Customer Behavioral Intentions for Online Purchases

f

h

0

d

www.elsevier.com/locate/dsw

Journal of Operations Management 24 (2006) 124–147

Customer behavioral intentions for online purchases:

An examination of fulfillment method and

customer experience level

Kenneth K. Boyer *, G. Tomas M. Hult 1

Eli Broad Graduate School of Management, Michigan State University, East Lansing, MI 48824-1122, USA

Received 20 December 2003; received in revised form 11 April 2005; accepted 20 April 2005

Available online 13 June 2005

Abstract

This study presents an analysis of the growing market for groceries and other foodstuffs ordered via the internet or telephone

for delivery to the customer’s home. This industry has been growing for the past 5 years at greater than 25% per year while the

overall market for foodstuffs has been largely stagnant. The research utilizes data from surveys of over 2100 customers of five

different home delivery grocers. The analysis utilizes two group variables (customer experience level and order picking method)

and five primary constructs (service quality, product quality, product freshness, time-savings and behavioral intentions). The

results indicate that customer perceptions of the primary constructs generally improve as they gain experience with this new

method of ordering and receiving groceries. Furthermore, the operational choice of picking method is also shown to have a large

impact on customer perceptions—in particular, more experienced customers generally rate the primary constructs higher for

distribution center (DC)-based picking than for store-based picking. The study provides support for the hypothesis that direct to

customer foodstuffs can be of better freshness and quality when picked from a DC because of the ability to shorten the supply

chain than from a store. The data suggest that a DC-based picking strategy is viable if grocers can re-shape customer perceptions

and master the numerous intricacies of the supply chain.

# 2005 Elsevier B.V. All rights reserved.

Keywords: Order fulfillment; E-commerce; Grocery home delivery; Service quality; Internet ordering

1. Introduction

Groceries are perhaps the most universal commod-

ity, thus competition often spurs supermarkets to go to

* Corresponding author. Tel.: +1 517 353 6381;

ax: +1 517 432 1112.

E-mail addresses: [email protected] (K.K. Boyer),

[email protected] (G.T.M. Hult).1 Tel.: +1 517 353 4336; fax: +1 517 432 1009.

272-6963/$ – see front matter # 2005 Elsevier B.V. All rights reserved

oi:10.1016/j.jom.2005.04.002

great lengths to develop new technologies and

methods of streamlining both their supply chain and

their marketing efforts. Supermarkets are well known

as a difficult business to compete in with net profit

margins typically about 1–2% of sales. The supply

chain challenges associated with supermarkets are

enormous: the average supermarket carries 30,000

plus SKUs which are in a constant state of flux and

prices that must match the competitor down the street.

.

K.K. Boyer, G.T.M. Hult / Journal of Operations Management 24 (2006) 124–147 125

In an effort to address some of the supply chain

challenges, starting in the early 1990s, the grocery

industry pursued a major initiative labeled efficient

consumer response (ECR) to streamline the supply

chain, yet the general consensus has been that changes

have been slow and far from successful (Frankel et al.,

2002). A 1996 Andersen Consulting study found that

manufacturers of packaged goods products spent 13%

of sales ($25 billion) on trade promotions and that

hidden costs due to supply chain volatility and

uncertainty accounted for $5–8 billion (Andersen

Consulting, 1997). If trade promotions are taken out

of the mix, supermarkets across the board lose

substantial amounts of money.

During the dot-com mania of the late 1990s and

2000, several start-ups promised to revolutionize the

way groceries are bought and sold. Companies like

Webvan, Streamline and Homegrocer promised

cheaper, more convenient methods of shopping.

Yet, Webvan became the standard bearer for dot-

bombs by burning through $1 billion in investor

capital in a little over a year before going bankrupt

in 2001 (Rizzo, 2001). There were numerous

problems with the first generation of online grocers,

including the challenges of extending the supply

chain from existing stores to customer homes and

changing customer behavior to embrace a new form

of shopping. Clearly these barriers were not met and

overcome by Webvan, Streamline or Homegrocer.

Yet, despite the relative dearth of publicity, there is

evidence that online grocery is alive and growing. A

recent estimate projected grocery sales on the web

would total $2.4 billion for 2004 and $6.5 billion in

2008 (Moran, 2004). In particular, at least 10

grocers currently offer online ordering for home

delivery and all have more than $50 million in sales.

These grocers include traditional bricks and mortar

stores: Tesco (the world leader with over $1 billion

in sales in 2004), Sainsbury’s, Safeway, Albertsons,

and stores that do not have physical stores but

choose to deliver from a central distribution center

(DC): FreshDirect, Ocado, Grocery Gateway,

SimonDelivers and Peapod (Hamilton, 2003;

McLaughlin, 2003; Moran, 2004). In short, despite

reports of its demise there is considerable life in this

sector.

The focus of the current research is on comparing

the two primary operational approaches for picking

and delivering customer orders. The most successful

online grocers to date, including Tesco, Safeway and

Albertsons have all chosen to pick grocery orders at

existing stores. This store-based picking approach has

the advantage of minimizing the cost of fixed

investments, leveraging existing facilities and being

close to the customer’s home. In contrast, several

grocers are bypassing the costs of physical stores and

delivering to customers straight from a central

distribution center. This DC-based picking approach

theoretically can reduce some costs by cutting a link

out of the supply chain and also offer fresher produce,

meats and dairy items due to the reduced length of the

supply chain. The DC-based approach requires large

initial investments to build the DC and thus requires

large volumes of business. Thus, one of the many

reasons for Webvan’s failure was an inability to attract

enough orders to run the DC at a profit. Current

grocers such as FreshDirect, Ocado and Grocery

Gateway have learned many lessons from earlier

failures and are approaching the problem in new ways.

Several of these grocers are approaching profitability

and have shown the ability to offer fresher produce

through this shortened supply chain (Laseter et al.,

2003).

At present, online grocers that pick from existing

stores are generally more successful than DC-based

grocers, yet there are reasons to believe that cutting

existing stores out of the supply chain offers a solution

to many of the grocery industry problems cited above.

Since the dot-com bubble burst, most grocers are

extremely reticent to release detailed financial or

operating details (Mnyandu, 2003). To examine

differences in customer perceptions of key aspects

of online grocery shopping, we report results from an

intensive study that includes a substantial portion of

the major players in this growth industry. In particular,

we focus on five grocers in our study that can clearly

be classified as employing either a pure store-based

picking approach or a pure DC-based picking

approach. The two grocers in our study that employ

a hybrid approach are excluded from this analysis.

This hybrid approach employed by two grocers

involves delivering orders from both a DC and

existing stores, thus they can not be clearly classified.

This approach allows us to rigorously control for

operational differences and evaluate customer percep-

tions and behaviors.

K.K. Boyer, G.T.M. Hult / Journal of Operations Management 24 (2006) 124–147126

Fig. 1. Research model.

2. Literature review and hypotheses

Our general research model is shown in Fig. 1. The

first step in our analysis is to examine two factors that

impact the online shopping experience: customer

experience and picking method. The basic premise of

this model holds that experience with online grocery

ordering is a critical factor due to the fundamental

differences from traditional, in-store shopping. Sev-

eral studies have noted that there is an adjustment

period or learning curve for customers to adjust to

online shopping (Boyer and Olson, 2002; Chen and

Hitt, 2002). Customers are often uncertain what to

expect during their first few orders from an online

retailer, but as they place successive orders, they

develop an increasing comfort level with the both the

ordering method and the company. Similarly, the

operational choice made by the company to pick/fulfill

orders from either a store or a distribution center will

affect numerous customer perceptions. The choice of

fulfillment methods can be expected to effect various

aspects of product quality, inventory availability,

range of product choice, etc. (Boyer and Hult, 2005).

Furthermore, there is often an interaction effect

wherein customers learn about the company and its

reliability in delivering quality products through

repeated experiences. As customers gain experience,

they may be more likely to perceive differences in

service or product quality resulting from differences in

picking method. Thus, as shown in Fig. 1, we will first

examine differences in customer perceptions based on

these factors. The odd numbered hypotheses (H1, H3,

H5, H7 and H9) will be tested using ANOVA to test for

main effects and interaction effects based on the

groups shown.

The second step in our analysis is to examine

relationships between direct factors (service quality,

product quality, product freshness, time savings) and

behavioral intentions of customers. This will be done

using linear regression as moderated by customer

experience and picking method. The even numbered

hypotheses (H2, H4, H6 and H8) will be tested using

regressions for both the entire sample and for each

sub-sample (i.e. new customers/store-based pick,

repeat customer/store-based pick, new customer/

DC-based pick and repeat customer/DC-based pick).

The following sections examine the existing

literature for each of the direct factors and develop

the specific hypotheses to be tested.

2.1. Service quality

Understanding the impact of service-encounter

constructs such as physical good quality, service


quality, and the servicescape on behavioral intentions

has preoccupied services researchers for more than

two decades (Lovelock, 1983; Shostack, 1977). In

addition to the vast amount of support found in service

quality literature for this link (e.g., Boulding et al.,

1999; Cronin and Taylor, 1992; Zeithaml et al., 1996),

the idea that customers prefer greater service quality is

intuitive, particularly if price and other cost elements

are held constant. Additionally, equity theory suggests

that customers who perceive an organization’s

delivery of service quality in conjunction with, for

example, superior groceries are likely to attribute

greater equity to the relationship with that organiza-

tion (Kelley and Davis, 1994).

Customer experience level and picking method are

both expected to affect perceptions of service quality.

