You are on page 1of 63

Accepted Manuscript

Title: The Role Ofphysical Distributionservicesas


Determinants Of Product Returns In Internet Retailing
Author: Shashank Rao Elliot Rabinovich Dheeraj Raju
PII:
DOI:
Reference:

S0272-6963(14)00049-7
http://dx.doi.org/doi:10.1016/j.jom.2014.06.005
OPEMAN 865

To appear in:

OPEMAN

Received date:
Revised date:
Accepted date:

20-5-2013
21-6-2014
25-6-2014

Please cite this article as: Rao, S., Rabinovich, E., Raju, D.,The Role Ofphysical
Distributionservicesas Determinants Of Product Returns In Internet Retailing, Journal
of Operations Management (2014), http://dx.doi.org/10.1016/j.jom.2014.06.005
This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.

us

cr

ShashankRao, Ph.D.*
Jim W. Thompson Assistant Professor of Business
Raymond J. HarbertCollege of Business
Auburn University
405 West Magnolia Avenue
Auburn, AL 36849
Shashank.Rao@auburn.edu
Ph. (334)844 6845

ip
t

THE ROLE OFPHYSICAL DISTRIBUTIONSERVICESAS DETERMINANTS OF


PRODUCT RETURNS IN INTERNET RETAILING

an

Elliot Rabinovich, Ph.D.


John G. and BarbaraA.BebblingProfessor of Business
ArizonaStateUniversity
W. P. CareySchool ofBusiness
Tempe, AZ85287-4706
Elliot.Rabinovich@asu.edu
Ph.(480)965 5398

te

Dheeraj Raju, PhD


Assistant Professor and Statistician
UAB School of Nursing
The University of Alabama at Birmingham
NB 1019F
1720 Second Avenue South
Birmingham, AL 35294-1210
Seeth001@uab.edu
Ph. (205) 975 5336

Ac
ce
p

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

* Corresponding Author

42
43

Page 1 of 62

32
33
34
35
36
37
38
39
40

te

an

us

cr

ip
t

ABSTRACT
Pressure continues to build on internet retailers to squeeze out inefficiencies from their day to
day operations. One major source of such inefficiencies is product returns. Indeed, product
returns in Internet retailing have been shown to be, on average, as high as 22% of sales. Yet,
most retailers accept them as a necessary cost of doing business. This is not surprising since
many retailers do not have a clear understanding of the causes of product returns. While it is
known that return policies of retailers, along with product attributes, are two important factors
related to product return incidents, little is known about which aspects of the online retail
transaction make such a purchase more return-prone. In the current study, we seek to address this
issue. We use a large data set of customer purchases and returns to identify how process
attributes in physical distribution service (PDS) influence product returns. The first attribute
involves perceptions of scarcity conditions in inventory availabilityamong consumers when
retailers reveal to consumers information on inventory levels for the products that they intend to
buy. Our results show that orders in which items are sold when these conditions are revealed to
shoppers have a higher likelihood of being returned than orders in which these conditions are not
revealed. While prior research has argued that inventory scarcity perceptions have an effect on
purchases, our findings suggest that they are also related to the likelihood of these purchases
being returned. The second attribute involves the reliability in the delivery of orders to
consumers. We find that the likelihood of orders being returned depends on the consistency
between retailer promises of timeliness in the delivery of orders and the actual delivery
performance of the orders. Moreover, we find that the effect that consistency in the delivery has
in the likelihood of returns, is stronger for orders that involve promises for expedited delivery
than for orders with less expeditious promises. That is, although the occurrence of returns
depends on the delays in the delivery of orders to consumers relative to the initial promises made
by the retailers, this effect is more notable for orders that involve promises of fast delivery.

Ac
ce
p

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

Keywords: Online Retailing; Electronic Commerce; Order Fulfillment; Returns; Physical Distribution

Acknowledgements The authors would like to sincerely acknowledge the assistance provided
by several colleagues during the development of this manuscript - C.CliffordDefee, Brian
Gibson, and Bill Hardgrave (all at Auburn University) for the academic assistance, and Scott
Talley and Randy Salley for the practitioner assistance. Any errors and omissions however, are
our own.

Page 2 of 62

THE ROLEOF PHYSICALDISTRIBUTIONSERVICESASDETERMINANTSOF


PRODUCT RETURNS IN INTERNET RETAILING
1. Introduction

ip
t

2
3
4
5
6

While the Internet has improved customers ability to search and find products best suited

to their needs (Brynjolfsson and Smith, 2000), important limitations still exist that prevent

shoppers from using the Web to efficiently fulfill their demands. Proof of this lies inthesizeable

us

cr

amount of product returns that retailers routinely receive from customers (Guide and van

11

Wassehnove, 2009). According to recent industry reports, average return rates for products

12

purchased online have reached levels as high as 22% (Demery, 2010). In contrast, product return

13

rates average a more reasonable 8.1% to 8.7% for traditional retailers, according to 2013 reports

14

by the Retail Equation and the National Retail Federation1. This difference in return rates

15

indicates that there are elements in online retail transactions that make the merchandise highly

16

susceptible to returns, relative to traditional retail transactions.

te

an

10

Despite such a clear preponderance of returns in online retail, there is limited extant

18

research on this issue. It has been argued within the Operations Management (OM), Supply

19

Chain Management (SCM), and Marketing literaturesthat there is a lack of theory-driven

20

research on returns,as a whole (Mollenkopf et al. 2011),and in online retailing contexts,in

21

particular (Griffis et al., 2012-a; Petersen and Kumar, 2009). Even the limited research that

22

exists on this subjecthas focusedlargely on policy issues like optimal return strategies for Internet

23

retailers (Wood 2001; Nasr-Bechwati and Siegel, 2005; Ofek et al., 2011) and onthe

Ac
ce
p

17

Please refer to:


https://www.theretailequation.com/Retailers/images/public/pdfs/white_papers/wp_TRE4010_WhitePaper_Optimizin
gValueofReturns_Jan2013.pdf

Page 3 of 62

management of returns once they occur (Mollenkopf et al., 2007; Griffis et al., 2012-a).

Someprior research hasbeen carried out into the causes of returns, and it has shown that returns

canoccur due to specific product characteristics (e.g., Petersen and Kumar, 2009), particularly on

the Web,where customers are limited in their ability to properly evaluate certain products

(e.g.,apparel) that require close inspection and evaluationprior to purchase (Peck and Childers,

2003).It has also been argued that it is difficult for customers to evaluate products that are newer

or are more obscure and are, therefore, less available for inspection and comparison on and off

the Web. As a result, such products aretypically more return-prone (Rabinovich et al., 2011).

cr

us

However,given that scholars have recognized that customer predispositions and

an

ip
t

behaviors(including returns)are shaped by a combination of market policies, product

11

attributes,and process (encounter-related) attributes (Humphreys and Williams, 1996), it would

12

be fair to argue that, along withreturnpolicies (e.g., Anderson et al., 2009; Mollenkopf et al.,

13

2007; Wood 2001) and product attributes outlined in prior researchin Internet retailing (e.g.,

14

Rabinovich et al. 2011; Petersen and Kumar, 2009), there are severalprocess attributes that may

15

predispose certain sales transactions between Internet retailers and customers to be more return-

16

prone than others. But, while market policies and product attributes have received some attention

17

in the literature as discussed earlier, process attributes have remained relatively

18

unexplored.These process attributes encompass the entirespan of purchase transactions on

19

theInternet starting at pre-salesbuyer-seller interactions and ending atpost-sales exchanges

20

after customers have paid for their products (Humpreys and Williams, 1996; Marshall et al.,

21

2003; Heim and Field, 2007).

22
23

Ac
ce
p

te

10

In this paper, we focus particularly on process attributes for physical distribution services
(PDS) to show that the disclosure of scarcity conditions in inventory availability to consumers

Page 4 of 62

prior to purchasing products and the post-sale deliveryreliability of purchased goods are

keypredictors of product returns in Internet retailing transactions. Our results show that orders in

which shoppers are aware of scarcity conditions in inventory availability, have a higher

likelihood of being returned than orders in which these conditions are not manifest. While prior

research has argued that perceptions of inventory scarcity by consumers have a strong effect

onpurchases, our findings suggest that they have a substantial effect on the likelihood of these

purchases being returnedas well. We also find that the likelihood of orders being returned

depends on the consistency between retailer promises of timeliness in the delivery of these orders

and their actual delivery performance. Moreover,we find that the effect that consistency in the

an

us

cr

ip
t

delivery has on the likelihood of returns is stronger for orders that involve promises of fast

11

delivery than for orders with promises ofless speedydelivery. Thus, although the occurrence of

12

returns is likely to result from delays in the delivery of orders to consumers relative to the initial

13

promises made by the retailers, this effect is more notable for orders that involve fast delivery

14

promises. This extends prior studies that have focused on the existence of a relationship between

15

quality in the delivery of orders to customers and perceptions of customer satisfaction to

16

articulate and validate the effect caused on actual product returns by inconsistencies between

17

retailer promises of timeliness in the delivery of Internet orders and actual delivery performance.

Ac
ce
p

te

10

18

Identifying such process attributes effects on the occurrence of product returns is an

19

issue that deserves focused research for one central reason- if these attributes arefound to be

20

related to a higher return-likelihood of a purchased item, then thiswould give online retailers

21

enhanced predictability with respect to product returns. Our focus on PDS attributes and their

22

effects on product returns extends OM research in online retailing that has studied the role of

23

order fulfillment and return operations and their implications for customer satisfaction, customer

24

loyalty, purchase behavior, and firm profitability (e.g., Rabinovich and Bailey, 2004; Rabinovich
5

Page 5 of 62

et al., 2007;Rabinovich et al., 2008; Boyer and Hult, 2005; Boyer and Hult, 2006; Boyer et al.,

2009; Thirumalai and Sinha, 2005; Griffis et al., 2012-a).In doing so, our work at a more general

level,also adds to research that has focused on studying relationships between OM and

Marketing (e.g., Sawhney and Piper, 2002; Malhotra and Sharma, 2002; Hausman et al., 2002).

The remainder of this paper is organized as follows. The next section provides a review

of the literature, an introduction to the theoretical background, and the hypotheses development.

The research methodsare then presented in the third section, followed by the data analyses in the

fourth section. Finally, the fifthsection presents a detailed a discussion of the

results,implications, and future research opportunities, while the sixth section concludes.

12

cr

us

an

2. Literature Review and Theoretical Background

10
11

ip
t

The study of productreturns in the OM and SCM literaturesspans the areas of product
recovery and closed loop supply chains (Ketzenberg and Zuidwijk, 2009).In these areas, the

14

activities and research insightsfall into three categories (Guide and van Wassenhove, 2009). The

15

first category deals with the front end of the supply chain and focuses on the customersproduct

16

return / return decision itself. The second consists of operational issues about remanufacturing,

17

while the third is devoted to market development of remanufactured products. Although research

18

has investigated operational issues regarding remanufacturing (e.g., Atasu et al., 2008; Souza et

19

al., 2002; Ketzenberg et al., 2003) and market development of remanufactured products (e.g.,

20

Guide and Li, 2010; Vorasayan and Ryan, 2006),research on customer return decisions is scant.

21

This is especially true in online retail contexts, where a substantial amount of research has

22

focused on forward (i.e., getting the product to the consumer), rather than reverse supply chain

23

management (Griffis et al., 2012-a; Petersen and Kumar, 2009). Thus, the remainderof

24

ourliteraturereview focuseson this direction.

Ac
ce
p

te

13

Page 6 of 62

As far as customer return decisionson or off the Internetareconcerned, two key

investigation-worthy issues jump to mind:why customers return products and what the value

proposition of these returns is in the customersminds.The latter question has been addressed,to a

large extent, by authors who have examinedhow the value of the returns proposition in the

customers eyesmay benefit the customer and the firm. For example, research has demonstrated

that product returns can result in increased follow-on spendingwith theretailer, thus suggesting

that the act of returningaproduct enhances customer confidence with the retailer (Griffis et al.

2012-a). Similarly, Petersen and Kumar (2009) haveshownthat this increased expenditure is not

just restricted to the short term, but can be extended to customers lifetime values as well. At a

10

more general level, Wood (2001) argues that leniency in return policies increases purchase rates

11

for customers in remote purchase situations. From a manufacturers standpoint, Padmanabhan

12

and Png (1997) show that lenient return policies also create sources of competitive advantage for

13

retailers and manufacturers, thus alluding to the notion of the value of the returns proposition in

14

customers minds.Extending this idea, recent research has focused on identifying return

15

strategies that provide optimal value to both buyers and sellers (e.g., Ketzenberg and Zuidwijk,

16

2009; Su, 2009; Shulman et al.,2009; 2011).

cr

us

an

te

Ac
ce
p

17

ip
t

The former question (understanding why customers return products) has alsoreceived

18

attention by prior research, but to a much lesser extent than the question about the value of

19

product returns for consumers. Scholars have recognized that customer behavior (including

20

returns) is shaped by a combination of product attributes and process (transaction-related)

21

attributes (Humphreys and Williams, 1996). Along the product-attributes front, research on

22

commercial returns of consumer goodshas demonstrated that returns can be driven by factors

23

such asdefects, product incompatibility with user needs, and deficiencies in product performance

Page 7 of 62

relative to customer expectations (Guide etal. 2006;Ferguson et al.2006). Researchon returns in

online retailing has also examined the role played by products valuations and popularity among

consumers (Rabinovich et al.,2011). In addition, among the process attributesof thetransactions,

it has been shown that attributes, such as the extent of disconfirming (negative) information prior

to purchase, affects the decision to return a product (Nasr-Bechwati and Siegal, 2005). But, to

the best of our knowledge, no research has considered how process attributes in the transaction

that are specifically related to PDSgenerate product returns, particularly in Internet settings.

cr

us

ip
t

Understanding such PDS process attributes effects on product returns is important for
online retailers because these firms are typically younger and less experienced in how to prevent

10

returns than conventional retailers (Rao et al., 2009).In addition, online retailers are more likely

11

to be exposed to false returns or return fraud (e.g.,when products are purchased purely for the

12

purpose of returning) something that is arguably also prevalent in conventional retail, but is

13

likely easier to control through efficient returns gatekeeping at the point of sale.Finally, PDS is

14

an extremelyimportant part of onlineretailsales sinceit is what completes the retailtransaction and

15

finallyputs the product in thehands of the endcustomer (Esper et al., 2003). Thus, online retailers

16

would benefit greatly from understanding what PDS attributes of the online retail transaction are

17

significantly relatedtoconsumers decisions about returns. The remainder of this

18

sectionaddresses this issue by building a theoretical model of product return antecedents in

19

online retailing. As argued earlier, our focus is on the PDS process attributes of the online retail

20

transaction rather than market policies or product attributes. Because the role of market policies

21

and product attributes in returns has received substantial attention in the literature, we treat them

22

as control factors in our study. We expand on these controls in our research methods section.

