· Published by: Labyrinth Publications Pottenbakkerstraat 15 – 17 2984 AX Ridderkerk The Netherlands Print: Offsetdrukkerij Ridderprint B.V., Ridderkerk ISBN 90-5335-106-X 978-90-

University of Groningen

Determining the cross-channel effects of informational web sitesTeerling, Marije Leonie

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite fromit. Please check the document version below.

Document VersionPublisher's PDF, also known as Version of record

Publication date:2007

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):Teerling, M. L. (2007). Determining the cross-channel effects of informational web sites. s.n.

CopyrightOther than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of theauthor(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons thenumber of authors shown on this cover page is limited to 10 maximum.

Download date: 11-06-2020

https://www.rug.nl/research/portal/en/publications/determining-the-crosschannel-effects-of-informational-web-sites(b3288b51-0306-4c4e-bdd1-41249339fe2d).html

https://www.rug.nl/research/portal/en/publications/determining-the-crosschannel-effects-of-informational-web-sites(b3288b51-0306-4c4e-bdd1-41249339fe2d).html

Determining the Cross-Channel

Effects of Informational Web Sites

Marije L. Teerling

Published by: Labyrinth Publications Pottenbakkerstraat 15 – 17 2984 AX Ridderkerk The Netherlands

Print: Offsetdrukkerij Ridderprint B.V., Ridderkerk ISBN 90-5335-106-X 978-90-5335-106-2 © 2007, M.L. Teerling Alle rechten voorbehouden. Niets uit deze uitgave mag worden

verveelvoudigd, opgeslagen in een geautomatiseerd gegevensbestand, of openbaar gemaakt, in enige vorm of op enige wijze, hetzij elektronisch, mechanisch, door fotokopieën, opnemen of enige andere manier, zonder voorafgaande schriftelijke toestemming van de auteur.

All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, including photocopying, recording, or otherwise, without prior written permission of the author.

RIJKSUNIVERSITEIT GRONINGEN

Determining the Cross-Channel Effects of Informational Web Sites

Proefschrift

ter verkrijging van het doctoraat in de Economische Wetenschappen

aan de Rijksuniversiteit Groningen op gezag van de

Rector Magnificus, dr. F. Zwarts, in het openbaar te verdedigen op

donderdag 15 maart 2007 om 13:15 uur

door

Marije Leonie Teerling

geboren op 29 september 1976 te Groningen

Promotor: Prof. Dr. P.S.H. Leeflang Copromotor: Dr. K.R.E. Huizingh Beoordelingscommissie: Prof. Dr. R.N. Bolton Prof. Dr. T.H.A. Bijmolt Prof. Dr. P.C. Verhoef ISBN 90-5335-106-X 978-90-5335-106-2

v

Preface The Internet has changed the world we live in. With blogs and

Wikipedia, everyday people increasingly add their own content to the Web. We determine when, where and which information we gather. From the comfort of our own home, we order products, use services, contact our loved ones or entertain ourselves in virtual worlds. Second Life, an example of a virtual world, is a game with unlimited possibilities and currently one of the biggest hypes. Internet is described as the marketing trend of 2006 (Financieel Dagblad 10/11/06). Overall the Internet provides everyday people with a plethora of opportunities. One of which is for instance the use of Web sites to search for information.

Fifteen Web sites have altered our lives (NRC Next 08/22/2006). Among these sites, for instance Nu.nl or Funda.nl, only one offers the possibility to purchase products. The majority offer information or entertainment. In my dissertation research, I study the impact of such an informational Web site on customer behavior. My interest in this area of research started with my Master thesis. Without the enthusiasm and coaching of my Master thesis supervisor, I would probably not have written this dissertation.

Even though the virtual world offers a lot of possibilities, I am indebted to many people who have supported and coached me on various aspects during my Ph.D. in the ‘traditional’ world. Primarily, I am deeply grateful to my advisors, Peter Leeflang and Eelko Huizingh for their continuous support and guidance. Peter: “I am grateful for the knowledge you shared with me, your insights into marketing research and model building. I am thankful for your patience with me during the first year of my Ph.D. period.” Eelko: “Your belief in my abilities and your enthusiasm were like the water and sun that a seedling needs to grow.” Most of all, I am indebted to both of you for your warmth and understanding during my final year of my dissertation.

Besides Peter and Eelko, I had the privilege of working with two more brilliant men. I would like to thank Erjen van Nierop for sharing his knowledge on Multivariate Probit models with me. I have really enjoyed working with him on both the study and other creative work. I hope that we will continue to do so in the future. As for Koen Pauwels from Dartmouth College, my time at The Tuck School of Business was probably the most productive of my entire Ph.D. I would like to extend

vi

my gratitude to him for making me feel welcome, helping me with time series modeling and for the conversations that had nothing to do with research.

Next, I extend special thanks to the members of my reading committee, professor Ruth Bolton from the Arizona State University and professors Tammo Bijmolt and Peter Verhoef from the University of Groningen. Working on their suggestions, comments and questions has definitely improved my dissertation.

Without the aid of a number of organizations, this research could not have taken place. Firstly, I am thankful for the willingness of a major Dutch retailler to participate in the research. Secondly, I express my gratitude to Acxiom for providing demographic data free of charge. Lastly, the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO) provided the financial resources to visit Dartmouth College, Tuck School of Business.

I am grateful for working in an environment with so many warm and inspiring colleagues. Special thanks go out to Liane Voerman for giving me the courage to explore my scientific boundaries. During emotional difficulties, especially in my final year, the support of colleagues like Liane Voerman, Janny Hoekstra, Jenny van Doorn, Jeanette Wiersema, Hanneke Tamling and Adriana Krawzcyk were very valuable to me. My thanks also go out to Peter Ebbes, Tessa Wouters, Ivan Orosa Paleo, Thijs Broekhuizen, Gijsbert Willenborg and Marcel Turkensteen. Lastly, I extend my thanks to Jennifer Jordan my office mate and most of all friend during my stay at Tuck.

My two paranimphs, Simone Teerling and Marbel Schoemaker are very special women in my life. I am grateful for their help in organizing my defense. Most of all, I am grateful for their support throughout the years and for their ability to lighten my spirits.

My family and friends have always supported me, even though at times I was absent (minded). The illness of my father that has partly characterized my Ph.D. period provides perspective. The unconditional support and love of my parents is one of the biggest forces in the completion of this dissertation. Mum: “Thank you for being my safe haven.” Dad: “Thank you for your wisdom and faith.”

Marije Leonie Teerling Enschede, January 2007

vii

Table of Contents Preface ......................................................................................... v Table of Contents ....................................................................... vii 1 Introduction ........................................................................ 1

1.1 Introduction.............................................................................1 1.2 Informational Web Sites ...........................................................2

1.2.1 Definition ................................................................................ 2 1.2.2 Use of Informational Web Sites ................................................ 3

1.3 Multichannel Environment.......................................................5 1.4 Information Search ..................................................................7 1.5 Problem Delineation.................................................................8 1.6 Definitions .............................................................................10 1.7 Empirical Setting ...................................................................11

1.7.1 Firm ...................................................................................... 11 1.7.2 Informational Web Site .......................................................... 12 1.7.3 Customer Panel ..................................................................... 13

1.8 Outline...................................................................................14 2 The Effects of an Informational Web Site on Customer Attitudes and Behavior ............................................................... 17

2.1 Introduction...........................................................................17 2.2 Literature Review ...................................................................18

2.2.1 Multichannel Setting ............................................................. 18 2.2.2 Proposed Model and Hypotheses............................................ 20 2.2.3 Moderating Effects of Customer Traits ................................... 22 2.2.4 Store/Site Attitude ................................................................ 24 2.2.5 Antecedents of Store/Site Attitude ......................................... 24

2.3 Empirical Setting ...................................................................25 2.3.1 Data ...................................................................................... 25

2.4 Proposed Methodology............................................................26 2.4.1 Specification: Measurement Part of the Model ........................ 27 2.4.2 Measurement Constructs....................................................... 28 2.4.3 Construct Validity ................................................................. 28 2.4.4 Specification: Structural Part of the Model ............................. 31 2.4.5 Longitudinal Design............................................................... 32 2.4.6 Measurement Moderators ...................................................... 33 2.4.7 Moderation and Validation Approach ..................................... 34

2.5 Findings.................................................................................35 2.5.1 Estimates for the Proposed Model .......................................... 35 2.5.2 Results Longitudinal Design .................................................. 36 2.5.3 Moderating Results of Customer Traits .................................. 38 2.5.4 Validation Results ................................................................. 41

2.6 Discussion .............................................................................41 2.6.1 Site Attitude Store Attitude................................................ 42

viii

2.6.2 Site Attitude Store Behavior............................................... 42 2.6.3 Site Behavior Store Behavior ............................................. 43

2.7 Conclusions ...........................................................................44 3 The Impact of an Informational Web Site on Offline Customer Buying Behavior ......................................................................... 47

3.1 Introduction...........................................................................47 3.2 Literature Review ...................................................................49 3.3 Proposed Methodology............................................................52

3.3.1 Specification.......................................................................... 53 3.3.2 Estimation............................................................................. 55

3.4 Empirical Setting ...................................................................56 3.4.1 Informational Web Site .......................................................... 56 3.4.2 Data ...................................................................................... 57 3.4.3 Explanatory Variables............................................................ 58 3.4.4 Exploratory Insights .............................................................. 59

3.5 Findings.................................................................................62 3.5.1 Number of Shopping Trips ..................................................... 62 3.5.2 Validation for the Number of Shopping Trips ......................... 64 3.5.3 Amount Spent per Trip per Category...................................... 64 3.5.4 Validation for Amount Spent per Trip .................................... 70 3.5.5 Further Investigation of the Individual Site Parameters .......... 70

3.6 Discussion .............................................................................72 3.7 Conclusions ...........................................................................74

4 Cross-Channel Behavior for an Informational Web Site and an Offline Store............................................................................... 77

4.1 Introduction...........................................................................77 4.2 Literature Review ...................................................................79

4.2.1 Multichannel Behavior........................................................... 79 4.2.2 Online Information ................................................................ 82 4.2.3 Decomposing Offline Buying and Online Search Behavior ...... 83

4.3 Hypotheses ............................................................................83 4.3.1 Hypotheses: Marketing Efforts ............................................... 84

4.4 Proposed Methodology............................................................85 4.4.1 Decomposition of Behavior .................................................... 86 4.4.2 Dynamics .............................................................................. 86 4.4.3 Model Calibration Steps......................................................... 88 4.4.4 Unit Root Testing Procedure .................................................. 88 4.4.5 Moderation ............................................................................ 90

4.5 Empirical Setting ...................................................................91 4.5.1 Data ...................................................................................... 91

4.6 Aggregate Level Findings ........................................................91 4.6.1 Multichannel Behavior........................................................... 92 4.6.2 Cross-Channel Marketing Efforts.......................................... 95

4.7 Median Split Findings ............................................................97 4.7.1 Product Type ......................................................................... 98

ix

4.7.2 Flow Median Split ................................................................ 100 4.7.3 Frequency of Site Visits ....................................................... 102 4.7.4 Cross-Channel Marketing Efforts for Median Splits.............. 103

4.8 Discussion ...........................................................................104 4.8.1 Cross-Channel Behavior ...................................................... 105 4.8.2 Cross-Channel Marketing Efforts......................................... 106 4.8.3 Context Characteristics ....................................................... 106

4.9 Conclusions .........................................................................107 5 Discussion and Conclusions ............................................. 111

5.1 Introduction.........................................................................111 5.2 Summary .............................................................................112

5.2.1 Attitudinal Framework (Chapter 2)....................................... 112 5.2.2 Individual Customer Behavior (Chapter 3) ........................... 113 5.2.3 Cross-Channel Effects (Chapter 4) ....................................... 114

5.3 Insights................................................................................115 5.4 Managerial Implications .......................................................119 5.5 Limitations and Future Research .........................................121

5.5.1 Cross-Channel Effects ......................................................... 121 5.5.2 Generalizations.................................................................... 123 5.5.3 Omitted Variables................................................................ 123

Appendix I. Multichannel Studies ........................................ 125 Appendix II. Survey Chapter 2 .............................................. 128 Appendix III. Full Conditional Posterior Distributions ............ 129 Appendix IV. Descriptives Chapter 3 ...................................... 134 Appendix V. T-Test Comparison ............................................ 135 Appendix VI. MVP Model Selection......................................... 136 Appendix VII. Post-Hoc Comparison ........................................ 138 Appendix VIII. Description Variables Chapter 4 ........................ 139 Appendix IX. Moderation Variables......................................... 140 Appendix X. Moderation Results ........................................... 141 References................................................................................ 143 Author Index ............................................................................ 157 Subject Index ........................................................................... 163 Samenvatting ........................................................................... 167

1

1 Introduction This dissertation investigates the effects of an informational Web site on offline behavior, specifically with regard to customer buying behavior in a “traditional” store. This chapter serves to introduce the literature on informational Web sites (§1.2) and the multichannel environment (§1.3). It also highlights that this dissertation contributes to existing literature by providing insights into (1) the effects of informational Web sites, (2) the sequential process of search and purchase in multiple channels, (3) differences in the process across segments and product categories, (4) the effects of free-riding behavior, and (5) methodologies that can be used to measure these effects (§1.4). We provide a number of definitions in Section 1.5 that we use throughout this dissertation. In Section 1.6, we describe the empirical setting of this dissertation. The introductory chapter ends with an overall outline of the dissertation and an overview of the research issues to which it pertains (§1.7).

1.1 INTRODUCTION

One of the biggest changes to society during the past decade has been the introduction the Internet. Initially, most companies started their Internet activities in a rush rather than as the result of deliberate strategic planning. To determine how to use the Internet as part of a (multichannel) strategy, insight is needed into the possible functions, e.g., communication, transaction and distribution, and the effects on customer behavior. Over the last seven years (1998-2005) an impressive body of marketing literature on the topic has appeared. Most of these top marketing journal publications focus on the transactional function of the Internet.

Regardless of this academic focus, firms sooner adopt an informational Web site than a transactional Web site due to the capabilities and investments required (Lee & Grewal 2004). Research also shows that consumers experience obstacles to buying online, such as privacy, trust and technology anxiety (e.g., Meuter, Ostrom, Bitner & Roundtree 2003; Schlosser, Barnett White & Lloyd 2006). Gupta, Su and Walter (2004) conclude that many companies make it very difficult for visitors to search their Web sites for information and then conduct a transaction. Moreover, the relative size of online purchases reflects the strong position of offline transactional channels. Even though e-commerce retail sales in the United States slowly increase, the percentage of retail sales originating from the Internet is still only 2.8 % in the third quarter of 2006 (US Census Bureau). The percentage of

The Cross-Channel Effects of Informational Web Sites

2

retail sales originating from the Internet has risen with 2% over the last 6 years. In 2004, of retail sales in the Netherlands, only 1.7% originated from the Internet (CBS 2006).

Some studies have focused on the comparison and the combined effect of online and offline transaction channels (e.g., Gensler, Dekimpe & Skiera 2007), but not so much on the effects of online informational channels. A few recent studies (Nicholson, Clarke & Blakemore 2002; Verhoef, Neslin & Vroomen 2007) show that customers tend to use online and offline channels as complements, in many cases customers search online and purchase offline. To our knowledge, the effects of using an informational Web site on offline customer behavior are uninvestigated.

1.2 INFORMATIONAL WEB SITES

1.2.1 Definition

For firms, the Internet provides a new channel for communication, transaction, or distribution. A plethora of firms in different industries currently uses the Internet to sell their products or services. Moreover, most firms provide a lot of information, both commercial and noncommercial, via the Internet (e.g., Huizingh 2000). Peterson, Balasubramanian and Bronnenberg (1997) indicate that the Internet is best suited for communication, in that it has been designed to deliver information and foster connectivity.

Web sites can therefore be classified on the basis of the functionality of Internet usage for a given firm. Hoffman, Novak and Chatterjee (1995) similarly distinguish Web sites into six categories of commercial activity: • Online storefronts are Web sites that offer direct sales through an

electronic channel. • Internet presence sites are those that provide a virtual presence for a

firm and its offerings. • Content sites offer particular content, such as news, for which the

visitor have to pay (distinguished as fee-based versus sponsored content sites).

• Malls are collections of online storefronts, each of which may contain many different categories of goods for sale.

• Incentive sites try to attract customers to visit the commercial site behind it.

Chapter 1: Introduction

3

• Search agents, like Google, offer visitors the opportunity to search the Web for other Web sites with particular information. Hoffman and Novak (1996) indicate that Internet presence sites, or

informational Web sites (Dann & Dann 2001), dominate commercial activity by providing detailed information about a firm’s offerings or creating an image to attempt to build ongoing relationships with customers. Hoffman and Novak claim that informational Web sites represent a form of nonintrusive advertising through which the customer actively chooses to visit and interact with the firm’s marketing communication efforts.

Similarly, Teo and Pian (2004) provide a Web adoption model in which they classify Web sites according to four levels. The third and fourth level, business integration and transformation, of their model are transactional Web sites. The first and second level, Web presence and prospecting, are informational Web sites. They describe prospecting as Web sites that provide extensive information about the organization and its products, feedback forms, e-mail support, and simple search but not the possibility to purchase (Teo & Pian 2004). Their research shows that the majority of firms are still in levels one and two and that especially larger firms have high levels of Web adoption (Teo & Pian 2004).

Lee and Grewal (2004) show that firm performance (i.e. Tobin’s Q) improves from adopting the Internet as an informational channel. The adoption of the Internet as a transactional channel is beneficiary to firms that have preexisting catalog operations (Lee & Grewal 2004). Considering these previous studies, we define an informational Web site (or channel) as follows.

Informational Web sites provide information about the firm, the firm’s offering, create an image, and/or attempt to build an ongoing relationship but do not offer the customer

any purchase opportunities.

1.2.2 Use of Informational Web Sites

Informational Web sites make sense as a response to consumers’ reasons for using a Web site, namely, to collect information and find an offline channel to buy a product. In general, the drivers of online


4

search/offline purchase may relate to (1) the product, (2) situational influences, (3) the consumer, and/or (4) the channel.

The product. Several studies show, in both more traditional and Internet-related settings, that product categories are associated with particular channels (e.g., Schoenbachler & Gordon 2002; Inman, Shankar, & Ferraro 2004). Prior research provides several product classifications that marketers may use to determine whether products are suitable for online selling (e.g., Peterson et al. 1997). For example, Broekhuizen (2006) indicates that the sale of physical products depends on factors such as additional delivery time and difficulties in returning faulty merchandise, which can inhibit consumers from buying online.

Situational influences. Nicholson, Clarke, and Blakemore (2002) argue that the reasons consumers use a particular channel depend on five dimensions of situational influence. First, the physical setting refers to the environment of the channel, such as store atmospherics or Web design. Second, the social setting focuses on the absence or presence of others, together with their social roles. Third, the temporal perspective refers to time constraints that affect shopping behavior. Fourth, the task definition is specific to the person and includes cognitive and motivational elements of the shopping situation. Fifth, antecedent states represent temporary conditions, such as moods, that may influence the use of a particular channel.

The consumer. Various consumer-related factors also might influence the decision to buy online. These factors include risk perceptions, preference for an offline channel, technology anxiety, and/or the need for tactile information. Forsythe and Shi (2003) show that Internet browsers (who do not buy online) are much more sensitive to the risks associated with Internet shopping. Hoffman, Novak, and Peralta (1999), Roy and Ghose (2006), and Schlosser, Barnett White, and Lloyd (2006) also indicate that privacy and trust are major barriers to purchases over the Internet. Järveläinen and Puhakainen (2004) demonstrate the influence of a consumer’s distrust in his or her own Web skills, as well as the effect of being accustomed to and satisfied with service through traditional channels. Keen, Wetzels, De Ruyter, and Feinberg (2004) suggest that for consumers, traditional offline stores remain the first choice to buy products. Technology anxiety, or managing technology paradoxes, provides another major barrier to consumers using self-service technologies, which includes shopping online (Mick & Fournier 1998; Meuter, Ostrom, Bitner, & Roundtree


5

2003). Citrin, Stem, Spangenberg, and Clark (2003) posit that consumers with high levels of tactile needs mainly use the Internet to browse, not to shop.

