Professional Documents
Culture Documents
net/publication/274956630
CITATIONS READS
33 3,167
3 authors, including:
All content following this page was uploaded by Marcus Cunha on 14 April 2015.
Marketing Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
. http://dx.doi.org/10.1287/mksc.2015.0906
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Articles in Advance, pp. 1–16
ISSN 0732-2399 (print) ISSN 1526-548X (online) http://dx.doi.org/10.1287/mksc.2015.0906
© 2015 INFORMS
Jeffrey D. Shulman
Michael G. Foster School of Business, University of Washington, Seattle, Washington 98195, jshulman@uw.edu
T his research examines how prepurchase information that reduces consumer uncertainty about a product or
service can affect consumer decisions to reverse an initial product purchase or service enrollment decision.
One belief commonly held by retailers is that provision of greater amounts of information before the purchase
reduces decision reversals. We provide theory and evidence showing conditions under which uncertainty-
reducing information provided before the purchase decision can actually increase the number of decision rever-
sals. Predictions generated from an analytical model of consumer behavior incorporating behavioral theory of
reference-dependence are complemented by empirical evidence from both a controlled behavioral experiment
and econometric analysis of archival data. Combined, the theory and evidence suggest that managers should be
aware that their information provision decisions taken to reduce decision reversals may actually increase them.
Data, as supplemental material, are available at http://dx.doi.org/10.1287/mksc.2015.0906.
Keywords: decision reversal; product returns; service cancellations; reference-dependence
History: Received: June 3, 2013; accepted: October 26, 2014; Preyas Desai served as the editor-in-chief and
Dmitri Kuksov served as associate editor for this article. Published online in Articles in Advance.
physical dimensions and observe how clothing fits on underutilization of capacity and salvage costs associ-
a virtual model (Boone and Kurtz 2007). ated with refund policies. Prior literature has exam-
In each of these cases, company information provi- ined the optimal refund policy in addressing the con-
sion actions are aimed at resolving a level of uncer- sumer’s decision reversal. There is extensive research
tainty surrounding the utility a consumer will derive on the optimal refund policy in dealing with oppor-
from owning the product (e.g., after receiving the tunistic consumers who purchase the product with
information, the consumer learns that the camera the intent to return it after “free renting" (e.g., Davis
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
takes good quality indoor pictures or that the frame et al. 1995, 1998; Chu et al. 1998; Hess et al. 1996). Xie
of the eyeglasses will nicely complement her facial and Gerstner (2007) and Guo (2009) establish equilib-
features). However, even with the growing amount rium refund policies for service providers. Che (1996)
of upfront information retailers provide, often con- and Shulman et al. (2009, 2010, 2011) find the profit
sumers only fully ascertain the actual utility of a prod- maximizing partial refund for product returns when
uct after purchase. For example, consumers may not consumers are uncertain about the product fit with
determine the actual fit of the clothes (or eyeglasses), preferences before purchase.
the texture of the materials, or the hue of the color A body of research looks at the effect of con-
until they receive the item at home even if a website sumer uncertainty about product fit on firm decisions
can offer a better approximation of certain features via (e.g., Iyer and Kuksov 2012, Gu and Liu 2013). Pre-
the use of a computer model. Thus, whereas prepur- vious research has examined the effect of full uncer-
chase information provided by retailers may reduce tainty resolution, before the consumer decision, on
consumer uncertainty, actual utility is often not fully firm profits. Shulman et al. (2009) find that informa-
revealed until after purchase. tion that eliminates returns can also reduce profit. Xie
We contribute to the literature on decision rever- and Shugan (2001) and Shugan and Xie (2005) show
sals by developing an analytical model that uniquely how advanced selling when consumers lack informa-
incorporates behavioral theories to predict the effect tion about fit with preferences can be more profitable
of a firm providing truthful, uncertainty-reducing than spot selling at a time consumers know the fit
information on consumers’ decisions to reverse a pur- with preferences. Kuksov and Lin (2010) and Gu and
chase. By contrast to previous analytical models that Xie (2013) find the equilibrium information revela-
predict that decision reversals will be reduced as con- tion decisions by competing high- and low-quality
sumer uncertainty is eliminated (e.g., Anderson et al. firms. Ofek et al. (2011) examine how the online chan-
2009, Shulman et al. 2009), we draw on behavioral nel affects the level of in-store assistance provided by
theory of reference-dependence to show that there brick-and-mortar retailers. Whereas information that
may be instances in which there is an increase in deci- reduces product returns has been shown to reduce
sion reversals when consumer uncertainty is reduced profit via its effect on pricing, we examine whether
yet not fully eliminated before the purchase. To our information can actually increase product returns.
knowledge, this research is the first to incorporate This research focuses on the consumer decision rather
behavioral theory of information processing into an than the firm decision.
analytical model of decision reversals. In addition to Although a large body of research examines prod-
the analytical model, we provide empirical evidence uct returns from a theoretical perspective, there is
both from experimental and archival data that sup- growing interest from empirical researchers to exam-
port the novel prediction that uncertainty-reducing ine this marketing problem. For instance, Anderson
information can increase the likelihood of decision et al. (2009) use a structural econometric model to esti-
reversals. The model and data suggest that informa- mate the utility a consumer derives from having the
tion provided to reduce decision reversals may actu- option to reverse the purchase decision. In a behav-
ally increase them. ioral study, Wood (2001) shows that lenient return
The results imply that firms should more care- policies lead to more positive ratings of product qual-
fully evaluate their information provision strategies. ity and an increase in the number of people who
Whereas information that can fully resolve uncer- order and ultimately keep a purchase. Bechwati and
tainty can reduce decision reversals (e.g., product Siegal (2005) find that simultaneous versus sequential
returns or service cancellations), information that only presentation of products affects product return likeli-
partially reduces uncertainty can lead to an increase hood by generating comparative versus noncompara-
in decision reversals. We identify conditions that will tive thoughts.
lead to this result. Our paper contributes to the literature on deci-
sion reversals by studying consumers’ decision rever-
2. Literature Review sal when the initial decision is made with uncer-
Decision reversals can have important implications tainty about the utility that will be derived from using
for firms’ profitability owing to their impact on the product or service. A decision reversal can lead
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS 3
to service cancellation or product returns. Product (Kahneman and Tversky 1979), consumer judgments
returns made for reasons other than uncertainty res- are sensitive to a point of reference. Reference-
olution such as intrachannel returns between chan- dependent models of information processing (e.g., Ho
nel partners (e.g., Cachon 2003, Gumus et al. 2013), et al. 2006, Lattin and Bucklin 1989) have been shown
durable good buy-backs (e.g., Desai and Purohit 1998, to successfully account for (1) how consumers eval-
Desai et al. 2004, Bruce et al. 2006, Shulman and uate discounts in product bundles (Janiszewski and
Coughlan 2007, Yin et al. 2010), and product-failure Cunha Jr 2004), (2) why consumers prefer discounts
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
(e.g., Moorthy and Srinivasan 1995, Balachander 2001, on a product whose actual price deviates most from
Ferguson et al. 2006) are outside the scope of this the point of reference (Saini et al. 2010), (3) how
research. consumers judge changes in brand attributes rela-
Our research also adds to the growing body of tive to initial multiattribute reference points (Hardie
multimethod research that develops analytical models et al. 1993), and (4) which brands will be preferred in
incorporating behavioral theory and empirically tests sequential purchases by consumers who are new to a
the model predictions (e.g., Amaldoss and Jain 2005, market (Heilman et al. 2000).