First, it appears to be intuitive that customers that

experience better service quality will be more loyal

and likely to continue purchasing with a given

company. However, in an e-commerce setting in

which the buying process is dis-intermediated or de-

personalized, the importance of the service quality can

be expected to change (Kaynama and Black, 2000;

Meuter et al., 2000). The measurement and impact of

service quality in electronic commerce applications

has received a great deal of attention over the past few

years (Parasuraman et al., 2004; Rabinovich and

Bailey, 2003). Generally speaking, service quality can

be expected to differ for electronic commerce between

aspects of the transaction involving placing the order)

and aspects of the transaction involving physical

interaction (the process of receiving the order). It is

intuitively logical that customers will perceive service

quality received via online ordering in a different

manner than service quality received through physi-

cally shopping in a store. Clearly, there is likely to be a

customer experience effect, since customers that are

new to online ordering of groceries must learn a new

way to shop.

Second, the picking method should affect service

quality by virtue of differences in the operational

execution by picking in-store or in a distribution

center. In a typology of service organizations, Bitner

(1992) highlighted the importance of complexity of

the provider’s operations as a major factor in customer

perceptions of service quality. Selecting items for

customers involves making choices as to what types of

produce or meat the customer will prefer. The different

picking methods are believed to effect customer

perceptions of service quality based on different levels

of performance relative to product quality and

freshness, as will be discussed below. In addition,

deliveries from a far away DC may be less likely to

occur on time than deliveries from a closer store

because of the greater time required to travel the

distance (Yrjola, 2001). Delivery drivers can also

generally make more deliveries from a closer store

than from a centralized DC (Delaney-Klinger et al.,

2003) and may be able to spend more time with

customers rather than on the road. Finally, some store-

based grocers offer a pickup option in which

customers can pick their order up at the store – this

saves the grocer money on deliveries but may impact

the perception of service quality – either negatively or

positively.

Based on the above arguments, we examine the

following hypotheses regarding service quality,

customer experience level, picking method and

behavioral intentions:

H1(A): Customer perceptions of service quality will

differ by customer experience level.

H1(B): Customer perceptions of service quality will

differ based on the picking method for selecting items

(i.e. store or DC-based).

H1(C): There is an interaction effect between

customer experience level and picking method for

customer perceptions of service quality.

H2: Service quality is correlated with increased

customer behavioral intentions.

2.2. Product quality

Product quality is always an important aspect of the

purchasing decision, but the importance generally is

intensified when purchasing over the Internet.

Numerous researchers have argued that online

markets facilitate increased competition and create

relatively ‘‘friction free’’ markets (Malone et al.,

1987; Bakos, 1991, 1997). Yet, while quality can be

judged purely via information available online for

intangible products and services such as travel,

software or music, tangible products require physical

handling and evaluation by the consumer. For

example, Clemons et al. (2002) find substantial

differences in ticket quality offered by online travel


agents, yet these product quality differences can be

evaluated online according to criteria such as the

number of connections or the accuracy of match

between requested and delivered departure/return time.

In contrast, tangible products present increased chal-

lenges involved in handling and shipping the product—

one of the reasons that Amazon started off selling books

was because books are a fairly easy to handle and ship

commodity (Rabinovich and Bailey, 2003).

For tangible products sold in an online channel,

Koch and Cebula (2002) point out four categories of

products where consumer perception of product

quality over the internet is likely to be quite variable:

(1) products which involve touch, taste or smell, (2)

the sale requires custom fitting, (3) the sale is from a

catalog or (4) the sale is accompanied by advice or

counsel. This makes intuitive sense since the

evaluation of commodity products commonly sold

over the Internet such as books, consumer electronics

or toys is largely based on brand perception and

loyalty since the same product is available at

numerous outlets. In contrast, products such as

apparel suffer from product quality perception

problems since consumers are unable to physically

try on items (Vickery and Agins, 2001). Similarly,

grocery items are vulnerable to customer mispercep-

tions of product quality because when buying the

product online the customer sacrifices the ability to

select their own merchandise.

Several examinations of the Internet grocery

retailing point out that customer perceptions of

product quality for groceries are likely to be

influenced substantially by both the experience level

of the customer and the method of picking/assembling

the order. First, customers new to the Internet retailing

face a hurdle in terms of becoming comfortable with a

new way of purchasing a very personal item—after all

everyone has different ‘‘ideal’’ characteristics for a

piece of fruit such as a pear. More experienced

customers are more likely to rate product quality

higher if they become comfortable over numerous

transactions that the retailer will deliver products that

match their specifications (Tanskanen et al., 2002;

Ellis, 2003). In other words, grocers are ‘‘on trial’’ for

the first few orders and, if they perform up to customer

expectations, the customer is more comfortable with

giving up control of their product selection and more

likely to cut the grocer some slack when problems do

occur. Second, the method used to pick orders will

directly impact quality. Store-based picking is, at best,

able to offer comparable or worse actual quality

levels—after all, the food has followed all the same

steps in the supply chain, including display on store

shelves, but the customer has given up control of the

process and trusts the personal shopper to select items

for them. In contrast, DC-based picking offers

potentially better quality for fresh items because it

can cut a link out of the supply chain (instead of a

producer ! DC ! store ! customer supply chain,

online ordering allows producer ! DC ! customer)

according to Delaney-Klinger et al. (2003). In both

models (store or DC-based picking), the customer has

given up the ability to directly select their own goods,

thus potentially causing poor perceptions of product

quality, but picking from a DC where the array of

available products is potentially fresher would seem to

offer a way to offset this effect. There may also be an

interaction between customer experience and picking

method, since customers who are new to a DC-based

grocer have no past experience with that grocer (since

it does not have physical stores), hence the grocer must

work to build brand credibility. Over time, the

potential advantages of fresher products should win

customer’s trust.

Based on these arguments, we examine the

following hypotheses, which examine the effects of

customer experience level and order picking method

on product quality perceptions and the relationship

between product quality and behavioral intentions:

H3(A): Customer perceptions of product quality will


H3(B): Customer perceptions of product quality will



H3(C): There is an interaction effect between product

experience level and picking method for customer

perceptions of service quality.

H4: Product quality is correlated with increased


2.3. Product freshness

A broad method of categorizing food products sold

in supermarkets considers consumer packaged goods

(CPG) as products that are manufactured and


packaged for mass consumption, while fresh products

include fruits, vegetables, meats, cheeses, etc. While

there are overlaps between these two groups, in

general CPG products are more commodity oriented

and more price sensitive. For example, the Coca Cola,

Nabisco Oreos or Pringles that you buy at the

supermarket are identical no matter where they are

bought—thus price tends to be the primary differ-

entiator for shoppers. In contrast, fresh products vary

substantially in both quality and appearance. Over the

past 5–10 years many supermarkets are placing

increased emphasis on offering high quality fresh

products and organic foods and see these as a way to

differentiate themselves from the Wal-Marts and other

low-price retailers (Rigby and Haas, 2004). Product

freshness has been found to correlate with customer

satisfaction for restaurants (Homburg et al., 2005) and

a recent study found that providing fresh and

appealing fruits and vegetables was the single most

important factor influencing consumer behavior for

supermarkets (Drake, 2001). While customers often

find supermarkets to be a convenient option for buying

produce, many also believe supermarket produce is

not as fresh as that found at farm stands and specialty

retailers (Supermarket News, 2004b). As a result of

these factors, supermarkets are investing in methods

such as providing samples of fresh fruit and vegetable

for taste-testing and increasing the visibility of

produce associates in order to encourage more

purchases of fresh produce (Supermarket News,

2004a). In a study of perishable product shelf life,

Tsiros and Heilman (2005) found that the relationship

between shelf life (days before product reaches its

expiration) and willingness to pay (WTP) was linearly

decreasing for lettuce, carrots, yogurt and milk, while

there was an exponential decrease for meat products

such as beef and chicken. Tsiros and Heilman (2005)

found that customers were willing to pay almost the

full list price for chicken and beef with 7 days of shelf

life remaining, but only 45% of the list price for these

same products with 5 days of shelf life remaining.

Lettuce, carrots, yogurt and milk had a more gradual

drop in WTP, with customers willing to pay 50% or

more of the list price until these products had 3 or less

days of shelf life remaining.

Given the importance of product freshness for

supermarkets, we turn our discussion to how online

ordering for home delivery is likely to effect customer

perceptions of freshness. First, as argued above for

product quality, fulfilling orders from a DC offers a

shorter, more direct supply chain than fulfilling orders

from a store. The effect on fresh products should by

even stronger than for CPG products, thus we expect

DC-based picking to have higher product freshness.

While this argument is intuitively appealing, being

able to effectively operate a DC to deliver fresh

products to individual customers is a nightmare

logistically. Produce wholesalers typically ship pallet

loads or truck loads of a particular product to stores,

whereas individual customers order a single head of

lettuce or bunch of carrots. In short the challenges

involved in forecasting, planning and executing direct

picking from a DC to a customer’s home are immense.

The failure to execute properly was a primary reason

behind the failure of Webvan, home grocer and other

early online grocers (Tanskanen et al., 2002; Delaney-

Klinger et al., 2003).

Customer experience should impact perceptions of

product freshness because initial online orders involve

a degree of faith that the grocer will select products

that meet the customers standards. As customers gain

experience through repeated orders, they should

develop a degree of trust and a set of reasonable

expectations regarding product freshness. The percep-

tion of freshness is also likely to vary between

customers of store-based versus DC-based grocers

since there may be an initial mistrust of companies

without an established brand name. This is one of the

reasons why FreshDirect (New York city) offers $50

off the first order a customer places and Ocado

(London) offers 10£ off of each of the first 5 orders a

customer places. Thus, we expect customer percep-

tions of freshness to change at differing rates as they

gain experience with either store-based or DC-based

grocers.

