inventory risk from a broader financial perspective using the Capital
Asset Pricing Model. Gotoh and Takano (2007) use conditional value at risk minimiza-
tion as optimization criterion. In managerial practice, budgets and performance targets
are more frequently used than optimization criteria in a mathematical sense. In the news-
vendor context a potential objective is to maximize the probability of exceeding a pre-
specified target profit level. Parlar and Weng (2003) consider an aspiration level criterion
using a moving target profit level depending on the order quantity. All these approaches
have in common that the prediction that the order quantity is lower than the risk-neutral
benchmark q*. Although risk-aversion can hardly be refuted normatively, it is thus de-
scriptively inaccurate in the context of the newsvendor problem.
III. Behavioral Explanations
In this section we present behavioral biases that can account for the observed decisions
and sketch their underlying psychological principles.
1. The Regretting Newsvendor
It is implicit in the standard formulation (1) that how the decision maker feels about an
order decision q in retrospect is independent from other options of the choice set available
at the time the decision was made. Rather, after a certain state of the world has material-
ized, many decision makers experience psychological sensations in the sense of “what
might have been” had one chosen differently (Loomes and Sugden, 1987; Loomes, 1988).
A newsvendor might experience regret from not having ordered realized demand D=d,
which is naturally the optimal order quantity ex-post. Anticipating potential disutility
from an ex-post inventory error |q-d|, the decision maker chooses an order quantity q that
maximizes total expected utility
(2) ED[u(q, D)] = ED[(q, D)] – ED[|q – D|)].
Schweitzer and Cachon (2000) show that the order quantity that maximizes (2) will
always be between q* and . The preference for minimizing ex-post inventory error thus
offers a viable regret-theoretic explanation for empirically observed newsvendor behav-
ior but it competes with different anchoring heuristics.
2. The Anchoring Newsvendor
When facing cognitively challenging problems, people’s decisions tend to be biased
towards salient anchor values suggested by the particular frame of the problem at hand
(Slovic and Lichtenstein, 1971; Kahneman and Tversky, 1974). The newsvendor problem
provides highly salient, anchorable information cues associated with the demand distribu-
ZfB-Special Issue 4/2008 87
Consider first the mean anchor heuristic which assumes that decision makers anchor
on mean demand and then insufficiently adjust towards the optimum, implying the same
too-low-too-high prediction as the ex-post inventory error minimization. With repeated
newsvendor decisions, the mean anchor heuristic predicts initial orders q0 to be close to
mean demand , followed by an insufficient convergence towards the optimum q*, for-
(3) qt = (t) + (1 – (t))q*
with ′(t) < 0 and 0 < (t) ≤ 1. Schweitzer and Cachon (2000) find that first round order
quantities are closer to mean demand than average order quantities across all rounds.
Similar findings are made in a number of follow-up studies (Benzion et al., 2005; Bolton
and Katok, 2007; Katok and Wu, 2007). Empirical evidence for the mean anchor heuris-
tic obviously exists, but it cannot fully explain why the adjustment process on the popula-
tion level remains strikingly insufficient.
3. The Chasing Newsvendor
Similar to the psychology of the mean anchor heuristic, decision makers might anchor on
prior order quantities and adjust towards prior demand realizations. Unlike the mean
anchor heuristic, the resulting chasing demand bias makes no formal claim regarding
the relationship between mean demand and an individual order decision qt in period t
(Schweitzer and Cachon, 2000). However, it does for the decision maker’s average order
quantity over N periods, ˉqN. Consider a simple model of the chasing demand heuristic
with the newsvendor adapting his previous order qt-1 towards the previous demand real-
ization dt-1 in order to choose his period t order quantity
(4) qt = qt–1 + (dt–1 – qt–1),
with 0 < ≤ 1. The average order quantity ˉqN then converges to mean demand as N grows
large (Kremer et al., 2007). The chasing demand heuristic can be viewed as a hybrid deci-
sion strategy. It encompasses a belief in positive correlation between independent demand
draws as well as a regret for past inventory errors. Learning about Dt-1 induces the experi-
ence of inventory errors |dt-1-qt-1|. Since past results cannot be changed in hindsight and
should not matter for future decisions, minimizing past regret from inventory errors by
adjusting the previous order qt-1 towards previous demand dt-1 is normatively incorrect.