The channel. Verhoef, Neslin, and Vroomen (2007) provide three channel-related explanations for the searching online/buying offline phenomenon. First, the Internet tends to have a strong search attribute advantage, whereas a traditional store has a strong purchase attribute advantage. Noble, Griffith, and Weinberger (2005) confirm this search advantage of the Internet. Second, the Internet’s lack of channel lock-in—or a channel’s ability to retain customers across different stages, including search and purchase, during their decision-making processes—might prompt the searching online/buying offline phenomenon. Third, cross-channel synergies might occur between the Internet and traditional stores. Cross-channel synergy takes place if attitudes or behavior in one channel have a positive effect on attitudes or behavior in another. These three reasons offer focus specifically on multichannel environments.

1.3 MULTICHANNEL ENVIRONMENT

As a result of continuous technological progress, consumers can use a variety of channels during their decision-making processes. In addition to the traditional retail stores, consumers have access to the Internet, home shopping networks, catalogs, call centers, and kiosks. This increase in the number of channels produces new challenges with respect to managing a multichannel environment (for a review, see Neslin, Grewal, Leghorn, Shankar, Teerling, Thomas & Verhoef 2006), which has resulted in a new stream of research focused on multichannel customer management. In this section, we provide an overview of relevant studies in this area. We indicate in subsequent chapters how our study contributes to this literature. In this context, we define a “channel” as a customer contact point or medium through which the firm and the customer interact (Neslin et al. 2006). As such, an informational Web site is also considered as a channel.

Multichannel consumer behavior reflects how consumers choose channels during the different stages of their decision-making processes as well as how this choice affects their buying behavior. We consider informational Web sites as informational channels that influence consumer behavior in the offline (traditional store) channel. Hence, previous research in the area of multichannel consumer behavior


6

provides the framework for interpreting our findings. Research in this area focuses on channel choice, channel migration, multichannel search, and buying behavior and the effects of acquisition channels. Appendix I provides an overview of selected multichannel consumer behavior studies.

The studies into channel choice suggest that multichannel customers are more valuable than single channel customers and that the choice of a particular channel depends on demographics, shopping traits, perceived channel integration and marketing efforts (Bendoly, Blocher, Bretthauer, Krishnan & Venkataramanan 2005; Knox 2005; Kushwaha & Shankar 2005). However, according to studies focused on channel migration, customer use of the Internet may lead to fewer purchases (Ansari, Mela & Neslin 2006; Gensler, Dekimpe & Skiera 2007). Moreover, the use of a combination of channels does not automatically signal profitability for the firm (Sullivan & Thomas 2004). This area of research also indicates that distinct segments prefer one channel over another (Thomas & Sullivan 2005). In line with this finding, Dholakia, Zhao and Dholakia (2005) demonstrate that customers prefer to add an Internet channel rather than replace their old (usually offline) channel.

Studies investigating multichannel search and buying behavior are usually forced to work with survey data, and collecting actual customer information about actual search behavior in one channel and buying behavior in another channel remains a significant challenge (Sullivan & Thomas 2004). Previous research in this field shows that the majority of consumers combine channels during their decision-making process (Nicholson et al. 2002) and prefer the Internet for information search (Burke 2002). Furthermore, customers can free ride (i.e., use the provided information in their decision-making process but purchase elsewhere) on online-provided information (Van Baal & Dach 2005).

Multichannel consumer behavior is an emerging field of research, but extant studies indicate some major knowledge gaps. For example, marketers lack a clear understanding of actual usage behavior across channels (Nicholson et al. 2002; Montoya-Weiss, Voss, & Grewal 2003; Dholakia et al. 2005). Several studies have emphasized the need to measure and manage cross-channel influences when search occurs in one channel but purchase takes place in another (Sullivan & Thomas 2004; Thomas & Sullivan 2005; Kushwaha & Shankar 2005). Another area open to further investigation is determining differences in behavior


7

across customer segments or product categories (Gupta, Su, & Walter 2004). Finally, marketers require further research into multichannel behavior from a single firm perspective, as well as into the sequential process of search and purchase in different channels (Verhoef et al. 2007).

1.4 INFORMATION SEARCH

Information search research demonstrates that the average search is limited and that it continues until the value of an additional unit of information equals its cost (e.g., Newman & Staelin 1972). Consumers increase their search if the purchase is important, there is a need to learn more or if the information is easily obtained (Punj & Staelin 1983). Also, previous research makes the distinction between internal search, i.e. retrieving previously acquired information form memory, and external search, i.e. retrieving new information (Beales, Mazis, Salop & Staelin 1981). It is also shown that (1) seller provided information is the least costly external source, (2) it is inherently one-sided and (3) it combines factual data with nonfactual persuasive appeals (Beales et al. 1981). Apparent from these references, information search research is the foundation of current internet-related research.

Wu and Rangaswamy (2003) demonstrate how web site features can either decrease or increase the amount of search and subsequently influence consumers’ consideration sets. The availability of the Internet can also lead to less information search in offline sources. Ratchford, Lee and Talukdar (2003) test this for the automotive industry. They show that consumers, who searched on the net, spent substantially less time with not only the dealer but also with other sources.

With respect to searching online, two types of visitors are distinguished, namely goal-directed and experiential Internet users (e.g., Hoffman & Novak 1996). Goal-directed consumers focus on the task at hand and try to find the required information as efficiently as possible. These customers are more likely to be driven by external search. Experiential consumers browse the Internet for fun and may accidentally run into content, which is of interest to them. It is likely that experiential customers at some later point in time use internal search to access this information. Hence, the two types of consumers have completely different search strategies for the Internet that might be a result from the internal search versus external search disctinction.


8

1.5 PROBLEM DELINEATION

This dissertation contributes to fill gaps in existing knowledge about multichannel customer behavior by providing insights into the following aspects:

1 The effects of using informational Web sites on customer behavior,

2 The customer’s sequential process of search and purchase,

3 The specific effects of the introduction of an informational Web site for different customer segments and product categories,

4 Customer free-riding behavior, and

5 A methodology that can be used to measure these effects.

Effect from an informational Web site. Determining the effects of

using an informational Web site is a crucial effort, because customers may sense a necessity to buy offline depending on (1) the product, (2) situational influences, (3) the consumer or (4) the channel. Moreover, with informational Web sites customers will have to go elsewhere (another channel or firm) to buy products. Given that most firms use the Internet for information and communication purposes (see e.g., Dutta & Segev 1999; Huizingh 2000; Carroll 2002), insight into these types of Web sites are valuable. We therefore investigate both attitudinal and behavioral effects of the use of an informational Web site.

Sequential process of search and purchase. We measure the sequential process of search and purchase over time and study actual behavior in multiple channels, i.e. in an informational Web site and an offline store. We determine the cross-channel effects of searching in the online channel on buying in an offline channel. Furthermore, we determine the cross-channel effects of marketing efforts.

Customer segments and product categories. We describe the effects of using an informational Web site and the sequential process of search and purchase for different (1) customer segments, (2) product types (sensory versus nonsensory) and (3) product categories (e.g., clothing versus books). Given previous studies on transactional Web sites, we are interested in determining if previous findings hold in this particular setting.

Free-riding behavior. With our single-firm perspective, we show how free riding on online information affects offline customer buying behavior. Van Baal and Dach (2005) show, using survey data, that providing online information stimulates free-riding behavior, i.e.


9

customers use the online information but buy elsewhere. We add to this new phenomenon by investigating the free-riding behavior of customers through behavioral data.

Methodology. We provide several different methods to capture the effects of using an informational Web site on offline attitudes and behavior. Furthermore, we provide a method to capture the cross-channel effects over time, specifically, the sequential process of search and purchase.

The contributions of this dissertation emerge from the answers to the following main research questions:

1 How do online attitudes influence offline attitudes and behavior, and what are the effects of moderators on these relationships (Chapter 2)?

2 How does online behavior influence individual offline customer behavior (Chapter 3)?

3 What sequential cross-channel effects take place, and given an informational Web site/ offline store setting, what are the cross-channel effects of marketing efforts? How do these effects differ across customer segments and product types (Chapter 4)?

The first study, which we present in chapter 2, examines the effects

of an informational Web site on customer attitudes and behavior. With this study, we want to answer the question how online attitudes, collected some time after the introduction of the Web site, effect offline attitudes and behavior. The use of attitudes and behavior collected a year after the introduction of the Web site reduces a potential novelty bias. Besides this main effect, we are interested in possible moderators of these relationships. The Internet provides managers with many opportunities to target specific customer segments. Testing for moderators shows whether an informational Web site is more suitable for certain customer segments. Hence, this study aims to offer insight into how attitudes towards the informational Web site influence attitudes towards and behavior in the traditional store, how behavior in the informational Web site influences behavior in the traditional store and how these relationships are moderated by certain customer traits.

In chapter 3, we present the second study, which details how online behavior influences offline customer behavior, with a focus on shopping trips and money spent at the category level. The underlying question is


10

what happens at an individual customer level if an organization provides online information to its customers. Do customers spend more or less and do they go to the store more often or not? Previous research shows both positive and negative effects at the firm performance level (e.g., Biyalogorski & Naik 2003; Lee & Grewal 2004), however results at the individual customer level are scarce. For transactional Web sites, effects on spending may vary depending on product category (Gensler et al. 2007). Hence, we investigate whether customers change their spending across product categories due to the use of an informational Web site. This study offers further insights into the same-period effects of an informational Web site, effects at the product category level and free-riding behavior.

The third study gives more insight into cross-channel behavior and appears in Chapter 4. More specifically, we investigate the sequential process of search and purchase in the context of an informational Web site and a traditional store. We also investigate how the marketing efforts affect this process. Lastly, is it possible to distinguish context characteristics that influence these cross-channel effects? With this third and final study, we describe (1) the cross-channel effects in the sequential process of search and purchase in different channels, (2) how customer segments and product types influence cross-channel effects, and (3) how firms can manage cross-channel effects.

1.6 DEFINITIONS

We use the following definitions throughout this dissertation: • Channel = customer contact point or medium through which the

firm and the customer interact (Neslin et al. 2006). • Channel cannibalization = the extent to which the sales in channel

a erode sales in channel b. • Channel integration = the extent to which customers perceive the

different channels of a firm to be linked and experience seamless interactions across these channels.

• Channel lock-in = the ability of a channel to retain customers during the different stages (i.e. search and purchase) of the decision-making process.

• Cross-channel effects = the effects of a change in behavior in channel a on behavior in channel b.


11

• Cross-channel synergy = when attitudes or behavior in channel a have a positive effect on attitudes or behavior in channel b.

• Incumbent (traditional) channel = the channel the firm originally used to sell its products or services.

• Informational channel = a channel that offers information, possibly related to products, brands, and/or services but no opportunity to buy.

• Multichannel customer behavior (or multichannel behavior) = customers’ choice and use of multiple channels during the decision-making process.

• Multichannel environment (setting) = a variety of channels available to customers during the different stages of the decision-making process.

• Offline attitudes (or store attitudes) = customers’ perceptions of the “traditional” store.

• Offline behavior (or store/buying behavior) = the buying behavior of customers in traditional stores.

• Online attitudes (or site attitudes) = customers’ perceptions of the informational Web site.

• Online behavior (or site/search behavior) = customers’ (search) behavior on the informational Web site.

• Same-channel effects = the extent to which activities in channel a influence customer attitudes or behavior in channel a.

• Transaction channel = channel that offers information and the possibility to purchase products or services.

1.7 EMPIRICAL SETTING

1.7.1 Firm

The research project described herein pertains to a retail environment. The available data were collected from customers of a large, well-known, national retailer in the Netherlands that offers 68 department stores in major urban areas. In general, each outlet comprises 13 different departments, such as clothing, interior design, books, and cosmetics. The data are used for the empirical studies described in Chapters 2 – 4.

This firm is a member of a national joint loyalty program of 21 partner firms in the Netherlands. Customers collect credits by


12

purchasing from these different firms, which range from retail stores to banks to gasoline stations. In turn, they may exchange these credits to receive discounts on products sold by the member firms or theatre or airline tickets. This popular program was established in the early 1990s, and at the time of the data collection, half of all Dutch households were members of this loyalty program.

1.7.2 Informational Web Site

The focal firm introduced an informational Web site in March 2001. Prior to then, the firm had focused solely on a single channel, namely, its offline stores. The introduction of the Web site was its first activity in terms of a multichannel approach. With respect to communication, the firm previously relied on mass advertising, such as door-to-door, television, and outdoor advertising and occasionally some direct mail. The informational Web site was designed to support the firm’s offline activities, improve store image, and increase the likelihood that customers would buy in the various stores of the national retailer. To gain access to the site, customers had to register during their first online visit using their loyalty card number. Table 1-1 shows an overview of typical Web pages, information, types of products, and features of the informational Web site.

The site specifically provides customers with information about lifestyle issues, products offered in the stores, and promotions. Furthermore, it is not simply an electronic brochure but a theme-oriented site. The theme orientation translates into specific topics, such as personal care, interior design, and events. In addition, it offers opportunities for entertainment (e.g., sending an e-mail card to a friend). The site also offers customers the possibility to generate ideas about desired products online and then fulfill these desires in the store.

In addition to the permanent themes in Table 1-1, the site communicates price promotions, such as the annual Christmas promotions. The major promotions the department store offers during the year are also communicated by temporary site pages, which supplement traditional door-to-door features.


13

TABLE 1-1 OVERVIEW OF THEMES, INFORMATION AND FEATURES ON THE INFORMATIONAL WEB SITE

Theme Type of Information Examples of Products

Features

Welcome Overview of new information.

Diverse Help function, Web poll, personal information

Fashion and beauty

From fashion trends to optimizing your closet space.

Clothing, cosmetics

Children From children fashion to a day out.

Clothing, toys

Holidays From new destinations to packing a suitcase.

Suitcases, camping gear

Feeling fit From sports to relaxation.

Equipment, clothing

Enjoying your house

From organizing to enjoying your home.

All interior design products

Present planner Party and present tips for all kinds of occasions.

None except for present suggestions

Calendar, notice of events, present suggestions, and possibility to send an e-card.

Online brochure No product unrelated information

Varies Selection of products for a specific customer shopping list and product database.

Entertainment N.A. N.A. Puzzles, games, sweepstakes, and polls.

1.7.3 Customer Panel

To answer the research questions formulated in Section 1.4, we collected extensive information about the customers of the department store. We collected both offline and online behavior as well as attitudes, and we gathered these data over an extended period.

The offline behavior—specifically, offline purchases by 8,847 customers—is available from January 2000 to May 2002. The online behavior—site visits by 6,594 of the 8,847 customers—is available for March 2001 to May 2002. Therefore, our data show that during this period, 2,253 customers did not use the Web site. To collect online and offline attitudes, we surveyed customers three months after the introduction of the Web site in May 2001 and again a year later, in May 2002. In total, approximately 749 customers completed both versions of


14

the online survey. In May 2001, 3,128 customers completed the online survey; in May 2002, 4,865 customers did so. The surveys also provide customer demographic information (6,289 customers), supplemented with data from Acxiom1 at the zip code level (8,647 customers).2

Even though extensive information is collected about the customers of the department store, we acknowledge a number of limitations in the available information. Any organization is subject to competition, on which we have no data. Because we deal with many product categories sold in the department store, the potential competitors are many. Furthermore, the competitive profiles of the 58 outlets of the department store differ, which makes it impossible to collect data about all potential competitors. Besides given the department store its restrictions on the length of the survey, we were unable to collect potentially interesting variables such as customers their technology preferences. We do emphasize however that the aim of this dissertation is not to explain differences in the use of the Web site but to determine if using the Web site has an impact on the buying behavior in the offline stores.

1.8 OUTLINE

The remainder of this dissertation pertains to three main chapters in which we address the research questions. Table 1-2 displays the research issues studied in the empirical chapters (Chapters 2–4). In Chapter 5, we draw conclusions on the basis of the studies described in these chapters and discuss some implications and further research.

1 We express our gratitude to Acxiom for providing these data free of charge. 2 All the figures mentioned in this section pertain to precleaned data. Therefore, these figures may deviate in subsequent chapters.


15

TABLE 1-2 OVERVIEW OF RESEARCH ISSUES STUDIES IN THIS DISSERTATION

Chapter 2: Attitudinal Framework

Chapter 3: Individual Customer Behavior

Chapter 4: Cross-Channel Effects

Effects of informational Web site • Attitudinal • Behavioral

√

√

√

Free-riding behavior √ √ √

Decomposition of behavior √ √

Sequential process of search and purchase

√

Product categories (types) • Category-specific effects

√ √

√

Moderators (segments) √ √

Marketing efforts • Cross-channel marketing effects

√ √ √

17

2 The Effects of an Informational Web Site on Customer Attitudes and Behavior3

This study aims to determine the effects of using an informational Web site on offline attitudes and behavior. We are especially interested in how attitudes toward the informational Web site contribute to offline attitudes and behavior. Using structural equation modeling, we test a model that determines the relationships between attitudes toward two channels—the store and the Web site—and actual behavior in those channels. The results provide evidence of synergy effects between the store and the Web site in terms of attitudes, as well as negative effects of site attitude on offline behavior.

2.1 INTRODUCTION

In the context of multichannel environments, the recent evidence of cross-channel effects should come as no surprise. Peterson et al. (1997) were among the first to recognize that the marketing implications of the Internet involve integrated marketing activities. In this sense, integration implies that companies do not develop a specific e-commerce strategy; rather, their use of the Internet becomes part of their overall marketing strategy, which covers multiple informational and transactional channels.

Recent research provides several insights into the multichannel environment from organizational (e.g., Biyalogorsky & Naik 2003) and consumer (e.g., Montoya-Weiss et al. 2003) points of view. Most studies investigating multichannel environments from a consumer point of view find positive attitudinal effects, such as improved customer satisfaction (Shankar, Smith, & Rangaswamy 2003). However, more recent work reports negative effects (e.g., Ansari et al. 2006; Gensler et al. 2007). Gensler et al. (2007) indicate, for example, that given a multichannel environment, the number of nonloyal customers who opt to use none of the firm’s channels increases over time.

In response to these contradictory results, the call for generalizable insights into multichannel behavior has rung out (Rangaswamy & Van

3 This chapter is based on Teerling, Marije L. and Eelko K.R.E. Huizingh (2003),

How about Synergy: The Impact of Online Activities on Store Satisfaction and Loyalty, Working Paper 04F08, SOM, Groningen.


18

Bruggen 2005). Such generalizability cannot be reached by focusing solely on transactional channels; to understand customer multichannel behavior fully, research also needs to include customer behavior in informational channels, that is, channels that do not allow the customer to buy anything. This realization is emphasized in studies that indicate the difficulties of obtaining such unique but necessary data (Montoya-Weiss et al. 2003; Sullivan & Thomas 2004; Van Baal & Dach 2005). Therefore, this study focuses on the relationships between an informational Web site and an offline store by investigating: How site attitudes and behavior influence store attitudes and

behavior, How these relationships hold up given a longitudinal design, and Whether and how these relationships may be moderated.

The key concepts we study are site attitude, store attitude, site

behavior (i.e., site visits), and store behavior (i.e., store purchases). Specifically, we investigate how site attitudes influence store attitudes and behavior and how site behavior influences store behavior. We formulate hypotheses about the relationships among these concepts and test them empirically using data from 2,877 customers. For 422 customers, we investigate the relationships given a longitudinal design. Lastly, we perform median splits to determine whether the relationships are moderated by sociodemographics, for example.

This chapter is structured as follows: First, we present our framework and the background for the hypotheses. Second, we discuss methodology, including the data available to test our framework, our measure development, and aspects such as unidimensionality, validity, and reliability. Third, the subsequent section presents the findings. Fourth, we end this chapter with a discussion of the main findings and conclusions.