H5(A): Customer perceptions of product freshness

will differ by customer experience level.

H5(B): Customer perceptions of product freshness

will differ based on the picking method for selecting

items (i.e. store or DC-based).



perceptions of product freshness.

H6: Product freshness is correlated with increased



2.4. Time savings

One of the most commonly cited reasons for

shopping for products on the Internet is to save time.

The mass media repeatedly stress the theme that

people are continually busy and are looking for ways

to save time. Internet ordering is often profiled as one

of those potential time savers, since customers can

order anytime, anywhere and dressed anyway.

Researchers have broadly supported these statements,

with Bhatnagar et al. (2000) and Donthu and Garcia

(1999) both finding that Internet stores were particu-

larly attractive for time-starved consumers. Research-

ers such as Eastlick and Feinberg (1999), Bellman

et al. (1999) and Alreck and Settle (2002) have

examined both customer perceptions of the time

savings and potential explanatory factors such as

demographics, comfort level with computers and type

of shopping.

In contrast to these studies, our examination of

online grocery purchases addresses two factors that

have not been addressed in any depth: measurements

of the effects of learning or repeat purchases on time

savings and differences in operational execution.

Internet ordering certainly can be a time saver, but

many people tend to overlook the effects of learning

and repeat experience. For example, Amazon is well

known for its development and attempted patenting of

its one-click ordering system—certainly for repeat

customers this offers a significant time savings.

However, first time customers of Amazon may

actually have to spend more time placing their order

due to the need to enter data such as their name,

address, billing information, etc.—none of which

information is required when buying a book in a

physical bookstore (Boyer, 2001). Thus, there are

likely to be differences between new and repeat online

buyers.

The difference between new and repeat purchasers

is likely to be more substantial for groceries due to

the extreme difference in shopping methods—many

people have trouble with online orders for items like

cereal since they have a hard time visualizing

groceries without handling them. One report by Ellis

(2003) indicated that the average time for customers

to place their first online order for groceries was

70 min, while the average for the fifth order was

approximately 30 min. Internet ordering of groceries

involves a switch in activities from the customer

doing their own shopping to a paid employee doing it

for them. While the time to place an order online for

groceries may seem long, this represents most of the

time to order and receive an order when the grocer is

picking the order. In contrast, the total time to shop

for groceries in a physical store includes the time to

compile a list, travel to the store, shop in the store,

checkout and travel home. Yrjola (2001) estimates

that the value of customers’ time spent shopping

represents 20% of the value of grocery products, or

roughly $90 billion a year in the US, given that the

annual sales of grocery products are $450 billion per

year. In an earlier paper on the online book and CD

market, Brynjolfsson and Smith (2000) used a similar

method for estimating the time spent shopping in a

physical store. However, given the more time

intensive nature (more items to select and the need

to shop more often) of shopping for groceries and the

highly varied opinions of grocery shopping it would

be difficult to develop an accurate estimate. There-

fore, while estimates such as those used by

Brynjolfsson and Smith (2000) and Yrjola provide

good insight into the general costs (both financial and

time), we believe that customer perceptions are more

important since they are the ones who must pay for

the service.

There is likely to be a customer experience level

effect on order time since customers become

comfortable with online ordering systems over time.

Every online grocer in business today offers some type

of discount or rebate for the first 1–5 orders.

Discussions with numerous executives indicate a

universal belief that it takes three to five orders for the

customer to really become comfortable with the

system. However, there is no widely available data to

support this claim and the shape/magnitude of the

improvement in time savings is not known.

There also is likely to be a pick method effect on

customer perceptions of time savings for two reasons.

First, companies delivering from a store generally are

closer to the customer due to a broader choice of stores

to deliver from than when delivering from a DC. Thus,

according to Delaney-Klinger et al. (2003), deliveries

from a store are likely to be more reliable and quicker.

Second, some grocers offer customer pickup of the

order at the store—while this is operationally easier

and of lower cost, it may not be perceived as the same


potential time savings as orders delivered directly to a

customer’s home (DC-based grocers have a hard time

offering pickup because the one, centralized facility

is large and not located convenient to customers’

homes).

Based on the above discussion, we examine

customer perceptions of time savings, as effected by

experience level and picking method. We also

examine the relationship between perceived time

savings and behavioral intentions:

H7(A): Customer perceptions of time savings will


H7(B): Customer perceptions of time savings will





perceptions of time savings.

H8: Time savings is correlated with increased


2.5. Behavioral intentions

We have already discussed the direct effects of each

of the factors on behavioral intentions. We will also

examine the moderating effects of customer experi-

ence level and picking method on behavioral

intentions. Zeithaml et al. (1996) suggest that positive

behavioral intentions are reflected in the service

provider’s ability to get its customers to: (a and b)

remain loyal to them, (c) pay price premiums, (d)

communicate concerns to other customers and (e)

communicate concerns to the company. Clearly, it is

likely that there is a relationship between customer

experience and behavioral intentions, since more

experienced customers have already expressed their

behavioral intentions by making repeat purchases.

However, a repeat customer is not necessarily

completely satisfied—there are degrees of customer

loyalty and the relationship is not necessarily linear.

We also examine the relationship between picking

method and behavioral intentions to evaluate if

operational differences based on picking method

effect behavioral intentions.

H9(A): Behavioral intentions will differ based on

customer experience level.

H9(B): Customer perceptions of behavioral intentions

will differ based on the picking method for selecting

items (i.e. store or DC-based).


experience level and picking method for behavioral

intentions.

3. Methods

3.1. Sample

The sample consists of customers of five online/

home delivery grocers—two in the US, two in the

UK and one in Canada. All five firms were generous

with their time and allowing us access to their

customers, but prefer not to be identified by name

given the dynamic nature of the home delivery

grocery industry. Thus, we will describe the firms in

a general manner while assigning fictional names to

each grocer. We are also limited in the degree to

which we can describe the individual sales and

financial characteristics of these firms due to the

highly competitive and developing nature of the

industry. We can say that, in aggregate, the firms

have annual online/home delivery sales of well over

$200 million through over 200 bricks and mortar

stores. The customer base of the firms, in aggregate,

is well over 200,000 customers, with at least 50,000

loyal or repeat customers that purchase over $500 in

groceries per year each. The people in our contact

sample account for a total of over 38,000 purchases,

and $4.75 million in home delivery sales.

While there are numerous differences, large and

small, in the techniques these grocers utilize to take,

assemble/pick and deliver grocery orders to custo-

mers’ homes, we are limited in what can be revealed

by the sensitive nature of the business. Therefore,

we will focus on the choice of method to pick

grocery orders: either in an existing store or in a

distribution center. While there are numerous other

operational decisions to be made, this choice is a

fundamental strategic decision that forms the

foundation for each grocer’s operations strategy.

Grocers A and B both pick customer orders from

existing stores, while grocers C–E have all built

dedicated distribution centers for picking customer

orders.


Table 1

Description of participating companies and data collection methods

Grocer A Grocer B Grocer C Grocer D Grocer E

Data collection methods

Sample selection Stratified Random Stratified Stratified Stratified

Invitation to customer Email Opt-in at checkout Email Written letter Email

Incentive Yes No Yes Yes Yes

Survey method Web survey Web survey Web survey Written survey Web survey

Follow-up invitation Yes No Yes Yes No

Customers contacted (16577) 1159 10418 2500 500 2000

Responses (2985) 396 1066 690 373 460

Response rate (18.0%) 34.2 8.6 27.6 74.6 23.0

Pick method Store Store DC DC DC

3.2. Data collection

Table 1 provides a summary of the data collection

techniques for the individual grocers in our sample,

the number of customers contacted from each

company, the number of responses and the response

rate. Prior to data collection, we assessed the face and

content validity of the scale items and the general

quality of the research design via a pretest involving

three operations management academics, three mar-

keting academics, and seven Internet grocery execu-

tives. This pretest resulted in minor modifications to

thewording of some of the items as well as revisions to

parts of the instructions to the survey respondents. All

of the customers contacted had purchased groceries

online for home delivery at least once.

Our goals for data collection were to receive at least

300 responses for each of the firms in the sample,

stratified by the experience level of the customers. In

designing the sample of customers to be contacted for

each firm, we split each sample into 1/3 brand new

customers (those that had placed one or two orders

online from that grocer), 1/3 repeat customers (three to

six online orders) and 1/3 experienced customers

(seven or more online orders). In general, the

principles advocated by Dillman’s (1978) total design

method for survey data collection were followed:

initial contact with follow-up reminders, a small

incentive for completing the survey and the promise of

anonymity in survey responses.

Unfortunately, one of the difficulties involved with

working with companies directly to contact their

customers is that each company wants data collection

handled in a separate manner. In general, companies

are leery of allowing outsiders to contact customers

due to recent publicity about revealing customer

sensitive information. Furthermore, companies offer-

ing online grocery ordering are extra sensitive due to

the dual need to build product awareness/trust and the

desire to avoid being classified as ‘‘another internet

startup’’ like Webvan. We thus had to negotiate with

each individual company to balance their desires to

protect their customers from undue spamming with

our desire to employ identical methods across multiple

companies. As can be seen in Table 1, we were able to

employ substantially similar data collection methods

albeit with some minor exceptions. There are five

basic data collection methods that differed across

grocers: sample selection, invitation to customer,

incentive offered, survey method and follow-up

invitation.