23

2.1. Conceptual Framework

Ac
ce
p

te

an

Page 8 of 62

When assessing the outcomes of a transaction, customers go through an evaluation


process where they judge the extent to which the outcome from the transaction has delivered

upon their desires. Depending on whether this outcome meetsthese desires, customers will

experience positive or negative emotions and will develop coping responses that are manifested

in future behaviors (Bagozzi, 1992). This process was originally introduced by Lazarus (1991)

who proposed a framework to explain the process by which consumers assess a transactionor an

exchange and develop future intentions and behaviors. This framework consists of three stages:

(1) customers appraisal, (2) service providers response, and (3) customers coping behaviors.

In the context of retail exchanges on the Internet, the appraisal stage focuses on

an

us

cr

ip
t

customers assessments of excellence of the products or services that are to be provided by

11

retailers (Griffis et al., 2012-b). The assessments during this stage depend oncustomers

12

perceptions of expected quality of products or services (Cronin and Taylor, 1992). During this

13

stage, customers assess different attributes in the retailers offers to form expectations, prior to

14

placing their purchases, aboutthe goods orservicesthat theywillreceive (Griffis et al., 2012-b).

te

10

Moving onto the response stage, researchers have embraced Olivers (1981, 1997) view

16

of customer satisfaction, defined as the emotional assessment of the ability of a service provider

17

to deliver a pleasurable amount of consumption-related fulfillment. Based on this definition,

18

measures of satisfaction have proven effective to account for emotional responses of customers

19

in the assessment of retail exchanges (Cronin et al., 2000). It is during the response stage that

20

customers get to experience and evaluate attributes in the retailers offers and develop a measure

21

of satisfaction based on initial expectations formed during the appraisal stage (Griffis et al.,

22

2012-b). In this sense, Internet retail transactions are different from traditional retail transactions,

Ac
ce
p

15

Page 9 of 62

given that there is usually a time difference between the appraisal and response stages (when

customers await delivery of theiritems), which may not be present in traditional retail.

The response stage is followed by the coping stage. During this stage, customers get to
react to what they have observed and experienced in the previous two stages. These reactions are

reflected in the customers conduct or behavioral intentions actions such as deciding to return

products, purchase again from the retailers, or recommend the retailers to others (Nyer, 1997).

Taken together, thesethreestagesform the foundation for thedevelopment of our

us

cr

ip
t

hypotheses. Based on thesehypotheses, wewillexaminethe role played by PDS process attributes

in online retail sales during the appraisal and response stages in shaping customers coping
behavior in the form of the occurrence of returns.

11

2.2. Hypotheses

PDS involveprocessesaimed at ensuring that customers receive their orders in an accurate

12

10

an

and timely manner (Bowersox and Closs, 1996; Perreault and Russ, 1976; Bienstock et al.,

14

1997).In online retailing, PDS are characterized by (1) inventory availability, (2) timeliness(i.e.,

15

order delivery cycle time) expectation,and (3) reliability in order delivery (Rabinovich and

16

Bailey, 2004).Forour hypotheses,wefirstconsidercustomer perceptions of inventoryavailability

17

and customerexpectations of delivery timeliness (the first two dimensions of PDS). These

18

perceptions and expectations are formed in the appraisal stage of the online retail exchange

19

model discussed in Section 2.1. We subsequently focus on the response stage of the retail

20

exchange, where customers get to evaluate thethird attribute in PDS: reliability in order delivery.

21

Based on our conceptual development, we then relate the PDS attributes in the first two stages to

22

the outcome variable of interestin the copingstage ofthe retail exchange model (productreturns).

Ac
ce
p

te

13

10

Page 10 of 62

InventoryAvailability. In traditional retail settings, customers perceptions of inventory


availability are based on environmental cues involving product displays, shelf space size, and

customer traffic(Jacoby, 2002). Without access to such cues in online retailing, customers are

circumscribed to assess availability using only textual and visual descriptions in the retailers

websites, informing them whether products are in-stock or not or specifying the number of

products left in inventory. Limiting customers access to this information makes it difficult for

them to evaluate inventory availability relative to product demand rates. This, in turn, can

confound their assessment of products true availability and stock-out risks.

cr

us

Moreover, customers sole reliance on textualdescriptions about products in-stock

an

ip
t

availability on the Internet makes them easily susceptible to scarcity heuristic effects in their

11

appraisal of Internet retailer offers.The scarcity heuristic can be understood as a manifestation of

12

the belief that items and opportunities [to buy them] become more valuable as they become

13

less available (Cialdini, 2001; p. 78). Ditto andJemmott (1989) argue that if consumers perceive

14

that an item is in short supply, they will assume that it is worth buying and that this leads to

15

systematic errors in judgingthe true value of products. The principle is founded on two

16

beliefs:(1) items that are difficult to obtain are typically more valuable because the availability of

17

an item (or the lack of it) can serve as a short-cut cue to its popularity / quality (Lynn, 1989) and

18

(2) when a product becomes less available, free choice is limited or threatened, and this makes

19

people want the product significantly more than they would have otherwise (Brehm, 1966).

20

Ac
ce
p

te

10

The scarcity heuristic has oftentimes been demonstrated in the literature. For example,

21

Verhallen (1982) and Verhallen and Robben (1994) found greater preference for books when

22

they were perceived as scarce. Similarly, Lynn (1989) found that paintings perceived as scarce

23

were more desirable than paintings perceived as readily available. In a meta-analysis, Lynn

11

Page 11 of 62

(1991) alsofound a positive relationship between perceived scarcity and value perceptions. The

use of the scarcity heuristic in the marketplaceis also well documented from a practitioner

standpoint. For example,companies in the hospitality and travelindustries often advertise that

availabilities are filling fast or that seats are limited. Similarly, retail advertisements for

products often contain time limits on offers (e.g., 1 week only sale, or this weekend only) or

quantity limits(e.g., maximum 2 items per customer) (Inman et al., 1997).

cr

We propose that, in an online environment, the scarcity heuristic will

us

ip
t

becomemanifestwhen retailers disclose information about on-hand inventory levels for products.

Indeed, some retailers disclose the number of units theyhaveininventory whenthis amount is

an

limited (thus creating a sense of scarcity). For instance, Amazon.comadvertises some items as

11

having limited availability (with an associated number of units in stock, e.g., only 3 available).

12

The reasons for this communication could be many: from wanting to clear out the last few units

13

of inventory and make way for new stock to encouraging purchases by playing on the scarcity

14

heuristic.

te

This communication of scarcity received by the consumer can then be seen as part of the

Ac
ce
p

15

10

16

overall appraisal process (Bagozzi, 1992; Cialdini, 2001), similar to how design features of a

17

website affect the appraisal process (Eroglu et al., 2003).This appraisal would then trigger an

18

increased value perception about the item in the consumers mind, which would make the item a

19

highly attractive purchase(Cialdini, 2001; Bagozzi, 1992).This is consistent with the findings of

20

Inman et al. (1997, p. 68), who argued that the presence of a restriction operates to activate a

21

cognitive resource that renders judgments regarding the favorableness of the offering.

22

Note also that in most online retail purchases, customers firstassess the retail offers

23

and,after waiting for delivery, get to experience and evaluate the products and services that

24

theyreceived.Thus the development ofa complete measure of satisfaction based on the


12

Page 12 of 62

initialexpectations is delayed, because there is often a timegap between the assessment of the

online offer and the experiencingof the product. On account of deliverylead times, customers

willonlybeable tofullyand objectively assess theirpurchases upon deliverywhich is decoupled

from the purchaseprocess itself(Peck and Childers 2003).We build on the existence of this

timegap and the scarcity heuristic principle to propose that theenhanced product value

communicatedby original perceptions of scarcityin online offers stands to besubstantially

reduced once customersactually receive possession of and experience the products. This is

because once they receive their products, consumers will be in a position to evaluate the

products intrinsic qualities without the enhanced value communicated by way of the scarcity

an

us

cr

ip
t

heuristic (once the item is in physical possession, there should be a reduced perception of

11

scarcity).One outcome to such adecreased assessmentof the value proposition of

12

productspurchased under perceived scarcity conditions would be an increase in the likelihood of

13

product returns during the coping stage.This reasoning is consistent with prior research which

14

proposes that when people buy on impulse, they are likely to be less happy with their purchases

15

(Gardner and Rook, 1988). Thus, weposit that products soldunder perceived scarcityconditionsin

16

availability have a higher likelihood of being returned than productssold under non-scarcity

17

conditions. This leads to our firsthypothesis.

te

Ac
ce
p

18
19
20
21
22

10

H1: Online retail orders in which items are sold under perceived scarcity conditions
in availability have a higher return likelihood than orders in which items are sold
under non-scarcity conditions in availability.
Order Delivery Reliability. Customers assessment of order delivery reliability in Internet

23

retail settings is a function of actual delivery performance (observed during the response stage)

24

in relation to the expectationsthat are formed during the appraisal stage (Rabinovich and Bailey,

25

2004). This is consistent with prior research in long-distance retailing where delivery reliability

13

Page 13 of 62

depends on the extent to which retailers can match actual delivery performance to the delivery

performance expectation signaled to buyers (Hutchinson and Stolle, 1968; Gilmour, 1977). In

addition, this conceptualization follows the service reliability construct as outlined in the

SERVQUAL literature, wherein reliability is understood as the ability to perform promised

services dependably and accurately (e.g., Berry and Parasuraman, 1991; Berry et al., 1994). The

relation between order delivery reliability, or the lack of it, and returns behavior is important to

investigate for one central reason.Prior research on customers redress -seeking behavior has

argued that customers will seek redress to a dissatisfying service encounter depending on their

impressions of locus and controllability (Weiner, 1985; Weiner, 2000). What this implies isthat

an

us

cr

ip
t

customers are likely to demonstrate redress seeking behavior depending on whether they feel

11

they can attribute blame exclusively to a single party for the failure in the encounter (locus) and

12

whether they feel that the party was in a position to avoid the failure (controllability) (Weiner,

13

1985; Folkes, 1984; Folkes et al., 1987; Bitner, 1990). The Internet retailing environment is

14

unique in this regard because mostretailers are not autonomously incharge of performing order

15

deliveries to customersand instead, share this responsibility withthird party logistics providers.

16

Thus, it is unclear whether customers will feel comfortable assigning the blame (i.e., locus and

17

controllability) for low reliability in deliveries purely to the retailers (given that delivery

18

reliability is at least partially controlled by logistics providers).As aresult, it is notobvious

19

whether customers will generate product returns (i.e., demonstrate redress seeking behavior) in

20

the face of delivery reliabilityfailures, since they will be unclear as to whether theycan

21

attributethe blameto the retailers or not (Weiner, 1985). This could bepartly why, to thebest of

22

ourknowledge,this issue has not been addressed in theliterature.

Ac
ce
p

te

10

14

Page 14 of 62

While prior research on redress seeking behaviordoes come up short in predicting return
likelihood in the face of delivery reliability failures in an Internet retailing context, the

Expectation Confirmation Theory (ECT) (Oliver, 1980) proves a useful resource to investigate

this relationship. According to the ECT, when a customers actual service outcome meets or

exceeds her expected outcome, there is positive disconfirmation which results in satisfaction

(Bearden and Teel, 1983; Oliver and DeSarbo, 1988; Susarla et al., 2003). Conversely, when

expectations are not met, the customer will experience negative disconfirmation and

dissatisfaction (Oliver, 1980, 1997; Bearden and Teel, 1983). Indeed, research(e.g., Bitner, 1990;

Bitner and Hubbert, 1994; Boulding et al., 1993; Kelley et al., 1993; Smith et al., 1999; Harris et

an

us

cr

ip
t

al., 2006) hasshown that service encounter satisfaction (dissatisfaction) leads to greater (lower)

11

overall perceived service satisfaction. Similar to the SERVQUAL literature (e.g.,Parasuraman et

12

al., 1994; Zeithaml et al., 1996; Zeithaml, 2000), the ECT paradigm visualizes the delivery gap

13

as the difference between the apriori expectations of service, and the levels of service actually

14

received (Susarla et al., 2003; Venkatesh and Goyal, 2010).

te

ECT provides a three-stepflow model to explain responses by customers to service

Ac
ce
p

15

10

16

performance relative to expectations. First,exposure to information describing a

17

servicesperformance characteristics leads to the formation of service specificbeliefs or

18

expectations by the customer (Olson andDover 1979). Second, a cognitive comparison between

19

expectationsand actual experiences leads to a subjective calculationof disconfirmation by the

20

customer (Oliver et al. 1994). Third, the combinationof expectations and disconfirmation

21

determines a satisfactionlevel that, in turn, influences subsequent behavior by the customer

22

(Venkatesh and Goyal, 2010).

23
24

At abroad level, this three-step flow model is consistent with theappraisal response
coping model ofLazarus (1991).This flow model provides a reference to understand customers
15

Page 15 of 62

assessments of order delivery reliability in the online retail exchange context. Consistent with

Lazarus (1991) framework, these assessments are a function of customers cognitive

comparisons of actual delivery experiences during the response stage relative to expectations

formedby customers when receiving information during the appraisal stage about delivery

performance promises.It has been argued that, in an online retail setting, customers face

asymmetric information on delivery performance because when they buy from retailers they can

only estimate the expected (not the actual) deliveryperformance that they will receive through

their interaction with the retailers websites (Rabinovich and Bailey, 2004).Thus, because

customers cannot ascertain actual delivery performance prior to giving retailers their business,

10

this delivery performance measuredthrough order cycle times plays a key role in customers

11

assessments of their exchanges with Internet retailers (Griffis et al., 2012-b).

an

us

cr

ip
t

This explains why order cycle times havebecome a key component of the overall value

13

proposition provided by online retailers. While location remains a key determinant of success

14

for conventional retailers, the importance of location in an online context is substantially reduced

15

(Esper et al., 2003). Ratherthanlocation, customersatisfactionoften hinges

16

uponhowwellonlineorders are delivered to their destinations (Alba et al., 1997; Boyer and Hult,

17

2005; Rabinovich and Bailey,2004). When orders are not delivered within their expected times,

18

customers consider them service failures by Internet retailers(Rao et al., 2011) and when such

19

service failures occur, the customers perceived value of their service encounters is lowered

20

(Smith and Bolton, 1998; Bolton, 1998), consistent with ECT.This suggests that a decrease in

21

performanceindeliveryreliability the extent to which actualorder deliveryperformance falls short

22

of expected deliveryperformance shouldgeneratedissatisfactionwith the retailersamong

23

customers.It remains tobedetermined whether,as inferred byReinartzand Kumar (2003),Stock et

24

al. (2006), andDaughertyetal. (2003),customer dissatisfactiondirected atretailersis a cause

Ac
ce
p

te

12

16

Page 16 of 62

ofproduct returnswhen the focus is on deliveryreliabilityin online purchases by consumers.If this

werethe case,adecreasein performancein deliveryreliabilityshouldincrease the risk that online

customers willreturn theirproducts. Oursecond hypothesisencapsulates this potential effect.

ip
t

H2: Delivery reliability for an online retail order is negatively related to the orders
return likelihood.
Expectation of Delivery Timeliness. Past empirical applications of the ECT havelead

cr

4
5
6
7

generally to linear operationalizations of expectation-confirmation gaps, suggesting that actual

service performance levels should increase proportionately withexpectations(e.g.,Szajna and

10

Scamell, 1993). However, recent studies have questioned the validity of this assumption and

11

haveargued in favor of the existence of a non-linear relationship between service performance

12

levels and expectations (Edwards, 2002; Staples et al., 2001; Brown et al., 2013). These studies,

13

which have drawn from the Assimilation-Contrast Model (Anderson, 1973), posit that the more

14

extreme an experienced stimulusis vis--vis the expected stimulus, the more likely it is that

15

positive or negative reactions will be experienced and that the magnitude of these reactionswill

16

be non-linear. Moreover, according to Chebat et al. (2005), the extent to which customers seek

17

redress in the face of a service failure depends on the magnitude of the failure itself. Thus,as

18

customers begin to seek redress (i.e., returning the product), the extent to which they engage in

19

such behavior will depend on the extent of deviation from the expected service.As service

20

expectations increase, a proportional increase in actual service performance will not be enough to

21

preserve satisfaction; instead, service performance mustimproveat an increasing rate as a priori

22

expectations go up(Brown et al., 2013; Bearden and Teel, 1983; Szymanski and Heard, 2000).