2.2 LITERATURE REVIEW

2.2.1 Multichannel Setting

The number of studies investigating multiple channels has increased substantially. Some studies focus on the performance of multiple channels from an organizational point of view, and several of these provide frameworks and guidelines for organizations that want to integrate their offline and online activities (e.g., Gulati & Garino 2000;


19

Otto & Chung 2000; Bahn & Fischer 2003). Geyskens, Gielens, and Dekimpe (2002), for example, demonstrate the benefits of an informational Web site on the overall financial performance of an organization, as measured by stock prices.

Other studies focus on different aspects of consumer behavior in the various channels. For instance, Danaher, Wilson, and Davis (2003) compare loyalty in online and offline environments and show that Internet usage benefits brands with a strong offline presence. Andrews and Currim (2004) investigate the behavioral differences between consumers attracted to online versus offline shopping. Still other researchers consider the reasons consumers prefer and use one channel over another and how this preference influences an overall relationship with the organization (Montoya-Weiss et al. 2003).

On the basis of how consumers use them, channels can be classified as transactional or informational. Consumers use informational channels to search for and retrieve information related to products, brands, and/or services, whereas transactional channels offer them the opportunity to purchase items in addition to information. Similarly, we can group current research into studies that consider (1) the effect of an additional transactional channel on customer behavior and (2) the effect of an additional informational channel on customer behavior. Table 2-1 gives an overview of multichannel studies, based on this classification.

TABLE 2-1 CLASSIFICATION OF CUSTOMER BEHAVIOR STUDIES RELATED TO MULTICHANNEL BEHAVIOR

Channel Addition Transactional Channel Informational Channel

Current transactional channel (store, Internet, catalog)

Ansari et al. 2006 Bendoly et al. 2005 Biyalogorsky & Naik 2003 Dholakia et al. 2005 Gensler et al. 2007 Montoya-Weiss et al. 2003 Shankar et al. 2003 Thomas & Sullivan 2005 Wallace et al. 2004

Current study

Most studies tend to focus on multiple transactional channels (see

Table 2-1). Several studies focus on determining the attitudinal effects of adding a transactional channel (e.g., Montoya-Weiss et al. 2003; Shankar et al. 2003; Wallace, Giese & Johnson 2004; Bendoly et al.


20

2005). Shankar et al. (2003) and Wallace et al. (2004) show improved firm loyalty as a result from offering customers multiple channels. Montoya-Weiss et al. (2003) show a positive effect on overall satisfaction. However, they show that satisfaction with the service quality in the current (traditional) channel can have a negative influence on the use of the additional (online) channel (Montoya-Weiss et al. 2003). This results might be explained by the preference for the traditional (offline) channel (e.g., Järveläinen & Puhakainen 2004; Keen et al. 2004). Overall, it seems that adding a transactional channel improves customer attitudes towards the firm.

2.2.2 Proposed Model and Hypotheses

Figure 2-1 shows the proposed model, which depicts how the effects of an informational Web site can be determined. Specifically, the proposed model determines the effects by focusing on (1) channel attitudes, (2) channel behavior, and (3) moderators that influence the cross-channel effects. With informational Web sites, the firm can only benefit if customers go offline to buy items. Hence, we focus on the effect of the online attitudes and behavior on offline attitudes and behavior. In addition to the main constructs and moderators, the model takes several antecedents into account.

Store attitude

Site attitude Site behavior

Store behaviorAntecedents

Antecedents

Moderators:*Demographics (age, gender, education)*Involvement*Channel integration

H1:+H2:+

H3:+

FIGURE 2-1 CONCEPTUAL MODEL: THE INFLUENCE OF AN INFORMATIONAL

WEB SITE ON STORE ATTITUDES AND BEHAVIOR Notes: Dashed lines indicate hypotheses, + indicate an expected positive result.


21

The first and second hypotheses (H1 and H2) relate site attitude to store attitude and store behavior, respectively. Shankar et al. (2003) argue that customer attitudes toward a service provides differ between its online and offline channels by showing the strengthening effect of an online channel on overall company satisfaction and loyalty. Montoya-Weiss et al. (2003) and Wallace et al. (2004) also indicate positive attitudinal effects of the additional online channel. Finally, Internet search attractiveness positively influences the choice or intention to purchase through the offline channel (Verhoef et al. 2007). Therefore, we expect positive relationships between site attitude and store attitude as well as between site behavior and store behavior. We also expect that customers with a positive site attitude engage more in positive buying behavior; for instance, they may shop more often or buy more products.

The third hypothesis (H3) relates site behavior to store behavior. Insights into this relationship thus far have been mixed. Ansari et al. (2006) and Gensler et al. (2007) both find that online behavior can have a negative effect on offline behavior. Ansari et al. (2006) indicate that customers who use the Internet more report lower purchase quantities, which implies that online behavior may lead to less offline purchase behavior. However, evidence from practice indicates that multichannel customers spend more than single-channel customers (DoubleClick 2004), which would indicate a positive relationship. Kushwaha and Shankar (2005) also report that multichannel customers buy more often, buy more items, and spend more than do single-channel customers. Because a purely informational site requires customers to purchase offline, we hypothesize that site behavior has a positive effect on store behavior.

Table 2-2 provides an overview of the cross-channel relationships found in previous studies, most of which apparently focus on the relationships between attitudes or between behaviors. In the case of attitudes, the findings are consistent, but for behavior, research has found both negative and positive effects. Only a few studies consider the cross-channel relationship between attitudes and behavior; in particular, Verhoef et al. (2007) capture this relationship through search attractiveness and purchase intentions. Our study contributes to this emerging field by capturing the relationship between site attitude and actual offline buying behavior. We formally propose the following hypotheses:

H1. Site attitude has a positive effect on store attitude.


22

H2. Site attitude has a positive effect on store behavior.

H3. Site behavior has a positive effect on store behavior.

TABLE 2-2 OVERVIEW RESULTS FROM PREVIOUS STUDIES ON THE MAIN THREE HYPOTHESES

Site Attitude Store Attitude (H1)

Site Attitude Store Behavior (H2)

Site Behavior Store Behavior (H3)

Positive relationship found by:

Burke 2002 Shankar et al. 2003 Montoya-Weiss et al. 2003 Wallace et al. 2004

Verhoef et al. 2007 Kushwaha & Shankar 2005 Verhoef et al. 2007 Dholakia et al. 2005

Negative relationship found by:

Ansari et al. 2006 Van Baal & Dach 2005 Ratchford et al. 2003

Expectation: + + + Notes: H1 = hypothesis 1, H2 = hypothesis 2, H3 = hypothesis 3.

2.2.3 Moderating Effects of Customer Traits

Several studies indicate differences across customer traits in the use and effects of multiple channels (e.g., Burke 2002; Balasubramanian, Raghunathan & Mahajan 2005). We investigate whether the relationships between the channels are moderated in case of an informational channel.

The proposed relationships (H1-H3) can be viewed from two angles. Firstly, we consider the relationships between attitudes (H1) versus those between attitudes and actual behavior (H2). The relationship between attitudes and actual behavior is investigated in studies not related to multichannel behavior (e.g., Mittal & Kamakura 2001 Seiders, Voss, Grewal & Godfrey 2005). Besides this angle, the relationships (H1 and H2) are investigated in multichannel studies (e.g., Burke 2002). To determine any possible moderators, we review previous studies that employ either of these angles.

Mittal and Kamakura (2001) found in a general setting that the relationship between attitudes and behavior is weaker for male, more educated younger adults. Research in a multichannel setting also shows that these subjects are more interested in technology (Burke 2002). On the basis of Burke’s (2002) work, we expect the hypothesized relationships to be stronger for male, more educated younger customers. However, their greater interest in technology might make


23

them more likely to order online. Finally, ordering online is not possible through an informational Web site, a limitation that might weaken the proposed relationships for male, more educated younger customers.

H4. The proposed relationships are weaker for

H4a. males versus females.

H4b. more versus less educated customers

H4c. younger versus older customers.

Highly accessible attitudes likely bias interpretations of relevant information and strengthen the relationship between attitudes and behavior (Olson & Zanna 1993). With stronger involvement, information gets processed more carefully. Seiders et al. (2005) show for instance that more involved customers spend even more when their satisfaction is high. In a multichannel setting, Wallace et al. (2004) show that customers with greater purchase involvement, experience a stronger relationship between multichannel service value and positive disconfirmation than do customers with a less purchase involvement. Therefore, we expect involvement to strengthen the proposed relationships.

H5. The proposed relationships are stronger for highly involved customers than for less involved customers.

With the advent of multiple channels, firms can choose to operate along a spectrum of possibilities varying from completely separate channels to fully integrated channels (Gulati & Garino 2000). With independent channels, customers may not perceive the channels as part of the same organization if the firm tries to provide customers with multiple channels that work together to fulfill their needs (Bendoly et al. 2005). Perceived channel integration, or the extent to which customers perceive integration among channels, such as the ability to order a product online and pick it up offline, may create cross-channel synergies. Further research is needed to investigate this relationship (Verhoef et al. 2007), but Bendoly et al. (2005) indicate that perceived channel integration may be associated with a reduced likelihood of firm switching when the initial channel fails. Perceived channel integration also promotes loyalty and within-firm channel switching. We expect that, especially for informational Web sites, perceived channel integration strengthens the proposed relationships.


24

H6. The proposed relationships are stronger for customers who perceive higher channel integration than for those who perceive less channel integration.

2.2.4 Store/Site Attitude

Attitudes are characterized as evaluations of a particular entity with some degree of favor or disfavor, which are represented in memory (Olson & Zanna, 1993). Usually, researchers distinguish among the affective, cognitive, and behavioral antecedents and consequences of attitudes. Engel, Blackwell and Miniard (1993) define attitudes as overall evaluations that result from direct contact with the entity. Furthermore, they indicate that (1) the affective component represents a person’s like or dislike of an entity, (2) the cognitive component represents his or her knowledge and beliefs about the entity, and (3) the conative component refers to the person’s behavioral tendencies (Engel et al. 1993).

Similarly, in marketing research customer satisfaction is usually defined as the overall evaluation of an entity, after a purchase or consumption experience (e.g., Oliver 1999). Specific to the setting of this dissertation, Macintosh and Lockshin (1997) define store satisfaction as the customer's overall evaluation of the store experience.

Loyalty consists of two dimensions, namely attitudinal and behavioral loyalty (see e.g., Mittal & Kamakura 2001). Attitudinal loyalty relates to the feelings customers have toward a product/service or organization (affective component), whereas behavioral loyalty relates to behavioral intentions to continue buying the same product/service from that organization or store (Srinivasan, Anderson & Ponnavolu 2002).

Taylor and Hunter (2002) show that the conceptual domains of satisfaction and loyalty in an online setting are similar to those in an offline setting. Therefore, store and site attitude in this study reflect overall evaluations, future intentions and preferences for a channel.

2.2.5 Antecedents of Store/Site Attitude

Store attitude antecedents range from physical characteristics, such as store interior, price, and merchandise, to softer characteristics, such as personnel, service, employee courtesy, store personality and pleasure (Van Kenhove, De Wulf & Van Waterschoot 1999). Sirgy, Grewal and Mangleburg (2000) and Tang, Bell and Ho (2001) focus on more physical store attributes, whereas others concentrate more on the


25

softer characteristics (e.g., Kelly & Smith 1983; Donovan, Rossiter, Marcoolyn & Nesdale 1994). Baker, Parasuraman and Voss (2002) have developed a model of store choice and patronage intentions that captures both physical and softer characteristics. In this study, we focus on a mix of these antecedents: store interior or design perceptions, monetary price perceptions, merchandise, and personnel or store employee perceptions.

Internet research has proven that several factors related to the content and design (e.g., navigation, graphic style) of a Web site have a positive influence on site satisfaction or usage (e.g., Venkatesh & Davis 2000; Montoya-Weiss et al. 2003; Van der Heijden 2003). With his Technology Acceptance Model (TAM), Davis (1989) predicts the adoption and use of information technologies and shows that perceived usefulness (content) and ease of use (design) relate to the overall acceptance of information systems. Because, e-commerce is highly technology driven, TAM also may be applied in an Internet setting (e.g. Pavlou & Fygenson 2006). In our study, we focus on the two main antecedents of site attitude: content and design.


To test the proposed model, we collected data among customers of a large, well-known, national retailer in the Netherlands, as discussed in Section 1.6.

The informational Web site is theme oriented, supporting offline activities with the aim to increase the likelihood of purchases in the stores of the retailer. The Web site was introduced in March 2001. To gain access to it, customers registered with the loyalty program identification number that designates their loyalty card, which they use to collect credits when buying products in stores.

2.3.1 Data

Attitudes, store behavior, and site behavior may be linked at the individual customer level by the loyalty card identification number. Customers were asked by e-mail to participate in an online survey and offered a small incentive in return for their participation. The survey, held in May 2001 and May 2002, included rule-based checks and a direct link to the online database to ensure fewer errors in the data collection process. The e-mail message contained the Web address of the survey, which linked to a username- and password-protected Web


26

site. Both the username and password appeared in the initial e-mail. From the 25,000 customers who visited the Web site, we collected 4,787 rows of data for the May 2002 survey. After careful screening of the data (to eliminate incomplete responses, yea- and nay-saying, erroneous loyalty numbers, and multiple entries), we emerge with a final sample (May 2002) that contains only those 2,877 customers whose channel attitudes and behaviors are available. For 422 customers, data for both surveys and behavior for the period March 2001- May 2002 is available.

2.4 PROPOSED METHODOLOGY

The proposed model includes several antecedents, overall attitudes, and actual behavior. The antecedents with respect to the store are store interior, personnel, merchandise, and price, and those that pertain to the site are content and design. Actual behavior (site visits and store purchases) is observed directly. We acknowledge that the antecedents and attitudes cannot be observed directly but require observable indicators, which suffer from measurement error (Steenkamp & Baumgartner 2000). However, the aim of our proposed model is not to predict actual behavior but to explain the relationships between the channels in terms of customer attitudes and behavior. Because of these characteristics and our aim to gain insight into the observed marketing phenomena, we employ the structural modeling approach.

The proposed model is estimated with structural equation modeling (SEM) using LISREL (version 8.54). Anderson and Gerbing (1988) recommend a two-stage approach with fixed loadings and error variances when A tentative theory underlies the constructs (Hair, Anderson, Tatham

& Black 1998), Measures are less reliable (Hair et al. 1998), and The fit of the structural model deteriorates because of a relatively

large number of parameters in the measurement model (Steenkamp & Van Trijp 1991). For both the measurement and structural parts of the model, we

use the Chi-square ( 2χ ) index to assess model fit, along with other fit

indices, such as the root mean square error of approximation (RMSEA), non-normed fit index (NNFI), and comparative fit index (CFI) (Byrne 1998). The generally accepted standards — insignificant Chi-square,


27

Chi-square index ( df/2χ ) between 2 and 5, RMSEA < .08, square root

mean residual (SRMR) < .05, NNFI, CFI, and goodness-of-fit index (GFI) ≥ .90 — were taken into account to determine the model fit (Jöreskog & Sörbom 1993; Byrne 1998). The standardized coefficients in the structural model indicate the relative importance of the relationships and closely approximate effect sizes comparable to standardized beta coefficients in regression analyses (Hair et al. 1998).

2.4.1 Specification: Measurement Part of the Model

The measurement model includes relationships between the latent constructs and their observed items (i.e., a factor-analytical model). The specification generally used in LISREL includes an equation for both exogenous (i.e., X-variables) and endogenous (i.e., Y-variables) variables (Byrne 1998). For our proposed model, we include six latent exogenous variables that represent the store and site antecedents, measured with 21 items in total. We include two latent endogenous variables, namely, store attitude and site attitude, measured with 6 items. The specification of the measurement model is as follows:

(1) xX ξ δ= Λ + , and

(2) yY η ε= Λ + ,

where

X = a (21 x 1) vector of observed exogenous items for the store antecedents and site antecedents;

xΛ = the (21 x 6) matrix of loadings, showing which observed exogenous item loads on which latent exogenous variable;

ξ = a (6 x 1) vector of latent exogenous variables, including personnel, store interior, price, merchandise, content, and design;

δ = a vector of error terms with expectation zero and uncorrelated with ξ ;

Y = a (6 x 1) vector of observed endogenous items for store attitude and site attitude;

yΛ = the (6 x 2) matrix of loadings, showing which observed endogenous item loads on which latent endogenous variable;

η = a (2 x 1) vector of latent endogenous variables, including store


28

attitude and site attitude; and ε = a vector of error terms with expectation zero and uncorrelated

with η .

2.4.2 Measurement Constructs

We use scales derived from previous studies (see Table 2-3). All items are measured on a five-point Likert scale ranging from “completely disagree” to “completely agree”.

TABLE 2-3 MEASUREMENT AND ORIGIN OF THE SCALES Previous Studies Sample Scale Item Store interior (3) The interior of X is attractive. Personnel (3) The salespeople at X know their

products well. Merchandise (3) X has different varieties of the same

product. Price (3)

Maddox 1982 Van Kenhove et al. 1999 The prices X charges are

reasonable.

Store attitude (3) Reynolds & Beatty 1999

I enjoy shopping at X.

Content (4) I find the Web site informative. Design (5)

Davis 1989

The layout of the Web site is attractive.

Site attitude (3) Van der Heijden 2003

I intend to visit X on a regular basis.

Notes: The number of items used to measure the scale appear in brackets.

With a pre-test, we confirm the proper translation, quality, and applicability of the scales to this study. Scales unrelated to the Web site (i.e., store attitude and its antecedents) were tested using a convenience sample of master’s students and university staff (n=50). The scales measuring online attitudes were not tested among this sample because of the private character of the Web site. With data from the first 100 customers who participated in the study, we tested these latter scales. Descriptive analyses and frequencies were performed to test for normality of the variables. Overall, the scales meet the criterion of normality.

2.4.3 Construct Validity

To purify the measures, we use both traditional approaches, such as exploratory factor analyses and Cronbach’s alpha, and LISREL (Steenkamp & Van Trijp 1991).


29

Traditional approach. The exploratory factor analyses performed on both the test and the final sample suggests unidimensionality. With regard to reliability, the Cronbach’s alpha reveals the consistency of the scales. In response to these tests, some of the scales were altered. In the final sample, all Cronbach’s alphas exceed .70 (see Table 2-4).

For the test samples (offline n = 50 and online n = 100), and the final sample, we analyze two other measures of internal consistency: item-to-total, and interitem correlations (e.g., Churchill 1979; Hair et al. 1998). All the correlations for the final sample are significant at a .01 significance level. Most scale items meet the required standards for internal consistency (item-to-total correlations >.50, inter-item correlations > .30). In Appendix II, we provide the inter-item and item-to-total correlations.

LISREL approach. LISREL estimated a confirmatory factor model

and the fit of this model is within the specified norms ( df/2χ = 4.39;

RMSEA = .049; NNFI = .92 and CFI = .93). Unidimensionality is determined through the model fit in terms of the standardized residuals (Steenkamp & Van Trijp 1991). The results show that the standardized residuals do not exceed the misspecification value of 2.58 (maximum value = 1.08). Table 2-4 depicts various measures that can assess other aspects of construct validity.

Convergent validity is established according to the lambdas of the measurement model. All item loadings are significant, indicating convergent validity (Anderson & Gerbing 1988). For merchandise, one item does not exceed the minimum loading of .70, but we chose not to remove it, because doing so does not improve the scale. The average variance extracted (AVE) exceeds the recommended level of .50 for all constructs, indicating that the variance captured by the constructs is greater than the variance due to errors (Fornell & Larcker 1981).