2010; Cui and Mallucci 2013). Specifically, we incor-
porate the behavioral theory of reference-dependence 3.1. Model Description
to the analytical model, a theory that has been shown Let Ui = i + i + i denote the actual utility of
to be successful in analyzing blocks in a price con- ownership where i is known before purchase, i is
tract (Lim and Ho 2007), framing a fixed fee (Ho and unknown at the purchase decision, and i is the part
Zhang 2008), the labor supply (Farber 2008), prod- of utility that becomes known with information provi-
uct line design (Orhun 2009), newsvendor models sion and is otherwise unknown at the purchase deci-
(Ho et al. 2010), and innovation strategy (Narasimhan sion. In the absence of information, consumers have
and Turut 2013, Chen and Turut 2013). This paper rational beliefs that i follows a distribution with
provides further evidence that modeling reference- mean a, variance a2 , and probability density func-
dependency can have a substantive impact on the tion f 4i 5. Consumers also have rational beliefs that
effectiveness of the marketing mix elements. i ∼ U 601 17. The uniform distribution of i allows
To summarize the contribution of this research rel- for parsimonious closed-form solutions as to return-
ative to the existing literature, our model is the first probability for a given consumer. The insights of the
to incorporate the behavioral theory of reference- model are robust to the assumptions about the distri-
dependence in examining the effect on consumer bution of i .
decision reversals of prepurchase information that We operationalize reference-dependence by allow-
reduces, but does not fully eliminate, consumer uncer- ing perceived utility to be a function of prepurchase
tainty. The novel prediction resulting from the ana- expectations. This is consistent with the formulation
lytical model is that such information may lead to in Kőszegi and Rabin (2006), which is based on empir-
an increase in the number of decision reversals. This ical evidence in Mellers et al. (1999) and Breiter et al.
prediction is empirically supported by a controlled (2001). Let UiE denote the ex post perceived utility and
experiment and empirical analysis of archival data of E6Ui 7 denote the ex ante expected utility of ownership
decision reversals. (or service use) in the first-stage decision. Ex post per-
ceived utility can be written as
3. Analytical Model of (
Ui + L 4Ui − E6Ui 75 if Ui < E6Ui 7
E
Decision Reversals Ui = (1)
In this section, we build an analytical model of con- Ui + G 4Ui − E6Ui 75 if Ui > E6Ui 71
sumer behavior to predict how uncertainty-reducing
where L is a measure of reference-dependence in
information affects consumer decision-making as to
losses and G is the equivalent measure in gains.1
the purchase and return of a single product (or enroll-
We assume that the second stage decision is based
ment and cancellation of a single service). The model
on perceived rather than actual utility. The second
is consistent with prior analytical work in product
stage decision rule is described in the following
returns and service cancellations in many respects.
assumption.
Specifically, we model a two-stage decision process.
In the first stage, consumers decide whether to pur-
1
chase the offering. In the second stage, consumers Anderson and Sullivan (1993) also start with a model in which
decide whether to reverse this purchase decision. gains or losses associated with perceived utility relative to expected
utility affect outcomes. However, their model of expectation dis-
As mentioned above, the notable exception is that confirmation involves hypothesis testing, which is affected by the
our model incorporates the behavioral concept of variance of expectations. We thank the associate editor for raising
reference-dependence. According to prospect theory this point.
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
4 Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS
Assumption. A consumer who has made a purchase indifferent between keeping and returning the pur-
will choose to reverse the purchase decision if and only if chased item, we must consider whether this marginal
perceived utility is less than the anticipated refund. consumer experiences a loss or gain relative to the
expected utility of ownership. The former is true
This assumption is consistent with behavioral find-
if i > P − a − 1/2; otherwise, the latter is true.
ings that perceived quality depends on expectations
Collecting terms and integrating over possible val-
rather than on performance alone (Hoch and Ha
ues of i , the probability of a return for consumer i
1986). Whereas previous literature treats the return
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
to the low-information condition. By logic similar to Figure 1 Return Probability for Intermediate Known Value
the low-information condition, it is straightforward to P – a – 1/2 > i > P – max{i} – 1/2
i
show that EHi Info 6purchase7 > 0 if and only if i + i +
i
Return probability
Intuitively, the lower bound implies that the known
component of utility is high enough such that there –1
is at least one consumer who will keep the pur- – 1/(1+ L)
chased item even though its actual utility is less than
Increase in
its expected utility in the high-information condi- returns
tion. The upper bound implies consumers in the low-
information condition only keep the purchased item
if it represents a gain relative to expected utility. For
a given i > P − i − 1/2, the effect of information on
return probability for any i can be written as
Figure 2 Return Probability for High Known Value low-information conditions: P − i − i . By contrast
i > P – min{i} – 1/2 to Figure 2, we see that for high i the increase in
i return probability for high i is offset by a decrease in
the return probability for low i , whereas for low i ,
the return probability is the same in each information
condition.
High info
Proposition 1. In the basic model in which all con-
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
Low info
sumers buy initially regardless of information condition,
Decrease
the effect of the known utility component on the relation-
Return probability
in returns
ship between information and returns is nonmonotonic.
Specifically, prepurchase uncertainty-reducing information
Increase in increases returns if P − mini 8i 9 − 1/2 > i > P −
returns maxi 8i 9 − 1/2 and has no effect on returns if i >
P − mini 8i 9 − 1/2 or if i < P − maxi 8i 9 − 1/2. The
effect of information on returns is positive only if L > 0.
See the appendix for the proof.
Proposition 1, to our knowledge, is the first in
the literature to establish that prepurchase informa-
tion can increase returns. The intuition for the result
a Ψi is as follows. Information will reveal that, for some
consumers, a utility component is high. This raises
consumers with below average i , regardless of the the expected utility of ownership and thus increases
loss-aversion parameter. the probability that these consumers experience a
Next consider i < P − max8i 9 − 1/2. This condi- loss relative to ex ante expectations. With loss aver-
tion represents the case wherein the known compo- sion, ex post utility is diminished and thus returns
nent utility is low enough such that the only way a are increased. Note that the effect of information
consumer keeps the product, regardless of the infor- on returns is positive only if L > 0. Moreover, the
mation, is if she experiences a gain relative to the increase in returns is an increasing function of L . This
expected utility. The return probability is equal across effect outweighs the effect on returns from consumers
who learn that a utility component is low because
conditions.
losses loom larger than gains.