23

Ac
ce
p

te

an

us

Given that in online purchases customers self-select into different categories of delivery

24

service (e.g., next day delivery, 7 day delivery) when choosing how to get their orders sent to

25

their destinations, customers who self-select into premium servicecategories (e.g., next day

17

Page 17 of 62

delivery) will have higher expectations of delivery speed than those who self-select into other

categories. Consequently, the deleterious impact of an equal delay in order delivery will be

higher for the former group than it will be for the latter group.
These arguments suggest that customers a priori expectations ofdeliverytimeliness (e.g.,

ip
t

the number of days promised for delivery)will interact with actual delivery reliability to

influence their likelihood of engaging in product returns, as stated in H2.Specifically, an increase

in the level of customers initial expectations of delivery timeliness for their orders will

attenuateany impact that the orders delivery reliability may have on the likelihood of the orders

being returned. This leads to our third hypothesis.

us

an

H3: Customer expectations of order delivery timelinessmoderate negatively


(attenuate)the effect byactual order delivery reliability on product returns.

10
11
12

cr

3. Research Methods

14
15
16

3.1. Data Source

17

online retail exchanges introduced in Section 2.1), we use transaction level data from a retailer

18

based in the United States.The retaileroperates channels based on the Internet, catalogs, and

19

physical stores and specializes in selling jewelry products, time-pieces,andothersuchproducts,

20

allof which fall within the personal accessory category. For the purposes of our study, we

21

focused exclusively on the retailers Internet channel. At the time of our study, this channel had

22

been in operation for over 5 years and had sold in excess of 100,000 items. Annual sales through

23

this channel positioned this retailer among the largest 500 online merchants in the U.S.,

24

according to rankings by Internet Retailer magazine.

25

13

Ac
ce
p

te

To empirically test our model (illustrated in Figure 1 along with the different stages in

[Insert Figure 1 Here]

18

Page 18 of 62

Our decision to focus on a single firm is consistent with prior empirical research on
Internet retailing (e.g., Olson and Boyer, 2003; Rabinovich et al., 2011; Rao et al., 2011). Given

the nature of our research, where the focus is on individual customer behavior rather than on firm

conduct, the use of a focal-company setting does not compromise the validity of the results

because our unit of analysis is at the customer transaction level, and not at the firm

level.Furthermore, the use of a focal company setting allows us to test our hypotheses without

firm-level confounding issues involving disparate product return policies and different levels of

brand recognition and loyalty among consumers that may influence the likelihood of product

returns.

cr

us

an

10

ip
t

Because the retailer sells fashion intensive, specialty merchandise, product returns in our
empirical setting tend to bedriven by product pricing, quality, and aesthetics considerations just

12

as much as they are bythe product availability and deliveryfactors at the center of our model.

13

This condition provides us with a critical settingto test our hypotheses. Arguably, if we can

14

validate our hypotheses in this setting, where attributes other than those involved in the PDS

15

process itself are key determinants in customers product return-decisions, this validationislikely

16

to hold in other online retail settings involving products like groceries, where factors such as

17

fashion are less importantwhile product availability and delivery are more critical.

te

Ac
ce
p

18

11

Moreover,our focus on fashion intensive, specialty products contributed to generate

19

results that carry substantive validity relative to other contexts in which firms sell commoditized

20

goods (e.g., books) that are mass marketed across many retail sites and for which scarcity is not a

21

common occurrence. The products in our study are also less subject to high rates of false returns

22

causedby productobsolescence or bybuyersinability touse the products dueto theircomplexity.

23

These false-return incidents are more common when items like electronics are involved.

19

Page 19 of 62

A final consideration in testing the hypotheses involves the nature of the retailers return
policy at the time of our study. As part of its policy,the retaileroffered an exchange or a refund

for the value of the products (but not the cost of shipping and handling)on all returns received

within 30 days of the purchase. Moreover the retailer did not pay for return shipping.The

stringency of this policy is about average for the industry segment in which the retailer competes.

For example,threeof the retailers four largest competitors within its product category also offer

return windows of 30 days, with the fourth one offering 45 days. Moreover, none of them

reimburse shipping and handling costs associated with the initial purchase or the return of the

products.However, this return policy is somewhat strict, when compared to policies followed by

an

us

cr

ip
t

well-known retailers in other industry segments. For example, retailers like Zappos.comhave

11

instituted policies in which customers face no time restrictions to return products and are

12

refundedboth the financial value of the returned goods, and the shipping and handling fees

13

incurred in initially buying and returning these goods. We contend that therelative stringency in

14

the return policy instituted by the retailer in our study provides us with an opportunity to test our

15

model in a conservativeenvironmentin which indiscriminate product returnsareatypical This

16

allows for anassessmentof our hypotheses withoutan excessiveinterferenceof falsereturns.

17

3.2. Data Structure

te

Ac
ce
p

18

10

The data we used for this study involves 6,732 Internet transactions between the retailer

19

and its customers during the entire calendar year of 2010. To ensure uniformity in our evaluation

20

of delivery reliability, we considered transactions that involvedorders fulfilled exclusively within

21

the U.S. We also limited our analysis to orders involving customers who placed only one order in

22

the 2010 calendar year. This guaranteed that we were not dealing with potential confounding

23

effects, where purchases were being made by customers in one time period to replace products

24

that they may have returned earlier. This also ensured that the data included only one transaction
20

Page 20 of 62

per customer and, consequently, that the data observations in the analyses were statistically

independent from each other across customers

Finally, we included orders that involved only one item. In so doing, we removed from
the sample orders placed by customers who purchased multiple products topossibly resell them

or to try them out and return some of them later. Arguably, those types of orders (especially the

latter) would have an element of return bias already embedded in them, given that items may

have been purchased with the express intent of returning at least some of them.

us

cr

ip
t

8
9
10

3.3. Main Constructs Operationalization

11

independent constructs, Scarcity Condition in Availability, Order Delivery Reliability, and

12

Expected Order Delivery Timeliness.We expand on these constructs operationalization below.

13

Table 1 summarizestheir operationalization as well as the labels used for the constructs in the

14

empirical analysis and theirdescriptive statistics. Table 2 lists their correlation coefficients.

an

te

[Insert Tables 1 and 2 Here]

Product Return.We operationalized the dependent construct in the model by using a

Ac
ce
p

15
16
17

The model in Figure 1involves a dependent construct, Product Return, and three

18

dummy variable that denotes whether the product in each of the Internet transactions in our data

19

was returned (value of 1) or not(value of 0). A product return was logged into the system when

20

previously purchased merchandisewas actually delivered back to the retailer,after the retailer was

21

able to process the corresponding return merchandise authorization (RMA) form submitted by

22

the customer. As shown inTable 1, the return rate across the orders in our sampleis 15%.

23

Scarcity Condition in Availability.The retailer in our study discloses to customers

24

information on inventory levels for stock keeping units (SKUs) only when those levels are less

25

than or equal to 10 units. We propose that such a disclosure of information regarding a limited

21

Page 21 of 62

number of items in stock (e.g., 10 or fewer units) creates perceptions of product scarcity in

customers minds. Thus, we use an array of binary dummy variables to operationalize perceived

scarcity conditions in availability among consumers. In effect, since consumers can access

information on inventory across ten different stock levels (from 1 to 10 units) at the time of

purchase, we usedten variables to capture scarcity conditions in availabilityin each of these

inventory levels. Each of these ten variableslabeled as Scarcity1, Scarcity2,, Scarcity10was

assigned a value of 1 if the amount of inventory disclosed to consumers at the time of purchase

was 1, 2,,or 10 units, respectively, or a value or 0 otherwise. To capture instances in which

SKUs were not sold under perceptions of scarcityconditions in availability, weused anadditional

10

dummy variable (Scarcity11) that was assigned a value of 1 when there was no communication of

11

scarcity received by consumers at the time of purchase in the form of inventory level information

12

(because stocks exceeded 10 units)or a value of 0 otherwise.To test H1, we will use Scarcity11as

13

the baseline against which we will measure the joint effects on the likelihood of returns across

14

the first ten dummies (Scarcity1 through Scarcity10) capturing scarcity conditions in availability.

cr

us

an

te

Order Delivery Reliability.We operationalize this construct as the extent to which (in

Ac
ce
p

15

ip
t

16

percentage terms)orders are delivered ahead of (or later than) promised delivery times. For

17

example, if a particular orders estimated delivery timeis 4 days, but the order is delivered in 3

18

days, its reliability measure is (4-3)/4 = 0.25 or 25%. Similarly, if the same order is delivered in5

19

days, its reliability measure is (4-5)/4 = -25%. The retailer inour study provides customers with a

20

promised delivery time based on where the orders ship from, where they ship to,the orders

21

chosen delivery modes, and the corresponding number of non-businessdays involved in

22

delivering the orders. The actual amount of time spent in delivering the orders is available from

23

the delivery receipts that the retailer obtains from its third party logistics providers.

22

Page 22 of 62

Expected Order Delivery Timeliness.As explained above, customers expectations of


order delivery timeliness areoperationalizedusing the promised delivery times for their orders.

For example, customers who are promised overnight delivery (next-day air shipping) would have

an order delivery timeliness expectation score of 1 day, while those who are promised a7 -day

delivery would have a delivery timeliness expectation score of 7. In effect, a lower score on this

measure implies a higher expectation of delivery timeliness.

7
8
9

3.4. Control Variables

us

cr

ip
t

In our analysis, we account for effects on product returnscaused by several control


variables. We expand on these variables and their expected effects on returnsbelow. Table 1

11

summarizes the variables, the labels used for them in the results, and their descriptive statistics.

12

Table 2 lists their correlations.

14

Failures Exogenous to PDS.It is likely that some customers will engage in product

te

13

an

10

returns due to failuresthat are external to PDS in our model. Prior work on commercial returns of

16

consumer goods by Guide et al. (2006) and Ferguson et al. (2006) argues that customers are

17

likely to return products if the products delivered are broken or damaged. Moreover, customers

18

arelikely to return products if the information describing the products online is incorrect or

19

flawed. This is because these information errors may induce customers to buy products that will

20

not match their needs or expectations.To account for these contingencies, we follow the lead by

21

Rabinovich et al. (2011) and include two binary (dummy) control variables in which we capture

22

respectively whether a product return in our sample was caused by products that were delivered

23

damaged or broken (1= yes, 0 = no) or by flawed product descriptions on the retailers Internet

24

site (1= yes, 0 = no), as reported by customers in their RMAs.

Ac
ce
p

15

23

Page 23 of 62

Product Price Leadership.Returns also depend on the competitiveness of prices offered


by the retailer for each of its products. The online retail business model is heavily price driven

and product price is one of the main factors that determine customers buying decisions. This is

because, on the Web, customers search costs are greatly reduced (Brynjolfsson and Smith,

2000) and price comparisons across retailers are but a few mouse-clicks away (Srinivasan et al.,

2002). Thus, we expect that items sold by a retailer at prices below competition are less likely to

be returned than items priced above competition. Customers are more likely to return items that

are priced above competition because these items carry lower transaction values than items sold

at bargain prices. We use a dummy variable to specify which orders involve items priced by the

10

retailerin our studyabove (value of 0) andat or below (value of 1) competition. We expect to find

11

anegative link between this variableand thelikelihood of returns.

an

us

cr

ip
t

Product Price. Beyond the competitiveness of prices charged by retailers for their

13

products, we expect that high-priced products will be more likely to get returned as compared to

14

low-priced products. This is because consumers will have a stronger motivation to recoup their

15

expendituresand send back their products if theseitems carry higher monetary values (Anderson

16

et al., 2009). Thus, we control for the purchase price of products and expect that as product price

17

increases, return likelihood will increase.

Ac
ce
p

te

12

18

Shipping and Handling (S&H) Price. The fees charged by a retailer to deliver its

19

products to customers may also influence the likelihood of product returns. But the nature of this

20

effectdepends on whether the retailer refunds these fees as part of its return policy or not. Since

21

the return policy by the retailer in our study does not include the reimbursement of S&H fees,

22

weexpect these fees to act as a deterrent against returns. As these fees increase, customers will be

23

less likely to return their products because they will not be able to recoup these expenses.

24

Page 24 of 62

Therefore, we expect to see a lower return likelihood for products for which customers pay high

S&H prices than for products for which customers pay low S&H prices.

Purchase History.Customers are likely to be more at ease when appraising offers from
online retailers that they have used in the past, rather than from unfamiliar retailers (Petersen and

Kumar, 2009). Moreover,customers ease in their relationships with retailers should translate into

more frequent purchasing encounters. Thus, customers with on-going relationships with a retailer

are likely to buy items more frequently than brand new customers (Reichheld, 1996; Reinartz

and Kumar; 2002; Petersen and Kumar, 2009). By extension, given that customers with on-going

relationships buy items more frequently from retailers, as compared to new customers, we would

an

us

cr

ip
t

also expect that the likelihood of observing product returns for a purchase exchange is higher

11

among existing customers than among new customers. To control for this effect, we use a

12

dummy variable to indicate which customers in our data had purchased products from the

13

Internet retailer in the past (value of 0) and which ones had not (value of 1). We identified

14

whether customers had purchased products from the retailer in the past depending on whether

15

their names and addresses had been used in prior purchases from the retailer, before they placed

16

their orders in our data set

Ac
ce
p

te

10

17

Gifting.Gift receivers usually assign emotional and social value to gifts, thus increasing

18

their utility (Areni et al., 1998; Sheth et al., 1991). In gifting situations, customer perceptions of

19

retail transactions comprise not only a utilitarian consideration (Sherry, 1983), but also a social

20

and emotional component. Sheth et al. (1991, p. 161) define this component as perceived utility

21

acquired from anassociation with one or more specific social groups. As a result, in making

22

product return decisions, gift receivers will take into consideration the social and emotional value

23

obtained from gifts, in addition to their transactional value. Therefore, because purchases

25

Page 25 of 62

involving gifts carry not only a utilitarian but also an emotional and social value, we expect that

they will have a lower likelihood of being returned than purchases in which no gifts are involved.