Ideally every company would have a stratified

contact sample with equal numbers of new (one or two

orders), repeat (three to six orders) and experienced

(seven or more) customers. In addition to facilitating

comparison of customers based on experience level

(i.e. number of purchases), stratified samples also

allow us to track specific customer responses so that

post-hoc, longitudinal information can be gathered at a

later date. This was done in three out of five cases by

the grocers contacting the customer directly and

referring them to our independent website. There,

customers were asked to input an ID number that

would allow us to match up their past/future


purchasing history without knowing their personal

information (i.e. name, address or email address). One

of five companies (grocer B) was unwilling to identify

customers in any manner (even using a single-blind

approach as described above where only personnel

at the grocer would be able to identify specific

customers). Thus, grocer B employed an opt-in

approach where customers were invited to participate

in the survey at checkout for their order. This method

resulted in a randomized sample, since all customers

received the invitation to participate. We added a

question on this survey asking customers how many

times they had shopped with grocer B, thus we were

able to stratify this sample in a post hoc manner based

on customer responses.

Another minor methodological difference can be

seen with grocer D. This grocer did not want to be

perceived as an internet startup, preferring to approach

customers as a grocery company that happened to take

orders over the Internet. Therefore, grocer D asked us

to send a written letter to customers with a written

survey. The standard techniques of two follow-up

letters and a pre-paid, business reply envelope

(Dillman, 1978) were employed for this survey. It is

interesting that this technique resulted in by far the

highest response rate (74.6% for grocer D versus

34.2% for the next highest company).

The two final methodological differences relate to

the offering of a small incentive for survey

completion and follow-up invitations to participate

in the study. Four of the five grocers provided a small

incentive to customers to participate in the study.

These incentives ranged from a company hat for all

customers filling out a survey, to one free delivery, to

a raffle for 20 pairs of movie theatre tickets. In all, the

incentives were all worth less than $10 per customer,

but these companies felt that it was important to

compensate their busy customers for their time and

valuable feedback. With regard to follow-up invita-

tions, three of the five grocers sent a reminder (either

by email or by written mail) to customers 1 week

after the initial invitation to participate in the study.

Grocer B was unwilling to identify customers in any

manner or to directly contact customers, thus there

was no follow-up at all. In contrast, grocer E sent the

initial invitation to 2000 customers via email, but felt

that they did not want to risk upsetting customers

with another reminder.

As shown in Table 1, the overall response rate for

the entire sample is 2152 responses out of 16,577

customers contacted, or 18.0%. All of the data was

collected in the period August, 2002–May, 2003. This

compares very favorably to the response rate in similar

studies (Duray et al., 2000; Papke-Shields et al.,

2002). With the exception of grocer B, all of the

individual company response rates were well above

20%. Grocer B had a substantially lower response rate

due to the different data collection methodologies

employed (opt-in rather than special invite, no

incentive and no follow-up invitation). Excluding

grocer B from the sample, the overall response rate is

31.2%. To assess non-response bias, we conducted

chi-square tests on the proportion of positive

responses for the number of orders placed online

with the sponsoring grocer. Grocers A and D indicated

no potential for bias, whereas grocers C and E had

significant chi-square statistics ( p < 0.05). The data

indicate that customers that have placed more orders

with a grocer were more likely to complete the survey,

a result which is both intuitively logical and has been

observed in the literature. This finding is consistent

with our primary reason for stratifying our contact

samples by the number of orders placed: to examine

customer differences based on usage. Therefore, our

analysis of the data will take this into account and

perform separate analyses on fairly new (one to four

orders) and fairly experienced (seven or more orders)

users.

Several tests were made across the different sub-

samples to test for potential biases due to the different

data collection methods (physical mail, email or the

Internet) and none of the tests suggested the presence

of a bias. This result is consistent with the findings of

Couper (2000) and Klassen and Jacobs (2001) that

surveys can be administered via physical mail,

electronic mail or the Internet with no cause for

concern as long as the research design is solid and the

questionnaire is consistent.

3.3. Scales

This section describes the scales used to measure

the various components of Fig. 1. We used existing

scales where possible, but also tried to develop

customized scales where appropriate to capture the

dynamic and customized nature of the online grocery


shopping. Each of the factors shown in Fig. 1 is

described below, while the following section will

describe the measurement analysis. The individual

items included in each scale are shown in Appendix A.

The construct of service quality has been studied and

debated for the last decade (e.g., Cronin and Taylor,

1992; Teas, 1993; Van Dyke et al., 1999). In general,

recent evidence supports the use of performance

perceptions in measures of service quality (Zeithaml

et al., 1996). Because of the need to ensure construct

and measurement equivalence across multiple grocers

in multiple countries, it is especially important to use a

broad range of scale items that can be generalizable

across the grocers and countries while at the same time

keeping the items to amanageable number. As such, we

devised a scale composed of ten items based on

Parasuraman et al. (1985) ten original dimensions of

service quality. Similar scales have been used by

Kettinger et al. (1995).

The product quality scale was developed to assess

customer perceptions of product quality relative towhat

they can get in the store. This scalemeasures customers’

view of the general brand quality of the physical

products as well as items relating to range of product

choices and the number of substitutions for out of stock

items. These issues repeatedly came up during our

interviews with managers at all of the grocers in our

study as key concerns of customers as well as key

challenges for the company to execute well. The items

havebeen employed inprior research and shown tohave

good reliability and validity (Boyer and Hult, 2005).

Product freshness focuses on perishable products

such as produce and meats. While product quality is

important for all firms, product freshness is of particular

importance for grocery stores since offering higher

quality perishable products is a strategy of increasing

importance for grocers attempting to differentiate from

their low price competitors (Miller, 2005). In contrast to

the product quality items, product freshness items

specifically refer to the quality of fresh produce and

meats. Fresh products are generally considered to be

substantially different and require different handling

methods than consumer packaged goods.

Time savings is measured using the ratio of two

questions: (1) what is the time (in minutes) for your

most recent order?, divided by (2) what is the time (in

minutes) for your first order with grocer X? The

expectation is that the time to place orders will shrink

as customers place more orders (Ellis, 2003). Thus, as

shown in Table 2, the average time savings ratio is 0.51

for the entire sample—indicating that on average the

time for the most recent order was 51% of the time for

the first order placed by that customer. The average

times for the entire sample are 30.91 min for the most

recent order (question 1) and 64.95 min for the

customer’s first order (question 2). This objective

measure of time savings correlates strongly with

customers’ more perceptual measure (customers were

asked to rate the question ‘‘The more I shop with

grocer X the less time it takes to place an order’’ from

1 = strongly disagree to 7 = strongly agree) with a

correlation of �0.33 ( p < 0.01).

The indicators of behavioral intentions represent

the outcome measures in this study. Zeithaml et al.

(1996) suggest that positive behavioral intentions are

reflected in the service provider’s ability to get its

customers to: (a and b) remain loyal to them, (c) pay

price premiums, (d) communicate concerns to other

customers and (e) communicate concerns to the

company. Based on the theoretical foundation by

Zeithaml et al. (1996), we adopted the behavioral

intentions scale used by Cronin et al. (2000).

4. Analysis and results

4.1. Measurement analysis

Prior to hypothesis testing, the multi-attribute

measures were assessed using a rigorous five-step

process. The following analyses were conducted: (1)

item-level robustness across the four segmented

samples, (2) fit of the measurement models, (3)

reliability, (4) discriminant validity, and (5) common

method variance (CMV) testing. The four segmented

samples (i.e., new customer/store-based, repeat

customer/store-based, new customer/DC-based, and

repeat customer/DC-based) were tested in three ways:

combined sample analysis, multi-group analysis, and

segmented samples.

Tables 2 and6present the results of themeasurement

assessment. Table 2 summarizes the variables’ means,

standard deviations, and correlations. Table 3 reports

the results of the item-level analysis across the four

segmented samples (i.e., the assessment of measure-

ment equivalence across the sample types). Table 4


Table 2

Correlations for the overall sample and each of the segmented samples

Mean Standard

deviation

Service

quality

Product

quality

Product

freshness

Behavioral

intentions

Time

savings

Combined sample (n = 2152)

Service quality 5.87 0.98 1.00

Product quality 5.15 1.07 0.48 1.00

Product freshness 4.66 1.28 0.01 0.23 1.00

Time savings 0.51 0.29 �0.05 �0.04 0.02 1.00

Behavioral intentions 4.85 1.23 0.48 .55 0.20 �0.15 1.00

All correlations �0.05 are significant at the p < 0.05 level

P0-C0 sample (n = 181)




Time savings 0.70 0.38 �0.17 �0.11 0.12 1.00

Behavioral intentions 4.62 1.30 0.58 0.62 0.11 �0.20 1.00






Time savings 0.46 0.27 0.00 0.02 0.00 1.00







Time savings 0.59 0.27 �0.06 �0.08 0.03 1.00



P1-C1 Sample (n = 472)




Time savings 0.41 0.22 �0.05 �0.05 0.11 1.00



P0 = order fulfillment from stores; P1 = order fulfillment from DC; C0 = customer with 1 to 4 prior orders; C1 = customer with 7 or more prior

orders.

presents the analysis of the measures in the combined

samples as well as the multi-group analysis, including

the averagevariances extracted, highest shared variance

between pairs of constructs, composite reliabilities, and

factor loadings. Table 5 presents the corresponding

measurement results for each of the four segmented

samples. Table 6 provides the results of the discriminant

validity assessments in the combined sample as well as

in each of the four segmented samples. Overall, the 19

purified and perceptually-based reflective items (of 24

original items) and their corresponding four latent

constructs were found to be reliable and valid in the

context of this research. The remainder of this section

discusses the measurement testing in detail.