TABLE 2-4 RELIABILITY MEASURES OF THE MEASUREMENT MODEL

Construct reliability (>0.70)

Average loading

Lowest loading

Variance extracted (>0.50)

Cronbach’s alpha

Store interior .89 .85 .74 .72 .84 Price .90 .87 .78 .75 .85 Personnel .86 .81 .70 .67 .81 Merchandise .76 .72 .60 .52 .71 Content .93 .87 .81 .76 .90 Design .93 .85 .87 .73 .91 Store attitude .90 .87 .78 .76 .86 Site attitude .90 .86 .77 .74 .86


30

Discriminant validity is achieved when: The correlations between the constructs differ from unity (i.e., the

confidence interval for each pairwise correlation does not include the value of 1.0),

The 2χ difference test indicates that two constructs are not

perfectly correlated when collapsed into a single construct (e.g., Anderson & Gerbing 1988; Steenkamp & Van Trijp 1991) and,

The squared correlation between two constructs exceeds the AVE for each of the two constructs (Hair et al. 1998). Table 2-5 shows the correlations among the latent factors.

TABLE 2-5 CORRELATIONS AMONG LATENT CONSTRUCTS

1 2 3 4 5 6 7

1. Store Interior 1

2. Personnel 0.5 1

(.10)

3. Merchandise 0.64 0.56 1

(.08) (.11)

4. Price 0.42 0.46 0.59 1

(-.11) (.12) (.11)

5. Store Attitude 0.51 0.55 0.74 0.57 1

(.09) (.10) (.07) (.09)

6. Content 0.43 0.42 0.52 0.32 0.47 1

(.09) (.10) (.09) (.12) (.09)

7. Design 0.5 0.45 0.53 0.39 0.51 0.87 1

(.09) (.11) (.11) (.13) (.10) (.02)

8. Site Attitude 0.43 0.42 0.53 0.33 0.53 0.84 0.82

(.10) (.11) (10) (.13) (-.09) (.03) (.04) Notes: Standard errors appear in brackets.

The results show that each of the constructs satisfies the criterion of pairwise correlation, though the correlations between some

constructs are reasonably high. Furthermore, the 2χ difference test

confirms the pairwise tests. All models with a collapsed pair of constructs reveal a worse fit than the original model. The third criterion is not satisfied in two cases. For merchandise, the AVE is .03 lower than the squared correlation with store attitude. For design, the AVE is


31

.03 lower than the squared correlation with content. For all other constructs, this criterion of discriminant validity is satisfied. Because most of the tests satisfy the conditions of construct validity, we proceed with the structural model.

2.4.4 Specification: Structural Part of the Model

The structural part of the model defines the relationships among the unobserved (or latent) attitudinal variables and observed behavioral variables. It specifies which variable directly or indirectly influences changes in the values of other variables (Byrne 1998). It is generally specified as follows:

(3) ζηΒξΓη ++= ,

where

η = a (m x 1) vector of unobserved endogenous variables, Γ = an (m x n) matrix of coefficients that relates the n exogenous

factors with the m endogenous factors, ξ = a (n x 1) vector of latent exogenous variables, Β = an (m x m) matrix of coefficients that relates the m endogenous

factors to one another, and ζ = an (m x 1) vector of residuals with zero expectation, and

uncorrelated with η and ξ .

In this study, there are four endogenous variables: store attitude

( 1η ), site attitude ( 2η ), store behavior ( 3η and site behavior ( 4η ). We

include six exogenous variables, namely personnel (ξ1 ), store interior

(ξ2 ), price (ξ3 ), merchandise ( 4ξ ), content ( 5ξ ) and design ( 6ξ ). The

equations of the structural part of the model are represented as:

(4)

11,1 1,2 1,3 1,4 21

2 35,2 6,23 4

546

1,2 1 12 2

3,1 3,2 3,4 3 33 44,2

0 00 0 0 00 0 0 0 0 00 0 0 0 0 0

0 0 00 0 0 0

00 0 0

ξτ τ τ τ ξη

η ξτ τη ξ

ξηξ

β η ςη ς

β β β η ςη ςβ

⎡ ⎤⎡ ⎤ ⎢ ⎥⎡ ⎤⎢ ⎥ ⎢ ⎥⎢ ⎥ = ⎢ ⎥ ⎢ ⎥⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎢ ⎥

⎣ ⎦⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥+ +⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦⎣ ⎦⎣ ⎦


32

Because an informational Web site does not provide customers with opportunities to purchase, buying takes place in offline stores. Store behavior reflects total customer buying since the introduction of the Web site, over the period March 2001 to May 2002, measured in total money spent. The same holds for site behavior, though it is measured in total pages viewed online. For the variables measuring store behavior and site behavior, we use the logarithmic transformation.

2.4.5 Longitudinal Design

A structural equation model including measurements over time, i.e. a longitudinal design, is estimated to validate whether the relationships between attitudes and behavior hold up in the case of temporal ordering (see Rindfleisch, Malter, Ganesan, & Moorman 2006 on causal inference). For 422 customers, data are available from both surveys (May 2001 and May 2002) with regard to store and site behavior (March 2001 – May 2002). We estimate the relationships (from figure 2-1) with these data for which Figure 2-2 shows the path diagram.

Store behaviorξ1

Site behaviorξ2

Store behaviorη3

Site behaviorη4

Store attitudeη1

Site attitude η2

Store attitude η5

Site attitude η6

Site antecedentsξ7 – ξ8

Site antecedentsη11 – η12

Store antecedentsξ3 – ξ6

Store antecedentsη7 – η10

β1,2

τ1,3:1,4:1,5:1,6

β1,1

β5,1

β5,7:5,8:5,9:5,10

β5,6

β5,3

β6,2β6,11:6,12

β6,4

β3,1

β3,2

β4,2

β3,4

τ2,7:2,8

τ2,2

τ7,3:8,4:9,5:10,6

τ11,7:12,8

τ1,2

τ4,2

β3,1

March ‘01 May ‘01 April ‘02 May ‘02

FIGURE 2-2 PATH DIAGRAM FOR THE LONGITUDINAL DESIGN WITH TIMELINE


33

The model includes the main variables of interest, namely store attitude, site attitude, store behavior and site behavior, each measured for two points in time. The number of endogenous and exogenous variables is different from equation (4), given that the number of variables has changed and the conceptual model is estimated in accordance with the timeline. Nevertheless, the main endogenous variables (η1 – η4) from equation (4) are the same in the longitudinal design.

The six antecedents of store and site attitude (personnel, store interior, price, merchandise, content and design), are also measured at two points in time. In Figure 2-2, we specify the structural relationships among the latent variables. We leave out the specification for the measurement model for clarity purposes. As is apparent in Figure 2-2, the antecedents measured in May ’01 or May ’02 influence site and store attitude at May ‘01 respectively May ‘02. Moreover, we estimate the effect from for instance the price perception in May 2001 on the price perception in May 2002. Thus, the antecedents measured in the survey from May ’01 also influence the antecedents measured in the survey from May ’01, which is common in longitudinal designs for structural equation models (Jöreskog & Sörbom 1993).

Besides the temporal ordering validation, a common method factor could show if common method variance from the surveys affects the estimated coefficients (e.g., Rindfleisch et al. 2006). We attempted to estimate the common method factor, however the model suffered from identification problems due to only two similar items in both surveys for the antecedent ‘price’.

2.4.6 Measurement Moderators

H4-6 propose that the relationships in the conceptual model are moderated by various factors. We perform a median split to test this moderation. H4 centers on sociodemographics: age, gender, and education. The median age in the panel is 38.59 years, a little more than half (55.4%) of the customers are female, and 46.5% of them have a college education.

The other two moderators (H5 and H6) are measured with multi-item constructs, namely, involvement and channel integration. We define involvement as the extent to which customers have a deep interest in a product or activity, which in our setting refers to the store and its products. Channel integration is the extent to which customers


34

can use multiple modes of fulfillment during their decision-making process (Bendoly et al. 2005).

We measure involvement as the number of product categories in which the customer indicates an interest. The mean number of product categories in which an interest was indicated is 4.4. Customers can indicate an interest in at most 15 product categories. The median is five product categories, 33.4% of customers indicated no interest in any of the categories.

We measure channel integration through three items, which indicate whether customers use the online information to shop offline. The exploratory factor analysis shows that the three items explain 61% of the variation. The Cronbach’s alpha is .69, the interitem correlation ranges from .39 to .43 and the item-to-total correlation ranges from .48 to .52.

2.4.7 Moderation and Validation Approach

For the moderation and validation approaches a similar test is used, namely the Chi-square difference test proposed by Dabholkar and Bagozzi (2002). For the moderation, we compare two groups based on a median split. For the validation, we perform a random split analysis, and a self-selection split analysis.

Before calibrating and estimating the proposed model, we randomly split the data set into an estimation (n = 1,433) and a validation (n = 1444) sample. To test for any significant differences between them, we follow the rigorous procedure described by Dabholkar and Bagozzi (2002). This procedure uses a Chi-square difference test to determine the presence of significant differences. It is adapted from Jöreskog and Sörbom’s (1993) procedure for verifying whether factor loadings are essentially the same or significantly different across two groups. Therefore, this procedure can be used for the random split or the self-selection split, as well as to determine whether moderating variables have a significant effect on the relationships.

We conduct two tests for each split/moderator variable using four models (A–D) for each split (Dabholkar & Bagozzi 2002). In model A, all factor loadings and error variances of the items for endogenous variables are constrained across groups. Model B leaves the factor loadings free but constrains the error variances. Model C frees both the factor loadings and the error variances. Finally, model D constrains the factor loadings but frees the error variances.


35

The first test compares models A and B (and models D and C). Differences between models A and B are due to factor loadings. The second test compares models A and D (and models B and C), and the differences between models A and D are attributable to error variances

in the dependent variables. The difference reflects the 2χ difference

between the two models, divided by the change in degrees of freedom

(i.e. Χ2 difference for one degree of freedom, dfΔΔ /2χ ).4

2.5 FINDINGS

In this section, we report the findings for four analyses. First, we discuss the findings for the proposed model. Second, we report the results for the longitudinal design. Third, we discuss the results for the moderators. Fourth and finally, we validate the proposed model.

2.5.1 Estimates for the Proposed Model

Table 2-6 shows the results for the proposed model. The fit measures show a reasonable fit for the model. Although, the Chi-square is significant, this result is expected because the statistic is sensitive to large sample sizes (e.g., Hair et al. 1998). The ratio of the Chi-square to

the degrees of freedom ( df/2χ ) is greater than the norm. The RMSEA,

SRMR, NNFI, and CFI are all within recommended levels. All standardized coefficients, except for store interior, are

significant. The proposed model explains more than 60% of the variation in site and store attitude. The attitudes explain a small part of the variation in the behavior variables, which is not surprising considering that previous studies show that the relationship between attitudes and actual behavior is extremely hard to capture (e.g., Mittal & Kamakura 2001; Chandon, Morwitz, & Reinartz 2005).

The results show that on an attitudinal level, the relationship between site and store attitude is positive. This relationship, compared to the store antecedents, is also of reasonable magnitude. The relationship between site attitude and behavior is positive when it concerns the site and negative when it concerns the store. Customers with a positive attitude towards the site have less store behavior than customers with a less positive site attitude. The relationship between

4 For a more extensive explanation of the test for moderating effects, see

Dabholkar and Bagozzi (2002).


36

store attitude and store behavior is positive and has the biggest impact. Lastly, site behavior has a positive relationship with store behavior. Hence, customers with more site behavior also have more store behavior.

TABLE 2-6 RESULTS FROM PROPOSED STRUCTURAL EQUATION MODEL (N = 1,433)

Dependent variables

Explanatory variables Site

Attitude Store

Attitude Site

Behavior Store

Behavior

Store Interior 0.00

Personnel 0.14

Price 0.18

Merchandise 0.47

Content 0.53

Design 0.36

Site Attitude 0.17 0.15 -0.14 Site Behavior 0.08 Store Attitude 0.20

R2 0.75 0.61 0.02 0.04 2χ 2464.12

Degrees of freedom 399 df/2χ 6.18

RMSEA 0.06

SRMR 0.046

NNFI / CFI 0.93 / 0.93

GFI 0.89 Notes: Bold parameter estimates are significant at the 5% level.

2.5.2 Results Longitudinal Design

The fit of the longitudinal design model5 is worse than that of the cross-sectional model6 and, for some fit indices, falls below the norm

5 The longitudinal design model determines to what extent the cross-sectional

model can be validated given the temporal aggregation of the cross-sectional model. Figure 2-2 shows the path diagram for this model.

6 The cross-sectional model tests the conceptual model (see Figure 2-1).


37

( df/2χ = 3.6; RMSEA = .078; SRMR = .13; NNFI = .74; CFI = .74; GFI =

.65). The R2 for store attitude in both years is .66 and .84 respectively. For site attitude, the R2 is .87 for 2001 and .96 for 2002. The R2 of store behavior for both periods is .01 and .07 respectively, for site behavior the R2 for the second period is .03.

Even though the fit of the model falls below the norm, a review of these coefficients provides additional insights into the reliability of the cross-sectional model. For example, the relationship between site attitude and store attitude holds up across time with standardized coefficients of .10 (t-value = 2.52) in 2001 and .17 (t-value = 4.81) in 2002. For the relationship between site attitude and store behavior, which is in temporal order for the longitudinal model, we find a coefficient of -.17 (t-value = 3.13), which confirms the result found in the cross-sectional model. For the relationship from site behavior to store behavior, we find a coefficient of .09 (t-value = 1.76) in 2001 and .11 (t-value = 2.37) in 2002.

The (standardized) carry-over coefficients for the main variables (β3,1:β5,1:β6,2:τ4,2 from Figure 2-2), i.e. the effect of for instance store attitude in 2001 on store attitude in 2002 are all significant expect for the effect of site attitude in 2001 on site attitude in 2002 (coefficient = .03, t-value = 1.15). This might be caused by the relative newness of the Web site in the survey of 2001 (i.e. only 3 months online). The effect of store behavior in the first period on the second period is .20 (t-value of 4.17). For store attitude the carry-over coefficient is .20 (t-value = 4.37). For site behavior the carry-over coefficient is .14 (t-value = 2.93). The carry-over coefficients for the store antecedents (τ7,3:8,4:9,5:10,6 from Figure 2-2) are all significant and range from .52 - .64 (t-value from 11.02 – 13.11). The carry-over coefficients for the site antecedents (τ11,7:12,8 from Figure 2-2) are all significant and are .52 and .67 (t-value respectively 11.49 and 16.61). Overall, the results of the longitudinal model confirm the results of the cross-sectional model.

Table 2-7 shows an overview of the expectations and results

pertaining to the three main hypotheses. The first hypothesis argues that site attitude relates positively to store attitude, and the results confirm a positive cross-channel relationship, in support of findings from previous studies such as Shankar et al. (2003) and Montoya-Weiss et al. (2003). The second hypothesis states that site attitude relates positively to store behavior, but this hypothesis cannot be confirmed,


38

because the results indicate a negative relationship. Verhoef et al. (2007) find a positive effect from Internet search on store purchase. The setting of both studies may explain the difference in results. Verhoef et al. (2007) study these effects in a non-specific setting. When customers are reviewing channels in this more general setting, e.g. before buying a product offline, did you search for information online, the likelihood of a positive relationship is higher. The third hypothesis posits that site behavior relates positively to store behavior and is confirmed by the findings. Ratchford et al. (2003) find a negative effect from the internet on information search behavior, not necessarily on buying behavior, in the traditional channel. Our findings may deviate from Van Baal and Dach (2005) considering that we study aggregate behavior over a year, instead of a single purchase occasion.

TABLE 2-7 OVERVIEW OF RESULTS FOR MAIN HYPOTHESES

Site Attitude Store Attitude

(H1)

Site Attitude Store Behavior (H2)

Site Behavior Store Behavior (H3)

Positive relationship

Burke 2002 Shankar et al. 2003 Montoya-Weiss et al. 2003 Wallace et al. 2004

Verhoef et al. 2007 Kushwaha & Shankar 2005 Verhoef et al. 2007 Dholakia et al. 2005

Negative relationship

Ansari et al. 2006 Van Baal & Dach 2005 Ratchford et al. 2003

Expectation + + + Result + - +

2.5.3 Moderating Results of Customer Traits

Certain variables may moderate the proposed relationships, as we discussed in Section 2.2.3. We test the moderation effects by applying the Dabholkar and Bagozzi (2002) method (see Section 2.4.7), which we also use to validate the proposed model. Table 2-8 shows which of the moderators involves a significant moderation effect due to factor loadings or error variances. For each moderator, we estimate the proposed model on two samples, high versus low (or male versus female in case of gender) (see Section 2.4.7).


39

TABLE 2-8 CHI-SQUARE DIFFERENCE RESULTS FOR THE MODERATING EFFECTS

Age

Gender

Education

Involvement Channel

Integration 1.73 Difference due to

factor loadings 5.86 4.41 1.39 4.19

4.67 Difference due to error variances

9.43 1.93 5.86 7.46

Notes: Chi-square test values in bold are significant at the .05 level

The test results show that age and involvement do not moderate the relationships of the proposed model (Figure 2-1), because the differences due to the factor loadings are insignificant. Considering Seiders et al. (2005), the result for customer involvement is somewhat surprising. The explanation most likely has to do with the type of measurement. Seiders et al. (2005) measure customer involvement through a number of items directly related to the retail chain. In this study, we measure involvement in the number of product categories, not with the store.

Apparently, the relationships between site and store attitudes and behavior are similar for young (i.e., younger than 38 years) and old (i.e., older than 38 years) customers. In addition, the extent to which customers are interested in different product categories does not affect the strength of the relationships. For the three remaining moderators (gender, education and channel integration), Table 2-9 shows the coefficients of the main relationships for each of the median splits.

TABLE 2-9 SEM COEFFICIENTS FOR MAIN RELATIONSHIPS OF THE PROPOSED MODEL FOR EACH MEDIAN SPLIT

Gender Education

Channel Integration

M F L H L H

SiA → StA 0.15 0.21 0.24 0.15 0.10 0.23 SiA → StB -0.07 -0.04 -0.03 -0.06 -0.07 0.02

SiB → StB 0.05 0.04 0.03 0.05 -0.03 0.01 Notes: SiA = site attitude, StB = store attitude, SiB = site behavior, StB = store behavior. For gender: M = male, F = female; for education: L = high school education, H = at least college education; for channel integration (measured on a scale from 1-5): low (L) ≤ 3.33, high (H) > 3.33. Bold parameter estimates are significant at the 5% level


40

The results show that the moderators mainly affect the relationship between site attitude and store attitude. The relationship between site attitude and store behavior and that between site behavior and store behavior, in most cases, is not significant according to the median split. This somewhat unexpected result could be due to (1) smaller sample sizes and (2) homogenous groups per analysis. Considering the validation results (see section 2.5.4), which show no differences between the estimation and validation samples, the homogeneity of the groups seems to be the more likely reason. Only in the case of low perceived channel integration do we find a significant relationship between site attitude and store behavior. The negative effect falls in line with the results from Table 2-6. We therefore review these results in light of the proposed hypotheses.

H4a thru H4c state that the proposed relationships will be stronger for male, more educated, younger customers. We test for the moderation of three sociodemographic variables and find that age has no moderating effect. Hence, we reject hypothesis H4c. For gender (H4a), we find that the relationship between site attitude and store attitude is weaker for men than for females. With respect to education (H4b), the results indicate a weaker relationship between site attitude and store attitude for customers that are more educated. Therefore, we accept hypotheses H4a en b. Females and less educated customers have a stronger relationship between site attitude and store attitude than do males and customers that are more educated.

Our fifth hypothesis, states that highly involved customers experience the proposed relationships with greater strength. The results show that involvement has no significant moderating effect, and therefore reject H5.

For customers who perceive higher channel integration, H6 suggests the proposed relationships will be stronger than they are for customers with low perceptions of channel integration. Our results support this hypothesis for the relationship between site attitude and store attitude, but channel integration does not moderate the other two proposed relationships. That is, customers with higher perceptions of channel integration show a stronger relationship only between site attitude and store attitude compared with those with lower perceptions of channel integration.