This expectation is consistent with findings by Petersen and Kumar (2009). Thus, we
follow these authors lead and control for the effect that gifts have on the likelihood of returns by

using a dummy variable to indicate which purchases in our data involved gifts (value of 1) and

which ones did not (value of 0). We expect to observe a negative relationship between this

variable and the likelihood of product returns. We identified those orders in our dataset that

involved gifts based on whether the orders billing names and addresses matched their mailing

names and addresses. If the two did not match, we inferred that the orders involved gifts.

cr

us

an

10

ip
t

Seasonality.In our setting, customers return decisions have seasonality built into them,
especially around the holidays. We expect a higher likelihood of returns during the holidays

12

because there ishistorically a greater risk that shoppers will reconsider their purchases and send

13

back products bought during this time of the year (Petersen and Kumar, 2009). As a result, we

14

control for seasonality effects on returns involving orders placed between the week before

15

Thanksgiving 2010 and December 31, 2010. We use a dummy variable to categorize seasonal

16

orders (with a value of1)and non-seasonal orders (with a value of 0).

te

Ac
ce
p

17

11

Product Popularity.As such, products that are sold more often will most likely be

18

returned more often (Rabinovich et al., 2011). Therefore orders that involve more popular

19

products are likely to carry a greater return likelihood as well. We operationalize product

20

popularity as a function of each products annual sales in the year prior to the time the order was

21

placed. Because sales for a small fraction of items were disproportionately larger than the rest of

22

the items, we rescaled this measurement using a natural log. This operationalization for

23

popularity is consistent with Rabinovich et al.s (2011) operationalization of a similar variable.

26

Page 26 of 62

1
2
3
4

4. Data Analysis and Results


As described earlier, we specified whether each item in the orders in our datasetwas
eventually returned or not and created a binary dependent variable,Return.Because weusea

binarydependent variable to operationalizethe occurrenceof product returns,we choseto

usealogistic regression to test thehypotheses. Moreover,following prescribed best practices in the

study of moderated models (c.f., Aguinis et al, 2013), we tested the hypotheses using hierarchical

multiple regression (Cohen and Cohen, 1983; Davison et al.,2002). Aguinis et al, (2013) propose

10

a multi-step approach to statistically test such type of models. Based on this approach, one must

11

first test a controls only model in which the dependent variable (Return) is regressed upon all

12

control variables. Subsequently, one must build a direct effects model that includes first-order

13

paths linking all independent variables (including all control variables and moderators) with the

14

dependent variable. Finally, one must build a model in which interaction variablesare added as a

15

function of the product of their first-order variables. The statistical significance of these

16

interaction variables would indicate moderation and complete the test in the overallmodel.

17

To test our hypotheses, we used a similar three-step approach. We first entered the

18

control variables in the model (Model 1).Subsequently, we introducedto the model (Model 2) the

19

controls and the variables that account forthedirect effects in H1 and H2 (i.e.,Scarcity1,

20

Scarcity2,,Scarcity10, and Reliability) along with a direct effect by the moderator

21

(i.e.,Timeliness Expectation). Finally, we tested the model with the interaction effect proposed in

22

H3, while keeping the direct effects of H1 and H2, along with the controls, in the model (Model

23

3).The results are presented in Table 3. The significance level of the Hosmer and Lemeshow

24

testfor all modelswas above 0.05, indicating that lack of fit is not a concern with any of the

25

models.Therefore, we proceed with a discussion of the modelsexplanatory and predictive

Ac
ce
p

te

an

us

cr

ip
t

27

Page 27 of 62

validity tests and the results from the independent variables coefficientsand their support for the

hypotheses.
[Insert Table 3 Here]

4.1. Explanatory Validity Tests

ip
t

As reported in Table 3, Models 1, 2, and 3 have statistically significant 2 coefficients (p<


0.01). Moreover, the pair-wise differences ofthese coefficients between Model 2 and Model 1

and between Model 3 and Model 2 show statistically significant improvements in the models

goodness of fit afterthe inclusion of the hypotheses direct and interaction effects(as we

transition from Model 1 to Model 2 and from Model 2 to Model 3).Please refer to Table 4.

an

us

cr

Table 3 also presents values for Nagelkerke Pseudo R-Squared rates in Models 1, 2, and

11

3, which compare favorably with those reported in other recentconsumer behaviorresearch (e.g.,

12

Dixon and Verma, 2013). Our Pseudo R-Squared values also increase as the analyses transition

13

from Model 1 to Model 2 and to Model 3. Moreover, based on the F-test procedures proposed by

14

Cohen (1968), we found statistically significant improvements in the Pseudo R-Squared values

15

between Model 2 and Model 1 and between Model 3 and Model 2. TheF-test results regarding

16

the difference in the Pseudo R-Squared values between Model 2 and Model 1 (F = 22.18; p

17

<0.001) suggestthat the addition of the hypotheses predictors in Model 2 makes a statistically

18

significant contribution in explaining our dependent variables variance, above and beyond the

19

contribution made by the control variables.Similarly, the F-test results pertaining to the

20

difference in the Pseudo R-Squared values between Model 3 and Model 2 (F = 10.2; p < 0.001)

21

demonstratea statistically significant improvement by the addition of the interaction effect from

22

H3 in Model 3 in explaining our dependent variables variance.Table 5 presents results from

23

these tests.Based on these results, we can conclude that the addition in Model 2 of the direct

24

effects in H1 and H2 contributes significant explanatory power over and above the control

Ac
ce
p

te

10

28

Page 28 of 62

variables. Also, the addition in Model 3 of the interaction effect hypothesized in H3 adds

significant explanatory power over and above the terms included in both Model 1 and Model 2.
[Insert Tables 4 and 5 Here]

4.2. Predictive Validity Tests

are classified correctly. More importantly, 27.6% of the overall returns are correctly predicted.

This is substantially better than the 15% that would be predicted purely by guessing. Moreover,

the effects of our hypotheses determinantsin Model 3 provide a statistically significant

ip
t

4
5
6

us

cr

The classification matrix for Model 3 (Table 6) shows that 89% of all our observations

improvement to our ability to predict the outcomes of our dependent variable in relation to the

11

predictive contribution provided by the control variables in Model 1. To evaluate these

12

improvements, we followed Hosmer and Lemeshows (2000) guidelines and calculated a C-

13

Statisticindexfor the Receiver Operating Characteristic (ROC) curve in Model 1 and in Model 3

14

and then statistically compared these indices.The value of each C-Statistic index corresponds to

15

the area under the ROC curve for its corresponding model (see Figure 2 for the corresponding

16

ROC curves). This valuewill range from 0 to 1 and provide a measure of each models ability to

17

discriminate between those observations that display the outcome of interest (product return) and

18

those that do not. For Model 1, we obtained a statistically significant C-Statistic value of 0.68

19

(p< 0.01), while the C-statistic value for Model 3 was 0.73 (also statistically significant with p<

20

0.01).

21

Ac
ce
p

te

an

10

Together, the values for Model 1 and Model 3 show that a total improvement of 5% in

22

predictive power can be attributed to the inclusion of our hypothesis variables in Model 3. This

23

improvement is statistically significant (at p= 0.05), as denoted by the lack of overlap between

24

the asymptotic 95% confidence interval for the C-statistic values for Model 1 (0.662, 0.702) and

25

Model 3 (0.709, 0.747). Moreover, according to the C-Statistic benchmarks put forth by Hosmer
29

Page 29 of 62

and Lemeshow (2000) to evaluate the predictive power of logistic regression analyses, we can

ascertain that the controls-only model (i.e., Model 1), with a C-statistic value of 0.68, is

considered to provide only a marginal level of discriminating power. However, with a C-statistic

value of 0.73, Model 3 is considered to offer good discriminating power.


[Insert Table 6 Here]

[Insert Figure 2 Here]


4.3. Regression Results andHypotheses Validation

us

cr

ip
t

As shown in Table 3, the coefficients in Model 3 show that four of the control variables

(Price Lead, New Customer, Gift, and Popularity) have statistically significant effects (p 0.10)

10

on the likelihood of products being returned. The signs of these coefficients in Model 3 are also

11

consistent with those in Models 1 and 2 as well as with our expectations.

12

an

To test H1,wemustevaluatethejoint(notthe separate) effects on the likelihood of returns


across the variables (Scarcity1, Scarcity2,, Scarcity10) that captured, individually, scarcity

14

conditions in availabilityperceived by consumers atvarious inventory levels (1,2,,or 10 units)

15

disclosed by the retailerat the time of purchase. These effects were measured in relation to the

16

likelihood of returns for orders in which consumers had no access toinformation oninventory

17

levels and, thus, could not form perceptions of scarcity conditions in availability (please refer to

18

Section 3.3).We mustevaluate theseeffects jointly because, as explained in the development of

19

H1, the scarcityheuristicin anonlinesettingmaybecomemanifest at any stock levelbelow the

20

threshold chosen by the retailer to serve as the limit at and above which inventory levelswill not

21

be shared with consumers.In our setting, the scarcity heuristicmay become manifest at any

22

inventory level (1, or 2,,or 10 units) below the threshold of 11 units set by the retailer.

Ac
ce
p

te

13

23

Thus, fromH1s statement,support for this hypothesis will depend on finding statistically

24

significantevidenceamongModel3s resultstorejectthenullhypothesis(H10)thatonline retailorders


30

Page 30 of 62

in which items are sold under perceived scarcity conditions in availability (at either 1, 2,,or 10

units of inventory) donothavea higher return likelihood than orders in which items are

soldundernon-

scarcityconditionsinavailability.Sinceeachcoefficient(B)inthearrayofScarcityvariables

(Scarcity1, Scarcity2,,Scarcity10) in Model 3measures the effect on the likelihood of

returnsbyscarcityconditionsin availabilityperceived byconsumersinrelationtothelikelihood

ofreturnsforordersinwhichSKUsaresoldundernoperceptionsofscarcityconditionsinavailability, H1

will be confirmed if H10: BScarcity1= 0 andBScarcity2,= 0andBScarcity10= 0 is rejected.

cr

us

The results in Model 3 show that the coefficients for Scarcity1through Scarcity9 have a

an

ip
t

positive and statistically significant (p 0.02) effect on Return. The values of these coefficients

11

are important because theysupport H1s expectation that online retail orders sold under scarcity

12

conditions in availability do have a higher return likelihood than orders in which items are sold

13

under non-scarcity conditionsinavailability. However, to fully validate H1, we must evaluate the

14

joint effect on Returnby all the variables in the Scarcity array through a test that will determine

15

whether we can reject H10. This test followedguidelines by Greene (1999), Murray (2005),

16

andLutsetal.(2011)for the evaluation of joint effectsby variable arrays, such as Scarcity1,

17

Scarcity2,, Scarcity10, in logistic regressions andother maximum likelihood estimation

18

analyses. The testevaluated whether the goodness of fit in Model 3 (based onitsChi-

19

Squaredvalue)statisticallyworsenedafterwesetBScarcity1=BScarcity2==BScarcity10=0, in accordance to

20

H10. The test results showed that Model 3s Chi-Squared value fell from 1290.58 (22 df and p<

21

0.01) to 1178.74(10 df and p< 0.01) after we set these coefficients values to zero. This

22

corresponds toa statistically significant decay in the goodness of fit in Model 3,reflected in a

23

drop of 111.84 (12 df, and p< 0.01)inModel3s Chi-Squared value. Based on this result, we can

Ac
ce
p

te

10

31

Page 31 of 62

reject H10 and, incombination with the positive and statistically significant Scarcity coefficients

identified above,validate H1.

As theorized in H2, Model 3 shows that Reliability has a negative and statistically
significant (p<0.05) effect on Return.Moreover, because the coefficient for the interaction term

(Reliability * Timeliness Expectation) is positive and significantly different from zero (p<0.05),

Model 3 provides empirical support for the moderation effect proposed in H3 (Cohen and Cohen,

1993).Recall that in the coding for expected delivery timeliness, a lower score in fact meant a

higher apriori expectation. Thus, the positive sign of the interaction regression coefficient

actually indicates a negativemoderation effect(or attenuation) by Timeliness Expectation on the

an

us

cr

ip
t

link between Reliability and Return, consistent with H3.Table 7 summarizes our results support

11

for H1, H2, and H3. Please note that the results in Model 3 and their support for these hypotheses

12

are consistent with those in Model 2.

13

10

[Insert Table 7 Here]

Post-hoc evaluations of the coefficients in the Scarcity array show that, whilethe

15

Scarcity10coefficientwaspositive (as expected in H1), it was the only coefficient inthis array that

16

was not statistically different from zero (p=0.13). While this result does not disconfirm H1, it

17

does provide insufficient evidence to conclude that perceived scarcity conditions among

18

consumers whose orders are placed under inventory level disclosures of 10 unitsgenerate a

19

higher risk of return than that caused when information aboutstock levels is not available.

Ac
ce
p

20

te

14

The post-hoc analyses also revealed that, according to the Scarcity variablesmeans

21

(Table 1),orders placed under inventory level disclosures ranging from 1 to 9 units and orders

22

placed under no inventory disclosurejointly account for 95% of our observations. Thus,the

23

statistical support obtained for H1 applies to a substantially large portion (95%) of our sample.

32

Page 32 of 62

Further examination of the statistically significant coefficients for the array of Scarcity

dummies (Scarcity1 through Scarcity9) in Model 3 (as well as Model 2) showed that their values

have a tendency to be inversely related to the inventory amounts disclosed to consumers at the

time of purchase. As inventory levels disclosed go down (i.e., from 9 units all the way to 1 unit),

the coefficients increase. Statistical analyses comparing the coefficients fromScarcity1 through

Scarcity9in Model 3 also show that the Scarcity1 coefficient is statistically higher than each of the

coefficients for Scarcity4 through Scarcity9. As shown in Figure 3, the 95 percent asymptotic

confidence interval for the coefficient for Scarcity1does not overlap with any of the intervals for

the coefficients for Scarcity4 through Scarcity9. These results suggest that the effect on the

an

us

cr

ip
t

likelihood of returns is statistically higher when inventory is disclosed at 1 unit at the time of

11

sale(implying a severely low inventory level) than when inventory is disclosed at 4 units or

12

higher.

These post-hoc findings complement the results in support of H1. Theyshow that online

13

10

orders in which SKUs are sold under perceived scarcity conditions in availability (specifically

15

when information shared with consumers revealsstock levels ranging from 1 to 9 units) have a

16

higher return risk than orders in which items are not sold under perceptions of scarcity conditions

17

in availability (when there is no disclosure of stock levelsto consumers). The results also show

18

that there is a statistically higher return risk when stock levels disclosedto consumers at purchase

19

comprise only1unitthanwhenthese levels are 4 units orhigher. Thus, while Scarcity and its overall

20

effect on returns is statistically higher when inventory is disclosed to consumers across different

21

inventory levels than when inventory is not shared with consumers, this effect becomes

22

statistically stronger when stock levels are disclosed at their lowest amount possible (1 unit).