Figure 3 demonstrates that when the known util-
We now briefly consider how the results are
ity component is sufficiently low, the return probabil-
affected by a partial return policy. If refund R is less
ity as a function of i is the same for the high- and
than price P , then PriLo Info 4Return i 1 purchase5 =
max842R − 2i + L 41 + 2a55/42 + 2L 5 − i 1 09 if
Figure 3 Return Probability for Low Known Value i > R − a − 1/2 and PriLo Info 4Return i 1 purchase5 =
i < P – max{i} – 1/2 max8R − i − i 1 09 otherwise. In the high-information
i condition PriHi Info 4Return i 1 purchase5 = max842R −
2i + L − 2i 5/42 + 2L 51 09 if i > R − i −
1/2 and PriHi Info 4Return i 1 purchase5 = max8R −
i − i 1 09. Note that in each information condition,
Pri 4Return i 1 purchase5 decreases as R decreases.
High info Thus, given conditions such that all consumers ini-
Low info
tially buy the product, a partial refund results in
Return probability
to make a purchase depending on the realization of Now suppose P − maxi 8i 9 − 1/2 < i < min8P −
i . Because of the infinitely small hassle cost of pur- a − 1/21 P − 1 − mini 8i 99. In the low-information
chase, a consumer will make a purchase if and only condition, the PriLo Info 4Return i 1 purchase5 = 1 if
if the expected utility of purchase is strictly positive. i < P − i − 1 and PriLo Info 4Return i 1 purchase5 =
Suppose P − 1 − maxi 8i 9 < i < P − 1 − min8i 9. As max8P − i − i 1 09 if i > P − i − 1. In the high-
shown above, the lower bound implies that all con- information condition, consumers with i < P − i − 1
sumers buy in stage 1 in the low-information con- will not purchase in stage 1. Consumers with P −
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
dition. However, the upper bound implies that not i − 1/2 > i > P − i − 1 have PriHi Info 4Return i 1
all consumers will buy in the high-information condi- purchase5 = max8P − i − i 1 09. Consumers with i >
tion. We first analyze returns in the low-information P − i − 1/2 have PriHi Info 4Return i 1 purchase5 =
condition. max842P − 2i + L − 2i1 5/42 + 2L 51 09. The effect of
For a given i , realized after purchase, we use information on returns is shown in Figure 5.
the established logic to identify PriLo Info 4Return i 1 Thus for a given i , information R P −1−i results in a
purchase5. If P − 1 − maxi 8i 9 < i < P − maxi 8i 9 − change
R in returns equal to − −
f 4i 5 di + 0 +
1/2, then the marginal consumer who is indiffer- P −i −1/2
4 L /41 + L 554 i − P + i + 1/25f 4i 5 di , where
ent between keeping and returning will experience a the first term is negative and the final term is posi-
gain relative to expectations. In the low-information tive. Note that this expression is increasing in i . This
condition, consumers with i < P − 1 − i have leads to the following proposition.
PriLo Info 4Return i 1 purchase5 = 1 and consumers with
Proposition 2. There exists ¯ such that information
i > P − 1 − i have PriLo Info 4Return i 1 purchase5 =
will weakly decrease returns if i < .
¯
P − i − i . All consumers in the low-information
condition buy because i is unknown at the time of Proposition 2 shows the conditions under which the
purchase and P − 1 − maxi 8i 9 < i ensures a posi- standard result occurs, i.e., that information reduces
tive probability of experiencing i + i + i > P . How- returns. If the consumer’s known utility component is
ever, in the high-information condition, consumers sufficiently low, then some consumers will find that,
are aware of i before purchase. Consumers with i < with additional information, a purchase would result
P − 1 − i , find zero probability of i + i1 + i > P in a return. As such, both purchases and returns are
and thus do not buy in stage 1. Consumers with decreased by uncertainty-reducing information. This
i > P − 1 − i have PriHi Info 4Return i 1 purchase5 = proposition shows that the results of prior literature
PriLo Info 4Return i 1 purchase5. Thus for a given i , can be nested within our modeling framework.
information results in a change in returns equal to In §3.2, we summarize the conditions that lead to
R R−1−
− − i f 4i 5 di < 0. Figure 4 shows the effect of our novel result (Proposition 1) versus the conditions
information on returns. that lead to the standard result (Proposition 2).
Figure 4 Return Probability for Low Known Value When Purchase Figure 5 Return Probability for Intermediate Known Value When
Depends on Information Purchase Depends on Information
P – 1 – max{i} < i < P – max{i} – 1/2 P – 1 – max{i} < i < P – max{i} – 1/2
i i i i
High info
High info Low info
Low info
Return probability
Return probability
Decrease in Increase in
Decrease in
returns returns
returns
P – 1– i Ψi P – 1– i P – 1/2 – i Ψi
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
8 Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS
Purchase prevention effect of Marginal loss aversion effect Net effect of information
Range of known values information on returns Subrange of known value of information on returns on returns
Note. ¯ = P − 1 − mini 8i 9, ˘ = P − a − 1/2, ˆ = P − maxi 8i 9 − 1/2, ˜ = P − mini 8i 9 − 1/2.
3.2. Summary of Analytical Findings Table 1 highlights when each of the established
The analytical model demonstrates a tension in the effects are in play. If consumers are homogeneous
effect of information on returns. On one hand, infor- in i , then Table 1 summarizes the condition on
mation may reveal a poor fit before purchase and pre- this parameter such that information increases or
vent purchases from consumers who would otherwise decreases returns. If consumers are heterogeneous
reverse the purchase decision. This purchase prevention in i , then one can infer from Table 1 that information
effect is present for consumers with sufficiently low i . increases returns on an aggregate level if i is skewed
On the other hand, information may increase the such that there is a greater density of consumers for
expected utility of purchase and increase returns due whom i > ¯ than i < . ¯
to diminished perceived value post-purchase. This Observation. Uncertainty-reducing information is
marginal loss aversion effect is less straightforward. more likely to increase returns with a left-skewed dis-
As can be seen in Figures 1–5, the marginal loss tribution of the known-utility component.
aversion effect occurs when the return probability as
a function of i is kinked. In other words, the effect We note that, in the interest of parsimony, our model
exists if knowing i changes whether the marginal abstracts from another information effect. Under a par-
consumer experiences a gain or loss relative to expec- tial return policy, information can reduce the risk asso-
tations. If the return probability function is not kinked ciated with purchase. As can be concluded from Desai
because i is low, then the return probability for each et al. (2008) and Roberts and Urban (1988), consumers
realization of i is the same across information condi- have a preference for an option that allows them
tions. If the return probability function is not kinked to reduce risk. Thus, prepurchase information may
due to high i , then any increase in the return prob- increase purchases by risk-averse consumers. Coupled
ability for high i is offset by the decrease in return with the effect of reference-dependent judgments after
probability for low i . purchase, this would result in a greater number of
This intuition facilitates exploring the robustness returns than in our current model.