4.1.1. Item-level analysis across the four

segmented samples

As an initial assessment of the items used to

measure the four latent constructs, we assessed each

item’s robustness via multi-group analysis using


Table 3

Multi-group item analysis of segmented samples

Item x2free Dx2

ðd:f:¼3Þ Sign

SQ1 6369.30 2.42 ns

SQ2 6370.68 1.04 ns

SQ3 6369.33 2.39 ns

SQ4 6364.48 7.24 ns

SQ5 6369.24 2.48 ns

SQ6 6371.21 0.51 ns

SQ7 6371.30 0.42 ns

SQ8 6347.94 23.78 p < 0.05

SQ9 6367.37 4.35 ns

SQ10 6370.96 0.76 ns

PQ1 6369.15 2.57 ns

PQ2 6369.77 1.95 ns

PQ3 6359.67 12.05 p < 0.05

PQ4 6369.14 2.58 ns

PQ5 6365.45 6.27 ns

PQ6 6361.46 10.26 p < 0.05

FP1 6371.56 0.16 ns

FP2 6371.12 0.60 ns

FP3 6368.43 3.29 ns

BI1 6369.25 2.47 ns

BI2 6370.47 1.25 ns

BI3 6366.69 5.03 ns

BI4 6356.72 15.00 p < 0.05

BI5 6356.86 14.86 p < 0.05

x2fixed ¼ 6371:72, d.f.fixed = 1146, d.f.free = 1143.

LISREL 8.71 (Joreskog et al., 2000). Specifically, sets

of b estimates were constrained, one parameter at a

time, to be equal and different across the four samples

(see Table 3 for complete results). The significance of

the resulting x2-change was examined (Anderson,

1987). In this analysis, we found five items that were

Table 4

Overall analysis of the measures (n = 2152)

Construct Variance extracted Highest shar

Combined sample analysis

Service quality 68.8% 23.0%

Product quality 50.3% 23.0%

Product freshness 80.3% 5.3%

Behavioral intentions 49.3% 30.3%

Fit statistics: x2 = 2445.45, d.f. = 146, Delta2 = 0.96, CFI = 0.96, RNI

Multi-group analysis (parameters constrained to be equal across groups)

Service quality 66.8% 23.0%

Product quality 50.0% 23.0%

Product freshness 80.3% 5.3%

Behavioral intentions 50.7% 30.3%

Fit statistics: x2 = 4151.28, d.f. = 716, Delta2 = 0.95, CFI = 0.95, RNI

not statistically robust across each of the four

segmented samples (SQ8, PQ3, PQ6, BI4, and BI5).

Each of these items, when allowed to vary freely

across the four samples, resulted in the Dx2’sexceeding the maximum limit (Dx2

d:f:¼3 ¼ 7:81) to

be considered robust (ranging from a Dx2 of 10.26 to

23.78, p < 0.05). The five items were consequently

removed from further analysis. The remaining items

were robust, with the Dx2 ranging from 0.16 to 7.24.

4.1.2. Fit of the measurement model

The model fit for the combined sample, the multi-

group analysis, and each segmented sample analysis

was evaluatedusing a series of indices recommendedby

Gerbing and Anderson (1992) and Hu and Bentler

(1999)—the DELTA2, relative noncentrality (RNI),

comparative fit (CFI), Tucker–Lewis (TLI), and the root

mean square error of approximation (RMSEA) indices.

After removing the inadequate items based on the item-

level analysis across the four segmented samples, an

excellent fit to the data was achieved for each CFA. For

the combined sample, we achieved fit statistics of

DELTA2, RNI, CFI, and TLI all being 0.96, and

RMSEA = 0.09 (x2 = 2445.45, d.f. = 146). Similarly,

the multi-group CFA resulted in fit statistics of

DELTA2, RNI, CFI, and TLI all being 0.95, and

RMSEA = 0.09 (x2 = 4151.28, d.f. = 716). Complete

results for both the combined sample and multi-group

analyses can be found in Table 4. Corresponding results

for the four segmented samples can be found in Table 5

(with fit statistics ranging from 0.95 to 0.97 for CFI,

Delta2, and RNI, and from 0.09 to 0.10 for RMSEA).

ed variance Composite reliability Factor loadings

0.95 0.71–0.90

0.80 0.58–0.83

0.92 0.84–0.93

0.74 0.57–0.76

= 0.96, RMSEA = 0.09

0.95 0.70–0.89

0.80 0.59–0.82

0.91 0.84–0.93

0.75 0.55–0.81

= 0.95, RMSEA = 0.09


Table 5

Segmented sample analyses of the measures

Construct Variance extracted Highest shared variance Composite reliability Factor loadings


Service quality 72.2% 51.8% 0.96 0.73–0.94

Product quality 62.3% 51.8% 0.87 0.71–0.86

Product freshness 80.3% 1.4% 0.92 0.84–0.90

Behavioral intentions 58.7% 38.4% 0.81 0.65–0.85

Fit statistics: x2 = 360.35, d.f. = 146, Delta2 = 0.97, CFI = 0.97, RNI = 0.97, RMSEA = 0.09


Service quality 74.0% 38.4% 0.96 0.74–0.91

Product quality 60.5% 42.3% 0.86 0.67–0.88





Service quality 61.8% 11.6% 0.93 0.66–0.87

Product quality 46.8% 24.0% 0.77 0.51–0.83





Service quality 63.1% 22.1% 0.94 0.69–0.88

Product quality 37.0% 22.1% 0.70 0.48–0.75





orders.

4.1.3. Composite reliability

We assessed the latent factors’ reliability by

calculating a composite reliability for each construct

(Fornell and Larcker, 1981). To be thorough, we

calculated reliabilities for each scale in the combined

sample, the multi-group analysis, and in each of the

four segmented samples. The composite reliability

was calculated as:

CRh ¼ðP

lgiÞ2

ðP

lgiÞ2 þ ð

PeiÞ

;

where CRh = composite reliability for scale h; lgi ¼standardized loading for scale item gi, and ei = mea-

surement error for scale item gi. Along with the

reliability calculations, we also examined the para-

meter estimates and their associated t-values as well

as the average variances extracted (Anderson and

Gerbing, 1988). Again, we report the average variances

extracted for each scale in the combined sample, the

multi-group analysis, and in each of the four segmented

samples. Average variance extracted was calculated

as:

Vh ¼ðP

lgiÞ2

ðP

lgiÞ2 þ ð

PeiÞ

;

where Vh = average variance extracted for h; lgi ¼standardized loading for scale item gi, and ei = mea-

surement error for scale item gi. The scales’reliabil-

ities ranged from 0.70 to 0.96, the factor loadings

ranged from 0.47 to 0.94 ( p < 0.01), and the average

variances extracted ranged from 37.0% to 84.0% (see

Tables 4 and 5 for complete results). The 19 purified

items were also found to be reliable and valid when

evaluated based on each item’ error variance, mod-

ification index, and residual covariation. In addition,

we found no evidence of skewness or kurtosis.


Table 6

Discriminant validity assessment: pairwise cfas in the overall sample and each of the segmented samples

Sample and pair of constructs x2free x2

fixed Dx2ðd:f:¼1Þ Sign

Combined sample p < 0.01

Service quality Product quality 2096.08 3716.07 1619.99 p < 0.01

Service quality Behavioral intentions 1689.43 2571.53 882.10 p < 0.01

Service quality Product freshness 1667.39 6713.63 5046.24 p < 0.01

Product quality Behavioral intentions 341.20 850.10 508.90 p < 0.01

Product quality Product freshness 270.91 2709.62 2438.71 p < 0.01

Behavioral intentions Product freshness 12.18 1362.56 1350.38 p < 0.01

P0-C0 sample







P0-C1 sample







P1-C0 sample







P1-C1 sample








orders.

4.1.4. Discriminant validity

Following the reliability analysis, we established

discriminant validity by two independent methods.

First, we calculated the shared variance between each

pair of constructs and verified that it was lower than the

variances extracted for the involved constructs (Fornell

and Larcker, 1981). Shared variance was calculated as:

g2 ¼ 1� c

where g2 = shared variance between constructs, and

with the diagonal element of c indicating the amount

of unexplained variance. Because h and e were stan-

dardized, g2 was equal to the squared correlation

between the two constructs. As shown in Tables 4

and 5, the average variances extracted were higher

than the associated shared variance in all cases.

Second, we examined all possible pairs of

constructs, as suggested by Anderson (1987) and

Bagozzi and Phillips (1982), in a series of two-factor

CFA models using LISREL 8.71 (Joreskog et al.,

2000). Specifically, each pairwise CFA model was run

twice—first, constraining the f coefficient to unity and


second, allowing f to vary freely. Based on the results

of a x2-difference test, the unconstrained model

performed significantly better than the associated

constrained model when f = 1 (i.e., Dx2ð1Þ > 3:84 was

exceeded in all cases). The lowest Dx2ð1Þ was found

between the product quality and behavioral intentions

scales in the P0-C0 sample ðDx2ð1Þ ¼ 64:51Þ. All other

pairwise tests in the combined sample and each of the

four segmented samples resulted in higher Dx2. As

such, each latent construct exhibited discriminant

validity vis-a-vis all other scales.