Overall, the moderating effects generally take place at the attitudinal level, that is, the relationship between site attitude and store


41

attitude. The moderating effects found for the relationships between site attitude and store behavior and between site and store behavior are minor in comparison.

2.5.4 Validation Results

The validation test, a comparison of the estimation and validation sample, employs the Chi-square difference test described by Dabholkar and Bagozzi (2002) (see Section 2.4.7). The results show no significant difference between the estimation and validation sample according to

the factor loadings ( dfΔΔ /2χ = 1.03, p-value = .31). The test also shows

that there are no significant differences based in the error variances. Overall, these results validate the proposed model presented in Figure 2-1.

Besides the validation based on a random sample split, we perform a self-selection split to see if the model relationships are stronger for predisposed loyal customers. More specifically, to test if heavy users show stronger relationships between the channels, as heavy users may be more likely to use all the channels. We use the purchase history, that is, the year before the introduction of the Web site, of customers to determine which customers are the heavy shoppers. We perform a median split based on the basis of store behavior to determine whether the findings are consistent for heavy and light shoppers at the department store. The results of the median split show no significant difference between both groups according to the factor loadings

( dfΔΔ /2χ =1.54, p-value = .22).

2.6 DISCUSSION

In this chapter, we propose a framework to evaluate the effects of an informational Web site in a multichannel setting. Three hypotheses related to both customer attitudes and behavior are formulated to determine the effects of the informational Web site. Three additional hypotheses function to investigate whether these relationships may be moderated. Two of the three main hypotheses receive support, and two of the three hypotheses pertaining to moderators are supported with regard to the relationship between site attitude and store attitude. We discuss each of these findings in light of existing multichannel literature.


42

2.6.1 Site Attitude Store Attitude

Montoya-Weiss et al. (2003) show that both competitive and complementary effects can exist among the various channels of a provider. The competitive effect results in channel preference when the perceived service quality of channel a is higher than that of channel b. In contrast, complementary effects arise when the higher perceived service quality of all channels leads to higher overall customer satisfaction.

We find that customers with a positive site attitude generally also have a positive store attitude. Providing customers with additional online information about products and related background topics, such as the latest fashion trends, improves their satisfaction with shopping at the store. After visiting the site, customers may save time and reduce their cognitive efforts by gathering a better understanding of the products available in the store or their own preferences. In this way, a complementary effect occurs when a positively site attitudes leads to improved store attitude.

This relationship is attenuated for the typical Internet user—male, highly educated customers. Previous research indicates that these customers have more interest in technology and a weaker relationship between attitudes and behavior in general (Mittal & Kamakura 2001; Burke 2002). Our research suggests that typical Internet users also evince a weaker cross-channel relationship between attitudes, because these male, highly educated customers perceive a stronger separation between the channels. For such customers, the informational Web site does not reinforce the cross-channel relationship as strongly as it does for female, less educated customers. The median split based on perceived channel integration provides a similar result: Customers who perceive less channel integration experience a weaker relationship between the channels than do customers who perceive more channel integration.

2.6.2 Site Attitude Store Behavior

Verhoef et al. (2007) show that the search attractiveness of the online channel has a positive impact on choosing to purchase through the offline channel, but our findings indicate a negative relationship in the case of actual behavior. We find that attitude toward the site, or the overall evaluation of and preference for the site, has a negative impact


43

on actual shopping behavior. The difference between these studies may lie in the setting and the type of data collected.

The study by Verhoef et al. (2007) focuses on the effect of the Internet in general instead of one specific Web site or firm. They investigate whether consumers enjoy searching online, at no particular Web site, and then buying offline, again at no particular store. As the majority of consumers still prefer the offline channel to purchase, but value the Internet as an information search channel, a positive relationship seems logical. In a firm-specific setting, searching in a firm’s online channel and buying in its offline channel, a positive relationship is less logical. Factors such as loyalty, free-riding behavior and competition influence the customer in this setting and may lead to the opposite finding.

The other explanation for the difference lies in the data used in the two studies. Verhoef et al. (2007) use survey data, that is, attractiveness and purchase intentions for a particular channel. Our study combines survey data (attitudes) and behavior (offline purchases). Mittal and Kamakura (2001) indicate that the use of intention ratings alone could be misleading.

Our results might indicate that the Web site has become a substitution channel for some customers, for whom it offers increased efficiency. That is, the Web site provides these customers with a tool to determine their consideration set without visiting the store, which might prompt their positive attitudes toward the site (Balasubramanian et al. 2005). The associated decrease in offline buying behavior might be the result of improved efficiency and fewer impulse buying trips.

2.6.3 Site Behavior Store Behavior

Various studies show that customers using multiple channels buy more than do single-channel customers (Kushwaha & Shankar 2005). These studies generally refer to multiple transactional channels. However, our study demonstrates that for an offline transactional channel and an online informational channel, customers who view more online pages also spend more offline. In advertising research, Tellis (1988) shows that repeated advertising exposures, i.e. between 1 and 3 exposures have a positive effect on buyers. Even though previous advertising research provides a basis for the interpretation of the results, there is a noticeable difference. A Web site is not a single advertisement, nor are the Web pages equal. A customer is not


44

repeatedly exposed to the same ad. Hence, informational Web sites can have a different effect than found by Tellis (1988).

Because our data are aggregated over 1½ years, this finding illustrates that customers who use one channel intensively likely use the other intensively as well. The findings do not clarify how this behavior evolves over time and the association we find between site behavior and store behavior is relatively small and insignificant in the case of some of the median splits. Nevertheless, this result is consistent with previous results, such as Verhoef et al.’s (2007) finding of a comparable association between Internet search and store purchases.

2.7 CONCLUSIONS

The Internet forms an integral part of our society. The role it plays in information exchange, communication, transaction, and distribution likely will continue to increase. Learning how to apply this “novel” channel effectively is essential for all kinds of businesses, but this study shows that the use of informational Web sites can lead to counterintuitive results. Our results indicate that an informational Web site does not necessarily provide benefits to the organization alone but that customers also benefit by using the information to make their decision-making process more efficient or by free riding on the provided information. Our distinction between attitudes and actual behavior, as well as the inclusion of moderators, provides new insights into how customers react in a multichannel setting. Moreover, we show that these findings are consistent over time.

When trying to reach a different target audience or tune the various channels to different buying situations, managers should focus on creating substitution effects. When they want to establish positive effects between an online and an offline channel, managers should focus on coordinating their channel strategies. Our results regarding the strengthening influence of channel integration seem to confirm the benefits of channel coordination. That is, the channels should have the same look and feel and stimulate cross-channel behavior. Coordination of channel strategies can be achieved by inviting site visitors to test products in the store and reminding store customers of specific, useful features on the Web site. We also emphasize that it is possible to obtain offline benefits by introducing an informational Web site, though these effects vary according to the target group and the level of perceived channel integration. For example, our results show that cross-channel


45

relationships are not as strong among the Internet-savvy customers. Moreover, organizations should keep in mind that providing an informational Web site, or any type of Web site, likely improves customer efficiency and lower impulse buying.

This study should be viewed as a first step in determining the offline effects of an informational Web site on customer attitudes and behavior. As does any study, it contains several limitations. First, our sample may suffer from a self-selection bias related to both the Web site and the survey. Nevertheless, the split sample according to store behavior prior to the site introduction shows no significant differences in terms of behavioral loyalty. Second, the attitudinal measure may suffer from common method variance, which may have affected the relationship between site attitude and store attitude. Although the knowledge that common method variance (i.e., variance attributable to the measurement method [Podsakoff, Podsakoff & Lee 2003]) does not necessarily lead to biased interpretations (Rindfleisch et al. 2006) offers some relief. The inclusion of actual behavior besides the attitudinal measures also provides additional confidence in the effects of the informational Web site on the offline channel. Further research should ensure that common method bias between the attitudinal measures is minimal. Third, our results for the relationships between site attitude and store behavior and that between site behavior and store behavior do not seem very stable and the variation explained in store behavior is low. These relationships warrant further investigation.

This study also shows how an informational Web site influences both attitudes and behavior in an offline transactional channel. To generalize cross-channel findings pertaining to both transactional and informational channels, more research is needed. Studies might elaborate on the effects of different types of Web sites, such as transactional versus informational sites, differences across branches, or differences due to site design. In addition, an interesting avenue for future research is the extent to which previous advertising research, such as by Tellis (1988), is applicable to informational Web sites.

The generalizability of our model also should be tested for other industries, other channels, or Web sites with different purposes, such as Web communities. Whereas our model testing involves a two-channel situation, it could easily be extended to include more channels, such as catalogs or telemarketing, which might confirm the strength of the model.


46

Realizing the value of an informational Web site depends on three criteria. First, value is achieved mostly at the attitudinal level, through improved satisfaction and loyalty toward the store. Second, improved attitude toward the store is achieved mostly from within a particular segment, namely, female, less educated customers. Third, the effect on store attitude can be increased if the customer perceives the channels as highly integrated. However, organizations should be aware that customers likely will spend less offline as a result of an informational Web site, especially if they like it. In conclusion, adding an informational Web site to a traditional channel provides benefits in the form of improved customer attitudes that get manifested offline.

47

3 The Impact of an Informational Web Site on Offline Customer Buying Behavior7

In this chapter, we study the effects of the use of an informational Web site on offline customer buying behavior for a large national retailer. We model the effects of individual online behavior on offline buying behavior at both the overall purchase level and the category level. The results show that informational Web sites may entail more bad than good news, because Web site visitors become more efficient in their offline shopping; that is, they engage in fewer shopping trips and spend less in all categories.

3.1 INTRODUCTION

With the commercialization of the World Wide Web, more and more companies offer information via their Web sites. The vast increase of available information has made it much easier for customers to compare offers of competing products and merchants and make well-informed decisions. Academic research regarding the effectiveness of Web sites in terms of customer buying behavior has focused mainly on the impact of transactional Web sites (see e.g., Moe & Fader 2004; Sismeiro & Bucklin 2004). However, Forrester Research also shows that most Internet users conduct research online before buying offline (see e.g., Kelley, Delhagen & Denton 2002; Mendelsohn, Johnson & Meyer 2006). Similarly, the Boston Consultancy Group reports that 88% of Internet users browse online before buying offline (Rasch & Lintner 2001). What these studies do not show is whether online search increases or decreases individual offline buying behavior.

Online information search can lead to various subsequent customer actions. For example, it may be followed by a desire to buy the product via the Web site on which the information was found, though online conversion rates rarely exceed 5% (Moe & Fader 2004). Customers also may decide to leave the site and buy the product at a competitor’s Web site. A third option involves online search followed by offline purchase.

7 This chapter is based on Teerling, M. L., J.E.M. van Nierop, P.S.H. Leeflang and

K.R.E. Huizingh (2006), The Impact of an Informational Web Site on Offline Customer Buying Behavior, Working Paper, University of Groningen.


48

This option is quite likely for products with predominantly search aspects (e.g., Alba & Lynch 1997; Citrin et al. 2003) or in case of customers with technology anxiety or trust issues (see e.g. Hoffman et al. 1999; Meuter et al. 2003; Roy & Ghose 2006).

Various studies investigate the effects of the “new” Internet channel on either aggregate firm performance or individual customer behavior. At the aggregate level, the Internet channel does not appear to cannibalize from the existing channel (see e.g. Deleersnyder, Geyskens et al. 2002; Biyalogorsky & Naik 2003), and the effects of an informational channel may even be positive (Lee & Grewal 2004). At the individual customer level, research thus far has been able to determine only the effects of an online transactional channel, which it posits as both positive and negative effects (Ansari et al. 2006; Kushwaha & Shankar 2005). As Neslin et al. (2006) recommend, more research into the contribution of various channels is needed. In the case of an online informational channel, no empirical results are available at the individual customer level.

The question therefore is what happens at an individual customer level if an organization provides online information to its customers. Do they become more interested in the organization and, as a result shop more often? Alternatively, does the information allow customers to know what the organization and its competitors have to offer, resulting in less frequent shopping trips? From Gensler et al. (2007) and Knox (2005) we know that the effects of an additional transactional channel can vary depending on the product category and that certain product categories are more suitable for online buying than others. So do customers increase or decrease their spending across product categories as a result of an informational Web site? More generally, does the use of an informational Web site alter spending in certain product categories.

This study aims to determine the effects of informational Web sites by investigating how a site influences offline shopping trips (i.e. shopping trips during which customers purchase at least one product) and product category purchases at the individual customer level. We decompose buying behavior in our attempt to answer the following questions: • Does the use of an informational Web site change the number of

offline shopping trips conducted by individual customers? • Does the use of an informational Web site alter the purchases of

individual customers in different product categories?


49

We link online search behavior with the offline buying behavior of individual customers of a large retailer using customer panel data. From these data, we know to what extent, over time, the customers visit the Web site, what they do online in terms of the frequency and depth of their site visits, and how much and how often they shop at the offline stores. We also can study how the offline buying behavior of customers has changed with the use of the Web site, because purchase data are available for both before and after the implementation of the Web site.

We first review the relevant literature. After discussing the methodology, we describe the data. Then, we present the findings and discuss them in light of previous studies. We end with a summary of the main conclusions.


Table 3-1 shows an overview of selected multichannel studies that attempt to determine the impact of the addition of an Internet channel on firm performance. We distinguish two possible Internet channel additions: a transactional channel or an informational channel. The impact of either of these additions has been investigated at the aggregate (firm) sales level and at the individual customer level. However, as far as we know, this study is the first to consider the effects from an additional informational channel on the individual customer level.

TABLE 3-1 OVERVIEW SELECTED MULTICHANNEL BEHAVIOR STUDIES

Level of the Data Analysis Aggregate Individual

Transactional

Biyalogorsky & Naik 2003 Coelho et al. 2003

Ansari et al. 2006 Danaher et al. 2003 Dholakia et al. 2005 Gensler et al. 2007 Knox 2005 Kushwaha & Shankar 2005

Internet Channel Addition

Informational

Deleersnyder et al. 2002 Geyskens et al. 2002 Lee & Grewal 2004

This study

The impact of an additional online transactional channel on

aggregate sales appears in two studies. First, Biyalogorsky and Naik (2003) investigate Tower Records’ sales figures during the period 1989-


50

1999 to determine to what extent the added online transactional channel cannibalizes offline sales. They find a cannibalization rate of 2.8% from online sales, indicating negligible contemporaneous cannibalization (Biyalogorsky & Naik 2003). Second, Coelho, Easingwood and Coelho (2003) show that when a company starts using a new channel, it can expect stronger sales growth from this channel than from its traditional channel. Growth opportunities likely occur because the firm reaches new customer segments, but these authors also indicate that as penetration into these segments increases, growth diminishes and cannibalization might exist between channels (Coelho et al. 2003).

Several other studies investigate the effect of an additional online informational channel at the aggregate sales level. Deleersnyder et al. (2002) find in the newspaper industry hardly any cannibalization between the online and offline channels, possibly because of the different market segments ; depending on the positioning of the channel portfolio, cannibalization or synergy between the channels is possible. Similarly, Geyskens et al. (2002) investigate the effect of an additional online channel on stock market responses in the newspaper industry and show that the online channel addition has a positive effect on stock price. Considering that the newspaper industry can easily take advantage of the special economics of information goods delivered over the Internet, these results may differ for retailers (Bakos & Brynjolfsson 2000). In a retail setting, Lee and Grewal (2004) show that adding the Internet as a transactional channel does not have an effect on the Tobin’s Q (a stock market-based measure of firm performance). However, adding the Internet as a communication or informational channel has a positive impact (Lee & Grewal 2004).

Studies investigating the effect of an additional Internet channel on individual customer behavior mostly have focused on transactional channels. This focus probably was probably the only alternative, because individual customer data for online informational and offline transactional channels are rarely available (see e.g., Sullivan & Thomas 2004). However, from these studies, we can infer that (1) marketing efforts can migrate customers to a particular channel (Ansari et al. 2006; Knox 2005), (2) most customers become multichannelers after the addition of an Internet-channel (Dholakia et al. 2005; Gensler et al. 2007), and (3) adding a transactional Internet channel may either


51

decrease (Ansari et al. 2006; Gensler et al. 2007) or increase (Kushwaha & Shankar 2005) customer buying behavior.

With this study, we try to fill the gap in the existing literature by determining the impact of an added informational Internet channel on individual customer buying behavior. Firms invest in (informational) Web sites because they expect positive effects on information needs, brand perceptions and buying behavior and negative effects on switching behavior and search time. However, extant research demonstrates that these effects are not always so evident, nor do they always have the expected and intended directions.

Informational Web sites may improve offline buying behavior due to (1) synergy effects between channels, (2) marketing effects, (3) improved brand/product awareness, and (4) increased loyalty. Verhoef et al. (2007) indicate that research shopping, or searching in channel a but buying in channel b, may create synergy effects, possibly as a result of economic benefits such as better choices informed by improved brand/product awareness or a sense of being a smart shopper.

Neslin et al. (2006) indicate that marketing efforts may provide another explanation for increased buying in the case of multichannel customers. This argument may also hold in the case of an informational Web site, because customers who use both channels are exposed to more marketing efforts than are those who use a single channel. Ansari et al. (2006) and Kumar and Venkatesan (2005) provide empirical support for this effect, and Wallace et al. (2004) show that retailers may receive a loyalty payoff because customers perceive an enhanced portfolio of service outputs provided by the multiple channels.

One of the main reasons informational Web sites may have negative effects on offline buying behavior relates to the decision-making process. Wu and Rangaswamy (2003) demonstrate that Web site features can either decrease or increase the amount of search and thereby influence consumers’ consideration sets. Although the Internet is assumed to facilitate better and more efficient decision making (e.g., Alba & Lynch 1997; Mick & Fournier 1998), its availability also can lead to less information search in offline sources. Ratchford et al. (2003) test this claim for the automotive industry and find that consumers gain efficiency, increased information, and bargaining power from an Internet channel.

Shiv and Fedorikhin (1999) indicate that impulse decisions might be reduced through the greater availability of processing resources. An


52

informational Web site that does not offer transaction capabilities, and therefore eliminates impulse buying, stimulates extensive processing of information. Hence, Shiv and Fedorikhin (1999) offer another possible explanation for a negative effect of an informational Web site on offline buying.

In addition to efficiency in decision making and reduced impulse buying, an informational Web site may reduce switching costs. Porter (2001) indicates that on the Internet buyers can easily switch suppliers with just a few mouse clicks. Because informational Web sites do not offer the possibility to transact, customers may switch to competitors’ Web sites that provide both information and the ability to make purchases (Neslin et al. 2006).

The benefit of informational Web sites for customers lies in the information provided and their ability to use this information in their decision making. Online information dissemination and communication not only require a different approach than does an offline setting, but the consequences also can differ (e.g., Hoque & Lohse 1999; Stewart & Pavlou 2002). Specifically, customers may make different choices or better informed decisions given the information they find online (Hoque & Lohse 1999). Therefore, it becomes crucial to investigate the effects of an additional online informational channel on offline buying behavior.


Store choice is frequently modeled with a logit or a probit model determining the utility for the customer of shopping at a particular store (e.g., Rust & Donthu 1995). These types of models explain the preference of a customer for a particular store through variables such as distance and the competitors in the area. In this study, we are interested in determining the effects of the introduction and use of an informational Web site on different elements of store behavior. Hence, a decomposition approach is more suitable.

Van Heerde, Leeflang and Wittink (2004) provide an extensive literature review of decomposition approaches, including the many new insights that stem from decompositions, ranging from front traffic, store entry ratios, closing rations, and average spending (Lam, Vandenbosch, Hulland & Pearce 2001) to purchase incidence, brand choice, and quantity (Gupta 1988; Zhang & Krishnamurthi 2004). In this study, we decompose individual buying behavior for both shopping


53

trips at the overall level and the amount of money spent at the category level.

We focus our analyses at the product category level instead of overall monetary value to determine (1) if an informational Web site has different effects across categories and (2) what the impact of category-specific site pages might be.