However, the salience of the recent demand realization dt-1 fuels the psychology of regret
(Zeelenberg, 1999), especially since dt-1 minimizes experienced regret.
4. The Randomizing Newsvendor
The three decision strategies discussed above can be loosely classified as a preference for
regret avoidance (inventory error minimization), a judgment bias (mean anchoring), or
both (demand chasing). Unlike these strategies, a fourth potential explanation for the too-
low-too-high order pattern is rooted in a more general notion of bounded rationality.
88 ZfB-Special Issue 4/2008
Consider a decision maker who strives after the expected profit maximal solution but, due
to bounded rationality, considers all possible order quantities as candidates for selection
with better alternatives being chosen with larger probabilities. Su (2007) captures this
logic in a multinomial logit model where a decision is not deterministic but rather the
realization of a probabilistic choice reasoning. While q* remains the most likely decision,
Su (2007) shows that average choices converge towards mean demand. The intuition be-
hind this model-based result is that, loosely put, there is more room to err towards the
mean and beyond, than there is room to err away from it.
Su’s theoretical argument finds empirical support in Kremer et al. (2007). Their ex-
perimental study includes a “neutral frame” along with a standard representation of the
newsvendor problem, the only difference being that the standard frame relates the
displayed profit distributions to the underlying combinations of “order quantities” and
“demand realizations”, while the “neutral frame” only displays a set of abstract lotteries.
Interestingly, subjects exhibit mean ordering behavior even in the “neutral frame”. Clearly,
neither the notion of inventory error, mean demand anchoring, nor demand chasing can
account for this result, because the “neutral frame” simply does not offer the necessary
5. The Multi-Attribute Newsvendor
While most single-attribute objective functions predict unilateral deviations from the risk-
neutral newsvendor solution, it is easy to construct combinations of preferences that do
imply the observed too-low-too-high order pattern (Schweitzer and Cachon, 2000). For
example, Parlar and Weng (2003) and Jammernegg and Kischka (2007) consider combi-
nations of expected profit maximization and an aspiration level or a conditional value at
risk criterion. Such combinations of criteria mix the normative risk neutral benchmark
and mean demand, and thus directly imply mean ordering behavior. However, lacking
convincing empirical support so far, such multi-attribute objectives remain technical arti-
facts without much descriptive validity.
C. Mean Ordering: Antecedents and Debiasing Strategies
This section discusses antecedent conditions and moderating variables for the decision
biases discussed above and, as the flip-side of the same coin, possible strategies to debias
flawed newsvendor decision making.
I. Task Complexity
By structure, the newsvendor problem is cognitively cumbersome to process accurately,
since it offers the decision maker a large set of order quantities each of which technically
corresponds to a distribution of risky profits. Even though some order options are clearly
better than others, a change in expected profit from an incremental change in the order
quantity is not salient, especially around the profit optimal quantity q*. Bolton and Katok
(2007) ease the cognitive burden and make profit differences between available options
ZfB-Special Issue 4/2008 89
more salient to the decision makers. Specifically, they thin out the set of order options (from
100 to 3), expecting performance to improve with fewer options available to the decision
maker. Interestingly, this manipulation alone has no positive impact on performance.
Kremer et al. (2007) enlarge on the impact of task complexity. They narrow the cardi-
nality of both the order choice and the demand space down to 7, 5, and 3 options. The task
is presented to subjects by way of decision matrices which display profit information for
every combination of order quantity and demand. Revealed choices on the population
level can be conveniently reconciled with the predictions of expected utility theory, inde-
pendent from the level of complexity. However, results on the individual level show the
large extent to which the mean ordering bias is driven by task complexity, even when the
latter is reduced to a minimum.