4.1.5. Assessment of common method variance

Finally, we assessed the potential of common

method variance influencing the data by using

Harmon’s one-factor test within a CFA setting. If

CMVposes a serious threat, a single latent factor would

account for all manifest variables (Podsakoff and

Organ, 1986). A worse fit for the one-factor model

provides support that CMV does not pose a serious

threat (Sanchez et al., 1995). We undertook the CMV

examination in six testing scenarios, including the

combined sample, the multi-group sample, and each of

the segmented samples. In the combined sample

analysis, the one-factor model resulted in a x2 =

10472.19 with d.f. = 152 (versus a x2 = 2445.45 and

d.f. = 146 for the measurement model). In the multi-

group analysis, the one-factor model resulted in a

x2 = 12298.52 with d.f. = 146 (versus a x2 = 4151.28

and d.f. = 716 for the measurement model). In the P0-

C0 sample, the one-factor model resulted in a








and d.f. = 146 for the measurement model). Finally, in

the P1-C1 sample, the one-factor model resulted in a


and d.f. = 146 for the measurement model). Based on

these CMVanalyses, the data used in this study do not

appear to be constrained by common method variance

thatwould effect the results of the hypothesis testing.As

such, the 19 purified items and their four latent

constructs are found to be reliable and valid in the

context of this study.

4.2. Results of hypothesis testing

Our results follow the general outline shown in

Fig. 1: the odd hypotheses (H1, H3, H5 and H7)

regarding customer experience level and picking

method are tested by means of ANOVA for both main

and interaction effects. The second step is to test the

relationship between each of our direct factors

(service quality, product quality, product freshness

and time savings) and outcomes (behavioral inten-

tions) using multiple regression for each of the four

combinations of customer experience level and

picking method.

5. Results

5.1. ANOVA comparison of groups

Table 7 shows the ANOVA results when Customer

group (new or repeat) and Pick method (DC-based or

store-based) are used as the independent variables.

Table 7 shows the means for the overall sample as well

as each of the four combinations of Customer and Pick

groups, as well as the F-statistics and significance

values for the main effects and interactions effects.

A quick review of Table 7 reveals that Customer

group has a significant effect for three of the five

dependent variables, Pick group is significant for three

out of five and that there is an interaction effect for

four out of five. Some interesting insights are found by

examining each independent variable separately.

Service quality is rated significantly higher by

customers of grocers employing a DC-based picking

model. There is also a small interaction effect, as

illustrated in Fig. 2. Customers of grocers employing a

DC-based picking model perceive an improvement in

service quality as they gain experience, while

customers of store-based grocers see a slight decline.

This finding is somewhat surprising because we can

think of no compelling reason why service quality

ought to be better for one picking model over another.

The results for product quality show that there is no

main effect for either Customer group or Pick group,

but there is a very strong interaction effect, as

illustrated in Fig. 3. This result indicates that as

customers gain experience with online ordering, there

is a major reorientation of opinions regarding product


Table 7

ANOVA comparison for customer type (New or Repeat) and order-picking method (DC or store)

Store-baseda DC-baseda F-statistics

Newb Repeatb Newb Repeatb Customer group Pick group Interaction

Overall sample

Service quality

Mean 5.87 5.65 5.59 5.99 6.14 0.69 78.69** 4.60*

St. Dev. 0.98 1.05 1.08 0.86 0.87

Product quality

Mean 5.15 5.31 5.00 5.14 5.33 1.18 2.34 20.78**

St. Dev. 1.07 1.17 1.15 1.03 0.92

Product freshness

Mean 4.66 4.62 4.73 4.56 4.77 9.41* 0.02 0.81

St. Dev. 1.28 1.09 1.28 1.24 1.38

Time savings

Mean 0.51 0.70 0.46 0.59 0.41 221.39** 31.98** 5.52*

St. Dev. 0.29 0.38 0.27 0.27 0.22

Behavioral intentions

Mean 4.85 4.62 4.86 4.62 5.34 88.21** 14.95** 22.78**

St. Dev. 1.23 1.30 1.25 1.21 1.06

N = 2152 overall N = 179 N = 702 N = 797 N = 472a Pick group.b Customer group.* p < 0.05.** p < 0.01.

quality. The interaction effect can be clearly seen in

Fig. 3, which shows that the ratings of repeat

customers increase substantially over new customers

when groceries are picked from a DC. In contrast, the

ratings of product quality decrease for repeat

customers of store-based grocers. This finding

supports the hypothesis that delivering groceries

straight from a DC can provide better quality, but

that customers may take some time to get used to this

Fig. 2. Interactions effects for service quality.

idea. In other words, the evidence supports the

hypothesis that there is a combined time/experience

and pick model effect.

The results for product freshness show a main

effect for Customer group, but no significant effects

for the Pick group or the interaction. The effect for

Customer group makes intuitive sense, since, as

argued previously, customers placing orders online

must adapt to a new method of shopping wherein they

Fig. 3. Interaction effect for product quality.


Table 8

Linear regression with behavioral intentions as dependent variable

Variable Model 1 Model 2

Constant 4.34 �0.31

Control variables

Customer experience level

(0 = New, 1 = Repeat)

0.513** 0.307**

Pick method (0 = store-based,

1 = DC-based)

0.336** 0.058

Number of prior orders 0.003 0.003**

Independent variables

Service quality 0.392**

Product quality 0.448**

Time savings �0.315**

Product freshness 0.107**

R2 0.045** 0.439**

DR2 0.394**

n 2131 1747

Sample sizes differ from those shown in Tables 1–7 due to missing

values for some of the control variables and for Time Savings. All

regressions are performed using listwise deletion for the sample

being analyzed. *p < 0.05.** p < 0.01

do not have direct control over their product selections

and instead must trust the store employees to make

‘‘good’’ selections. Thus, the increase in product

freshness rating indicates that repeat customers rate

product freshness higher. However, it is somewhat

surprising that there is not an interaction effect since

we previously argued that the shortened supply chain

for DC-based grocers should allow better product

freshness than when products are picked from stores.

Time savings has significant main effects for both

Customer group and Pick group, as well as an

interaction effect. Both pick methods result in

increases in time savings as customers gain experi-

ence—likely reflecting their increasing knowledge of

the ordering system and greater ability to place orders

quickly. It is somewhat surprising that there is a Pick

Method and Interaction effect, since there is no logical

reason for customers to perceive a time difference.

The data for behavioral intentions shown in Table 7

and Fig. 5 exhibit a very similar pattern. There is a

strong main effect for both Customer group and Pick

method as well as an interaction effect. As shown in

Fig. 5, The DC-Pick Method has much higher ratings

for repeat customers than the store-based Pick method.

To summarize, our data provide support for

Hypotheses 1(B and C), 3(C), 5(A), 7(A–C) and 9

(A–C). The data indicate that both customer experi-

ence level and picking method have strong effects on

customer perceptions of three aspects of quality, time

savings and behavioral intentions.

5.2. Multiple regression results—predicting

behavioral intentions

We test the relationships between the direct factors

shown in Fig. 1 and behavioral intentions in two steps.

We first test the overall model for the entire sample,

then we test individual models for each of the four

combinations of Customer group and Picking method.

This two-step method allows us to test for the presence

of group effects based on customer experience level

and pick method. Then, individual regressions are run

for each sub-group in order to get deeper insight into

how the relationship between each of the independent

variables and customer behavioral intentions varies

across groups.

Table 8 shows the regression equation derived for

the entire sample. We show the Beta weights for the

un-standardized data because this allows greater

interpretability. Model 1 includes just three control

variables (customer experience level = 0 or 1, pick

method = 0 or 1 and number of previous orders = 1 to

49). The control variables explain a significant

proportion of variance (R2 = 0.045), with both

Customer Experience Level and Pick Method being

significant. Model 2 adds the independent variables

service quality, product quality, product freshness and

time savings. Model 2 explains 43.9% of the variance

in behavioral intentions. All four of the independent

variables are significant at the p < 0.01 level. Based

on the magnitude of the coefficient, product quality

has the largest effect, followed by service quality and

time savings. This is interesting since increasing

service quality can often be considered as directly

conflicting with time savings in an operational setting,

since better service quality often requires spending

more time with customers. It is our belief that grocery

home delivery offers an opportunity to capitalize on a

clear moment of truth at the time of delivery wherein

high service quality can be offered in a time efficient

manner. We will discuss this possibility in the next

section. For now however, the results in Table 8

indicate that behavioral intentions can be predicted


Table 9

Linear regression by pick and customer type with behavioral intentions as dependent variable

New customers

(one to four orders)

Repeat customers

(seven or more orders)

Betaa Betaa

A. Regressions for store-based picking

Constant 0.36 0.30

Service quality 0.32** 0.40**

Product quality 0.47** 0.47**

Time savings �0.48* �0.49**

Product freshness 0.27 0.04

R2 0.44** 0.50**

N = 149 N = 584

B. Regressions for DC-based picking

Constant �0.18 0.05

Service quality 0.32** 0.48**

Product quality 0.45** 0.32**

Time savings �0.26 �0.12

Product freshness 0.17** 0.14**

R2 0.36** 0.37**

N = 608 N = 406

Sample sizes differ from those shown in Tables 1–7 due to missing values for some of the control variables and for Time Savings. All regressions

are performed using listwise deletion for the sample being analyzed.a Variable.* p < 0.05.** p < 0.01.

with a high degree of accuracy and that all four of the

independent variables make a contribution.

Table 9A and B show separate regression results for

Store-based Picking and DC-based Picking, respec-

tively, and for new and repeat customers within each

of these groups. These separate analyses are

performed since customer experience and pick method

were shown to have a significant effect as control

variables in the full model in Table 8. Individual

analyses of each sub-sample allow us to further

examine the specific relationship between the inde-

pendent and dependent variables within each sub-

sample. Given our fairly large sample sizes, we are

able to do this without a noticeable loss of power.

First, all four R2 values are significant and explain

over 36% of the variance in behavioral intentions.