3.3.1 Specification We specify the total purchases, or total money, computed as the

product of the number of shopping trips in which a customer purchases at least one product and the total amount of money spent in all categories. More specifically:

(1) 1 1 1itcit it

c it

MM V * ,i ,..., I ;c ,...,C;t ,...,T

V= = = =∑ ,

where

itM = the total amount of money spent by individual i during month t,

itV = the total number of shopping trips made by individual i during month

t,

itcM = the total amount of money spent by individual i during month t in category c,

I = the number of individuals,

C = the number of product categories, and

T = number of months. We assume that the first component of equation (1), Vit, follows a

Poisson process, so that the distribution of the number of shopping trips in any interval depends only on the length of that interval (e.g., Ehrenberg 1959; Leeflang, Wittink, Wedel & Naert 2000). The probability that vit shopping trips occur is given by

(2) ( )Pr!

itit vit

it itit

eV v

v

λ λ−

= = ,

where λit reflects the expected number of shopping trips for individual i in month t. The parameter λit is explained by regressors as follows:

(3) ( )ln it i it itXλ γ γ ε= + + .


54

The vector Xit may contain individual-specific sociodemographic covariates, period and individual-specific site-related covariates, promotional activities, period dummies to account for seasonal and trend effects, and lagged dependent variables. The parameter γ describes the average effect, and the parameter iγ accounts for

unobserved household specific heterogeneity.

The second component of equation (1), itc

it

MV

, is modeled with a

Type-II Tobit model (Chib 1992; Amemiya 1985, Fox, Montgomery & Lodish 2004; Franses & Paap 2001). We use the Type-II Tobit without correlation between the two stages, also known as a two-part model, to explain whether a customer buys in a particular category (purchase incidence), as well as the actual amount of money spent per category, given that the customer buys. We adopt the multivariate Type-II Tobit approach (Fox et al. 2004) because our application deals with multiple categories that may correlate.

For each individual i, we denote the decision to buy in category c in month t as Zitc. This variable equals 1 if the customer buys and 0 if not. The multivariate Probit model (MVP) that describes this decision is specified as:

(4) *

*1 if 00 if 0

itc itc

itc itc

Z ZZ Z

= >= ≤

where

(5) * ( )itc c ic itc itcZ Hα α ε= + + .

Here Zitc* is the latent utility for individual i of buying from category c in period t. If the utility Zitc* is greater than 0, a purchase is made. The vector Hitc contains (mostly) time-varying explanatory variables that may influence the decision to purchase in category c. The parameter cα

describes the average effect, and the parameter icα refers to the

household-specific effect. The error terms εitc follow a multivariate normal distribution, with mean 0 and covariance matrix Σ. For identification purposes, the diagonal elements of Σ are usually set to 1 (Manchanda, Ansari & Gupta 1999).

We denote the amount spent by individual i in category c in month t as Yitc. by


55

(6) ⎧ == ⎨⎩

* if 10 otherwise

itc itcitcY ZY ,

where the natural log of *itcY is explained by

(7) *ln ( )itc c ic itc itcY Gβ β η= + + .

The vector itcG also contains (mostly) time-varying explanatory

variables that may influence the amount of money spent in the various categories. The parameter cβ describes the average effects of these

explanatory variables, and the parameter icβ accounts for unobserved

heterogeneity. The error term ηitc follow a multivariate normal distribution with mean 0 and covariance matrix Ω.

The vector itcG contains category-specific variables that describe

the individual’s online behavior in the month under consideration, and explanatory variables that also occur in the matrix X in equation (3). The vectors H and G can be equal in both stages (decision to buy and the amount of money spent) of the Type-II Tobit or contain different explanatory variables depending on the expectations of each stage. Throughout, we concede that the effects of the explanatory variables on the decision to spend and spending levels differ. Therefore, promotional activities may have an insignificant effect on the decision to spend (Zitc* ) but significantly influence the amount spent (Yitc* ). The unobserved household-specific heterogeneity, modeled by adding icα and icβ to the

intercept parameters in both stages, draws from a multivariate normal distribution with means 0 and variance matrices αΣ and βΣ

respectively (e.g., Allenby & Rossi 1999). The multivariate error distributions for ε and η allow for

information from one category to influence the conditional predictions of other categories. We expect contemporaneous correlations of the disturbances across categories because excess expenditures in one category may result in less spending in other categories (substitution) or complementary sales.

3.3.2 Estimation We estimate the Poisson model that describes the number of

shopping trips with a log-link function. To incorporate heterogeneity, we estimate a random effects Poisson model, with individual-specific


56

constant and site visits parameter in MLwiN 2.02. The log-likelihood function is (Greene 2003)

(8) ( )'ln ln !it it it i itL v X vλ γ γ⎡ ⎤= − + + −∑ ⎢ ⎥⎣ ⎦.

For discrete response multilevel models, a maximum likelihood estimation is computationally intensive, and therefore MLwiN implements quasi-likelihood methods (Rasbash, Steele, Browne & Prosser 2004). The estimations of the equations are generated through a

second-order Penalized Quasi-Likelihood (PQL) procedure, using the first-order Marginal Quasi-Likelihood (MQL) estimates as starting values. Without further instructions, first-order MQL is used to estimate the coefficients, an estimation procedure that may lead to downwardly biased estimates (Rasbash et al. 2004).

The category-specific multivariate Type-II Tobit model for money spent is estimated using Markov chain Monte Carlo (MCMC) methodology in Gauss 5. To obtain posterior results, we use the Gibbs sampling technique of Geman and Geman (1984) with data augmentation (Tanner & Wong 1987). The idea of Gibbs sampling aims to sample iteratively from the full conditional posterior distributions of the model parameters, which creates a Markov chain that converges under mild conditions, such that the draws can be used as draws from the joint distribution (for an introduction, see Casella & George 1992 or Tierney 1994). The posterior means and standard deviations of the parameters of interest thus can be obtained. Appendix III provides the conditional posterior distributions. We use 10,000 draws for the burn-in and store 1 per each 10 of the next 10,000 draws for the inference (see e.g., Fox et al. 2004). We monitor the graphs of the sampled values to ensure convergence of the parameters.


To test the model, we collected data among customers of a large, well-known national retailer in the Netherlands, as described in Section 1.6. The organization has 58 outlets in all major urban areas in the country, and each outlet contains a range of different categories, including clothing, cosmetics, toys, books, furniture, and so forth.

3.4.1 Informational Web Site The informational Web site for this study is a theme-oriented site

that supports offline activities to increase the likelihood of purchase in


57

stores. It provides customers with information about lifestyle issues related to the different categories of the store, products offered in the stores, promotions, and the organization itself.

The theme orientation is apparent across various topics, such as fashion, interior design, and sports. For instance, the site contains four fashion pages: a general introduction page, with editorial articles and a summary of what can be found on the remaining pages, and three subject pages related to topics such as style, trends, and cosmetics. These pages also contain editorial articles and pictures of products, such as the latest fashion products.

Table 3-2 shows the various product categories from which customers may purchase offline and the corresponding categories featured on the Web. For each Web category, we provide examples of the site pages to offer a sense how the categories are promoted on the site. In this chapter, we consider six different categories.

TABLE 3-2 MATCHING OFFLINE PRODUCT CATEGORIES AND ONLINE THEMES

Site Theme Site Pages Fashiona Fashion Fashion and beauty related pagesb

Children’s products Children Child and family related pages, children’s play time

Accessories Accessories Accessories, trends and cosmetics Interior design products

Interior design Interior design, cooking and dining, bed and bath

Sports Sports Feeling in shape, sports and physical activities, care and relaxation

a. Fashion refers to both women’s and men’s fashion departments. b. These pages are customized according to the gender of the online visitor. For

example, a female visitor to the Web page views women’s clothing or beauty products.

3.4.2 Data We provided a description of the data in Section 1.6.3, in which we

noted the data are at the individual customer level. The panel contains 8,615 customers, ranging from those who have never used the Web site to frequent site visitors. Data about the offline buying behavior of these customers are available for 29 months, 14 months before the introduction of the Web site and 15 months after. That is, data on the shopping trips in which customers purchase at least one product is available. Data on the number of browsing trips, i.e. in which a customer does not purchase a product, in the offline stores is not available. We model purchase behavior on a monthly level instead of


58

weekly or biweekly because of the infrequency of department store buying behavior by individual customers. The total number of non-zero buying observations in the data set is 117,537. To keep the estimation of the model manageable, we draw a sample of 209 customers responsible for 3,181 non-zero behavior observations. Only customers who were active for at least 5 of the 29 periods either in the store or on Web site are selected. To validate the results, we also draw four random samples to check for consistency in the findings by comparing the estimates and the fit of the validation samples with the estimation sample.

3.4.3 Explanatory Variables A wide range of variables may influence customers in their

purchase decisions, including physical and social surroundings, time, goals and objectives of the task, and antecedent states (e.g., Belk 1975). In considering our data, we include the following explanatory variables: • Online behavior

For both models, we include the number of site visits by individual i in period t. For the Tobit-II model, we also include the number of site pages per category. Due to multicollinearity between the total number of site pages and the number of site visits, we eliminate the total number of site pages from the Poisson model. • Promotional activities by the department store

The organization has three major offline promotional activities during specific periods of the year: the holiday shopping season of November and December; general promotion discounts in all categories in the store; and promotions of fashion categories. • Individual customer characteristics

Customer characteristics at the zip code level, such as the percentages of loyal households or those in a particular life stage, are gathered from Acxiom8 In addition, we include the distance the customer has to travel to the nearest outlet of the department store to account for individual effort required to reach the store. Huff (1964) was one of the first to show that store choice depends on the distance between the store and the customer. • Past behavior

8 Acxiom is a data management services company that provides customer data and the infrastructure needed to manage and use customer and prospect information and thereby support business decisions and actions.


59

In both models, we include buying behavior during the previous month, using shopping trips for the Poisson model and average amount spent per category for the Tobit-II model, as well as two-year dummies to capture trends.

3.4.4 Exploratory Insights Appendix IV shows some descriptive statistics for the most

important variables in the data set and the estimation sample. The average number of shopping trips per month drops in the period after site implementation (from 2.6 to 2.1 for the entire data set; from 2.4 to 1.8 for the estimation sample). For most categories, the average amount of money spent also decreases in the period after site implementation for both the overall data set and the estimation sample. The averages of site visits may seem small, but these averages are for all customers, including those who never use the Web site. Appendix IV further shows that buying behavior in the estimation sample is slightly higher than that across the entire data set. Most likely, this finding relates to the selection procedure of the sample, which required customers to be active in at least 5 of the 29 periods.

Table 3-3 compares the months before (e.g., February 2001) and after (e.g., February 2002) the site introduction in terms of both site visitors and non-site visitors.

TABLE 3-3 COMPARISON OF MONTHS BEFORE VERSUS AFTER SITE INTRODUCTION WITHIN GROUPS FOR THE ENTIRE DATA SET

Non-Site Visitors Site Visitors

# of periods lower

after SI

# of periods higher after SI

# of periods no

change after SI

# of periods lower

after SI

# of periods higher after SI

# of periods no

change after SI

# Shopping trips 2 0 10 12 0 0 € Ladies fashion 0 2 10 5 0 7 € Men’s fashion 1 1 10 6 0 6 € Children's 0 2 10 7 0 5 € Accessories 0 1 11 7 0 5 € Interior design 0 1 11 6 0 6 € Sports 2 1 9 5 0 7

Total number of periods compared 12 Notes: SI refers to the Web site introduction.


60

The results show that the non-site visitors display virtually no differences in their offline buying behavior before and after site introduction. Among the site visitors, in at least half the cases, a significant decrease occurs in their offline buying behavior after the introduction (and their use) of the Web site.

With regard to the number of shopping trips, the results for the estimation samples are similar to those for the entire data set. For the other variables, we find more instances of “no change” for the estimation samples than for the entire data set. These results generally indicate that nonvisitors barely changed their behavior over time, whereas the buying behavior of visitors who started using the Web site significantly decreased.

We perform several self-selection analyses to ensure that the visitors to the Web site are not predisposed to be more loyal than nonvisitors. Figure 3-1 shows a comparison over time of both groups, in which the introduction of the Web site is indicated by the line marking period 15 (Appendix V provides the t-values of the comparison of both groups in each period).

FIGURE 3-1 COMPARISON OF SITE VISITORS AND NON-SITE VISITORS OVER

TIME FOR SHOPPING TRIPS

The number of shopping trips per period indicates similar pre-introduction behavior by both groups. After the introduction of the Web site, the site visitors decrease their number of shopping trips per period, as well as the money they spend per category. These results are

00.5

11.5

22.5

33.5

1 4 7 10 13 16 19 22 25 28

Periods

Stor

e vi

sits

Site users

Non-site users


61

consistent with the comparisons from Table 3-3. Overall, these analyses provide exploratory insights into the negative effects of site use.

In addition, in Table 3-4, we compare both groups in terms of various sociodemographics. Site visitors are slightly younger and more educated, and more men appear in the site visitors group than in the non-site visitors group. These findings seem to match the general Internet population at the time (2000–2002) of more educated, younger, male customers (e.g., Burke 2002)9.

Even though Table 3-4 shows differences between the groups, these differences likely are not strong enough to explain the variation over time in the number of shopping trips and money spent per category. Overall, these exploratory analyses show that before the Web site introduction, the buying behavior of both groups was similar. After the introduction, the customers using the Web site change their buying behavior, whereas the non-site visitors do not. Furthermore, a Poisson model of the number of site visits shows that the small differences in sociodemographics between visitors and non-site visitors do not explain online behavior.

TABLE 3-4 COMPARISON OF SOCIODEMOGRAPHICS BETWEEN SITE VISITORS AND NON-SITE VISITORS

Site

Visitors Non-Site Visitors Test Value

Age 39.5 42.7 3.189

Number of children 1.2 1.2 0.517

Number of adults 2.2 2.3 0.639

High school education 98.0% 97.5% 0.127

College education 45.7% 29.8% 12.138 Distance to closest store 6.5 6.2 -1.587

Gender: male 44.8% 23.1% 22.655 Gender: female 55.2% 76.9%

N 6594.0 951.0 Notes: Bold test values (t-value or χ2-value) are significant at the 5% level.

9 To determine whether sociodemographics explain online behavior, we ran

a Poisson regression of the number of site visits per month. For non-site visitors, the number of site visits is 0. In addition to the sociodemographics from Table 3-4, we include lagged Web visits, year dummies, and promotional activities. The results show that sociodemographics do not explain online behavior.


62

3.5 FINDINGS

3.5.1 Number of Shopping Trips

We test multiple sets of explanatory variables on the estimation sample and find that most of the individual customer characteristics in Table 3-4, with the exception of distance to the store, do not significantly influence the number of shopping trips. The total amount of time spent on the Web site also does not contribute significantly to the explanatory power of the model.

We obtain the best results from a parsimonious model, the results of which appear in Table 3-5. The first column in the table shows the effects of the explanatory variables on the parameter λit, and the second column provides the corresponding standard errors. We calculate so-called “partial effects” for each of the k independent variables Xk for 2001 and 2002. Because the expected number of shopping trips per period is given by

(9) ( ) ( )| | itXit it it it itE v X Var v X eγλ= = = ,

we can express the partial effect for Xk as:

(10) ( )|

itXit itk

kit k

E v Xe

Xγγ

λ γ

∂=

∂=

.

To capture the individual heterogeneity γk is extended with the

individual effect

(11) γ=∑

1

1 I

kiiI .

Because λit depends on all explanatory variables, this expression reveals that the effect of a particular explanatory variable depends on the value of the other explanatory variables. Except for the year dummies, the explanatory variables are set to their mean value. However, vector Xit contains two year dummies. By setting these dummies to 1, independent of each other, we can calculate the partial effects of the explanatory variables for each year; if we set both dummies to 0, we can determine the partial effects for 2000. However, because the Web site was not introduced until March 2001, we do not discuss these effects herein.


63

TABLE 3-5 PARAMETER ESTIMATES OF THE VARIABLES THAT EXPLAIN THE NUMBER OF SHOPPING TRIPS

Estimated Parameter

Std. Error Partial Effects 2001

Partial Effects 2002

Intercepta 0.625 0.057 -- --

Holiday promotion 0.192 0.044 0.383 0.341 General promotion 0.133 0.041 0.265 0.236 Fashion promotion 0.075 0.043 0.150 0.133

Dummy 2001 -0.134 0.035 -- -- Dummy 2002 -0.250 0.046 -- --

Site visitsa -0.456 0.070 -0.909 -0.809

Distance to closest outlet -0.013 0.005 -0.026 -0.023 Lagged shopping trips 0.072 0.009 0.144 0.128

Variance intercept 0.093 0.015 -- -- Variance site visits 0.508 0.074 0.570 0.508 a. The variable is estimated at the individual level. Notes: Bold parameter estimates are significant at the 5% level.

From the partial effects in 2001, we determine that the strongest influence on the number of shopping trips in a particular month is the number of site visits. For 2001, with an increase of 1 in the number of site visits, the expected number of monthly shopping trips decreases by .9 trips. The decrease in the number of shopping trips is slightly lower in 2002, with a decrease of .8. Hence, an important part of the decrease in the number of shopping trips over time can be attributed to the introduction and use of the Web site.

Regarding the other partial effects in 2001, the holiday promotion has a partial effect, namely, a .38 increase in the number of shopping trips. The general promotion has a slightly smaller positive influence (.27) than the holiday promotion, whereas the fashion promotion has no significant effect on the number of shopping trips. We find a similar pattern in the promotional activities for the partial effects in 2002.

The negative coefficients for the year dummies illustrate a trend effect that can be explained partly by a decline in the economy. The distance to the closest store has a negative influence on the number of shopping trips, such that the closer customers live to a store, the more often they shop. Finally, we find a positive partial relationship between the number of shopping trips in t and the shopping trips in t – 1. We


64

discuss the estimated variance for the site visits parameter in Section 3.5.5.

To determine the predictive power of the model, we calculate a hit rate for the estimation sample and allow for different margins of error. The hit rate describes the percentage of correctly estimated shopping trips per month compared with the actual number of shopping trips per month. The average number of shopping trips per month is 2.05 with a standard deviation of 1.88. When the margin of error is 1 store trip per month, the hit rate is 50.3%, whereas for an error margin of 2 shopping trips per month, the hit rate is 75.9%.

3.5.2 Validation for the Number of Shopping Trips To determine whether these results are driven by specific features

of the estimation sample, we draw four validation samples, estimate the model on each sample, and determine the extent to which the parameters of the validation samples are comparable with the corresponding parameters of the estimation sample. We use three comparisons: (1) the percentage of validation sample parameters that fall within the 95% confidence interval of the estimation sample parameters, (2) the percentage of validation sample parameters that have the same sign as the estimation sample parameters, and (3) the percentage of validation sample parameters that have the same sign and significance as the estimation sample parameters.

Of the validation sample parameters, 75% fall within the confidence interval. Of the 48 parameters, 47 (97.9%) have the same sign as the estimation sample parameter, and 43 (89.6%) have both the same sign and significance level as the estimation sample parameter. We also determine the percentage of customers with a positive coefficient for site visits across all samples; on average, 21% of customers across the five samples experience a positive effect from the Web site on their offline shopping trips. Overall, the results from the validation samples are consistent with the results of the estimation sample.

3.5.3 Amount Spent per Trip per Category We now consider the estimates for the multivariate Type-II Tobit

model, which we estimate for six categories simultaneously.10 To detect possible cross-category correlation s, we use minimal restrictions on the

10 We also estimate a Type-I Tobit of the total money spent in month t by

individual i. The parameter coefficients from this model are similar in terms of direction to the parameters presented for the MVP Type-II Tobit.


65

covariance matrices. The covariance matrices Σ and Ω are set to full, with 1 on the diagonal and full, respectively. The matrix Σ provides information about cross-category effects regarding the decision to buy, such that a high positive correlation indicates that purchases in two categories usually coincide. Matrix Ω indicates the variances and covariances between the error terms of the six categories for the amount spent. The covariance matrices of the unobserved heterogeneity αΣ and

βΣ are set as the diagonal matrices. Furthermore, βΣ must be

restricted to ones to ensure identification. The covariance matrix of unobserved heterogeneity ( αΣ ) provides information about the level of

heterogeneity in the sample, for instance minimal variance indicates a low level of heterogeneity. We summarize the settings for the covariance matrices in Table 3-6.