23

Ac
ce
p

te

14

[Insert Figure 3 Here]

33

Page 33 of 62

Regarding H3, the results in Model 3 also reveal that, since the direct effect of Reliability
on Return remained statistically different from 0 (p < 0.01) after the addition of the interaction

term in Model 3, the moderating effect by Timeliness Expectation on the ReliabilityReturn

relationship is partial.Thus,Timeliness Expectationpartially attenuates the impact

thatReliabilityhas on Return. This result gives further credence to the manifestation of the

Assimilation-Contrast model (Anderson, 1973) that we had proposed earlier, wherein service

performance needs to go up at an accelerating rate with increasing a priori expectations (Brown

et al., 2013; Bearden and Teel, 1983; Szymanski and Heard, 2000). According to our results,

customers who hold high a priori expectations regarding delivery servicesare more likely to

an

us

cr

ip
t

engage inreturns during the coping stage, as compared to other customers, whendelivery services

11

fall short of expectations.These findings are discussed in detail in the next section.

5.Discussion of Research Findings, Contributions, Implications, and Limitations


The advent of the Internet has improved customers ability to search for and find products

te

12
13
14
15

10

that are best suited for their needs (Brynjolfssonand Smith, 2000). But, the presence of

17

sizeableproduct return rates across different retail segments on the Webis evidence thatmany

18

limitations still exist. Research has shown that the type of return policies (Wood, 2001) and the

19

typeof products offered by retailersmay affect customers decisions to return items(Rabinovich et

20

al.,2011).The implication has been that returns constitute a necessary drawback of doing

21

business on the Web, and that retailers can lower returns only if they sell the right kind of

22

products(thosethat shoppers can easily search and evaluate online) orif they make their return

23

policies more stringent in order to dissuade customers from sending products back to them in the

24

first place (Griffis et al., 2012; Rogers et al., 2002).These approaches at lowering the incidence

25

of returns, however,do not constitute a practical option for retailers who have already committed

Ac
ce
p

16

34

Page 34 of 62

to selling products outside those categories or for retailers who do not want to alienate their

customers by toughening up their return policies. The OM literature hasaddressed this issue

bydeveloping return policies that offeroptimal value to buyers and sellers (e.g.,Ketzenbergand

Zuidwijk, 2009; Su, 2009; Shulman et al., 2009)and bystudyingtheeffect on

returnsbyproductdefects, product incompatibilitywith userneeds, andflaws in product

performance(Guideetal.,2006;Fergusonet al., 206).Yet, there is a limited understanding

abouttheroleplayed bybehavioral aspects among consumers and process attributes in consumers

transactionsin returns in anonline retailsetting.

cr

us

Our study has taken a step towards addressing this shortcomingin the literature

an

ip
t

byconsidering process attributes in PDSprovided by Internet retailers as part of their transactions

11

with consumers. In so doing, it identifies how retailers can reduce the likelihood of product

12

returns based on the perceptions that they instill among consumers aboutproduct availability and

13

based on the actual performance in their execution of delivery in PDS (i.e., timeliness

14

expectation and reliability) all of which are key elements of the online retail transaction.

te

10

Our first hypothesis proposed a positive relationship between scarcity conditions in

16

availability at the time of purchase and the likelihood of product returns. Our results offer

17

support for this hypothesis. Thus, items that are purchased under conditions of perceived scarcity

18

are more likely to be returned than those that are purchased devoid of such conditions. This

19

finding has key implications for online retailers. Rabinovich et al, (2011) argue that large virtual

20

shelf spaces available in online retailing sites give sellers the flexibility to stocklarge assortments

21

of products. This often gives rise to situations where retailers maintain broad productassortments

22

but limited depth of on-hand inventory, thereby creating slim stock levels for products available

23

for sale. In those cases, the prevalence of reduced levels of inventory (when disclosed), along

24

with the scarcity heuristic, are likely to induce customers to hurry and buy products to avoid

Ac
ce
p

15

35

Page 35 of 62

missing out in case of stockouts. However, according to our results, products that are bought

under these conditions are also increasingly likely to be returned. Thus, while retailers may

indeed succeed in generatingsales bypromoting scarcity perceptions among consumers (Gierl et

al., 2008), our results suggest that many such sales will not stick and products will be returned at

a higher rate than they would have been, had such an approach not been adopted in the first

place. This studyalsoextends research in traditional retail settings that has found that,when

people buy impulsively, theyareless likely to be happy with their purchases(Gardner and Rook,

1988). Thus, online retailers must consider carefully the value of operational approaches that use

such manipulations of inventory scarcity conditions to move goods.

cr

us

an

10

ip
t

Another finding relates to the communication of product scarcity to potential shoppers.


According to our post-hocanalysis, return risks increase significantly when information is

12

disclosed to consumers to show that there is a minimal inventory amount (1 unit) available for

13

sale. Traditional retailers have long opted to use this scarcity manipulation to influence consumer

14

purchase choice, thus utilizing the scarcity heuristic to their advantage (Parker and Lehmann,

15

2011; van Nierop et al., 2008). For example, the clothing retailer American Apparelhasrecently

16

tested displays of one item per SKU in its shelves in an attempt to create impressions of scarcity

17

among consumers (Swedberg, 2012). Indeed, even in the online space, we often see the

18

implementation of this strategy. Some retailers disclose information about product availability

19

when stock levels are low (and even add statements like Hurry, only 1 item remaining!). While

20

such approaches may have some success in generating sales (Parker and Lehmann, 2011), our

21

results indicate that these strategies may also have strong effects on the risk of product returns.

22

Thus, these strategies must be carefully scrutinized by online retailerssoastonot share inventory

23

availability to consumers in a way that will generate a large amount of returns in relation to the

Ac
ce
p

te

11

36

Page 36 of 62

sales added by consumers scarcity perceptions.Themostprominentcaseof this scarcity effect on

returns, of course,isseenunderextremely low inventory levels (i.e., only 1 item) where return

likelihoodwill be greater than at higher levels of inventory disclosed.


In addition to the scarcity effect, consumers assessment of deliveryservicesin their

ip
t

transactions with Internet retailers depends on experiential qualities that willonlybe evident to

consumers when theyreceive their orders. As aresult, the performance in the execution of these

deliveries ought to have a stronginfluence on the likelihood of product returns on theInternet.

Ourpaper confirms this influenceempirically: oursecond hypothesis proposed that delivery

reliability-measuredasthe extent to which ordersaredeliveredahead of (or laterthan) promised

10

deliverytimesis related to product returns. Statistically significant evidencewas obtained in

11

support of this hypothesis. Thus, thereisevidenceto suggest thatdeliveryreliabilitydoes,in fact,

12

mitigate the risk ofproduct returns.

us

an

This findingadds another dimension to the stream of literature that demonstrates the role

te

13

cr

of order fulfillment and operations in the success of online retailers. It has been argued that the

15

two key encounter-specific dimensions of online retailing that drive customer satisfaction and

16

retention are product performance and order fulfillment performance (Reichheld and Schefter,

17

2000). For online retailers, product performance often depends on their suppliers quality of

18

manufacturing, while order fulfillment performance often depends on the operational excellence

19

of third party logistics providers.

Ac
ce
p

14

20

This is an important finding - research has demonstrated that order delivery reliability in

21

online retailing is a strong indicator of repeat purchase intentions (Boyer and Hult, 2005; Boyer

22

and Hult, 2006), and referral behavior (Griffis et al., 2012-b). In addition, it has been shown that

23

unreliable order delivery decreases customer loyalty (Rao et al., 2011). While those earlier

24

findings have value, it has also been suggested that in an online environment, customers usually
37

Page 37 of 62

have low levels of loyalty to a particular retailer to start with (Kuttner, 1998; Brynjolfsson and

Smith, 2000) and thatloyalty among online shoppers is not something that retailers should rely

upon. Thus, in addition to measuring the follow-on impacts of order delivery reliability through

customer loyalty, we propose that it would be meaningful to measure its immediate impacts. Our

results indicate that the effects of order delivery reliability directly relate to the likelihood of

closing current sales, never mind generating subsequent sales.

cr

Our third hypothesis proposedthe existence of a moderationeffectbyaprioriexpectations of

us

ip
t

order delivery timeliness on the linkbetween actual orderdeliveryreliability and thelikelihood of

product returns. Our results suggest that the effect that reliabilityin thedeliveryofordershasin

an

decreasing the likelihood of returns is stronger for orders thatinvolveexpectations offastdelivery

11

thanfororders withlessexpeditiousdeliveryexpectations.Put differently, althoughtheoccurrence of

12

returns is a result of delays in the delivery of orders relative to the initial promises made by the

13

retailers, this effect is more notable for orders that involve short a priori delivery promises.

te

It is important to note that, in testing our hypotheses, we controlled for effects on the

Ac
ce
p

14

10

15

likelihood of product returns originating from failures exogenous to PDS as well as several

16

characteristics regarding the products and customers involved in the transactions between

17

customers and Internet retailers. Our results show significant effects by product and customer

18

characteristics on the likelihood of product returns. Regarding product characteristics, we found

19

that product popularity increases the likelihood of returns. However, this likelihood is lower

20

when products are sold at prices below competition. Regarding customer characteristics, we

21

found that the likelihood of product returns is higher for transactions in which customers have

22

previously established relationships with an Internet retailer than in transactions in which brand

38

Page 38 of 62

new customers are involved. On the other hand, this likelihood is lower for transactions in which

customers are gift receivers than in transactions in which no gift receivers are involved.
It is alsoimportant to note that several coefficients for other control variables turned out

to be not significantly different from zero in our regression analysis. Two of these coefficients

correspond to those for Misinformation (B = 33.14, p>0.99), and Damage (B = 33.39, p>0.99).

These coefficients stand in contrast to the high correlation coefficients that we observed between

these same variables and Returns. A reason why the regression coefficients are not significantly

different from zero is that the number of returns causedbyMisinformation or Damage is

relatively small (a total of only 289 out of 6,732 total observations). These numbers are not

10

surprising: according to the 2012 ComScore Online shopping survey, issues associated with

11

Misinformation or Damage were not even among the top 10 concerns that customers had with

12

online shopping. Unfortunately, a downside of having a small number of observations for these

13

variables is that statistical significance is hard to obtain for their effects on returns in the results,

14

after controlling for all other determinants of returns that have a much greater incidence.

cr

us

an

te

In addition, we found that the regression coefficient for S&His not statistically significant

Ac
ce
p

15

ip
t

16

(B = 0.004, p=0.30). We contend that this is because S&H feesin our data set are rather small

17

compared to the price of the product (the average S&H fees are less than 2% of the average

18

purchase price of the product). Thus these fees by themselves may not be enough to deter

19

customers from returning their products. It is likely that in data sets where S&H fees are a higher

20

percent of product price, they may still act as a deterrent against returns.

21

Finally, we found noevidenceto suggestthat holiday purchases were more likely to be

22

returned than purchases placed during other times during the year. This result differs from that in

23

aprior studyby Petersen and Kumar(2009).We contend that this discrepancy is due to differences

24

between the data set in our study and the data in the study by Petersen and Kumar (2009).
39

Page 39 of 62

Petersen and Kumar (2009) obtained data from a catalog retailer and had several different

product categories in their sample. Based on this, they contended that the increased holiday

returns were because customers were buying products in new/different categories during the

holiday season, than they did during other times of the year. Given that our data set involved

only one product category, we are not able to isolate this effect, which may explain why, in our

study, we did not obtain a statistically significant effect (B= -0.03, p= 0.75) for this variable.

7
8
9

5.1. Literature Contributions

us

cr

ip
t

Our study is positioned at the intersection among the SCM,OM, andMarketing literatures.
In the past, research at this intersectionhas considered pre-purchasing events to highlight how

11

Internet use by consumers yields lower search costs that allow them to find better suited products

12

and lower prices (Brynjolfsson and Smith, 2000). We expanded on these phenomena by going

13

past the initial pre-purchasing events to include returns. As a result, we can better explain why,

14

despite the existence of lower search costs (and most likely better search results) on the Internet,

15

consumers are still compelled to send back products purchased online.

te

an

10

This study also underscores the synergies accrued when OMand Marketing functions are

17

well integrated and the problems encountered when they are not. Researchers have long asserted

18

the value of linking business processes across functionsto achieve world-class supply chain

19

performance (Sawhney and Piper, 2002; Malhotra and Sharma, 2002; Hausman et al., 2002). Yet

20

few empirical studies have investigated the impact of one function on the other. As we have

21

shown, the OMfunction should not be viewed merely as a cost center, but rather as a center of

22

revenue creation with direct influence in the marketing function. This is critical in online

23

retailing, where competition is intense and firms are subject to a great deal of pressure to find

24

ways to raise revenues through innovative operations (Laseter and Rabinovich, 2011).

Ac
ce
p

16

40

Page 40 of 62

Forinstance, our study demonstrates thatpoorly performingOM practices (by way of

decreased order delivery reliability) can be problematicformarketing. While prior research has

shown that delivery glitches can curtail lifetime customer value for a retailer (Rao et al., 2011),

our study shows that these failures can decrease immediate value as well by increasingreturn

rates.

Moreover, ourfindings offer two extensions to Lazarus (1991) original theory of

cr

ip
t

exchange assessment. First, ourfindingsdemonstratethat elements within the appraisal and

response stagescan directly impact each other in addition to the outcomes in the coping stage.

This extends the work by Gotlieb et al. (1994) who first proposed that the appraisal of an

an

us

exchange may involve a sequence of evaluations. Second,our researchgoes beyond Lazarus

11

(1991) original focus on customers coping responsestoconsider the formation of customers

12

behavioral intentionsand demonstratesthat emotional responses to an exchange (coping) can

13

directlytrigger changes in actual behavior (product returns). In so doing, our studyoffers

14

advancements beyond theoriginal conceptualizations and subsequent empirical evaluations in

15

retailing and electronic commerce (e.g., Cronin et al. 2000; Griffis et al, 2012).

Ac
ce
p

te

10

16

Finally, we have extended the prior literature by incorporating principles from the

17

Assimilation-Contrast Theory and the ECT into the PDS evaluation framework in the context of

18

the exchange assessment (Lazarus, 1991) to show that customer expectations of order delivery

19

timeliness moderate negatively the effect by actual order delivery reliability on product returns.

20

While researchers have assumed that delivery reliability has a uniform impact on customer

21

satisfaction (Boyer and Hult, 2005 and 2006; Griffis et al., 2012), we find that, when product

22

returns are considered, this effect actuallydependson the initial delivery timeliness expectations

23

signaled to buyers:as the levels of these expectations increase, the delivery reliability provided to

24

customers should improve at a greater rate to prevent the return ofpreviouslypurchased items.
41

Page 41 of 62

1
2
3

5.2. Managerial Contributions

products will get returned. This is a key issue for managers - there is growing consensus in the

industry thatretailers need to do a better job in managing product returns if they are to ensure

profitable growth. According to a study, nearly 63 % of shoppers payattention to online retailers

return policieswhen making purchasing decisions (ComScore, 2012). As such then, returns are a

major issue for online retailers, with many managers just accepting them as a cost of business.

ip
t

cr

us

Our results offer retailers practical insightsto managePDSand curb thelikelihood that their

Yet, as we have demonstrated, there are simple actions that retailers can take, to prevent
returns. These include a careful assessment of the wisdom of revealing inventory level

11

information to consumers prior to purchase, especially when stocks are running at their lowest.

12

While the disclosure of low inventory levels may be attractive from the traditional viewpoint

13

held in the OM literature regarding the benefits of accelerating inventory turns by playing on

14

scarcity fears among customers, it is a double edged sword that can also contribute to create

15

unnecessary product returns. This is because these scarcity perceptions can artificially inflate the

16

value of products in the eyes of customers to significantly exceed the actual value that customers

17

will attain once they have bought and received theseproducts.