Second, comparing the beta weights for the different

regressions provides useful insights. In Table 9A, the

beta weights for the independent variables are fairly

consistent for new and repeat customers, with two

exceptions. First, the importance of service quality

increases from 0.32 for new customers to 0.40 for

repeat customers. This makes sense since we

previously had discussed how online customers give

up control of their product selections, thus it is logical

that the service quality they do receive – i.e. their

interaction with employees – is of increasing

importance as they gain experience with the service.

Second, product freshness has a weight of 0.27 for new

customers versus 0.04 for repeat customers. This may

be due, in part, to the smaller sample size for this

group, but it may also be because new customers do

not recognize that the products they receive are, at

best, no better than they get by selecting the product

themselves, and, at worst, much worse than they

would select for themselves.

Table 9B shows regression equations for DC-based

picking for both new and repeat customers. Both R2

values are quite high (0.36 and 0.37) and in both cases

three out of four independent variables are significant.

Similar to the data shown in Table 9A, the results for

DC-based picking (Table 9B) show that repeat

customers have a higher weight for service quality

(0.48) than do new customers (0.32). In contrast,


Fig. 4. Interaction effect for time savings.

product quality decreases in importance, dropping

from 0.45 for new customers to 0.32 for repeat

customers. The biggest difference between Table 9A

and B is that while time savings had a significant

coefficient for store-based picking, it was not

significant for DC-based picking, while product

freshness was significant for DC-based picking and

was insignificant for store-based picking. This can be

interpreted to support the argument that customers of

store-based grocers expect the service to save them a

substantial amount of time, but they do not expect their

produce and meats to be particularly fresh. In contrast,

customers of DC-based grocers place a high premium

on freshness but do not place as much importance on

time savings.

To summarize, the data shown in Tables 8 and 9

provide support for the hypotheses H2, H4, H6 and

H8 regarding the relationship between the direct

factors shown in Fig. 1 and behavioral intentions

of customers. In both the overall model and the

individual combinations of Customer level and Pick

group, the behavioral intentions of customers can be

predicted quite accurately. However, the relative

weight of each independent factor differs substantially

based on Customer and Pick group. We turn now to a

discussion of the implications of these findings for

grocery home delivery companies.

Fig. 5. Interaction effect for behavioral intentions.

6. Discussion

In examining home delivered groceries, our

evidence supports the twin propositions that customer

experience level and order-picking method have an

effect on several different dimensions of the overall

customer experience. In general, customers with more

experience rate the experience more highly—in terms

of service quality and product quality, time-savings

and behavioral intentions. This finding provides

support for the idea that customers must be re-trained

to accept a newmethod of shopping for groceries, after

all they have been shopping in-person and in the store

for their entire lives so it is logical that placing orders

online for home delivery, while convenient, may offer

some initial hurdles to adoption. Our second major

finding is that the method for picking customer orders

has a significant effect on service and product quality,

time-savings and behavioral intentions—either as a

main effect or as an interaction with customer

experience level. This finding supports the hypothesis

that picking from a distribution center can actually

provide better product quality because of the

shortened supply chain, but that customers may take

a while to realize advantages because of their concern

about dealing with a far off DC that they can not

physically see or touch.

As shown in Figs. 2 through 5, the interaction

effects for product quality and behavioral intentions

are quite substantial. From an observer’s point of view,

it is logical that picking orders from a DC offers

advantages due to a shortened supply chain—a la Dell

computers or Amazon books. Both Dell and Amazon

are widely respected for their ability to turn inventory

more quickly due to their business models that do not

include physical stores (Boyer, 2001). In comparison,

picking orders from existing stores simply adds

another link to the supply chain—rather than


customers selecting their own groceries, the store pays

an employee to select orders and then deliver them.

However, there are also numerous challenges in

picking orders from DCs, including convincing

customers of the utility of this approach and mastering

the numerous intricacies of piece picking orders in

high volume for a wide variety of items. Clearly, early

competitors such as Webvan or HomeGrocer were not

able to master these challenges before running out of

cash. In contrast, the second generation of DC-based

grocers (Ocado, FreshDirect, Grocery Gateway,

SimonDelivers) shows some signs of mastering the

learning curve. Our data generally indicate that DC-

based picking outperforms store-based picking on

several measures of customer perception.

In examining the ability to predict customer

behavioral intentions, all four of the scales assessed

in this study have a strong effect. Home delivery

grocers must simultaneously deliver service and

product quality as well as a substantial time-savings

and a high degree of product freshness. The relative

importance of each of these four factors differs

substantially for new versus repeat customers and

store-based versus DC-based picking. The impor-

tance of product freshness is substantial for DC-

based grocers, whereas time savings is relatively

unimportant. In contrast, time savings is of greater

importance for customers of store-based grocers

while product freshness is relatively important.

These findings give clear guidance for home delivery

grocers on where to focus their attention since

factors are of different importance for different

customers. In closing, we argue that home delivered

groceries are helping to re-shape the expectations of

customers, who are slowly becoming less tolerant of

the many inconveniences associated with ‘‘tradi-

tional’’ shopping in a supermarket and willing to pay

a slight premium for better service via online

ordering for home delivery. Furthermore, DC-based

grocers are showing that there is potential in this

business model.

Future research should continue to track develop-

ments in this nascent branch of the ubiquitous grocery

industry. In particular, researchers should continue to

examine three areas: how customer perceptions

evolve, how retailers manage the receipt and picking

of orders, and how deliveries to the final customer are

best managed. There is great potential to design new

methods of delivering/receiving orders and numerous

companies, including FedEx, UPS and Deutsche Post

are concerned that a large percentage of their

shipments to customer’s homes are not received

because the customer is not home or is otherwise

occupied. Also, while this research focused primarily

of the grocery industry, its insights should be

generalizable to some degree to other online

retailers—particularly those that deal with bulky

products, low value to size ratios and a high number

of items per order.

7. Limitations

All research has limitations and this study is

certainly not an exception. First, two of our sub-

samples, grocer C and E, exhibited evidence of a

response bias—more experienced customers were

more likely to complete our survey. However, as

shown in Table 1, the overall response rates were

generally quite high when compared to similar

research, thus the potential for bias is minimal.

Similarly, our research is prone to common method

bias since the dependent variable was answered by the

same respondents that answered the independent

variable, in a cross-sectional manner. However, we

believe this is not a large problem since our analysis of

some longitudinal data collected subsequently to our

initial survey shows that behavioral intentions

correlates highly with follow-up purchases made by

those customers in the 8–12 months following the

survey. Finally, there may some biases and limitations

due to the choice of questions and method of data

collection, which differed from company to company

due to influences beyond our control. However, the

differences in data collection are fairly minimal and

comparison of the data did not indicate any cause for

concern.

Acknowledgements

This research was conducted with Grant SES

0216839 from the National Science Foundation. We

appreciate the cooperation and assistance of three

grocers in the US, one grocer in Canada and one grocer

in the UK.


Appendix A. Scales

All measures used a Likert-type ratings scale

ranging from 1 = strongly disagree to 7 = strongly

agree except where noted.

A.1. Service quality (SQ)

1. Grocer X employees are reliable in providing the

service I expect

2. G
rocer X employees are understanding of my
service needs

3. G
rocer X employees are responsive to my service
requests

4. G
rocer X employees are competent in providing
the expected service

5. I
feel secure in service encounters with Grocer X
employees

6. G
rocer X employees are courteous in providing
me service

7. G
rocer X employees are available to answer my
service-related questions

8. T
he tangible (appearance of trucks, staff, pro-
ducts) aspects of Grocer X service are excellent*

9. G
rocer X has good credibility in providing the
service I need

10. I
have access to communicate with Grocer X
regarding my service needs

A.2. Product quality (PQ)

1. Grocer X has prestigious products

2. G
rocer X has an excellent assortment of products
3. G
rocer X products are among the best*
4. G
rocer X has a sufficient range of product choices
(I can get what I want)

5. T
he products are the same quality as I can get in the
store

6. T
he number of substitutions or out of stock items is
reasonable*

A.3. Product freshness (PF)

The specific question asked was: Please rate the

degree of change when using the Internet for ordering

groceries throughGrocer X in comparison to shopping

in a neighborhood store (i.e. an Albertsons, Publix,

Kroger, Safeway etc.). The question was rated on a

scale from 1 = Much worse than in-store shopping, to

4 = about the same to 7 = Much better than in-store

shopping.

1. T
he freshness of food
2. T
he quality of fresh produce
3. T
he quality of meats
A.4. Time savings (TS)

1. Time to place the most recent order in minutes

divided by the time to place the first order in

minutes

A.5. Behavioral intentions (BI)

1. I would classify myself as a loyal customer of

Grocer X

2. I
do not expect to switch to another online grocer to
get better service in the future

3. I
would continue to do business withGrocer X even
if I had to pay more

4. I
would complain to other customers if I
experienced a problem with Grocer X service*

5. I
would complain to Grocer X employees if I
experienced a problem with their service*

Notes: (*) indicates this item was dropped from the

scale based on analysis of inter-item variability. Insert

name of specific company whereverGrocer X appears.

References

Alreck, P.L., Settle, R.B., 2002. The hurried consumer: time-saving

perceptions of Internet and catalogue shopping. Journal of

Database Marketing 10 (1), 25–35.

Andersen Consulting, 1997. The Daunting Dilemma of Trade

Promotion.

Anderson, J.C., 1987. An approach for confirmatory measurement

and structural equation modeling of organizational properties.

Management Science 33, 525–541.

Anderson, J.C., Gerbing, D.W., 1988. Structural equation modeling

in practice: a review and recommended two-step approach.

Psychological Bulletin 103 (3), 411–423.

Bagozzi, R.P., Phillips, L.W., 1982. Representing and testing orga-

nizational theories: a holistic construal. Administrative Science

Quarterly 27, 459–489.