to reference dependence in gains (i.e., G > 05. If
G = L , then there will be no kink in the function. 4. Overview of the Empirical Testing
As such, the marginal loss aversion effect disappears. To our knowledge, empirical evidence supporting the
However, if 0 < G < L , as is generally accepted novel analytical prediction that uncertainty-reducing
in behavioral research (e.g., Kahneman and Tversky information may lead to an increased likelihood of
1979) based on the principle that losses loom larger decision reversals is, at best, lacking. We used a
than gains, the slope of the return probability func- two-prong approach to provide such empirical evi-
tion in gains is flatter than in the current model, dence. First, we designed a controlled experiment
but not as flat as the function in losses. Thus, the in which we manipulated levels of the amount of
marginal loss aversion effect would be dampened information provided at prepurchase and leniency
because information would decrease returns for low of the return policy in a scenario designed to be
realizations i . The increasing effect on returns per- consistent with higher levels of i . Second, we ana-
sists, however, because losses loom larger than gains. lyzed the archival enrollment data of actual deci-
Thus, the greater L − G is, the greater the increase in sions from a major university’s registration record to
returns associated with information. Table 1 summa- allow us to support the external validity of the link
rizes the cases wherein the purchase prevention effect between uncertainty-reducing information provisions
and the marginal loss aversion effect arise. and purchase-reversal decisions.
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS 9
To truly isolate the effect of the amount of informa- wherein attributes communicate meaningful benefits),
tion, we used unfamiliar technical attributes to con- attribute information was selected to be technical, and
trol for performance expectations based on attribute described in unfamiliar terms. This allowed us to
knowledge. decrease the effects of prior knowledge and to more
4.1.1. Method. Participants were 420 Amazon Me- closely examine the effect of the amount of informa-
chanical Turk workers (42.4% male, average age = tion in setting a reference point. The attribute levels
36 years) who participated in the experiment in for the seven attribute levels were: 120 CBM/hour,
exchange for monetary compensation. The design maximum of 18.6 pints, below 13 C, 0.70 gph, 3 gph,
of the experiment was a 3 information level (low 3 Kg/hour, and operates in temperatures below 30 C.
versus medium versus high) by 2 return policy Participants were asked to review the return pol-
leniency (high full refund versus low 15% restock- icy of the store, which was a full refund (i.e., if you
ing fee) between-subjects design. We manipulated the buy the humidifier and decide to return it later, you
return policy to observe whether the results are repli- will get the full price $199 back) or a partial refund
cated for partial refunds. Participants were randomly 15% restocking fee (i.e., if you buy the humidifier
assigned to one of six conditions.3 and decide to return it later, you will get $169 back).
Participants were told they would be making deci- These denote the high- and low-leniency conditions,
sions about the purchase of a humidifier. They were respectively. Participants were then asked to indicate
given information describing five health benefits of the likelihood that they would buy the humidifier on
using a humidifier and told to assume that they were a 101-point scale (ranging from 0—very unlikely, to
interested in these benefits4 as a way to induce a 100—very likely), and to rate how well they expected
higher value of i . They were told that experts gen- the humidifier to perform (on a 101-point scale rang-
erally agree that the key attributes found in high- ing from 0—not well at all, to 100—very well).
performing humidifiers are air flow, bucket capac- After an unrelated five-minute filler task designed
ity, coil temperature, evaporation rate, feed rate, mist to erase participants’ short memory, participants were
volume, and temperature operation. These consti- asked to assume that they had purchased the humid-
tuted the full set of information that defined the util- ifier and that they had the opportunity to take the
ity of the product. Next, participants were asked to humidifier home and learn more about the full set
assume that they had found a humidifier from a of attributes. This was designed to simulate a situ-
retailer they trusted at the price they were willing ation in which the value of i is high such that all
to pay for this type of product ($199). The amount consumers buy the humidifier (as in Proposition 1).
of uncertainty-reducing information was manipulated The picture of the humidifier along with the table fea-
by telling participants that they were unable to turing the fully disclosed attribute values was then
verify information about all seven attributes given presented and participants were asked to rate the
their own time constraints. Participants were then likelihood that they would return the humidifier on
presented with a two-column table with the seven
a 101-point scale (ranging from 0—very unlikely to
attributes in the left column (order of presentation
100—very likely). This was the key dependent mea-
randomized based on a random sequence gener-
sure in this experiment. A number of additional mea-
ated by random.org) and information about each
sures with respect to manipulation checks, attribute
attribute in the right column (varying according
familiarity, retailer intentions, attribute desirability,
and attention recall checks were also collected; these
3
Data was collected in two different points in time (approximately measures are described in the appendix.
five months apart) per request of the review team. Variable control-
ling for this time lag in data collection is included in all analyses 4.1.2. Results. The manipulation checks showed
reported hereafter. Given the lack of statistical significance of this
variable in the analyses, results are reported collapsed across the
that the manipulations were successful in affecting
two points in time. participants’ perception of the amount of information
4
Adapted from http://www.fitsugar.com/5-Reasons-Why-Important and the store return policy. The analysis also showed
-Own-Humidifier-14569814. that participants perceived the set of attributes to be
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
10 Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS
largely unfamiliar, increasing the odds that any effects ings. Given that participants were unfamiliar with
observed are likely to stem from the amount of infor- the product attributes, and were not provided with
mation rather than from attribute knowledge (the full posterior attribute performance information that dis-
set of results are presented in the appendix). confirms prepurchase information, an expectation-
We performed an analysis of covariance (ANCOVA) disconfirmation explanation is unlikely to account for
on the post-purchase likelihood-to-return measure our results.
using the information amount and return policy fac- Overall, the predictions of the model were con-
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
tors (and the data collection time variable) as inde- firmed within a controlled experiment test designed
pendent variables and using the likelihood to buy to isolate and manipulate the levels of the key amount
the humidifier measure as a control variable with of information variable of interest. One potential
likelihood to return as the dependent measure. This shortcoming stemming from this controlled test of
analysis revealed a statistically significant effect of the predictions is the extent to which this intended
the information amount (F 4214075 = 4038, p = 0001), decision-reversal behavior could also be observed for
refund policy (F 4114075 = 7021, p = 0001), and the actual behavior. We address this issue in §4.2.
likelihood to buy control variable (F 4114075 = 16082,
p = 0001) on the likelihood to return the product. 4.2. Empirical Analysis of Archival Returns Data
We also analyzed the data using only participants To test the prediction that uncertainty-reducing infor-
who correctly recalled the refund amount in the sce- mation can lead to an increase in actual decision
nario. This new analysis (n = 369) indicated that reversals, we collected archival enrollment data from
these participants were noise in the data as shown a major university’s registration records. In line with
by the strengthening of the results despite the 12% the analytical model, a university course is a bun-
drop in sample size. Again, the information amount dle of attributes (e.g., lecture materials, reading mate-
had a statistically significant effect on the likelihood rials, assignments, peer interaction, instructor teach-
to return the humidifier (F 4213565 = 6038, p < 0001). ing style and interpersonal communication skills, etc.)