Ac
ce
p

te

an

10

18

Our finding that the disclosure of inventory levels can contribute to generate scarcity

19

perceptions among customers and that these perceptions influencestronglythe risk of returns

20

should also serve as a motivation to deemphasize the use of centralized product recovery

21

networks in favor of decentralized systems. While centralized networks are highly cost-effective

22

in supporting the inspection and refurbishment of defective items, they do not offer the means to

23

rapidly capture returnscaused byscarcityperceptions, in which products present no flaws (Guide

24

et al. 2006). Retailers thatrely on the disclosure of stock level information to promote perceptions

42

Page 42 of 62

of scarcity among consumersmay,therefore, benefit from having access to decentralized

networks, provided commonly by thirdpartyproviders,torapidlyprocessthistypeoffalsereturns.


It is alsoimportant to note that wearenot proposingthat the disclosureofstock levels should

always be avoided. Whilesuchdisclosure can increasereturns, itis conceivablethatit mayalso

increasedemandforsome productsto amuch greater extent. Moreover,thedisclosure

ofinventorylevelsis somethingthatretailers couldembrace if profitmargins from products arehigh

enough tooffsetreturn costs.In addition, thedisclosureof stocklevels can be important for customer

service, particularly in cases involving products that are sold in multiple

units(suchashomeimprovement supplies) andfor which customers wouldwant to see available on-

10

hand inventories beforemakingtheirpurchases2. Thus, in addition totheir potential to boost sales,

11

the disclosureof inventorylevelsmaybeworthwhile, even in the faceof increases in returns, if

12

theprofit marginsforthe products involved arehighenough to absorb the cost ofthe added returns

13

or if customer servicemakes itnecessaryto sharesuchinformation with shoppers.

te

an

us

cr

ip
t

Our results alsodemonstrate that an increasingly reliable delivery, in accordance with

14

evermore demanding customer expectations, is critical in preventing incidents of productreturns.

16

There are several actions that online retailers can take to address this issue.For example, research

17

has revealed that one of the reasons for fulfillmentfailures among retailers is that there is often a

18

mismatch between the retailers actual inventory levels on hand and the inventory level data

19

onrecord (Fleisch and Telkamp, 2005; Delen et al., 2007). Indeed, factors such as product theft,

20

deterioration, and misplacement have the potential to create a mismatch between the stock-

21

leveldatathat a retailer has on-recordversusnumber of units onhand. This mismatchcan

22

underminereliability in order fulfillment. This problem becomes even more acute when on-hand

23

inventory levels are low to start with. In this case, if a retailer promises a customer a particular

Ac
ce
p

15

We thank a reviewer for suggesting this case

43

Page 43 of 62

level of delivery timeliness, and fails to deliver owingto amismatch between on-handand on-

recordinventory, delivery reliability will be affected. This, as we have demonstrated, has a

potential to lead to returns. Similarly, given the effect of expected delivery timeliness and

reliability on returns, our findings also highlight the importance of efficient interfacing between

the online retailer and third party logistics providers in charge of last-mile delivery. This further

demonstrates the value of efficient linkage of business functions across companies (Hausman et

al., 2002; Malhotra and Sharma, 2002).

cr

us

Our results also point to hidden drawbacksinvolved in efforts by Internet retailers to


enhancethe value of their services in the eyes of consumers. Since most online retailers have

an

ip
t

been facing increasingly smaller poolsof available first-time customers, a number of them have

11

shifted their efforts to capturea higher share-of-wallet from existing customers byupgrading,atno

12

charge,the PDS that they offer. A stronger attempt is thus being made to snag a higher financial

13

commitment from customers during the appraisal stage of the appraisal response coping

14

cycle itself. Our results should givethese retailers pause: theymust ensure that they are prepared

15

to back up their free upgrades with processes that are free of glitches to avoid poor service

16

experiencesthat could easily result in product returns during the coping phase.

te

Ac
ce
p

17

10

Finally, according to our results, PDS and pricing competencies should be tied in with an

18

appreciation for the various types of customers that retailers serve in order to improve

19

theirmanagement of product returns. We find, in particular, that new customers and customers

20

who receive gifts are less likely to engage in product returns relative to other customers.

21

Different risks of returns according to the type of customers involved in the transactions elevate

22

the need to tailor retailers PDS and prices in favor of those customers who have had long

23

relationships with the retailers and who are engaged in the purchase of products exclusively for

24

themselves.
44

Page 44 of 62

5.3. Limitations and Future Research Opportunities


As with all empirical studies, ours presents several limitations thatmay offer avenues for

future research. First, our studyfocused exclusively on returns, which may be seen as an extreme

case of customer dissatisfaction. Yet, it is possible that some customers may be highly

dissatisfied with their products but choose to not actually return them. Therefore, it is possible,

for example, that the effects ofPDS attributes on returns in our studymay have underestimated

the extent of dissatisfaction among customers. Future research may examine this issue.

cr

us

Our study also focused on a single retailer in order to control for potential biases in the
occurrence of product returns brought about by different return policies. As we explained earlier,

an

ip
t

this retailer had in place a fairly restrictive return policy, relative to those found among other

11

retailers. The stringency of our retailers policy could havecurtailed the incidence of returns

12

linked to PDS, making the testing of our hypotheses exceedingly demanding. Future work may

13

consider testing our hypotheses with the support of retailers that have more liberal return policies

14

in order to ascertain thesignificanceof our results in a setting inwhich returnsare morecommon.

15

Specifically, under this new setting, theeffect inH1 maytakeon agreaterstatisticalsignificance

16

dueto the easein whichcustomers maybe ableto return items boughtafterexperiencing conditions

17

of perceived scarcityin availabilityat the appraisal stage oftheir purchases.

te

Ac
ce
p

18

10

Future research may also extend our work by considering a wider range of products. Our

19

focus on only one type of products (personal accessories) enabled us to control for spurious

20

effects on returns caused by different product classifications. Typically, for personal accessories,

21

returns are driven by the pricing, quality, and overall look and feel of the products, all of which

22

are important in the appraisalstageoftheexchange framework,justas much as thePDS issues at the

23

core of our model. Thus, we contend that our setting provided for more robust tests of our

24

hypotheses than other settings in which returnsare not dependent on as manydeterminants. And
45

Page 45 of 62

while this contributes tothe validity of our results, the study of other settings could add further

insights by showing how the involvement of different product categories can strengthen our

results. We anticipate, for instance, that the effects in our empirical model will be stronger in

online retail settings where groceriesare involved because, in the minds of consumers, PDS is a

much more significant quality factor for these product transactions on the Internet than it is for

fashion intensive products such as ours.Moreover, theeffects ofscarcityconditionsinavailability

on the likelihood of returns for theonline grocery industry may differ from those in our study due

to the distinct patterns that exist among shoppers and SKUs in online grocery retailing.

us

cr

ip
t

6. Conclusion

12

business modelthat has gained a significant amount of attention over the past several years.

13

Nevertheless, several problems have continued to challenge managers in this industry product

14

returns being one of them. Many online retailers now compete not only with other web-based

15

counterparts, but also with traditional retailers, who are usually in a better position to control,

16

manage, and process returns. Because of this, and the fact that online retailing usually involves

17

remote purchase environments where customers do not have the opportunity to examine the

18

products they buy, online retailers are usually exposed to higher return rates than traditional

19

retailers. This has been recognized as an issue by several managers, who often view it as

20

something that they simply cannot control. The prevailing belief among managers has been that

21

returns are an inevitable part of the nature of the online retail environment itself.

22

an

9
10
11

Ac
ce
p

te

Online retailing offers a unique way of getting products to end customers, and is a

While there may be some truth to that, as our study has demonstrated, there are at least

23

some aspects of the online purchasing process that managers can in fact address, and which will

24

at least have some effect on returns. As we have shown, a purposeful management of the PDS

25

process, including availability, expected delivery timeliness, and reliability, can contribute to
46

Page 46 of 62

control the occurrence of returns. This is important because managers can effectively track PDS

activities through the use of objective measurements, which, as we demonstrate, should impact

the extent of returns that they receive.

ip
t

4
5

Ac
ce
p

te

an

us

cr

47

Page 47 of 62

References

3
4
5

Aguinis, H., R. K. Gottfredson, S. A. Culpepper, 2013. Best practice recommendations for


estimating cross-level interaction effects u sing multilevel modeling. Journal of Management, 39
(6), 1490-1528.

6
7
8

Alba, J., J. G. Lynch, B. Weitz, C. Janiszewski, R. Lutz, S.Wood, 1997. Interactive home
shopping: Consumer, retailer, and manufacturer incentives to participate in electronic
marketplaces. Journal of Marketing 61 (3), 3853.

cr

us

Anderson, R. E. 1973. Consumer dissatisfaction: The effect of disconfirmed expectancy on


perceived product performance. Journal of Marketing Research 10 (1), 3844.

an

Anderson, E. T., K. Hansen, D. Simester. 2009. The option value of returns: Theory and
empirical evidence. Marketing Science 28 (3), 405423.
Areni, C. S., P. Kiecher, K. M. Palan, 1998. Is it better to give than to receive? Exploring gender
differences in the meaning of memorable gifts. Psychology and Marketing 15 (1), 81-109.

Atasu, A., M. Sarvary, L. N. van Wassenhove, 2008. Remanufacturing as a marketing strategy.


Management Science, 54(9), 17311746.

Bagozzi, R.P. 1992. The self- regulation of attitudes, intentions, and behavior. Social Psychology
Quarterly 55(2), 178-204.

te

Bearden, W.O., J.E. Teel, 1983. Selected determinants of customer satisfaction and complaint
reports. Journal of Marketing Research, 20 (1), 21-28.
Berry, L.L., A. Parasuraman, 1991. Marketing services. Free Press, New York, NY.

Ac
ce
p

9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47

ip
t

Berry, L.L., A. Parasuraman, V. A. Zeithaml, 1994. Improving service quality in America:


lessons learned. Academy of Management Executive 8 (2), 3252.
Bienstock, C.C., T. J. Mentzer, M. M. Bird, 1997. Measuring physical distribution service
quality. Journal of Academy of Marketing Science 25 (1), 31-44.
Bitner, M. J., 1990. Evaluating service encounters: the effects of physical surroundings and
employee responses. Journal of Marketing 54(2), 69-82.
Bitner, M. J., B. H. Booms, M. S. Tetreault, 1990. The service encounter: Diagnosing favorable
and unfavorable. Journal of Marketing, 54(1), 71-84.
Bitner M. J., A. R. Hubbert, 1994. Encounter satisfaction versus overall satisfaction versus
quality: The customers voice. In: Rust RT, Oliver RL, editors. Service quality: new directions in
theory and practice. Thousand Oaks, CA, Sage Publications, 72-94.
Bolton, R. N., 1998. A dynamic model of the duration of the customers relationship with a
continuous service provider: The role of satisfaction. Marketing Science 17 (1), 4565.

48

Page 48 of 62

Boulding, W., A. Kalra, R. Staelin, V. A. Zeithaml, 1993. A dynamic process model of service
quality from expectation. Journal of Marketing Research 30(1), 7-27.
Bowersox, D.J., D.J. Closs, 1996. Logistical Management. McGraw-Hill, New York, NY.
Boyer, K.K., T.G. Hult, 2005. Extending the supply chain Integrating operations and marketing
in the grocery industry. Journal of Operations Management 23(6), 642661.

ip
t

Boyer, K.K., T.G. Hult, 2006. Customer behavioral intentions for online purchases: An
examination of fulfillment method and customer experience level. Journal of Operations
Management 23(6), 642661.

us

cr

Boyer, K.K., A.M. Prudhomme, C. Wenming, 2009. The last mile challenge: Evaluating the
effects of customer density and delivery window patterns. Journal of Business Logistics 30 (1),
185-201.
Brehm, W., 1966. A theory of psychological reactance. New York: Academic Press.

an

Brown, S., V.Venkatesh, S. Goyal, 2013. Expectation confirmation in technology use.


Information Systems Research (Articles in Advance) 1 - 14.

Brynjolfsson, E., M. Smith, 2000. Frictionless commerce? A comparison of internet and


conventional retailers. Management Science 46 (4), 563585.
Chebat, J.C., D. M. Davidow, and I. Codjovi, 2005. Silent voices: why some dissatisfied
consumers fail to complain. Journal of Service Research 7 (4), 328-342.

Cialdini, R. B. 2001. Influence: Science and practice (4th Ed.). Boston: Allyn & Bacon.

te

Cohen, J., 1968. Multiple regression as a general data-analytic system. Psychological


Bulletin 70 (6 Pt. 1), 426443, http://dx.doi.org/10.1037/h0026714
Cohen, J., P. Cohen, (1983). Applied multiple regression/correlation analysis for the behavioral
sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Ac
ce
p

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

ComScore 2012. Improving product returns presents biggest opportunity for retailers wishing to
increase online shopping satisfaction.
http://www.comscore.com/Insights/Press_Releases/2012/6/. Accessed May 10 2013
Cronin, J.J., S.A. Taylor, 1992. Measuring service quality: A re-examination and extension.
Journal of Marketing 56(3), 55-68.
Cronin, J.J., M. K. Brady, G. T. M. Hult, 2000. Assessing the effects of quality, value, and
customer satisfaction on consumer behavioral intentions in service environments. Journal of
Retailing 76(2), 193-218.
Daugherty, P., G. Ritchie, B. J. Hudgens, C. W. Autry, 2003. Reverse logistics in the automobile
aftermarket industry. International Journal of Logistics Management, 14 (1), 49-62.
Davison, M., N. Kwak, Y. S. Seo, J. Choi. Using hierarchical regression to examine moderator
effects: Person by organization interactions. Organizational Research Methods 5 (5), 231-254.

49

Page 49 of 62

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

Delen, D., B. C. Hardgrave, R. Sharda, 2007. RFID for better supply chain management through
enhanced information visibility. Production and Operations Management 16 (5), 613-624.

28
29

Fleisch, E., C. Telkamp. 2005. Inventory inaccuracy and supply chain performance: A simulation
study of a retail supply chain. International Journal of Production Economics 95 (3), 373385.

30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

Folkes, V. S., 1984. Consumer reactions to product failure: An attributional approach. Journal of
Consumer Research, 10(4), 398-409.

Demery, P., 2010. Return policies can win over customers, and theres more than one path to
success. Internet Retailer-April 2010. Available at
http://www.internetretailer.com/2010/03/31/get-back. Accessed May 9, 2013.

ip
t

Ditto, P. H., J. B. Jemmott, 1989. From rarity to evaluative extremity: Effects of prevalence
information on evaluations of positive and negative characteristics. Journal of Personality and
Social Psychology 57 (1), 16-26.

cr

Dixon, M., R. Verma 2013. Sequence effects in service bundles: implications for service
Design and Scheduling. Journal of Operations Management, 31(3), 138152.

an

us

Edwards, J. R. 2002. Alternatives to difference scores: Polynomial regression analysis and


response surface methodology, in Measuring and analyzing behavior in organizations:
Advances in measurement and data analysis, F. Drasgow and N. Schmidt (eds.), San Francisco:
Jossey-Bass/Pfeiffer, 350-400.
Eroglu S. A., K. A.Machleit, L. M. Davis, 2003. Empirical testing of a model of online
store atmospherics and shopper responses, Psychology & Marketing 20 (2), 139-150.