Bakos, J.Y., 1991. A strategic analysis of electronic marketplaces.

MIS Quarterly 15 (3), 295–310.


Bakos, J.Y., 1997. Reducing buyer search costs: implications for

electronic marketplaces. Management Science 43 (12), 1676–

1692.

Bellman, S., Lohse, G.L., Johnson, E.J., 1999. Predictors of online

buying behavior. Communications of the Association for Com-

puting Machinery 42 (12), 32–38.

Bhatnagar, A., Misra, S., Rao, H.R., 2000. On risk, convenience and

internet shopping behavior. Communications of the Association

for Computing Machinery 43 (11), 98–105.

Bitner, M.J., 1992. Servicescapes: the impact of physical surround-

ings on customers and employees. Journal of Marketing 56 (2),

57–701.

Boulding, W., Kalra, A., Staelin, R., 1999. The quality double

whammy. Marketing Science 18 (4), 463–484.

Boyer, K.K., 2001. E-Operations: A Guide to How the Internet

Streamlines Operations. Business Horizons 44 (1), 47–54.

Boyer, K.K., Hult, G.T.M., 2005. Extending the supply chain:

integrating operations and marketing in the online grocery

industry. Journal of Operations Management 23 (6), 642–661.

Boyer, K.K., Olson, J.R., 2002. Drivers of Internet purchasing.

Production and Operations Management 11 (4), 180–198.

Brynjolfsson, E., Smith, M.D., 2000. Frictionless commerce? A

comparison of Internet and conventional retailers. Management

Science 46 (4), 563–585.

Chen, P.Y., Hitt, L.M., 2002. Measuring switching costs and the

determinants of customer retention in internet enabled busi-

nesses: a study of the online brokerage industry. Information

Systems Research 13 (3), 255–274.

Clemons, E.K., Hann, I.H., Hitt, L.M., 2002. Price dispersion and

differentiation in online travel: an empirical investigation. Man-

agement Science 48 (4), 534–549.

Couper, M.P., 2000. Usability evaluation of computer-assisted

survey instruments. Social Science Computer Review 18 (4),

384–396.

Cronin, J.J., Brady, M.K., Hult, G.T.M., 2000. Assessing the effects

of quality, value, and customer satisfaction on consumer beha-

vioral intentions in service environments. Journal of Retailing 76

(2), 193–218.

Cronin, J.J., Taylor, S.A., 1992. Measuring service quality: a reex-

amination and extension. Journal of Marketing 56 (3), 55–68.

Delaney-Klinger, K., Boyer, K.K., Frohlich, M., 2003. The return of

online grocery shopping: a comparative analysis of Webvan and

Tesco’s operational methods. The TQM Magazine 15 (3), 187–

196.

Dillman, D.A., 1978. Mail and telephone surveys: the total design

method. Wiley, New York, NY.

Donthu, N., Garcia, A., 1999. The Internet shopper. Journal of

Advertising Research 39 (3), 52–58.

Drake, J., 2001. What consumers want: price? Convenience? Clean-

liness great product variety. Canadian Grocer 115 (6), 33–37.

Duray, R., Ward, P., Milligan, G., Berry, W., 2000. Approaches to

mass customization: configurations and empirical validation.

Journal of Operations Management 18 (6), 605–626.

Eastlick, M.A., Feinberg, R.A., 1999. Shopping motives for mail

catalog shopping. Journal of Business Research 45 (3), 281–290.

Ellis, C., 2003. Lessons from online groceries. Sloan Management

Review 44 (2), 8.

Fornell, C., Larcker, D.F., 1981. Evaluating structural equation

models with unobservable variables and measurement error.

Journal of Marketing Research 18, 39–50.

Frankel, R., Goldsby, T.J., Whipple, J.M., 2002. Grocery industry

collaboration in the wake of ECR. International Journal of

Logistics Management 13 (1), 57–72.

Gerbing, D.W., Anderson, J.C., 1992. Monte Carlo evaluations of

goodness of fit indices for structural equation models. Socio-

logical Methods and Research 21, 132–160.

Hamilton, A., 2003. What’s for dinner? Time 161 (20), 85.

Homburg, C., Kaschate, N., Hoyer, W.D., 2005. Do satisfied cus-

tomers really pay more? A study of the relationship between

customer satisfaction and willingness to pay. Journal of Market-

ing 69, 84–96.

Hu, L., Bentler, P.M., 1999. Cutoff criteria for fit indexes in

covariance structure analysis: conventional criteria versus new

alternatives. Structural Equation Modeling 6 (1), 1–55.

Joreskog, K.G., Sorbom, D., Du Toit, S., Du Toit, M., 2000. LISREL

8: new statistical features, second ed. Scientific Software

International Inc., Chicago, IL.

Kaynama, S.A., Black, C.L., 2000. A proposal to assess the service

quality of online travel agencies: an exploratory study. Journal of

Professional Services Marketing 21 (1), 63–72.

Kelley, S.W., Davis, M.A., 1994. Antecedents to customer expecta-

tions for service recovery. Journal of the Academy of Marketing

Science 22 (1), 52–61.

Kettinger, W.J., Lee, C.C., Lee, S., 1995. Global measures of

information service quality: a cross national study. Decision

Sciences 26 (5), 569–588.

Klassen, R.D., Jacobs, J., 2001. Experimental comparison of web,

electronic and mail survey technologies in operations manage-

ment. Journal of Operations Management 19 (6), 713–728.

Koch, J.V., Cebula, R.J., 2002. Price, quality and service on the

internet: sense and nonsense. Contemporary Economic Policy

20 (1), 25–37.

Laseter, T., Berg, B., Turner, M., 2003. What FreshDirect learned

from Dell. Strategy & Business 1–6.

Lovelock, C.H., 1983. Classifying services to gain strategic market-

ing insights. Journal of Marketing 47, 9–20.

Malone, T.W., Yates, J., Benjamin, R.I., 1987. Electronic market and

electronic hierarchies. Communications of the ACM 41 (1), 73–

80.

McLaughlin, K., 2003. Back from the dead: buying groceries online.

Wall Street Journal D.1.

Meuter, M.L., Ostrom, A.L., Roundtree, R.I., Bitner, M.J., 2000.

Self-service technologies: understanding customer satisfaction

with technology-based service encounters. Journal of Marketing

64 (3), 50–64.

Miller, L., 2005. Fresh difference; three retailers outrun the com-

petition in fresh foods. Supermarket News 53 (3), 35.

Mnyandu, E., 2003. Online grocers mum on profits. Chicago

Tribune.

Moran, J.M., 2004. Online grocery services are on rebound. Hartford

Courant.

Papke-Shields, K.E., Malhotra, M.K., Grover, V., 2002. Strategic

manufacturing planning systems and their linkage to planning

system success. Decision Sciences 33 (1), 1–30.


Parasuraman, A., Zeithaml, V.A., Berry, L.L., 1985. A conceptual

model of service quality and its implications for future research.

Journal of Marketing 49, 41–50.

Parasuraman, A., Zeithaml, V.A., Malhotra, A., 2004. ‘‘E-S-Qual: a

multiple-item scale for assessing electronic service quality

(04–112)’’, Marketing Science Institute, working paper series

No. 04-003.

Podsakoff, P.M., Organ, D.W., 1986. Self-reports in organizational

research: problems and prospects. Journal of Management 12

(4), 531–544.

Rabinovich, E., Bailey, J.P., 2003. Physical distribution service

quality in Internet retailing: service pricing, transaction attri-

butes, and firm attributes. Journal of Operations Management 21

(6), 651–672.

Rigby, D.K., Haas, D., 2004. Outsmarting Wal-Mart. Harvard

Business Review 82 (12), 22–27.

Rizzo, T., 2001. The death of Webvan. Internet World 1, 4–5.

Sanchez, J.I., Korbin, W.P., Viscarra, D.M., 1995. Corporate support

in the aftermath of a natural disaster: effects on employee strains.

Academy of Management Journal 38 (2), 504–521.

Shostack, G.L., 1977. Breaking free from product marketing.

Journal of Marketing 41, 73–80.

PMA Consumer Study Reveals Produce Purchase Drivers, 2004a.

Supermarket News, 52 (July 19(29)), p. 26.

Supermarket News, Supermarkets Convenient, but Farm Stand

Produce is Fresher, 2004b. Supermarket News, 52 (October

4(4)), p. 46.

Tanskanen, K., Yrjola, H., Holmstrom, J., 2002. The way to profit-

able internet grocery retailing—six lessons learned. Interna-

tional Journal of Retail & Distribution Management 30 (4),

169–178.

Teas, R.K., 1993. Expectations, performance evaluation, and con-

sumers’ perception of quality. Journal of Marketing 57, 18–34.

Tsiros, M., Heilman, C.M., 2005. The effect of expiration dates and

perceived risk on purchasing behavior in grocery store perish-

able categories. Journal of Marketing 69, 114–129.

Van Dyke, T.P., Prybutok, V.R., Kappelman, L.A., 1999. Cautions on

the use of the SERVQUAL measure to assess the quality of

information systems services. Decision Sciences 30 (3), 877–

891.

Vickery, L., Agins, T., 2001. Retailers find web apparel unprofitable.

Wall Street Journal B.6.

Yrjola, H., 2001. Physical distribution considerations for electronic

grocery shopping. International Journal of Physical Distribution

& Logistics Management 31 (10), 746–761.

Zeithaml, V.A., Berry, L.L., Parasuraman, A., 1996. The behavioral

consequences of service quality. Journal of Marketing 60, 31–

46.

Documents

Customer Behavioral Intentions for Online Purchases