Pairwise comparisons showed that participants in that a student may choose to purchase by enrolling
both the high-information (M = 34081) and medium- in the course. A student will reverse their decision
information (M = 36008) conditions reported a greater and drop the course if the utility derived from these
likelihood to return the product than their counter- attributes is less than the value of the net refund
parts in the low-information condition (M = 25026, resulting from the decision reversal. Here the net
both p-values < 0001).5 There was also a greater ten- refund captures the value to the student of the time
dency to return the product in the full-refund condi- and/or tuition dollars that are returned to the student
tion (M = 36000) than in the restocking fee condition or their parents. We examine how additional infor-
(M = 28011; F 4113565 = 8090, p < 0001). mation about the course available at the purchase
Supporting our hypothesis that information in- decision affects the number of decision reversals. Pre-
creases expectations against which actual performance purchase information can resolve uncertainty about
is judged, the expected product performance partici- certain attributes of the course (e.g., assignments, top-
pants held increased as the number of pieces of infor- ics covered), but cannot resolve all uncertainty (e.g.,
mation available at prepurchase increased (Mlow =45031, lecture materials, instructor’s teaching style, and the
Mmedium =52069, Mhigh = 58034; F 4214085 = 6079, p < 0001). quality of other students enrolled in the course) before
The interaction between information amount and the first day or week of classes. Because the refund
return policy (p > 0015) did not reach statistical value is uncorrelated with information provision, any
significance. effect on returns of prepurchase information will be
Consistent with the analytical model, we find that driven by the effect on perceptions of the value of
a greater provision of information at prepurchase keeping the course.
indeed leads to a greater likelihood to return the
purchased humidifier when the value i is high. 4.2.1. Data Description. The archival data was
This result holds despite participants’ low stated obtained from the university time schedule and
familiarity with the product attributes. Combined matched with aggregate purchase (i.e., adding into
with the finding that larger amounts of informa- the course) and decision reversal (i.e., dropping the
tion lead to increased perceived product perfor- course) data collected from the university registrar’s
mance, this result is inconsistent with an alterna- office. The time schedule, as viewed by students at
tive expectation-disconfirmation account for our find- the time of registration, contains the following infor-
mation used in our analysis: department, course level,
5
The same pattern of results is found when using the full sample
days of instruction, time of instruction, current enroll-
both in terms of means and statistical significance at an alpha level ment, and maximum enrollment. Whereas all courses
of 0.05. in the time schedule have titles and links to a brief
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS 11
course description (typically two or three sentences), (e.g., instructor confidence in material or rigor). In
some courses also provide an external link to a more the appendix, we fully describe the procedures used
detailed syllabus. The presence/absence of a link to to control for endogeneity and to create instrumental
this additional information for a course offering is the variables (PERCENTDEPT and PERCENTTIME).
key independent variable of interest as it replicates Given the characteristics of our data (count depen-
the situation wherein more (versus less) information dent and binary endogenous variables) we use
is available before purchase. Additional details about the instrumental variable Pseudo Poisson Maximum
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
the data collection procedure are presented in the Likelihood (PPML) estimator technique described in
appendix. detail in Windmeijer and Santos Silva (1997). Given
the vector Xj of exogenous covariates described above
4.2.2. Empirical Methodology. The unit of analy-
and in Equation (6), the model is specified as
sis is a course (j). The number of drops for a given
course during the first week of classes, DROPSj , is the DROPSj = exp4yj∗ + Xj0 5 + j 1
dependent variable of interest. Because the number
of drops is an integer greater than or equal to zero, yj∗ = 1 PERCENTDEPTj + 2 PERCENTTIMEj + j 1
we consider the following relationship between drops 2
1j 12j
and course characteristics: Cov4j 1 j 5 = 1
12j 1
DROPSj = exp + 1 INFOj + 2 DAYj
where yj∗ is a latent endogenous variable capturing the
+ 3 TIMEj + 4 YEARj + 5 REQj proclivity to post syllabus information and E6j Xj 1
PERCENTDEPTj 1 PERCENTTIMEj 7 = 0. Simultaneity
+ 6 LIMITj + 7 DEPTj + 8 LEVELj
arises through the correlation of j and j . The PPML
+j 0 (6) estimator is shown to be consistent without requir-
ing assumptions about the distribution of the error
The key variable of interest is INFO, which is terms (Windmeijer and Santos Silva 1997). In fact,
a binary variable indicating whether there was a the data do not have to be Poisson for the PPML
more detailed course description available at the estimator to be consistent (Gourieroux et al. 1984,
time of registration. The model also accounts for Santos Silva and Tenreyro 2006). The PPML approach
other factors that affect a student’s decision to drop has been shown to appropriately estimate parame-
a course. On the right-hand side of Equation (6), ters when the dependent variable has a large num-
we include dummy variables indicating whether the ber of zeroes (Santos Silva and Tenreyro 2010, 2011)
class met Mondays and Wednesdays or Tuesdays and thus is appropriate for our data for which 34 of
and Thursdays (DAY); whether the class met early 202 courses had no drops. The estimation was pro-
morning, i.e., before 10:30 a.m., late morning, i.e., grammed in STATA.
between 10:30 a.m. and 12:20 p.m., early afternoon,
4.2.4. Empirical Results. We first established a
i.e., between 1:30 p.m. and 3:20 p.m., late afternoon, i.e.,
benchmark against which we could observe whether
3:30 p.m. and after (TIME); the academic year in which
the fit of the model improved once we added the
the course was offered (YEAR); whether the course
information-level variable (i.e., model presented in
was required for a business degree (REQ); the depart-
Equation (5) excluding the INFO variable). This model
ment in which the course was offered (DEPT); and
was estimated using a one stage PPML estimation
whether the course level was 300 or 400 (LEVEL). We
(Pseudo R-square = 0036, adjusted R2 = 0030; Table 2).
also account for the course enrollment limit (LIMIT),
Instrumental Variable PPML Estimation0 The second
which ranges from 19 to 125 in our data. As a bench-
model estimated included the key variable of inter-
mark to determine the additional model fit of includ-
est (INFO) in addition to the variables in the bench-
ing INFO, we examine a modified version of Equa-
mark model (i.e., model presented in Equation (6)).
tion (6) in which the INFO variable is not included.