II. Learning and Experience
It comes at little surprise that humans, when facing a complex problem, make biased deci-
sions initially. But there is ample evidence that people sometimes can “learn the optimum”
over time (Erev and Haruvy, 2008). The notion of learning translates naturally to the three
decision strategies presented in Sections B.III.1 through B.III.3. Concerning the demand
chasing strategy, this decision bias roughly classifies as “learning false lessons from the
past”. As to the minimization of expected ex-post inventory errors, we would not expect a
decision maker with this objective to arrive at the expected profit optimal quantity q* even
after unlimited learning experience, but rather learn towards the expected inventory error
minimum q=. By contrast, the mean anchor heuristic implies initial orders close to mean
demand and then predicts choices to converge towards the profit maximum q*. Interest-
ingly, Schweitzer and Cachon (2000) observe in their study that choices do not change
significantly over time at all. Essentially, the subjects fail to learn.
Bolton and Katok (2007) build on this issue by more explicitly investigating the role of
feedback, experience and learning. They significantly extend the number of decision
rounds. Regression based estimates of the learning coefficient (t) in (3) support the con-
tention that a population of decision makers learns to move slowly towards the optimum
q*. However, while slowly approaching the profit maximizing solution with sufficiently
many repetitions, the average ordering behavior remains largely consistent with the too-
high/too-low pattern. Similar findings are provided by the study of Benzion et al. (2005).
Bostian et al. (2006) explicitly model the notion of bounded rationality and learning in
the newsvendor problem. Specifically, they assume that decision makers cannot ad hoc
solve the problem accurately. Instead, decision makers try to learn over time, which is
potentially hindered by limited precision (as to the profit function of a given order quan-
tity) and limited memory (as to the amount of historic information incorporated into the
learning process). In order to capture the high degree of decision inertia revealed in their
experimental data, the authors incorporate a reinforcement learning element into their
model, allowing the decision maker to learn from both factual payoffs from the order
quantity and counterfactual payoffs from order quantities not chosen. The model calibra-
tion provides a good fit with their data from a standard newsvendor experiment.
Pulling together the results from the above studies, there is evidence of learning in the
newsvendor model. But it is weak and, moreover, not even robust across different frames
90 ZfB-Special Issue 4/2008
of the newsvendor problem (e.g. Katok and Wu, 2007). The inability to learn is likely to
worsen in real world situations with managers operating in highly unstable environments
where past lessons may provide little guidance for future decisions (Schweitzer and Ca-
chon, 2000). Rather than passively relying on the ability of decision makers to learn the
optimal solution, the persistent biases rather call for more active approaches.
The decision strategies leading to the mean ordering pattern require context-specific in-
formation “to work on”. Controlling the availability and presentation of this information
thus is a potential means to mitigate biased decision making.
The mean demand anchor
Kremer et al. (2007) control for the anchorable demand-related information provided by
the newsvendor problem and observe significantly more mean anchoring in the newsven-
dor representation compared to the choices in a “neutral frame” which only displays a set
of profit distributions without reference to the terms “demand” and “order quantity”.
While removing demand information lessens the tendency to anchor on mean demand,
this experimental manipulation is rather an academic exercise to illustrate the strength of
the mean demand anchor, but does not represent a very viable strategy for practice. It is
rather likely that, even with the ambiguous information often found in practice, decision
makers form beliefs about the most likely outcome, and then anchor on it. For example,
we know it is widespread planning practice to use best-case, average-case, and worst-case
scenarios in strategic decisions under uncertainty, an approach which carries the notion of
“mean”. Likewise, most of today’s ERP and demand planning systems provide highly
salient point forecasts when there is a demand history (Wagner, 2002).