Esper, T., T. Jensen, F. Turnipseed, S. Burton, 2003. The last mile: An examination of effects of
online retail delivery strategies on consumers. Journal of Business Logistics 24 (2), 177-203.

Ac
ce
p

te

Ferguson, M., V. D. R. Guide, Jr., G. Souza. 2006. Supply chain coordination for false failure
returns. Manufacturing and Service Operations Management 8(4), 376393.

Folkes, V. S., Koletsky, S., Graham, J. L., 1987. A field study of causal inferences and consumer
reaction: The view from the airport. Journal of Consumer Research, 13(4), 534-539.
Gardner, M. P., D. W. Rook, 1988. Effects of impulse purchases on consumers affective states.
In M. J. Houston (Ed.), Advances in Consumer ResearchVol.15, 127-130. Association for
Consumer Research, Provo, UT.
Gierl, H., M. Plantsch, J. Schweidler, 2008. Scarcity effects on sales volume in retail. The
International Review of Retail, Distribution and Consumer Research 18 (1), 45-61.
Gilmour, P., 1977. Customer service: differentiating by market segment. International Journal of
Physical Distribution and Logistics Management 7 (3), 141148.
Gotlieb, J.B., D. Grewal, S. W. Brown, 1994, Consumer satisfaction and perceived quality:
Complementary or divergent constructs? Journal of Applied Psychology 79 (6), 875-885.
Greene, W.H. 1999. Econometric Analysis, 4th Edition. New York: Prentice Hall.

50

Page 50 of 62

Griffis, S. E., Rao, S., Goldsby, T. J., and Niranjan, T. T., 2012-a. The customer consequences of
returns in online retailing: an empirical analysis. Journal of Operations Management 30 (4), 282294.

ip
t

Griffis, S. E., S. S. Rao, T. J. Goldsby, C. Voorhees, D. Iyengar, 2012-b. Linking order


fulfillment performance to referrals in online retailing: An empirical analysis. Journal of
Business Logistics 33 (4), 279 294.
Guide, Jr., V. D. R., K. Li. 2010. The potential for cannibalization of new product sales by
remanufactured products. Decision Sciences 41 (3), 547-572.

cr

Guide, V. D. R., L. N. van Wassenhove. 2009. The evolution of closed-loop supply chain
research. Operations Research 57(1), 10-18.

us

Guide Jr., V.D.R., L.N. van Wassenhove, 2006. Closed loop supply chains: An introduction to
the feature issue (part 1). Production and Operations Management 15 (3), 345-350.

an

Guide, Jr., V. D. R., G. Souza, L. N. Van Wassenhove, J. D. Blackburn. 2006. Time value of
commercial product returns. Management Science 52(8), 12001214.

Harris, K. E., D. Grewal, L. Mohr, K. L. Bernhardt, 2006. Customer response to service


recovery strategies The moderating role of online versus offline environment. Journal of
Business Research 59 (4), 425 431.

Hausman, W., D. B. Montgomery, A. Roth, 2002. Why should Marketing and Manufacturing
work together? Some empirical evidence. Journal of Operations Management 20 (3) 241 257.

te

Heim, G. R., J. M. Field, 2007. Process drivers of e-service quality: Analysis of data from an
online rating site. Journal of Operations Management 25 (5), 962-984.
Hosmer, D.W., S. Lemeshow, 2000. Applied Logistic Regression. Wiley, 2nd Ed., Hoboken, NJ

Ac
ce
p

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

Hutchinson, W.M., J. F. Stolle, 1968. How to manage customer service. Harvard Business
Review 46 (6), 8596.
Humphreys, M., M. Williams, 1996. Exploring the relative effects of salesperson interpersonal
process attributes and technical product attributes on customer satisfaction. Journal of Personal
Selling and Sales Management 16 (3), 45-57.
Inman, J. J., A. C. Peter, P. Raghubir, 1997. Framing the deal: The role of restrictions in
accentuating deal value. Journal of Consumer Research 24 (3), 68-79.
Jacoby, J. 2002. Stimulus-organism-response reconsidered: An evolutionary step in modelling
consumer behavior. Journal of Consumer Psychology. 12 (1), 51-57.
Kelley S. W., D. K. Hoffman, M. A. Davis, 1993. A typology of retail failures and recoveries.
Journal of Retailing 69(4), 429-452.
Ketzenberg, M. E., G. C. Souza, V. D. R. Guide Jr., 2003. Mixed assembly and disassembly
operations for remanufacturing. Production and Operations Management, 12(3), 320335.

51

Page 51 of 62

Ketzenberg, M., R. Zuidwijk, 2009. Optimal ordering, pricing, and return policies for consumer
goods. Production and Operations Management, 18(3), 344-360.
Kuttner,R.1998.Thenet:amarkettooperfectfor profits.BusinessWeek,3577(May 11),20.

ip
t

Laseter, T., E. Rabinovich, 2011. Internet Retail Operations: Integrating Theory and Practice for
Managers (Supply Chain Integration Modeling, Optimization and Application. Taylor and
Francis, CRC Press, Boca Raton, FL.
Lazarus, R. S. 1991. Emotion and Adaptation, London: Oxford University Press.

us

cr

Luts, J., J.A. Roldan Nofuentes, J. Lunadel Castillo, S. VanHuffel 2011.


Asymptotichypothesistesttocomparelikelihoodratiosof multiplediagnostictestsinunpaireddesigns.
Journal of Statistical Planning and Inference 141, 3578-3594.
Lynn, M. 1989. Scarcity effects on desirability: Mediated by assumed expensiveness?
Journal of Economic Psychology 10 (2) 257-274.

an

Lynn, M. 1991. Scarcity effects on desirability: A quantitative review of the commodity theory
literature. Psychology & Marketing 8 (1), 43-57.

Malhotra, M., S. Sharma, 2002. Spanning the continuum between Marketing and Operations.
Journal of Operations Management 20 (3) 209 219.

Marshall, G. W., D. Goebel, W. C. Moncrief, 2003. Hiring for success at the buyer-seller
interface. Journal of Business Research 56 (4), 247 255.

te

Mollenkopf, D., R. Frankel, I. Russo, 2011. Creating value through returns management:
Exploring theMarketing-Operations interface.Journal ofOperationsManagement29(5),391-403.
Mollenkopf, D., E. Rabinovich, T. M. Laseter, K. K. Boyer, 2007. Managing internet product
returns: A focus on effective service operations. Decision Sciences 38 (2), 215-250.

Ac
ce
p

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53

Murray, M.P. 2005. Econometrics: A Modern Introduction. New York: Prentice Hall
Nasr-Bechwati, N. I., W. S. Siegal, 2005. The impact of the prechoice process on product
returns. Journal of Marketing Research 42 (3), 358-367.
Nyer, P. 1997. A study of relationships between cognitive appraisals and consumption
emotions. Journal of the Academy of Marketing Science 25 (4), 296-304.
Ofek, E., Katona, Z., M., Sarvary, 2011. The impacts of product returns on the strategies of
multi-channel retailers. Marketing Science 30 (1), 42 60.
Oliver R.L., 1980. A cognitive model of the antecedents and consequences of satisfaction
decisions. Journal of Marketing Research 17, 460-469.
Oliver, R.L. 1981. Measurement and evaluation of satisfaction processes in retail settings.
Journal of Retailing 57 (3), 25-48.
Oliver, R.L. 1997. Satisfaction: A behavioral perspective on the consumer. New York, NY:
Irwin/McGraw-Hill.

52

Page 52 of 62

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

Oliver, R. L., P.V. Balakrishnan, B. Barry, 1994. Outcome satisfaction in negotiation: A test of
expectancy disconfirmation, Organizational Behavior and Human Decision Processes 60 (2),
252-275.

20
21

Parker, J.D., D.R Lehmann, 2011. When shelf-based scarcity impacts consumer preferences.
Journal of Retailing. 87 (2), 142-155.

22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

Peck, J., T.L. Childers, 2003. To have and to hold: The influence of haptic information on
product judgments. Journal of Marketing, 67 (2), 35-48.

38
39
40

Rabinovich, E., M. Rungtusanatham, T.M. Laseter, 2008. Physical distribution service


performance and Internet retailer margins: The drop-shipping context. Journal of Operations
Management 26 (6), 767-780.

41
42

Rabinovich, E., R. Sinha, T. Laseter, 2011. Unlimited shelf space in internet supply chains:
Treasure trove or wasteland? Journal of Operations Management 29 (4), 305-317.

43
44
45
46

Rao, S. S., S. E. Griffis, T. J. Goldsby, 2011. Linking online order fulfillment glitches with
customer purchase patterns An empirical analysis, Journal of Operations Management 29 (7),
693-703.

Oliver, R. L., W. S. DeSarbo, 1988. Response determinants in satisfaction judgments. Journal of


Consumer Research 14 (4), 495-507.

ip
t

Olson, J.R., K. K. Boyer, 2003. Factors influencing the utilization of Internet purchasing in small
organizations. Journal of Operations Management 21 (2), 225245.

cr

Olson, J. C., P. A. Dover, 1979. Disconfirmation of consumer expectations through product trial,
Journal of Applied Psychology 64 (2), 179-189

us

Padmanabhan, V., I. P. Png, 1997. Manufacturers retail policies and retail competition.
Marketing Science 16 (1) 81 94.

an

Parasuraman, A., V. A. Zeithaml, L. L. Berry, 1994. Reassessment of expectations as a


comparison standard in measuring service quality: Implications for further research. Journal of
Marketing 58 (1), 111-124.

Perreault Jr., W.D., F. A. Russ, 1976. Physical distribution service in industrial purchase
decisions. Journal of Marketing 40 (4), 310.

te

Petersen, J. A., V. Kumar, 2009. Are product returns a necessary evil? Antecedents and
consequences. Journal of Marketing 73 (3), 35-51.

Ac
ce
p

Rabinovich, E., J.P.Bailey, 2004. Physical distribution service quality in internet retailing:
Service pricing, transaction attributes, and firm attributes. Journal of Operations Management 21
(6), 651-672.
Rabinovich, E., M. Knemeyer, C. Mayer, 2007. Why do Internet commerce firms incorporate
logistics service providers in their distribution channels? The role of transaction costs and
network strength. Journal of Operations Management 25(3), 661-681.

53

Page 53 of 62

Rao, S.S., D. Iyengar, T. J. Goldsby, 2009. The marketing and logistics efficacy of online sales
channels.InternationalJournalof PhysicalDistributionandLogisticsManagement39(2),106-130.
Reichheld, F., 1996. The loyalty effect: The hidden force behind growth, profits, and
lastingValue, Boston: Harvard Business School Press.

ip
t

Reichheld, F.F., Schefter, 2000. E-Loyalty: Your secret weapon on the web, Harvard Business
Review, 78(4) 105-113.

cr

Reinartz, W., Kumar, V., 2002. The mismanagement of customer loyalty. Harvard Business
Review, 80 (7), 86-94.

us

Rogers, D.S., D.M. Lambert, K.L. Croxton, S.J. Garcia-Dastugue, 2002. The returns
management process. International Journal of Logistics Management 13 (2), 1-18.

an

Sawhney, R., C. Piper, 2002. Value creation through enriched marketingoperations interfaces:
an empirical study in the printed circuit board industry. Journal of Operations Management
20 (3), 259-272.

Shulman, J., A. Coughlan, R. Savaskan, 2009. Optimal restocking fees and information provision
in an integrated demand-supply model of product returns. Manufacturing and Service
Operations Management 11 (4), 577-594.
Shulman, J., A. Coughlan, R. Savaskan, 2011. Managing consumer returns in a competitive
environment, Management Science, 57 (2), 347-362.

te

Sherry, J. F., 1983. Gift giving in anthropological perspective. Journal of Consumer Research
10 (2), 157-168.
Sheth, J., B. Newman, B. Gross, 1991. Consumption values and market choices, theory and
applications. South-Western Publishing Co, Cincinnati, USA.

Ac
ce
p

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

Smith A.K., R. N. Bolton, 1998, An experimental investigation of customer reactions to service


failures: Paradox or peril? Journal of Service Research 1 (1), 65-81.
Smith A. K., R. N. Bolton, J. Wagner, 1999. A model of customer satisfaction with service
encounters involving failure and recovery. Journal of Marketing Research 36 (3), 356 72.
Souza, G. C., M. E. Ketzenberg, V. D. R. Guide Jr., 2002. Capacitated remanufacturing with
service level constraints. Production and Operations Management, 11(2), 231248.
Srinivasan, S.S., R. Anderson, K. Ponnavolu, 2002. Consumer loyalty in E-commerce:
an exploration of its antecedents and consequences. Journal of Retailing 78 (1), 4150.
Staples, D. S., I. Wong, P. B. Seddon, 2001. Having expectations of information systems benefits
that match received benefits: Does it really matter? Information and Management 40 (2), 115131.
Stock, J., T. Speh, H. Shear, 2006. Managing product returns for competitive advantage. MIT
Sloan Management Review 48 (1), 57-62.

54

Page 54 of 62

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

Su, X. 2009. Consumer returns policies and supply chain performance, Manufacturing & Service
Operations Management 11 (4), 595-612.

24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

Venkatesh, V., S. Goyal, 2010. Expectation disconfirmation and technology adoption:


Polynomial modeling and response surface analysis. MIS Quarterly 34 (2), 281-303.

Susarla, A., A. Barua, A. B. Whinston, 2003. Understanding the service component of


application service provision: An empirical analysis of satisfaction with ASP services, MIS
Quarterly 27 (1), 91-123.

ip
t

Swedberg,C.2012.AmericanApparel adopting RFID at every store.RFIDJournal, Feb 8, 2012.


Available online at http://www.rfidjournal.com/articles/view?9202. Accessed May 10, 2013.

cr

Szymanski, D. M., D. H. Heard., 2000. E-satisfaction: An initial examination. Journal of


Retailing 76 (3), 309322.

us

Szajna, B., R. W. Scamell, 1993. The effects of information system user expectations on their
performance and perceptions, MIS Quarterly17 (4), 493-516.

an

Thirumalai, S., K.K. Sinha, 2005. Customer satisfaction with order fulfillment in retail supply
chains: Implications of product type in electronic B2C transactions. Journal of Operations
Management 23 (3-4), 291-303.

van Nierop, E., D. Fok, P. H. Franses, 2008. Interaction between shelf layout and marketing
effectiveness and its impact on optimizing shelf arrangements. Marketing Science 27(6), 10651082.

te

Verhallen, T.M.M., 1982. Scarcity and consumer choice behavior. Journal of Economic
Psychology 2 (4), 299-321.