The second stage of the regression (PPML) showed
4.2.3. Endogeneity and Estimation. Because the that the effect of the information level variable on
decision to post additional information during regis- the number of students who drop a class is posi-
tration is made by the instructor, it may be correlated tive and statistically significant ( = 0029; z = 2001,
with the error term (j 5 owing to a simultaneity bias. p = 0002; Table 2), indicating greater levels of decision
In other words, the number of drops may depend reversals (i.e., course drops) when more (versus less)
on the information provided by the instructor, and information was available at prepurchase. The model
the instructor’s information provision may depend on also showed a strong improvement in fit relative to
the (expected) number of students who will drop the the benchmark model that did not include the INFO
course or on an omitted factor that also affects drops variable (Pseudo R-square = 0064, adjusted R2 = 0060;
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
12 Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS
Course level 00067 (0.130) −00055 (0.127) chase information show a greater number of deci-
Year_2009 00288∗ (0.173) 00287∗ (0.169)
sion reversals (i.e., course drops) relative to courses
Year_2010 00206 (0.194) 00242 (0.177)
Year_2011 00297 (0.207) 00226 (0.203)
that offer less prepurchase information. Thus, the
Year_2012 00254 (0.195) 00400∗∗ (0.185) data from actual decisions support the prediction that
Finance 00780∗∗∗ (0.205) 00732∗∗∗ (0.211) information reducing uncertainty before purchase can
Management 00979∗∗∗ (0.212) 00913∗∗∗ (0.211) increase the number of service cancellations.
Marketing 00317 (0.206) 00270 (0.209)
ISOM 00165 (0.285) −00218 (0.261)
Late morning −00205 (0.155) −00232 (0.150) 5. General Discussion
Early afternoon −00014 (0.161) −00123 (0.164) This research examined how truthful prepurchase
Late afternoon 00285∗ (0.169) 00344∗∗ (0.155) information that decreases (but does not fully resolve)
Course required 00091 (0.211) 00216 (0.190)
uncertainty influences decision reversals. We draw on
Class enrollment limit 00014∗∗∗ (0.004) 00008∗∗ (0.003)
behavioral theory of reference-dependence to derive
Notes. We used the variance inflation factors (VIFs) from an ordinary least a model predicting that the provision of uncer-
squares (OLS) regression as a measure of multicollinearity. The highest VIF
tainty-reducing prepurchase information can actually
was 2.14 (mean VIF = 1068), which is well below the recommended thresh-
old of 10. Thus, multicollinearity does not appear to be an issue. Standard increase the likelihood that consumers will reverse
errors are in parentheses. their purchase decision. This research, the first to
∗
Significant at p < 0010; ∗∗ significant at p < 0005; ∗∗∗ significant at p < 0001. derive an analytical model of decision reversals from
behavioral theory and the first to establish the con-
sequences of reference-dependence in the context of
Table 2). The instruments were shown to be strong decision reversals, sheds light on the consumer’s
with large values of adjusted R2 (0.92) and partial R2 information processing when deciding whether to
(0.91) and an F -statistic (F 4211845 = 322043, p < 000001) reverse a decision as well as on the potential impli-
cations for marketers. The proposed model parsimo-
well above the threshold of 10 proposed by Staiger
niously predicts situations wherein greater levels of
and Stock (1997). This indicates that the estimates of
information reduces or increases decision reversals.
the coefficients are likely unbiased. Overidentification To our knowledge, the latter result is novel in the liter-
restrictions also do not seem to be an issue in our ature and at odds with current marketplace practices.
estimation (Hansen’s J 2 415 = 2002, p > 0010), support- In both validation studies, information about a
ing the assumption of exogeneity of the instruments dimension of the product was revealed before pur-
(Baum 2006). chase but full evaluation was not possible until after
We further tested the robustness of our results by purchase. Thus, our findings apply to the many mar-
capturing the likelihood of decision reversal as mea- ket situations wherein a consumer’s uncertainty is
sured by the proportion of students who dropped only partially resolved before purchase. Examples of
each course relative to the total number of students these situations are the three-dimensional computer
enrolled in each course on the first day of classes. models used by online retailers to reduce uncertainty
We estimated two models using a fractional response around the physical fit of clothing or accessories, the
probit model with instrumental variables (ivprobit test driving of cars, and the trial of sound equipment
syntax in STATA). The first model estimated was iden- in artificial settings designed to enhance sound per-
tical to the full PPML model except that the depen- formance. In each case, the information about a set
dent measure was the proportion of students who of attributes changes the expectations of overall prod-
uct utility and affects the overall evaluation after pur-
dropped the course. In the second fractional response
chase when the remaining dimensions are ultimately
model we removed the control class_limit because the
revealed in full. Thus, compared to a consumer who
number of students enrolled on the first day of classes
(the denominator in the proportion) could be strongly
6
Given that PPML estimation has been shown to be consistent for
influenced by the class-size limit. Both analyses repli-
any nonnegative distribution, we also estimated both models using
cated the overall pattern of results with the variable this estimation technique. The same pattern of results was found
of interest INFO rendering positive and statistically ( = 0044, Z = 2067, p < 0001, and = 0041, Z = 2042, p = 0002).
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS 13
information-amount factor. That is, participants believed The aggregate data were collected from the university’s
they were receiving larger amounts of information as the registrar office. For each of the course identification num-
number of attribute specifications available in the pre- bers, the registrar office provided the aggregate number of
purchase scenario increased (M low = 3008, M medium = 4030, students enrolled in the course on the first day of class and
M high = 5026; F 4214085 = 78090, p < 00001). All pairwise com- the aggregate number of students who reversed their deci-
parisons between the three cells also showed statistical sig- sion and dropped the course during the first week of class
nificance (all p-values < 00001). These ratings did not vary (when full tuition refunds are provided and new course reg-
as a function of the return policy manipulation (F 4114085 = istrations are permissible). The average number of drops
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
0001, p > 0085) nor as a function of the interaction between per course was 3.09 (standard deviation = 3003) and the
these two manipulated factors (F 4214085 = 1001, p > 0036). number of drops ranged from 0 to 20 across the 202 courses.