TABLE 3-6 COVARIANCE MATRIX SETTINGS

Yes/No Decision Amount Decision Covariance matrix of error term

Σ full with 1 on diagonal

Ω unrestricted

Covariance matrices of unobserved heterogeneity

αΣ diagonal βΣ identity

We test multiple sets of explanatory variables, and just as for the

model of shopping trips, we find that most of the customer characteristics do not significantly influence the money spent per category.

To determine the final model, we first estimate six different models11 with the specific elements pooled or unpooled across the six product categories. Appendix VI shows the fit criteria for the models and the results of the Chow test to determine which variables can be pooled. The fit criteria show that the models have a comparable fit, but the Chow test comparing model 1 with model 2 shows that pooling is not justified. However, comparing model 1 with models 3–6 indicates that pooling in these cases is justified. Comparing model 3 (only constant unpooled) with model 4 (constant and site visits unpooled)

11 Model 1 has all variables unpooled across the categories. In model 2, all

variables are pooled. In model 3, only the constant is unpooled. In model 4, both the constant and the number of Web site visits are unpooled. In model 5, the constant and the number of Web site pages per category are unpooled. In model 6, the constant, Web site visits, and number of Web site pages per category are unpooled.


66

shows that choosing not to pool site visits significantly improves the model, whereas models 5 and 6 (site pages unpooled) show no significant improvement. Therefore, we use model 4 with the constant and site visits unpooled.

We provide the estimation results of model 4 in Tables 3-7 and 3-8. The intercept and site visits are category specific; the “Yes/No” columns indicate the effect of each variable on whether someone purchases in a particular category. The “Amount” columns indicate the effect of each variable on the amount of money spent per shopping trip in a specific category, given that an individual buys from the category.

We used Morrison’s Proportional Chance Criterion (Morrison 1969) to determine whether the first stage of the model, i.e. the classification of the decision to buy, outperforms chance. For each of the categories the chance proportion is calculated (i.e. ladies’ fashion = .59; men’s fashion = .84; children = .68; accessories = .56; living = .78; sports = .87) with which the hit rate is compared. For the categories ladies, children and accessories the model did significantly better than expected by the chance proportion. For the categories men, living and sport the model did not do significantly better than the chance proportion. Most likely the low number of purchases in these categories causes this result.

The number of site visits significantly decreases purchase incidence (yes/no) and the amount spent for almost all categories, though the negative effect for decision to buy in the men’s fashion category is insignificant. These negative effects are consistent with those we found pertaining to site visits on offline shopping trips. The effect of the number of site pages viewed per category is insignificant.

The general promotion has a positive effect on the decision to buy and, given this decision, the amount of money spent, but the fashion promotion has a significant effect only on the amount of money spent. A similar effect occurs for the holiday promotion. Therefore, even though these promotional activities do not significantly increase the probability of deciding to buy, they increase the amount of money spent.

From the estimation results, we deduce that there is virtually no (economic) trend effect on the probability of buying and amount of money spent. The distance to the closest store has a negative effect on both stages, but it is only significant for the decision to buy. In other words, if the customer lives farther away, he or she is less likely to


67

decide to buy in the department store. Finally, the effect of the last purchase occasion is insignificant.

The variance of unobserved heterogeneity in the number of site visits indicates small individual differences in the effect of Web site visits. For the constant, the variances are slightly higher, which suggests that missing variables could reduce the differences between individuals. The largest unobserved heterogeneity occurs in the Children’s category.

TABLE 3-7 PARAMETER ESTIMATES OF VARIABLES THAT EXPLAIN THE BUYING DECISION AND AMOUNT SPENT FOR LADIES’ FASHION, MEN'S FASHION, AND

CHILDREN'S PRODUCTS

Ladies Fashion Men's Fashion Children Yes/No Amount Yes/No Amount Yes/No Amount Category-specific effects

Intercept -0.379 0.469 -1.087 -0.612 -0.840 -0.178Site visits -0.311 -0.901 -0.076 -0.449 -0.256 -0.816 Pooled effectsa

Promotions: • Holiday 0.059 0.165 • General 0.215 0.517 • Fashion 0.039 0.144 Year dummies:

• 2001 -0.008 0.040

• 2002 -0.075 -0.016

Distance -0.007 -0.014

Lagged spending 0.001 -0.004

Site pages -0.004 -0.017 Variance of unobserved heterogeneity (Σα,Σβ)b

Intercept 0.142 1 0.216 1 0.572 1

Site visits 0.051 1 0.056 1 0.054 1

Yes/No hit rate 0.71 0.85 0.82 Notes: The intercept and the site visits are category specific. Bold parameter estimates are significant at the 5% level (95% HPD region). a. Pooled effects indicate coefficients that are pooled across the (six) categories. b. For Σα the estimated diagonal elements are shown. For Σβ (an identity

matrix) the ones are shown. c. The multivariate Probit Yes/No hit rates show in what percentage of

occasions the model correctly predicts that an individual will buy for each category.


68

TABLE 3-8 PARAMETER ESTIMATES OF VARIABLES THAT EXPLAIN THE BUYING DECISION AND AMOUNT SPENT FOR ACCESSORIES, LIVING, AND SPORT

Accessories Living Sport Yes/No Amount Yes/No Amount Yes/No Amount Category-specific effects

Intercept -0.355 0.205 -0.809 -0.285 -1.142 -1.098 Site visits -0.349 -0.867 -0.348 -0.980 -0.190 -0.717 Pooled effectsa

Promotions: • Holiday 0.059 0.165 • General 0.215 0.517 • Fashion 0.039 0.144 Year dummies:

• 2001 -0.008 0.040

• 2002 -0.075 -0.016

Distance -0.007 -0.014

Lagged spending 0.001 -0.004

Site pages -0.004 -0.017 Variance of unobserved heterogeneity (Σα,Σβ)b

Intercept 0.205 1 0.100 1 0.115 1 Site visits 0.054 1 0.061 1 0.055 1 Yes/No hit ratec 0.71 0.78 0.87

Notes: The intercept and the site visits are category specific. Bold parameter estimates are significant at the 5% level (95% HPD region). a. Pooled effects indicate coefficients that are pooled across the (six) categories. b. For Σα the estimated diagonal elements are shown. For Σβ (an identity

matrix) the ones are shown. c. The multivariate Probit Yes/No hit rates show in what percentage of

occasions the model correctly predicts that an individual will buy for each category.

Table 3-9 offers the correlations among the different categories

estimated through the Type-II Tobit model; these correlations are not very strong. The ladies’ fashion, men’s fashion, accessories, and sports categories indicate significant correlations that suggest cross-category or co-occurrence effects (Manchanda et al. 1999). Categories can be considered complementary when promotional activities in category x lead to additional buying in category y (for instance, coffee and creamer), but they are substitutes if a significant negative relation characterizes the sales of x and y. Manchanda et al. (1999) also describe co-occurrence, which takes place if categories co-occur in the


69

same basket because they have similar purchase cycles, such that sales levels of category x influence the sales levels of category y. The strongest correlation appears for [Men’s Fashion, Sports] and [Ladies Fashion, Accessories]. Intuitively, the latter pair could be a matter of co-occurrence, but it also may be that when ladies’ fashion items are discounted, women are more inclined to buy accessories to match their new clothes, which implies cross-category effects. The expenditures for living do not correlate with expenditures in any other category.

TABLE 3-9 CORRELATION MATRIX FOR THE DECISION TO BUY PER CATEGORY

Ladies Men’s Children Access Living Sports Ladies 1 0.125 0.100 0.182 0.017 0.158 Men’s 0.125 1 0.031 0.145 0.035 0.202 Children 0.100 0.031 1 0.130 0.014 0.061 Accessories 0.182 0.145 0.130 1 0.068 0.090 Living 0.017 0.035 0.014 0.068 1 0.066 Sports 0.158 0.202 0.061 0.090 0.066 1 Notes: Bold parameter estimates are significant at the 5% level.

Table 3-10 shows the variance–covariance matrix Ω for the second stage of the Tobit model, spending. The highest variance occurs in the unexplained part of the ladies’ fashion category; in other words, behavior in this category is the hardest to explain with our set of explanatory variables. The largest covariances again occur for the pairs [men’s fashion, sports] and [ladies’ fashion, accessories], but we also find another relatively high covariance for [men’s fashion, accessories].

TABLE 3-10 VARIANCE–COVARIANCE MATRIX FOR THE AMOUNT OF MONEY SPENT

Ladies Men’s Children Access Living Sports Ladies 1.214 0.100 0.093 0.163 -0.006 0.095 Men’s 0.100 0.883 0.027 0.125 0.005 0.121 Children 0.093 0.027 0.828 0.092 0.009 0.028 Accessories 0.163 0.125 0.092 0.929 0.064 0.047 Living -0.006 0.005 0.009 0.064 0.954 0.033 Sports 0.095 0.121 0.028 0.047 0.033 0.765 Notes: Bold parameter estimates are significant at the 5% level.


70

3.5.4 Validation for Amount Spent per Trip To determine whether the results are driven by the composition of

the estimation sample, we estimate the multivariate Type-II Tobit model on four random validation samples. We use the same three comparisons as those we performed for Poisson model, namely (1) the percentage of validation sample parameters that fall within the 95% confidence interval, (2) the percentage of validation sample parameters that have the same sign, and (3) the percentage of validation sample parameters that have the same sign and significance.

With regard to the confidence interval, across both stages, 60.6% (65% for stage one, 56.3% for stage two) of the validation sample parameters fall within the confidence interval. Of the 160 total parameters, 153 (95.6%; 100% for stage one and 91.25% for stage two) have the same sign as the estimation sample parameter, and 119 parameters, or 74.38% (75% for stage one and 73.75% for stage two), have both the same sign and significance level as the estimation sample parameters. An inspection of the hit rate and the mean average percentage error (MAPE) also shows a comparable fit across samples. Overall, the results from the validation samples again are consistent with the estimation sample.

3.5.5 Further Investigation of the Individual Site Parameters

Although the effect of the number of site visits on shopping trips is negative, the Web site may have positive effects for at least some customers. The unobserved heterogeneity parameters for both the Poisson model and the MVP Type-II Tobit model enable a deeper investigation. In the Poisson model, it appears that approximately 20% of the customers in the estimation sample experience a positive effect of the Web site on the number of shopping trips. For other customers, the expression ( )iγ γ+ from equation (3) regarding the effects of site visits is

negative. The unobserved heterogeneity parameters for the Type-II Tobit

model suggest an investigation of the positive effects of the site visits on some customers’ decision to buy and the amount spent. With respect to the decision to buy (first stage of the Tobit-II model), in the estimation sample, 0–9% of the customers experience a positive effect from using the Web site. With regard to money spent, it ranges from 9% to 16%. We display the percentages for the estimation sample and the average percentages across all five samples in Table 3-11.


71

TABLE 3-11 PERCENTAGE OF CUSTOMERS WITH POSITIVE EFFECTS (POSITIVE COEFFICIENTS IN THE MVP) FOR THE NUMBER OF SITE VISITS

Estimation Sample Average: 5 Samples Yes/No

(%)( icα ) Spending (%)( icβ )

Yes/No (%)( icα )

Spending (%)( icβ )

Ladies 0 14 0 12 Men’s 9 16 2 8 Children 1 11 0 10 Accessories 0 10 0 10 Living 0 9 1 11 Sports 1 11 0 7 Average 2 12 0 10

Note that for the men’s category, a substantial number of customers in the estimation sample suggest that site use has a positive effect on their decision to buy and the amount of money they spend. In the other four samples, we find similar patterns, through not similarly high percentages.

We perform various post hoc analyses for the Type-II Tobit parameters (see Appendix VII for the results). Across the 5 samples, 341 of the 1,286 customers display positive site use coefficients. With this approach, a customer may have 12 positive coefficients (i.e., 2 stages, 6 categories per stage), and among the group of customers with positive coefficients, 92.4% reveal 3 positive coefficients at most.

Customers with at least one positive coefficient live closer to the store. At the zip code level, the group with at least one positive coefficient contains more single households, which buy significantly less from catalogs. None of the individual sociodemographics is significant except for distance. Therefore, these results indicate that sociodemographics do not convincingly contribute to explaining the different effects of the Web site.

Customers with at least one positive coefficient buy significantly more items, spend more money, and visit the store more often. At the category level, we find similar effects for all categories except for the living and sports categories. Customers with positive coefficients also have viewed significantly more Web pages.

Finally, we performed an overall calculation to determine how much more the customers with a positive effect from the Web site (i.e. approximately 20% of the sample for the Poisson model) had to visit the


72

Web site to make up for decrease in behavior by the customer with a negative effect from the Web site. The analysis shows that if both groups of customers (i.e. customers with a positive coefficient and customers with a negative coefficient) have an equal number of Web site visits, the total effect on offline buying behavior is negative. If the customers with a negative effect of the Web site only visit the Web site once, the total effect is positive. If these customers visit the Web site twice, the customers with a positive effect from the Web site have to visit the Web site at least seven times to counteract the negative effect. Hence, it seems that given the low percentage of customers with a positive effect from the Web site a total positive effect on offline buying behavior is unlikely.

3.6 DISCUSSION

For most customers, the effects of visiting the Web site on shopping trips, the decision to buy in a particular category, and the amount of money spent in that category are negative. Ansari et al. (2006) and Gensler et al. (2007) reveal similar negative effects on individual customer behavior in the case of an added transactional Web site. We extend their work by demonstrating that an added informational Web site causes the same negative effects on individual customer behavior. Our results also fall in line with those of Van Baal and Dach (2005), who find customer retention levels within the same organization of 10% across the online and offline channels.

Ansari et al. (2006) focus on transactional channels, Van Baal and Dach (2005) base their results on a survey, and our study is based on actual behavior given an informational channel. The differences among these studies clearly show that on the firm-specific level, multiple channels provide customers, but not necessarily firms, with benefits.

The negative effects for firms, and the positive effects for customers, may result from (1) better and more efficient customer decision-making processes, (2) reduced impulse buying, and/or (3) reduced switching costs. We review our findings in light of these three possible explanations.

Efficient decision-making processes. Customers pursuing economic goals are likely to use online channels to form their consideration sets, because it is easier to compare products in this medium, unless the hedonic dimension is important for the particular product category (Balasubramanian et al. 2005). Mathwick and Rigdon


73

(2004) indicate that attitudes toward a firm’s Web site and its brands and/or products may be enhanced when customers participate in an engaging, enjoyable online experience. Combining this finding with the distinction made by Hoffman and Novak (1996) between experiential and goal-directed Internet users; we posit that most customers in our panel use the site in a goal-directed manner. Therefore, on the one hand our respondents may become more efficient in their decision-making process. On the other hand, for experiential visitors, the Web site may not offer sufficiently compelling experience to enhance their attitudes and behavior toward the firm.

Reduced impulse buying. Shiv and Fedorikhin (1999) indicate that in situations in which processing resources are scarce, customers with higher impulsivity choose products on the basis of spontaneous evoked affect rather than cognitions. Providing highly impulsive customers with a medium like the informational Web site with which they can interact but not transact, makes their processing resources highly available. After gathering information online, they must consciously decide to go to the store, which minimizes the chances that they will engage in impulse buying. In other words, highly impulsive customers who have easy access to sufficient processing recourses will choose on the basis of cognitions rather than impulses. Our findings indicate that customers who use the Web site make fewer shopping trips, which means they have fewer opportunities to choose products on the basis of evoked affect as they walk through the store. Therefore, site visits may reduce impulse buying behavior.

Reduced switching costs. The last explanation relies on reduced switching costs, which might occur because competitors are just a click away (Porter 2001) or because the online context loosens psychological bonds (e.g., Neslin et al. 2006). An informational Web site always has a disadvantage, compared with competitors who offer transactional Web sites, because it cannot offer the customer a possibility to buy the product immediately. Informational Web sites therefore force the customer to visit the offline store if they wish to purchase a particular product. In this situation, a customer might just as easily click to a transactional Web site, especially for generic or nonsensory products such as books or compact discs.


74

3.7 CONCLUSIONS

In this chapter, we study the effects of informational Web sites on individual offline buying behavior. With our decomposition, we model two components to determine whether the use of the informational Web site changes the frequency of offline shopping trips or the average amount of money spent per shopping trip in six different categories. We also determine whether category-specific site pages affect the amount of money spent per category.

For most customers, use of the informational Web site (i.e., site visits) has a negative influence on offline shopping behavior, in terms of the number of shopping (offline) trips, the likelihood of buying, and the amount of money spent in a particular product category. Previous studies (e.g., Moe & Fader 2004) indicate that in a pure online setting, more online visits increase the propensity to buy online, but according to our results, this relationship does not hold in an online/offline setting; rather, more online visits decrease people’s propensity to shop and buy offline. However, we argue that this result is not driven by a negative perception of the Web site; the May 2002 survey reveals average satisfaction with the Web site of 3.43 and average satisfaction with the store of 3.78 on 5-point scales—quite satisfactory. Furthermore, as we discussed in Chapter 2, customers with positive site attitudes actually spend less money. The results also show that category-specific site pages do not significantly affect the money spent in a particular category.

For only a small percentage of customers, visiting the Web site has a positive impact on offline behavior. Specifically, 20% of customers visit the offline store more often and about 10% buy more products. Our post-hoc comparison shows that customers who experience a positive effect from using the Web site on average spend more at the department store. Therefore, an informational Web site can be beneficial among a small percentage of the company’s customers. If these customers spend more, an exclusive Web site makes sense. The overall calculation confirms that a Web site should be exclusive, as the small percentage of customer with a positive impact on their offline behavior cannot make up for the customers with a negative impact on offline behavior. Introducing an informational Web site to the majority of customers makes them more efficient and conscious of their decisions.

The results from studies focusing on a more general (non-firm specific) level indicate that consumers benefit from using multiple


75

channels (see e.g., Nicholson et al. 2002; Burke 2002). However, these studies seem to indicate that firms’ offline channels can also benefit from this behavior. The findings from our study and Gensler et al. 2007, Ansari et al. 2006 and Van Baal and Dach 2005 indicate that this might not be the case. Further research is needed to determine under which circumstances and/or conditions both firms and customers benefit from the multichannel environment.

Many variables may affect cross-channel behavior, but we restrict our consideration to those to which we had access. Most of these variables pertain to individual behavior either offline or online, whereas virtually no data are available for marketing instruments other than the Web site, such as pricing, features, displays, or competitors’ marketing efforts. Because we deal with many product categories sold in the department store, the potential competitors are also many. Furthermore, the competitive profiles of the 58 outlets of the department store differ, which makes it impossible to collect data about all potential competitors. The negative effects from visiting the Web site and its pages might be more logical in a competitive setting. Customers with a higher degree of technology readiness (e.g., Parasuraman 2000) might also visit competitors’ Web sites more frequently. If these competitors offer the possibility to buy online, then a negative effect between Web visits or page views and offline store patronage might be expected. An interesting avenue for future research would be not only to include competitive effects but also variables such as “share of Web visits or page views”.

Because of the few time-varying variables included in this study, our results are less suitable for forecasting purposes. Rather, they are intended primarily to provide deeper insights into how the use of an informational Web site may influence buying behavior in the offline store. We acknowledge that the fit of the model likely would increase with more time-varying variables.

Furthermore, our study is limited to one Web site for one organization, which makes it hard to indicate the extent to which these results are generalizable to other organizations or Web sites. With our customer panel, we also cannot determine if the Web site provided the organization with new customers but instead are limited to investigating its effect on existing customers. The likelihood that the Web site attracted new customers is small given that the Web site is only promoted through the offline stores. Our study does provide a


76

preliminary clarification of the impact an informational Web site can have on offline buying behavior, and the models we have developed easily can be applied to other situations in which both online and offline information about individual customers is available.