Decision makers also respond to reference points other than mean demand. As an exam-
ple, Gavirneni and Xia (2007) provide participants in their experiments with information
cues which lack any relevance for the profit optimal solution of the problem, like e.g. the
order quantity of a hypothetical competitor. Interestingly, providing such immaterial
anchor information suffices to weaken the impact of the mean demand anchor. While the
anchors provided by Gavirneni and Xia (2007) do not translate into improved perform-
ance (since the anchor values never equal the optimal solution in this particular study), the
implied managerial implication is to carefully select and provide anchors that guide deci-
sions in the right direction.
Information on past demands
Over repeated decisions, the newsvendor problem provides plenty of ex-post information
like previous demand, inventory error, or profit realizations. While information economics
and adaptive learning theories strongly suggest that more information is strictly better
ZfB-Special Issue 4/2008 91
than less, experimental results illustrate that decision makers in the newsvendor problem
frequently convert available ex-post information into flawed subsequent decisions. Care-
fully controlling these ex-post information cues thus offers opportunities for debiasing
flawed newsvendor behavior.
Simply providing the decision maker with the most recent demand realization is barely
helpful, though. Past demand draws bear no information about the critical fractile solution,
but potentially facilitate simple comparisons between realized demand d and the chosen
order quantity q. Such comparisons shade the correct logic of leftover units being more
costly than unmet demand in low profit situations (and reversed in high profit situations)
and fuel symmetric disutility from inventory errors, |q-d|, underlying both the ex-post
inventory error minimization objective and the demand chasing heuristic. Furthermore,
providing feedback only on past demand might lead the decision maker to confuse the
inventory control problem with the task to correctly “guess demand”.
Information on foregone profits
Rather than letting past demand realization become a potentially misleading anchor point,
it seems more promising to provide feedback on foregone payoffs from order options not
chosen. Bolton and Katok (2007) investigate this conjecture but find that decision makers
fail to exploit information on foregone profits efficiently. One potential reason for this
result is that hindsight knowledge of the profit optimal order quantity, which is by defini-
tion realized demand d, might tempt the decision maker to adjust towards it, further rein-
forcing the demand chasing strategy.
Frequency of feedback
The question of “what” information to provide thus easily poses a dilemma. The decision
on “how often” to provide information seems somewhat simpler. Lurie and Swaminathan
(2007) investigate the impact of feedback frequency on newsvendor behavior. Contrary
to what a normative account would suggest, their results show that those who receive
more frequent feedback actually have a lower performance. The reason is that less fre-
quent feedback keeps subjects from reading too much into variability. The results further
suggest a diminishing effect of reducing feedback frequency on performance. In a differ-
ent treatment, Lurie and Swaminathan (2007) show that decision makers acquire more of
the available information (past orders, demand, and associated profits), when given less
frequent feedback. Bolton and Katok (2007) constrain subjects to ordering the same
quantity for a sequence of 10 periods. The reduced feedback variability of this “standing
order” constraint finally drives order quantities significantly closer to the optimum, com-
pared to the base case where feedback is given period-by-period.
Overall, too frequent information is not helpful in the newsvendor problem and can
even degrade performance, contrary to what common managerial instinct as well as deci-
sion theory would suggest. In particular, less frequent feedback has proven to result in
less demand chasing: Indeed, if learning about inventory errors reinforces regret and sub-
sequently the chasing demand heuristic, it seems less surprising that too frequent feed-
back degrades performance (Bolton and Katok, 2007; Lurie and Swaminathan, 2007).
92 ZfB-Special Issue 4/2008
In the following we discuss incentive-driven debiasing strategies proposed or explored in
Change the critical ratio
Although they exhibit the too-high-too-low ordering pattern even with extended experi-
ence, decision makers in the newsvendor problem respond to incentives in a qualitatively
correct manner, i.e. they order more than mean demand of a high profit product and less
than mean demand of a low profit product. Katok and Wu (2007) investigate order deci-
sions in the standard newsvendor setting and contrast them with behavior under alterna-
tive risk-sharing contracts which theoretically should induce larger order quantities.