An ANOVA on the measure of perceived leniency of the Endogeneity and Instrumental Variables Procedures.
return policy showed a strong effect of the return-policy To identify appropriate instrumental variables to address
manipulation with participants perceiving a policy offering the endogeneity issue, we surveyed 34 instructors from the
a full refund as being statistically significantly more lenient same population as the enrollment data to better under-
than a policy charging a restocking fee (M full_refund = 3006, stand the decision to post additional information during
M restocking_fee = 6028; F 4114085 = 493011, p < 00001). These rat- registration. Participants were asked if they posted addi-
ings did not vary as a function of the information-amount tional information, if they were aware that the option to
manipulation (F 4214085 = 1042, p > 0020) nor as a func- post information was available to them, and to state their
tion of the interaction between these two manipulations agreement (or disagreement) (on a 7-point scale) with the
(F 4214085 = 0088, p > 0040). Overall, these results confirm following two statements: “Making the syllabus available
that the manipulations worked as expected. online during registration increases course enrollment” and
An ANOVA on the measure of familiarity with the “Making the syllabus available online before the start of
attributes showed that familiarity did not vary as a function the term decreases the number of students who drop the
of the information-amount manipulation (F 4214085 = 1008, course.” We tested the relationship between awareness and
p > 0030), return-policy manipulation (F 4114085 = 0035, agreement with each of the two statements on the decision
p > 0050), nor as a function of the interaction between to post the syllabus by analyzing the correlations between
these two manipulations (F 4214085 = 1015, p > 0030). the variables. The decision to provide prepurchase infor-
A one-sample t-test analysis, however, showed that, as mation statistically significantly correlated with awareness
intended by design, participants perceived the attributes of the option to post (r4345 = 0047; p = 00005). Neither the
to be largely unfamiliar given that the average familiar- belief that prepurchase information increases enrollment
ity rating (M = 3025) was statistically significantly lower (r4345 = −0013; p = 0046) nor the belief that it decreases
than the neutral point (rating of 4) of the familiar- drops (r4345 = 0020; p = 0025) statistically significantly corre-
ity scale (t44195 = −9022, p < 00001). Overall, participants lated with the decision to provide the information. Aware-
found retailers offering a full refund to be less misleading ness was not statistically significantly correlated with the
(M = 2049) than retailers offering a partial refund (M = 3011; degree of agreement with the two statements (both ps >
F 4114085 = 7037, p < 0001). The main effect of information 0050). The results suggest that awareness of the ability to
amount and the interaction between policy and information post has a strong influence on the decision to post.
amount did not statistically significantly affect the mislead- To control for potential endogeneity issues, we posit that
ing ratings (both p-values > 0005). there may be a peer effect in generating awareness about
the option to provide prepurchase information. One may
Archival Data learn from other members of their department with whom
scholarly interaction is more likely. One may also learn from
Data Collection Procedure. Research assistants, un- other instructors who are teaching during the same time
aware of the research hypotheses, collected the time sched- window. As such, we created two instrumental variables to
ule data and created a dummy variable equal to 1 for capture the peer effects in the decision to post.
courses providing additional course information (high- For a given course j in the sample, we created the
information level) at the time of registration and 0 other- PERCENTDEPTj (PERCENTTIMEj 5 variable equal to the
wise (low-information level). The data consisted of courses number of courses in the department (time of instruction)
offered in the Business School in the Fall term of five con- exclusive of course j that offered the additional informa-
secutive academic years (2008–2012). Fall term was used tion, divided by the total number of courses exclusive of
to temper network effects and information spillovers that course j in the department (time of instruction). Information
are likely when registration occurs shortly before the course made available before registration by an instructor’s peers is
begins. Courses for which multiple sections were offered exogenous to the instructor’s course drops in that this infor-
by different instructors were not sampled for two reasons. mation does not affect the value a student receives from the
First, such cases make it difficult for students to deter- instructor’s course and thus COV4PERCENTDEPTj 1 j 5 = 0
mine how much spillover there is from the information pro- and COV4PERCENTTIMEj 1 j 5 = 0.
vided by one section to the other sections. Second, course
drops from one section may be attributable to new open-
ings in other sections that were previously full. A total of References
202 courses met all of the criteria and 48 (23.76%) courses Amaldoss W, Jain S (2005) Conspicuous consumption and sophis-
provided additional information about the course. ticated thinking. Management Sci. 51(10):1449–1466.
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS 15
Amaldoss W, Jain S (2010) Reference groups and product line deci- Guo L (2009) Service cancellation and competitive refund policy.
sions: An experimental investigation of limited editions and Marketing Sci. 28(5):901–917.
product proliferation. Management Sci. 56(4):621–644. Hardie B, Johnson E, Fader P (1993) Modeling loss aversion and
Anderson E, Sullivan M (1993) The antecedants and consequences reference dependence effects on brand choice. Marketing Sci.
of customer satisfaction for firms. Marketing Sci. 12(2):125–143. 12(4):378–394.
Anderson E, Hansen K, Simester D (2009) The option value Heilman C, Bowman D, Wright G (2000) The evolution of brand
of returns: Theory and empirical evidence. Marketing Sci. preferences and choice behaviors of consumers new to a mar-
28(3):405–423. ket. J. Marketing Res. 37(2):139–155.
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
Balachander S (2001) Warranty signaling and reputation. Manage- Hess J, Chu W, Gerstner E (1996) Controlling product returns in
ment Sci. 49(7):1282–1289. direct marketing. Marketing Lett. 7(4):307–317.
Baum C (2006) An Introduction to Modern Econometrics Using Stata Ho T, Zhang J (2008) Designing pricing contracts for boundedly
(Stata Press, College Station, TX). rational customers: Does the framing of the fixed fee matter?
Bechwati N, Siegal W (2005) The impact of the prechoice process Management Sci. 54(4):686–700.
on product returns. J. Marketing Res. 42(3):358–367. Ho T, Lim N, Camerer C (2006) Modeling the psychology of con-
Boone L, Kurtz D (2007) Contemporary Marketing, 13th ed. (Thomson sumer and firm behavior with behavioral economics. J. Mar-
Learning, Mason, OH). keting Res. 43(3):307–331.
Breiter H, Aharon I, Kahneman D, Dale A, Shizgal P (2001) Func- Ho T, Lim N, Cui T (2010) Reference-dependence in multi-location
tional imaging of neural responses to expectancy and experi- newsvendor models: A structural analysis. Management Sci.
ence of monetary gains and losses. Neuron 30(2):619–639. 56(11):1891–1910.
Bruce N, Desai P, Staelin R (2006) Enabling the willing: Consumer Hoch S, Ha J (1986) Consumer learning: Advertising and the ambi-
rebates for durable goods. Marketing Sci. 25(4):350–366. guity of product experience. J. Consumer Res. 13(2):221–233.
Cachon G (2003) Supply chain coordination with contracts. Iyer G, Kuksov D (2012) Competition in consumer shopping expe-
Graves S, de Kok T, eds. Handbooks in Operations Research and rience. Marketing Sci. 31(6):913–933.
Management Science: Supply Chain Management (North Holland, Janiszewski C, Cunha M Jr (2004) The influence of price discount
Amsterdam), 227–339. framing on the evaluation of a product bundle. J. Consumer
Che Y (1996) Consumer return policies for experience goods. Res. 30(4):534–546.
J. Indust. Econom. 44(1):17–24. Kahneman D, Tversky A (1979) Prospect theory—Analysis of deci-
sion under risk. Econometrica 47(2):263–291.