Ac
ce
p

Verhallen, T.M.M., H. S. J. Robben, 1994. Scarcity and preference: An experiment on


unavailability and product evaluation. Journal of Economic Psychology 15 (2), 315-331.
Vorasayan, J., S. M. Ryan, 2006. Optimal price and quantity of refurbished products. Production
and Operations Management, 15(3), 369383.
Weiner, B., 1985. Human motivation, New York: Springer-Verlag.
Weiner, B., 2000. Attributional thoughts about consumer behavior. Journal of Consumer
Research, 27(3), 382-388.
Wood, S. 2001. Remote purchase environments: The influence of return policy leniency on twostage decision processes. Journal Marketing Research 38 (2), 157169.
Zeithaml, V.A. 2000. Service quality, profitability, and the economic worth of customers: What
we know and what we need to learn. Journal of the Academy of Marketing Science 28 (1),67-85.
Zeithaml, V.A., L. L. Berry, A. Parasuraman, 1996. The behavioral consequences of service
quality. Journal of Marketing 60 (2), 31-46.

49

55

Page 55 of 62

1
2

Figure 1: Theoretical Model


3

Stages in the Online Retail Exchange


Response Stage

Coping Stage

H2: (-)

Order Delivery
Reliability

ip
t

H1: (+)

Product Return
Likelihood

us

H3 (+):
(Attenuation)

an

Expected Order
Delivery Timeliness

Figure 2: ROC Curves for the 3 Models

7
8
9

Ac
ce
p

te

10
11

4
5

Scarcity Condition in
Availability

cr

Appraisal Stage

12
56
Page 56 of 62

Figure 3: Post-Hoc Statistical Comparison of Confidence Intervals for Scarcity Variable Coefficients

1.990

1.568

1.570

cr

1.500

1.177

an

0.504

0.913
0.758

0.614
0.500

1.047

0.949

0.000

4
5
6

Scarcity 2

Scarcity 3

Scarcity 4

Scarcity 5

0.231

Scarcity 6

Scarcity 7

0.119
Scarcity 8

0.064
Scarcity 9

Ac
ce
p

Scarcity 1

te

0.309
0.131

95% Confidence Interval for Scarcity


through Scarcity 9Coefficients

0.997

1.000

us

1.245

0.524

95% Confidence Interval


for Scarcity 1 Coefficient

2.000

ip
t

2
3

57
Page 57 of 62

1
2TTable 1: Operationalization of Variables and Descriptive Statistics
Std.
Dev.

Return

0.15

0.35

Scarcity1

0.02

0.15

Scarcity2

0.02

0.12

Scarcity3

0.02

0.13

Scarcity4

0.04

0.20

Scarcity5

0.05

0.21

Scarcity6

0.04

0.20

Scarcity7

0.02

0.15

Scarcity8

0.04

0.18

Scarcity9

0.05

0.21

Scarcity10

0.05

0.23

Scarcity11

0.65

0.48

Continuous
(Percentage)

Reliability

0.01

0.43

Continuous
(Number of days promised
for delivery)

Timeliness
Expectation

4.27

2.59

Binary Dummy
(1=Yes, 0 = No)

Misinformation

0.04

0.19

Binary Dummy
(1=Yes, 0 = No)

Damage

0.003

0.06

Binary Dummy
(0=Yes, 1 = No)

Price Lead

0.18

0.39

Continuous

Price

644.13

1334.1

The orders S&H fees


Did the customer name and address exist
in the database prior to the customer order
in the studys dataset?
Did the billing name and address match
the shipping name and address?
Was the purchase made in the holiday
season (November / December)?

Continuous

S&H

12.83

12.55

New Customer

0.15

0.36

Gift

0.16

0.37

Holiday

0.43

0.5

Ln of annual product sales

Continuous

Popularity

16.71

21.86

Scarcity conditions are communicated to


customers depending on whether there
were 10 or less units in stock for a
product when customers placed their
orders

Flawed Product
Description
Damaged /
Broken Goods
Product Price
Leadership
Product Price
S&H Price
Purchase History
Gifting
Seasonality
Product
Popularity

te

Expected Order
Delivery
Timeliness

The extent to which the order was


delivered earlier than promised (measured
in percentage terms)
The promised order delivery speed (based
on the delivery commitment provided
during purchase)
Was the product information provided
prior to purchase, deemed to be incorrect
by the customer?
Was the product deemed to be broken or
defective by the customer upon delivery?
Was the item available for a lower price
on the purchase date at a competing
retailer?
Price paid for the product at the time of
purchase.

Ac
ce
p

Order Delivery
Reliability

ip
t

Binary Dummy
(1=Yes, 0 = No)
11 Binary Dummies
One variable, Scarcity11,
identifies whether customers
had access to stock level
information at the time of
purchase (0=Yes, 1 =
No).Scarcity11 was the
baseline in the regression
analyses.The other 10
variables (Scarcity1,
Scarcity2,, Scarcity10)
identify the available
inventory level disclosed at
the time of purchase. For
example, if 7 units are
revealed to be in stock,
Scarcity7= 1 and the other
variables will equal 0

cr

Whether the product was returned (1) or


not (0)

Scarcity
Condition in
Availability

Mean

Operationalization

us

Product Return

Label

Description of Operationalization

an

Name

Binary Dummy
(0=Yes, 1 = No)
Binary Dummy
(0=Yes, 1 = No)
Binary Dummy
(1=Yes, 0 = No)

58
Page 58 of 62

ip
t
Holiday

Popularity

M
an

ce
pt

ed

Gift

.01
-.04
-.03
-.03
-.05
-.05
-.05
-.04
-.04
-.06
1

New
Customer

Scarcity10

.01
-.04
-.03
-.03
-.05
-.05
-.05
-.04
-.04
1

S&H

Scarcity9

.03
-.03
-.03
-.03
-.04
-.04
-.04
-.03
1

Price

Scarcity8

.02
-.02
-.02
-.02
-.03
-.03
-.03
1

Price Lead

Scarcity7

.03
-.03
-.03
-.03
-.04
-.05
1

Damage

Scarcity6

.04
-.03
-.03
-.03
-.05
1

Misinformati
on

Scarcity5

.05
-.03
-.03
-.03
1

cr

Scarcity4

.05
-0.02
-.02
1

Timeliness
Expectation

Scarcity3

.05
-.02
1

Reliability

Scarcity2

.10
1

-.04
-.01
-.03
-.03
.00
.00
-.01
.00
.02
.01
.00
1

-.01
.01
-.02
-.03
-.02
-.01
-.01
-.01
.02
.03
-.01
.62

.47
.05
.04
.04
-.03
.04
.01
.00
.03
.00
.02
-.03

.14
.01
-.01
.01
-.01
-.01
.00
.00
-.01
.00
-.02
.00

-.02
.00
.00
-.01
.02
-.01
.00
.00
.01
.00
-02
-.01

.00
.00
.01
.00
.00
.00
.01
.00
.00
-.01
.04
-.17

.02
.00
.01
.00
.01
.00
.00
.00
.01
-.01
.00
-.31

-.06
.00
.00
.00
.03
.03
-.03
.00
-.01
-.02
.01
-.01

-.04
.00
.00
.02
-.02
.01
-.02
.00
.03
-.01
.00
-.05

.00
.00
.02
.00
.03
.00
-.01
.02
.02
-.01
.00
-.01

.02
-.05
-.03
-.04
-.06
-.06
-.03
.02
.05
.06
.02
.06

-.01

-.01

.00

-.23

-.43

.00

-.08

.00

.12

-.01
1

-.01
-.01
1

.01
-.01
-.03
1

.01
.02
.00
.11
1

-.02
.00
-.01
.01
.01
1

-.03
.00
-.01
.04
.04
.01
1

-.01
.02
.00
.01
-.01
.00
.00
1

.00
.00
-.01
-.16
-.05
.00
-.02
.00
1

us

Scarcity1

Ac

Return
Scarcity1
Scarcity2
Scarcity3
Scarcity4
Scarcity5
Scarcity6
Scarcity7
Scarcity8
Scarcity9
Scarcity10
Reliability
Timeliness
Expectation
Misinformation
Damage
Price Lead
Price
S&H
New Customer
Gift
Holiday
Popularity

Return

Table 2: Correlation Coefficientsa

Bold indicates p < 0.05

60
Page 59 of 62

ip
t

Table 3: Regression Resultsa

2
Variables

Ac

ce
pt

Controls

Nagelkerke

Model 2

Model 3

(Direct Effects)

(Direct+Interaction Effects)

Variance
Inflation
Factor

B (Std. Err)

B (Std. Err)

B (Std. Err)

NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
23.23 (2493.69)
23.25 (8350.68)

0.99
0.99

1.62 (0.19)
1.05 (0.26)
1.09 (0.24)
0.84 (0.17)
0.65 (0.18)
0.59 (0.18)
0.59 (0.23)
0.51 (0.20)
0.41 (0.18)
0.26 (0.17)
-0.25 (0.11)
0.03 (0.02)
NA
23.15 (2454.24)
23.38 (8258.32)

<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
0.01
0.01
0.13
0.02
0.11
0.99
0.99

1.61 (0.19)
1.05 (0.27)
1.09 (0.24)
0.84 (0.17)
0.65 (0.18)
0.58 (0.18)
0.59 (0.23)
0.51 (0.20)
0.41 (0.18)
0.26 (0.17)
-0.49 (0.17)
-0.00 (0.03)
0.14 (0.07)
23.15 (2454.67)
23.39 (8258.23)

<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
0.01
0.02
0.12
<0.01
0.91
0.05
0.99
0.99

1.02
1.01
1.01
1.03
1.03
1.02
1.01
1.02
1.03
1.03
4.11
3.45
7.06
1.01
1.00

0.07
0.64
0.30
<0.01
<0.01
0.75
<0.01
<0.01

1.00
1.08
1.29
1.00
1.01
1.00
1.05

M
an

ed

Hypothesized

us

(Control Only)

Scarcity1
Scarcity2
Scarcity3
Scarcity4
Scarcity5
Scarcity6
Scarcity7
Scarcity8
Scarcity9
Scarcity10
Reliability
Timeliness Expectation
Reliability*Timeliness Expectation
Misinformation
Damage
Price Lead
Price
S&H
New Customer
Gift
Holiday
Popularity
Constant
- 2 Log Likelihood
Chi-Squared
Overall Significance
D.F. (p-value)
Predictive Power
C-Statistic (p-value)
Variance Explained Pseudo R- Cox and Snell
Squared Values

cr

Parameter Estimates

Model 1

-0.18 (0.11)
1.18E-5 (2.81E-5)
0.00 (0.00)
-0.42 (0.10)
-0.36 (0.12)
-0.00 (0.08)
0.05 (0.03)
-1.78 (0.13)
4565.64
1167.87
9 (<0.01)

0.08
0.68
0.11
<0.01
<0.01
>0.99
0.10
<0.01

-0.19 (0.10)
1.47E-5 (2.84E-5)
0.00 (0.00)
-0.46 (0.10)
-0.38 (0.12)
-0.03 (0.08)
0.09 (0.03)
-2.25 (0.17)
4464.46
1287.04
21 (<0.01)

0.08
0.6
0.14
<0.01
<0.01
0.75
<0.01
<0.01

-0.19 (0.10)
1.3E-05 (2.9E-05)
0.00 (0.00)
-0.46 (0.10)
-0.38 (0.12)
-0.03 (0.08)
0.09 (0.03)
-2.16 (0.18)
4442.91
1290.58
22 (<0.01)

0.682 (<0.01)

0.727 (<0.01)

0.728 (<0.01)

0.159

0.174

0.174

0.276

0.304

0.305

The sign for the Timeliness Expectation coefficient changes from 0.03 in Model 2 to -0.003 in Model 3. However, this change does not hinder H3s validation becauseit is not a
caused by multi-collinearity in Model 3 (as evidenced by the fact that all VIF values in this model are below 10). Moreover, the Timeliness Expectation coefficients across both
models are not statistically different from each other (at p< 0.05), as shown by the coefficients highly overlapping 0.95 confidence intervals: (-0.01, 0.07) in Model 2and (-0.06, 0.05)
in Model 3. Furthermore, both coefficients are not statistically different from zero, making it impossible to ascertain with statistical confidence what their signs actually are. For these
reasons, it would be inappropriate to infer that the coefficient for Timeliness Expectation in Model 2 is positive and different from the Timeliness Expectation coefficient in Model 3
and that the product between the Timeliness Expectation coefficient and the coefficient for Reliability in Model 2 will be negative (contrary to our expectation in H3).

61
Page 60 of 62

Table 4: 2 Difference Testsbetween Model 2 and Model 1 and Model 3 and Model 2
Model 2
(Direct Effects)
4446.46
1287.04
21 (<0.01)

Model 1

Measures of Relative
Goodness of Fit
Model 3 vs. Model 2

4446.46 - 4565.64 = -119.18


1287.04 - 1167.87 = 119.17a
12

4442.91 4446.46 = -3.55

1290.58 12874.04 = 3.54b

8
9
10

11
12
13
14
15

Comparing Model 2 with 1

Model 2
0.304
0.0276 a
22.18a
12a
6710a
<0.001a

Model 3
0.305
0.0010 b
10.2 b
1b
6709 b
0.001b

Comparing Model 3 with 2

Table 6: Classification Matrix for Model 3a

Ac
ce
p

te

an

3a This difference exceeds the critical Chi-squared value for 12 DF (21.02) at p=0.05
4bThis difference exceeds the critical Chi-squared value for 1 DF (3.53) at p=0.06
5
6
Table 5: R-Squared Change F Tests
7
Model 1
0.276
Nagelkerke Pseudo R- Squared Value
Change in Nagelkerke Pseudo R- Squared Value
F
df1
df2
p

ip
t

Measures of Relative
Goodness of Fit
Model 2 vs. Model 1

4565.64
1167.87
9 (<0.01)

-2 Log Likelihood
Chi-Squared Value
DF (p value)
-2 Log Likelihood Model 2
-2 Log Likelihood Model 1
Chi-Squared Model 2
Chi-Squared Model 1
dfModel 2 dfModel 1
-2 Log Likelihood Model 3
-2 Log Likelihood Model 2
Chi-Squared Model 3
Chi-Squared Model 2
dfModel 3 dfModel 2

us

Measures of Overall
Significance

Model 3
(Direct+Interaction)
4442.91
1290.58
22 (<0.01)

cr

1
2

Observed
Not Returned
Returned
Overall Percentage

Not Returned
5708
740

Predicted
Returned
Percentage Correct
2
100 %
282
27.6 %
a
89 %

aFollowing

the Associate Editors recommendation, we compared the full model's (i.e., Model 3) overall percentage
(i.e., 89% in this case) to the chance percentage (i.e., 50% given the binary data) plus 25%. Comparing that chance
percentage plus 25% (i.e., 75%) to the current Model 3's reported 89%, suggests that the full model is more than
capable of discriminating between returned and not returned items.

62
Page 61 of 62

Table 7: Summarized Hypotheses Tests

H1

Hypothesis
Online retail orders in which items are sold under perceived scarcity conditions in
availability have a higher return likelihood than orders in which items are sold under
non-scarcity conditions in availability

Support
Supported

H2

Delivery reliability for an online retail order is negatively related to its return likelihood

Supported

H3

Customer expectation of order delivery timeliness moderates negatively(attenuates) the


relation between delivery reliability and return rates

Supported

ip
t

1
2
3

Ac
ce
p

te

an

us

cr

63
Page 62 of 62