While the authors observe additional behavioral biases under the different contracts
studied, subjects indeed order more when they should.
This suggests that a stockout penalty or a subsidy for leftover inventory could be im-
posed in order to correct the mean ordering behavior for a high profit product, whereas an
excess inventory penalty would work in a low profit environment. The efficacy of this
approach is questionable from a practical perspective, though. First, it requires the firm to
know the optimal quantity q*, but then the firm could just implement q* instead of dele-
gating the order decision. Second, order behavior is widely heterogeneous in a population
of decision makers, and one incentive scheme is unlikely to fit them all. Third, correcting
behavior by monetary incentives might not even add to the firm’s bottom line when coor-
dinating bonus payments to inventory managers exceed the profit increase from ordering
towards q*. Lastly, monetary penalties and subsidies for leftovers and stockouts help little
in mitigating the chasing demand heuristic.
Impose costs of change
Since frequent changes of order quantities are ultimately detrimental in a stationary news-
vendor task, Lurie and Swaminathan (2007) investigate whether introducing costs for
changing order quantities can mitigate decision makers’ propensity to respond to random-
ness too heavily. Surprisingly, the authors find in their study that costs of change does not
improve performance, suggesting that decision makers respond to demand fluctuations
even when it is costly. However, introducing artificial costs of change is potentially dan-
gerous in many real settings anyway because it sets the wrong incentive when demand is
non-stationary and optimal inventory control obviously requires adjustments of order
Mitigate inventory error regret
Besides a fallacious belief in positively correlated demand, the demand chasing strategy
has a regret component. Kremer et al. (2007) investigate how the heuristic is moderated
by the costs of ex-post inventory errors |q-d|. The authors find an increased propensity to
chase demand among those subjects that are penalized for ex-post inventory errors the
ZfB-Special Issue 4/2008 93
most. In order to mitigate the detrimental impact of the chasing demand heuristic, it
seems good managerial advice to attenuate intra-firm incentives that foster the psychol-
ogy of regret from inventory errors. Simply penalizing operational decisions for being
wrong is clearly wrong, because it easily provides decision makers on the operational
level with good reasons to follow suboptimal but regret minimizing strategies like the
preference for minimizing ex-post inventory errors and the demand chasing heuristic.
This paper reviews the behavioral insights from recent empirical tests of the newsvendor
model, which has accumulated evidence that intuitive newsvendor behavior systemati-
cally deviates from normative prescriptions. Since decision making under uncertainty is
inherently difficult and the decision literature has documented a large variety of behavioral
deviations from normative accounts (Kahneman and Tversky, 2000), the mere existence
of decision biases in the newsvendor problem is of course not surprising per se. What
makes it interesting is the fact that the observed (mis)behavior is systematic, and guided
by aspects that are rather unique to the newsvendor context, requiring new approaches to
While existing behavioral theory can help to explain anomalies, most results to date
show that it is unlikely to translate to the operations domain in a simple way, if at all. For
example, Schweitzer and Cachon (2000) show that the well-established prospect theory
(Kahneman and Tversky, 1979) does not predict the observed mean ordering pattern. A
recent study by Schultz et al. (2007) further illustrates this point. They present the news-
vendor problem first in a gain frame and, to a different group of subjects, in a loss frame.
By arguing along the value function of prospect theory (risk aversion in the gain frame,
risk seeking in the loss frame) the authors expect smaller order quantities in the gain
frame and larger orders in the loss frame, as an analogy to the risk-reflection effect ob-
served in numerous choice problems outside the business context (Kuhberger, 1998). But
their data reveals no statistically significant order behavior between the two frames.
Therefore, the application of behavioral theories requires some care checking their con-
text-sensitivity in the operations management domain.