Chen Y, Turut O (2013) Context-dependent preferences and inno-
Kőszegi B, Rabin M (2006) A model of reference-dependent prefer-
vation strategy. Management Sci. 59(12):2747–2765.
ences. Quart. J. Econom. 121(4):1133–1165.
Chu W, Gerstner E, Hess J (1998) Managing dissatisfaction: How
Kuksov D, Lin Y (2010) Information provision in a vertically differ-
to decrease customer opportunism by partial refunds. J. Service
entiated competitive marketplace. Marketing Sci. 29(1):122–138.
Res. 1(2):140–155.
Lattin J, Bucklin R (1989) Reference effects of price and promotion
Cui T, Mallucci P (2013) Fairness ideals in distribution channels.
on brand choice behavior. J. Marketing Res. 26(3):299–310.
Working paper, Carlson School of Management, University of
Lim N, Ho T (2007) Designing price contracts for boundedly ratio-
Minnesota, Minneapolis.
nal customers: Does the number of blocks matter? Marketing
Davis S, Gerstner E, Hagerty M (1995) Money back guarantees in
Sci. 26(3):312–326.
retailing: Matching products to consumer tastes. J. Retailing
Mellers B, Schwartz A, Ritov I (1999) Emotion-based choice. J. Ex-
71(1):7–22.
periment. Psych.: General 128(3):332–345.
Davis S, Hagerty M, Gerstner E (1998) Return policies and the opti-
Miller C (2011) Defying conventional wisdom to sell glasses online.
mal level of “hassle.” J. Econom. Bus. 50(5):445–460.
New York Times (January 16):B1. http://www.nytimes.com/
Desai P, Purohit D (1998) Leasing and selling: Optimal strategies 2011/01/17/technology/17glasses.html?_r=0.
for a durable goods firm. Management Sci. 44(11):S19–S34.
Moorthy S, Srinivasan K (1995) Signaling quality with money-
Desai P, Kalra A, Murthi B (2008) When old is gold: The role of back guarantees: The role of transaction costs. Marketing Sci.
business longevity in risky situations. J. Marketing 72(1):95–107. 14(4):442–466.
Desai P, Koenigsberg O, Purohit D (2004) Strategic decentralization Narasimhan C, Turut O (2013) Differentiate or imitate? The role of
and channel coordination. Quant. Marketing Econom. 2(1):5–22. context-dependent preferences. Marketing Sci. 32(3):393–410.
Douthit D, Flach M, Agarwal V (2011) A “returning problem” Ofek E, Katona Z, Sarvary M (2011) Bricks and clicks: The impact
reducting the quantity and cost of product returns in con- of product returns on the strategies of multi-channel retailers.
sumer electronics. Accenture. http://www.accenture.com/ Marketing Sci. 40(1):42–60.
sitecollectiondocuments/pdf/accenture-reducing-the-quantity Orhun Y (2009) Optimal product line design when consumers
-and-cost-of-customerreturns.pdf. exhibit choice set-dependent preferences. Marketing Sci. 28(5):
Enright T (2003) Post-holiday logistics. Traffic WORLD (January 868–886.
6):20. Roberts J, Urban G (1988) Modeling multiattribute utility, risk, and
Farber H (2008) Reference-dependent preferences and labor supply: belief dynamics for new consumer durable brand choice. Man-
The case of New York City taxi drivers. Amer. Econom. Rev. agement Sci. 34(2):167–185.
98(3):1069–1082. Saini R, Rao R, Monga A (2010) Is that deal worth my time? The
Ferguson M, Guide VDR, Souza G (2006) Supply chain coordination interactive effect of relative and referent thinking on willing-
for false failure returns. Manufacturing Services Oper. Manage- ness to seek a bargain. J. Marketing 74(1):34–48.
ment 8(4):376–393. Santos Silva J, Tenreyro S (2006) The log of gravity. Rev. Econom.
Gourieroux C, Monfort A, Trognon A (1984) Pseudo maximum like- Statist. 88(4):641–658.
lihood methods: Applications to poisson models. Econometrica Santos Silva J, Tenreyro S (2010) On the existence of the maxi-
52(3):701–720. mum likelihood estimates in poisson regression. Econom. Lett.
Gu Z, Liu Y (2013) Consumer fit search, retailer shelf layout, and 107(2):310–312.
channel interaction. Marketing Sci. 32(4):652–668. Santos Silva J, Tenreyro S (2011) Further simulation evidence on
Gu Z, Xie Y (2013) Facilitating fit revelation in the competitive mar- the performance of the poisson pseudo-maximum likelihood
ket. Management Sci. 59(5):1196–1212. estimator. Econom. Lett. 112(2):220–222.
Gumus M, Ray S, Yin S (2013) Returns policies between channel Shugan S, Xie J (2005) Advance-selling as a competitive marketing
partners for durable products. Marketing Sci. 32(4):622–643. tool. Internat. J. Res. Marketing 22(3):351–373.
Shulman, Cunha Jr., and Saint Clair: Consumer Uncertainty and Decision Reversals
16 Marketing Science, Articles in Advance, pp. 1–16, © 2015 INFORMS
Shulman J, Coughlan A (2007) Used goods, not used bads: Prof- Windmeijer F, Santos Silva J (1997) Endogeneity in count data mod-
itable secondary market sales for a durable goods channel. els: An application to demand for health care. J. Appl. Econo-
Quant. Marketing Econom. 5(2):191–210. metrics 12(3):281–294.
Shulman J, Coughlan A, Savaskan R (2009) Optimal restocking fees Wood S (2001) Remote purchase environments: The influence of
and information provision in an integrated demand-supply return policy leniency on two-stage decision processes. J. Mar-
model of product returns. Manufacturing Service Oper. Manage- keting Res. 38(2):157–169.
ment 11(4):577–594. Xie J, Gerstner E (2007) Service escape: Profiting from customer
Shulman J, Coughlan A, Savaskan R (2010) Optimal reverse channel
cancellations. Marketing Sci. 26(1):18–30.
structure for consumer product returns. Marketing Sci. 29(6):
Downloaded from informs.org by [198.137.20.104] on 27 March 2015, at 10:23 . For personal use only, all rights reserved.
1071–1085. Xie J, Shugan S (2001) Electronic tickets, smart cards, and online
Shulman J, Coughlan A, Savaskan R (2011) Managing consumer prepayments: When and how to advance sell. Marketing Sci.
returns in a competitive environment. Management Sci. 57(2): 20(3):219–243.
347–362. Yin S, Ray S, Gurnani H, Animesh A (2010) Durable products with
Staiger D, Stock J (1997) Instrumental variables regression with multiple used goods markets: Product upgrade and retail pric-
weak instruments. Econometrica 65(3):557–586. ing implications. Marketing Sci. 29(3):540–560.