The results from experimental evidence on human behavior in the newsvendor prob-
lem have a common denominator: context matters. A second key observation concerns
heterogeneity of behavior. While much of the evidence on biased newsvendor decision
making has been provided on the population level, a breakdown to the individual level in
Bolton and Katok (2007) and Kremer et al. (2007) shows that behavior in a population of
decision makers is rather heterogeneous. When it comes to the provision of incentives and
information, knowledge of a populations’ average behavior might be insufficient, just like
knowledge of mean demand is insufficient for making optimal inventory decisions. When
multiple decision makers need to be debiased, a firm might design individual incentive
schemes tailored to each individual decision maker or a single incentive scheme that is in
some sense robust against behavioral imperfections. The behavioral heterogeneity ob-
served in newsvendor experiments thus entails interesting challenges for future modeling
work on mechanism design.
94 ZfB-Special Issue 4/2008
In the above light, human experiments and mathematical models can jointly advance
operations theory. But they encounter a common criticism, namely their potentially
limited relevance for managerial decision making in the field. Complaints have long ac-
cumulated that formal operations models and techniques often have an unsatisfactory
impact in practice (Corbett and Van Wassenhove, 1993). Since disregard of human
element is one potential reason for this explanatory gap, experimental research is one
vehicle to bridge it. Still, the value of the experimental method for providing insights
beyond mere laboratory artifacts remains yet to be proven. For example, it is sometimes
argued that the use of student samples lacks external validity (Bendoly et al., 2006). To
date, only few experimental studies provide leads concerning a possible student sample
bias in the operations domain. Thonemann et al. (2007) compare newsvendor behavior
of students and procurement professionals, with and without upfront training. The key
result is that managers overall perform worse than student subjects, although not statisti-
cally significant. The performance gap even increases when subjects received an ex-
tensive lecture on the newsvendor problem prior to the experiment, indicating that
managers are less susceptible to instructive learning. Such evidence from non-student
samples as well as careful extrapolation from the lab to the real world is clearly valuable.
Ultimately, we need to test behavioral operations theory in the field, which requires
methodological triangulation (see Corbett and Fransoo (2007) for a survey-based exam-
On a final note, and further speaking to the external validity of human experiments,
Behavioral Operations Management research needs to carefully address the unit of analy-
sis issue. For example, while most formal supply chain modeling approaches set the unit
of analysis to the level of “the firm”, in practice important decisions with a newsvendor
structure are often made in groups of individuals (e.g. Fisher et al., 1994). In a recent
study, Gavirneni and Xia (2007) contrast individual decision making with group decision
making. Surprisingly, group decisions are dispersed wider, contrary to the intuition that
group dynamics would tend to make individual preferences converge. Enlarging on multi-
person aspects of operations management appears to be one of the most promising ave-
nues for future empirical research.
Anvari, M. (1987). Optimality criteria and risk in inventory models: The case of the newsboy problem. Journal
of the Operational Research Society 38, 625–632.
Bendoly, A., K. Donohue, and K. Schultz (2006). Behavior in operations management: assessing recent findings
and revisiting old assumptions. Journal of Operations Management 24 (6), 747–752.
Benzion, U., Y. Cohen, R. Peled, and T. Shavit (2005). Decision-making and the newsvendor problem – an
experimental study. Working paper, Ben-Gurion University.
Bolton, G. E. and E. Katok (2007). Learning-by-doing in the newsvendor problem: A laboratory investigation
of the role of the experience. Manufacturing and Service Operations Management, forthcoming.
Bostian, A., C. Holt, and A. Smith (2006). The newsvendor “pull-to-center effect:” adaptive learning in a labora-
tory experiment. Working Paper, University of Virginia.
Boudreau, J. (2003). Commissioned paper on the interface between operations and human resources manage-
ment. Manufacturing and Service Operations Management 5 (3), 179–202.
Brown, A. O. and C. S. Tang (2006). The impact of alternative performance measures on single-period inven-
tory policy. Journal of Industrial and Management Optimization 2 (3), 297–318.
ZfB-Special Issue 4/2008 95
Cachon, G. (2003). Supply chain coordination with contracts. In S. C. Graves and A. De Kok (Eds.), Supply
Chain Management: Design, Coordination and Operation, Volume 11 of Handbooks in Operations Re-
search and Management Science, pp. 229–240. Amsterdam: Elsevier.
Chen, F. (2003). Information sharing and supply chain coordination. In S. C. Graves and A. De Kok (Eds.),
Supply chain Management: Design, Coordination and Operation, Volume 11 of Handbooks in Operations
Research and Management Science, pp. 341–421. Amsterdam: Elsevier.
Corbett, C. J. and J. C. Fransoo (2007). Entrepreneurs and newsvendors: Do small businesses follow the news-
vendor logic when making inventory decisions? Working paper, UCLA.
Corbett, C. J. and L. N. Van Wassenhove (1993). The natural drift: What happened to operations research?
Operations Research 41, 625–640.
Croson, R. and K. Donohue (2003). Impact of pos data sharing on supply chain management: An experimental
study. Production and Operations Management 12, 1–11.
Croson, R. and K. Donohue (2005). Upstream versus downstream information and its impact on the bullwhip
effect. System Dynamics Review 21, 249–260.
Croson, R. and K. Donohue (2006). Behavioral causes of the bullwhip effect and the observed value of inven-
tory information. Management Science 52 (3), 323–336.
Eeckhoudt, L., C. Gollier, and H. Schlesinger (2004). The risk-averse (and prudent) newsboy. Management
Science 41 (5), 786–794.
Erev, I. and E. Haruvy (2008). Learning and the economics of small decisions. Working paper Columbia Uni-
Fisher, M. L., J. H. Hammond, W. R. Obermeyer, and A. Raman (1994). Making supply meet demand in an
uncertain world. Harvard Business Review 72 (3), 83–93.
Gavirneni, S. and Y. Xia (2007). Anchor selection and group dynamics in newsvendor decision making: Results
from an experimental study. Working paper, Cornell University.
Gotoh, J.-Y. and Y. Takano (2007). Newsvendor solutions via conditional value-at-risk minimization. European
Journal of Operational Research 179, 80–96.
Jammernegg, W. and P. Kischka (2007). Risk-averse and risk-taking newsvendors: A conditional expected value
approach. Review of Managerial Science 1 (1), 93–110.
Kahneman, D. and A. Tversky (1974). Judgment under uncertainty: Heuristics and biases. Science 185, 1124–
Kahneman, D. and A. Tversky (1979). Prospect theory: An analysis of decision under risk. Econometrica 47,
Kahneman, D. and A. Tversky (2000). Choices, values, and frames. Cambridge University Press.
Katok, E. and D. Wu (2007). Contracting in supply chains: A laboratory investigation. Working paper, Penn
Khouja, B. (1999). The single period (newsvendor) problem: Literature review and suggestions for future
research. OMEGA 27, 537–553.
Kremer, M., S. Minner, and L. N. Van Wassenhove (2007). Anchoring and regret in the newsvendor problem –
the impact of task complexity and framing. Working paper, University of Mannheim.
Kuhberger, A. (1998). The influence of framing on risky decisions: A meta-analysis. Organizational Behavior
and Human Decision Processes 75 (1), 23–55.
Lau, H.-S. (1980). The newsboy problem under alternative optimization objectives. The Journal of the Opera-
tional Research Society 31 (6), 525–535.
Lee, H. and S. Whang (2002). The impact of the secondary market on the supply chain. Management Science
48 (6), 719–731.
Loomes, G. (1988). Further evidence of the impact of regret and disappointment in choice under uncertainty.
Economica 55, 47–62.
Loomes, G. and R. Sugden (1987). Testing for regret and disappointment in choice under uncertainty. The
Economic Journal 97, 118–129.
Lurie, N. H. and J. M. Swaminathan (2007). Is timely information always better? The effect of feedback
frequency on performance and knowledge acquisition. Working paper, Georgia Institute of Technology.
Parlar, M. and Z.Weng (2003). Balancing desirable but conflicting objectives in the newsvendor problem. IIE
Transactions 35 (2), 131–142.
Schultz, K. L., J. O. McClain, L. W. Robinson, and J. Thomas (2007). The use of framing in inventory decisions.
Working paper, Cornell University.
Schweitzer, M. E. and G. P. Cachon (2000). Decision bias in the newsvendor problem with a known demand
distribution: Experimental evidence. Management Science 46 (3), 404–420.
96 ZfB-Special Issue 4/2008
Slovic, P. and S. Lichtenstein (1971). Comparison of Bayesian and regression approaches to the study of infor-
mation processing under uncertainty. Organizational Behavior and Human Performance 6, 649–744.
Sterman, J. (1989). Modeling managerial misbehavior: Misperceptions of feedback in a dynamic decision
making experiment. Management Science 35 (3), 321–339.
Su, X. (2007). Bounded rationality in newsvendor models. Manufacturing and Service Operations Manage-
Thonemann, U., G. Bolton, and A. Ockenfels (2007). Experience and information in the newsvendor problem.
Working paper, University of Cologne.
Wagner, H. M. (2002). And then there were none. Operations Research 50 (1), 217–226.
Wu, D. and E. Katok (2005). Learning, communication, and the bullwhip effect. Journal of Operations Manage-
ment 24 (6), 839–850.
Zeelenberg, M. (1999). Anticipated regret, expected feedback and behavioral decision making. Journal of
Behavioral Decision Making 12, 93–106.
ZfB-Special Issue 4/2008 97
The Human Element in Inventory Decision Making under Uncertainty –
A Review of Experimental Evidence in the Newsvendor Model
It is a long-standing concern that formal models and techniques in operations often have
an unsatisfactory impact in practice, and sometimes lack descriptive power due to unreal-
istic assumptions. Since disregard of human element is one potential reason for this ex-
planatory gap, experiments on human behavior are one promising way to bridge theory
We review recent empirical evidence on human behavior in the newsvendor model
which is one of the centerpieces of inventory theory. The robust finding from all studies
is that decision makers systematically order closer to mean demand, relative to the bench-
mark of a risk-neutral decision maker. We elaborate on the psychology underlying inter-
twined decision strategies that imply the observed mean ordering behavior, and discuss
their implications for debiasing.
Der Faktor Mensch im Bestandsmanagement unter Unsicherheit –
Ein Überblick experimenteller Ergebnisse zum Zeitungsverkäuferproblem
Ein häufiger Vorbehalt gegen formale Modelle und Lösungsmethoden in Produktion und
Logistik ist die unbefriedigende praktische Nutzung sowie der teilweise nicht vorhandene
Erklärungsgehalt. Unrealistische Modellannahmen und die Vernachlässigung mensch-
lichen Entscheidungsverhaltens sind mögliche Ursachen für die zu beobachtende Erklä-
rungslücke zwischen Theorie und Praxis, zu deren Schließung kontrollierte Laborexperi-
mente ein vielversprechendes Instrument darstellen können.
Dieser Beitrag gibt einen Überblick zu empirischen Studien menschlichen Entschei-
dungsverhaltens im Zeitungsjungenproblem. In allen Studien zeigt sich eine systemati-
sche Abweichung des beobachteten Bestellverhaltens vom normativen, unter Risiko-
neutralität ermittelten, Benchmarks in Richtung des Erwartungswertes der Nachfrage.
Der Beitrag stellt psychologische Erklärungsansätze für das beobachtete Verhalten vor
und diskutiert Implikationen und Strategien zur Verringerung des Problems.