Consequently, we offer evidence that the use of an informational Web site influences customers’ buying behavior, especially when the site itself does not provide customers with the opportunity to buy online. Our research also demonstrates that implementing an informational Web site should be considered with great care, because the majority of customers will likely become more efficient as they gather the readily available online information.

77

4 Cross-Channel Behavior for an Informational Web Site and an Offline Store12

The objective of the study discussed in this chapter is to gain insight into the long-term cross-channel effects given an informational Web site and an offline store. In comparison to Chapter 3, we now consider long-term effects of cross-channel behavior at a more aggregate level. We explicitly consider various marketing efforts, such as the introduction of an informational Web site and online promotions on offline buying. We also consider feedback loops, including the way in which online search might trigger offline buying and vice versa, by estimating a vector autoregressive model. Cross-channel behavior varies with the context characteristics, such as product type and the frequency of site visits. Our findings show that at the aggregate level online and offline behavior do not effect each other strongly. When we split the data based on context characteristics, we find more effects between online and offline behavior. The results for the median splits are unexpectedly compared with findings from previous research. Moreover, online and offline marketing efforts do not necessarily have the same impact on the behavior in the channels.

4.1 INTRODUCTION

The Internet has captured practitioner and research attention during the past decade, leading to an impressive body of marketing literature on the topic. For instance, 6% of all articles in the top five marketing journals in the past seven years (1998-2005) deal with Internet channels. Most of these articles focus on transactional Web sites; only a handful (0.4%) study the effects of informational Web sites on customer behavior.

Most firm Web sites do not allow for transactions and thus may be deemed informational in nature (e.g. 70% in Carroll 2002). Informational Web sites are easier to implement, because they do not require integration with the follow-up processes demanded by online orders. Moreover, consumers predominantly prefer to use the Internet

12 This chapter is based on Teerling, M.L., K.H. Pauwels, P.S.H. Leeflang and

K.R.E. Huizingh (2006), Cross-Channel Behavior for an Informational Web Site and an Offline Store, Working Paper, University of Groningen.


78

for information searches and conduct their subsequent purchases offline. Such ‘Web-to-store’ shopping, or cross-channel behavior, represents about $1.70 of every dollar spent directly online (CrossMedia Services 2006). In another study, American multichannel retailers indicated that the Web influenced 20% of their in-store sales (Forrester Research 2005).

The inability to combine data regarding actual search behavior in one channel with actual buying behavior in another (Sullivan & Thomas 2004) has prevented insights. This study answers the call for further research (see, e.g. Verhoef et al. 2007) that studies actual channel use regarding search and purchase from a single firm perspective; it also incorporates the sequential process by observing the channel dynamics over time. Unlike Chapter 3 where we focus on immediate effects of online behavior, this study focuses on the long-run effects of cross-channel behavior. These effects might be quite different from the short-term effects in a similar way as has been found in studies measuring the short- and long-term effects of promotions (see e.g., Nijs, Dekimpe, Steenkamp & Hanssens 2001).

Therefore, the objective is to determine the long-term cross-channel effects, i.e., from online search to offline buying and vice versa. Furthermore, this study helps to determine the effects of marketing efforts in channel a on buying behavior in channel b (i.e., cross-channel effects of marketing efforts), as well as how context characteristics moderate cross-channel behavior and marketing effort effects.

We investigate cross-channel customer behavior in an informational Web site/offline store setting by explicitly considering feedback loops between offline buying and online search behavior. We focus on two channels, a traditional department store (offline channel) and an informational Web site (online channel). We estimate a vector autoregressive (VAR) model that includes four components of offline buying behavior (monetary value per product, products per shopping trip, shopping trips per customer, and number of store customers) and four components of online search behavior (time per page, pages per visit, site visits per visitor, and number of online visitors) for a period of 126 weeks. We also address various marketing efforts (offline promotions, site introduction, online promotions, and online communications) with the VAR model and determine moderating effects of context characteristics through median splits.


79

Given that we study the long-term impact of cross-channel behavior, we focus on the cumulative impact. The cumulative impact measures the total effect over a period of 26 weeks. For marketing efforts, we determine the immediate impact in the opposite channel, which involves effects in the same week. For this study, we are primarily interested in the immediate cross-channel effect of promotions, although it is known that marketing effects may effect sales in the weeks following the promotion (e.g., Van Heerde, Leeflang & Wittink 2001) We estimate the VAR model at two levels: the aggregate level (over the entire data set), and the median split level. To the best of our knowledge, this chapter describes the first study to

• Analyze sequential cross-channel customer behavior in terms of online search and offline buying for a given firm,

• Determine how context characteristics moderate cross-channel customer behavior, and

• Use a persistence modeling approach to capture dynamic effects and feedback loops between online search and offline buying. The remainder of this chapter is organized as follows: We describe

the relevant literature, then present several hypotheses related to marketing efforts and context characteristics that may influence multichannel behavior in this particular setting. In our methodology section, we focus on the persistence modeling approach for estimating long-term and feedback effects and show that this method can be used to determine effects for different moderators. After describing our data set, we present and discuss our findings and conclude with some avenues for further research.


We first discuss previous research in multichannel behavior and then detail relevant studies in online consumer information search. Given our research theme of cross-channel effects, we also discuss the decomposition of customer behavior.

4.2.1 Multichannel Behavior

The general assumption is that multichannel customers are better customers for a firm, because they generate more revenue and purchase more items. This is confirmed in a number of studies (see e.g.,


80

Thomas & Sullivan 2005). However, Sullivan and Thomas (2004) indicate that introducing the Internet as a transactional channel does not necessarily increase the customer’s spending amount. Ansari et al. (2006) and Gensler et al. (2007) both show that customers who increasingly use the Internet have lower behavioral loyalty, manifested as lower purchase quantities. Ansari et al. (2006) speculate that this effect results from the lower switching costs, easy comparison across firms, and less personal contact involved with the Internet, all of which loosen the psychological bonds between the customer and the firm. Another explanation of these results suggests that customers may free ride. This means that customers use online information in their decision-making process but purchase elsewhere (Van Baal & Dach 2005). These consumers are also called “research shoppers” (Verhoef et al. 2007).

The key question for these studies is how search behavior in one channel influences buying behavior in another. Both studies collect cross-sectional survey data, but Verhoef et al. (2007) do not focus on a particular firm, whereas Van Baal and Dach (2005) do. This difference probably explains why Verhoef et al. (2007) find that for 73% of consumers, Internet search leads to offline store purchases, whereas Van Baal and Dach (2005) find a similar relationship for only 10% of consumers. Furthermore, they indicate that multichannel companies lose more customers across channels than they retain (Van Baal & Dach 2005).

The choice of a particular channel depends on context characteristics, such as product attributes, consumer and situational characteristics (e.g., Neslin et al. 2006). These characteristics likely also influence cross-channel behavior. Verhoef et al. (2007) note that customers evaluate channels for the different stages in their decision-making processes, on the basis of channel attributes, such as search convenience and risk. We expect multichannel behavior to differ depending on the following characteristics: (1) product type, (2) level of flow experienced, and (3) frequency of site visits as we discuss in more detail below.

Product type. Empirical research demonstrates that product type can explain some variation in channel choice. Consumers evaluate sensory products using all their senses, especially touch, smell, and sound, before purchase (Degeratu, Rangaswamy, & Wu 2000). In contrast, they usually can assess the value of nonsensory products


81

objectively using readily available information conveyed by descriptions (Peterson et al. 1997; Degeratu et al. 2000).

Research shows that buying via the Internet is less suitable for products with more sensory attributes, such as clothing, for several reasons. First, the offline environment offers more total information about sensory products (Degeratu et al. 2000). Second, consumers prefer channels that accurately portray the characteristics of the product (Burke 2002). Third, consumers need more tactile information (Peck & Childers 2003; Citrin et al. 2003). That is, Peck & Childers (2003) indicate converting consumers with a high need for touch to non-touch media, such as the internet, will be difficult. For these customers, an integrated click and brick strategy is necessary (Peck & Childers 2003). Because nonsensory products are more suitable for online shopping, a consumer probably will not need an offline channel to select them, unlike sensory products. H7.

Level of flow experienced. The extent to which a customer experiences flow while visiting a web site influences cross-channel effects (e.g., Novak, Hoffman, & Yung 2000; Mathwick & Rigdon 2004). Hoffman and Novak (1996) define flow as a state that is characterized by a seamless sequence of responses facilitated by machine interactivity, intrinsically enjoyable, accompanied by a loss of self-consciousness, and self-reinforcing. This state enhances attitudes toward a firm’s Web site (Mathwick & Rigdon 2004), as well as behaviors such as depth of search and repeat visits (Hoffman & Novak 1996; Novak et al. 2000). Customers who experience higher levels of flow engage in more online activity and find the online channel more enjoyable, which suggests a high state of flow mainly drives same-channel effects.

Frequency of site visits. Consumers with a higher visiting frequency also have higher conversion rates online (e.g., Moe & Fader 2004) and tend to have greater loyalty toward or preference for a particular channel (e.g., Shankar et al. 2003). In the setting of an informational Web site, a conversion must take place offline, and from the firm perspective, more frequent Web visits should lead to increased offline behavior. However, as previous research indicates, it is more likely that a high frequency of Web visits signals preference for the online channel.


82

4.2.2 Online Information

The vast amount of easily accessible information on the Internet and the relative newness of the medium prompt questions about the type of information that can add value for the customer, how online experiences might influence the effects of that information, and whether crucial differences exist between online information and classical advertising.

We assume that the elements that determine advertising effectiveness also determine site effectiveness, on the basis of empirical outcomes from various studies. For example, Chen and Wells (1999) demonstrate that consumers evaluating Web sites consider entertainment and informativeness important just as they do when they evaluate traditional media. Similarly, Alpar, Porembski and Pickerodt (2001) show that Web sites with special-interest content tend to be more effective than those with general interest content. Gallagher, Parsons and Foster (2001) and Gallagher, Foster and Parsons (2001) show that given an equal opportunity for exposure, a print advertisement transferred to the Web is as effective as it has been in print, even when it does not take full advantage of the Internet’s capabilities.

However, we must note a crucial difference between the effects of online and traditional media. Traditional media are limited by consumers’ tendency to avoid advertising, the potentially limited relevance of messages at the time of exposure, and the nature of the advertising, which may not be worth the consumer’s attention (e.g., Ducoffe 1996). In contrast, online consumers have control over what they view and for how long, which implies they engage in much more active dealing with the available information. On the Web, consumers actively search for specific information about, for instance, the latest promotions, which implies a greater level of consumer involvement. Thus, differences in the effectiveness of each type of information are likely.

Regarding active search, Ratchford et al. (2003) demonstrate that the Internet reduces search for an item on average and that the presence of information on the Web leads to a substantial reduction in the search time devoted to offline sources, such as visiting a dealer. The value of the information depends on what the consumer is trying to accomplish, as well as the fit between the consumer’s shopping goals and the properties of the retail environment (Mathwick, Malhotra &


83

Rigdon 2002). However, given the relative ease of searching on the Internet, it seems surprising that current search levels remain low and that consumers tend to be loyal to just one site (Johnson, Moe, Fader, Bellman & Lohse 2004).

4.2.3 Decomposing Offline Buying and Online Search Behavior

To capture cross-channel behavior, we turn to previous studies on decomposing customer behavior into various components. Decomposition customer behavior can provide additional insights in a variety of research settings, including promotional effectiveness in both the short (e.g. Gupta 1988; Bell, Chiang & Padmanabhan 1999) and the long (e.g. Pauwels, Hanssens & Siddarth 2002; Van Heerde et al. 2004; Van Heerde & Bijmolt 2005) run. In a store context (see e.g., Lam et al. 2001), managers often have a keen interest in improving specific performance components, such as switching customer purchases to higher-revenue products, increasing the number of products bought per visit, increasing store trips by existing customers, or increasing the customer base overall. Ideally, site content and promotions can induce customers to cherry-pick and buy only cheap brands (Fox & Hoch 2005), or upgrade to higher-margin brands (Chandon, Wansink & Laurent 2000). We focus on two decompositions, namely offline buying behavior and online search. With regard to offline buying behavior, we focus on the money spent per product, products bought per shopping trip, shopping trips per customer, and the number of unique customers in the store per period. In the online search arena, we consider the time spent per page, pages seen per online visit, online visits per visitor, and number of unique visitors online per period.

4.3 HYPOTHESES

Our hypotheses involve the marketing efforts that influence offline buying and online search behavior. The marketing efforts differ according to the channel (offline versus online) and focus (price versus nonprice). We focus on the immediate impacts of marketing efforts, because previous research shows that promotions usually do not change sales structurally over time (e.g., Nijs, Dekimpe, Steenkamp & Hanssens 2001; Pauwels et al. 2002).


84

4.3.1 Hypotheses: Marketing Efforts

Following Kaul and Wittink (1995), we classify marketing efforts into price promotions (promotions) and nonprice promotions (communications). Promotions inform customers about the price and availability of a product, whereas communications inform the customer about product positioning and unique product characteristics. We consider the introduction of the web site as a communication decision.

Research on promotions demonstrates that temporal price discounts supported by featuring and/or advertising substantially increase sales of the promoted products (at the reduced prices) in the short term and lift store traffic, but they also decrease reference prices and increase price sensitivity (Blattberg, Briesch & Fox 1995; Nijs et al. 2001; Van Heerde, Leeflang & Wittink 2001; Van Heerde et al. 2004). Thus, we expect the effects of offline promotions on offline buying behavior to be similar to those found in numerous previous studies.

Offline promotions also may increase online search behavior. Customers might be triggered by offline promotions to use the informational Web site to find out more about promotions and determine if they are worth a trip to the store. However, if offline promotions provide customers with sufficient information, a visit to the Web site might become redundant, which would decrease the average number of site visits.

In this study, we are interested how online marketing efforts influence offline buying. We therefore formulate our hypotheses to pertain to the immediate effects of marketing efforts in the online channel on buying behavior in the offline channel.

Ansari et al. (2006) indicate that marketing efforts (both online and offline) have a positive effect on purchase volume, but mainly drive customers to the same channel; that is, these authors mostly report same-channel effects. We argue that the effects of online promotions may differ from those found for offline promotions, possibly as a result of the associated level of customer consciousness. Customers online tend to search more actively for specific information and therefore likely are more efficient. In turn, the Web site can facilitate less effort or time on their part, as well as fewer shopping trips, to fulfill their goals (Mick & Fournier 1998). We therefore formulate the following hypothesis for the immediate effect of online promotions.


85

H1. Online promotions decrease money spent per product, increase products purchased, decrease trips taken and increase the number of customers of the retailer.

In contrast to promotions, communications reduce price sensitivity (Kaul & Wittink 1995) because they are geared toward communicating unique brand or product features. Because online customers actively search for particular information, online communications likely increase the money they spend per product or lead them to upgrade to higher-margin products (Chandon et al. 2000). However, just as we hypothesized for online promotions, communications can facilitate consumer efficiency and thereby reduce the number of shopping trips they make (Mick & Fournier 1998).

That is, we expect online communications to increase the money spent, the products purchased, and the number of customers but decrease trips taken. Just introducing the focal site should have a similar effect to general online communications because of its theme orientation and general focus on communication. We formulate the following hypothesis for the immediate effect of online communications:

H2. Online communications and the site introduction increase money spent per product, products purchased, and number of customers but decrease trips taken.

Table 4-1 provides an overview of these hypotheses.

TABLE 4-1 OVERVIEW OF FORMULATED HYPOTHESES REGARDING MARKETING EFFORTS

From To Literature Expectation Money per product - - Products per trip + + Number of trips + -

Online promotions (H1)

Customers + +

Money per product + + Products per trip + + Number of trips + -

Online communications & site introduction (H2)

Customers + +


We describe the persistence modeling approach for estimating long-term marketing and feedback effects.


86

4.4.1 Decomposition of Behavior

We decompose both offline buying and online search behavior into various relevant components, as follows:

(1) = = * * *t t tt t t

tt t

M P TrOffline behavior Total money C

CP Tr,

where

tM = the monetary value spent in period t,

tP = the total number of products purchased in period t,

tTr = the total number of shopping trips in period t, and

tC = total number of customers in period t.

(2) * * *t t tt t t

tt t

Ti Pa VsOnline behavior Total time Vrs

VrsPa Vs= = ,

where

tTi = the total amount of time spent online in period t,

tPa = the total number of pages seen in period t,

tVs = the total number of online visits in period t, and

tVrs = total number of Web visitors in period t. In Appendix VIII, we provide a more precise description of the series. In

the rest of this chapter, we will refer to t

t

MP

as money ( tM ), t

t

PTr

as

products ( tP ), t

t

TrC

as trips ( tTr ) and tC as customers in terms of offline

buying behavior. For the online search behavior, we refer to t

t

TiPa

as

time ( tTi ), t

t

PaVs

as pages ( tPa ), t

t

VsVrs

as visits ( tVs ) and tVrs as visitors.

4.4.2 Dynamics

To test our hypotheses, we need a flexible model that can uncover dynamic marketing effects and feedback loops. Because we have little a priori knowledge about the signs and dynamics of those effects, we employ a persistence modeling framework (Dekimpe & Hanssens 1999)


87

and specify a VAR model to uncover important interrelationships among the series instead of determining them a priori (Sims 1980; Enders 1995). Equation (3) represents the VAR model:

(3)

= =

= =

= =

= =

+ Σ + Σ + ∂⎡ ⎤ + Σ + Σ + ∂⎢ ⎥ + Σ + Σ + ∂⎢ ⎥⎢ ⎥ + Σ + Σ + ∂

=⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎣ ⎦

4 20, 1 , , 1 , ,

4 20, 1 , , 1 , ,

4 20, 1 , , 1 , ,

4 20, 1 , , 1 , ,

0,

M y y M y t s s M s t M

t P y y P y t s s P s t Pt

Tr y y Tr y t s s Tr s t Trt

C y y C y t s s C s t Ctt

ttt

a a YD a SD tM a a YD a SD tP a a YD a SD tTr

a a YD a SD tCTi aPaVsVrs

β

= =

= =

= =

= =

⎡ ⎤⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥⎢ ⎥ +

+ Σ + Σ + ∂⎢ ⎥⎢ ⎥+ Σ + Σ + ∂⎢ ⎥

+ Σ + Σ + ∂⎢ ⎥⎢ ⎥+ Σ + Σ + ∂⎣ ⎦

1,1

4 21 , , 1 , ,

4 20, 1 , , 1 , ,

4 20, 1 , , 1 , ,

4 20, 1 , , 1 , ,

.i

Ti y y Ti y t s s Ti s t Ti

Pa y y Pa y t s s Pa s t Pa

Vs y y Vs y t s s Vs s t Vs

Vrs y y Vrs y t s s Vrs s t Vrs

a YD a SD ta a YD a SD ta a YD a SD ta a YD a SD t

ββ ββ ββ ββ ββ ββ ββ β

γ γγ

−

−

−

−

−=−

−

−

⎡ ⎤⎢ ⎥ ⎡ ⎤⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

+⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

⎣ ⎦⎢ ⎥⎢ ⎥⎣ ⎦

∑

1,8

2,1 2,8

3,1 3,8

4,1 4,8

1 5,1 5,8

6,1 6,8

7,1 7,8

8,1 8,8

1,1 1,4

2,1

.......................

....

i

i it i

i i t it ii iKt i

i it ii

i i t it ii it ii i

MPTrCTiPaVsVrs

εγ ε

γ γ εγ γ εγ γ εγ γ εγ γ εγ γ ε

⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎡ ⎤⎢ ⎥ ⎢ ⎥⎢ ⎥ +⎢ ⎥ ⎢ ⎥⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥⎣ ⎦⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦

,

2,4 ,

3,1 3,4 ,1,4,1 4,4 ,2,5,1 5,4 ,

6,1 6,4 ,

7,1 7,4 ,

8,1 8,4 ,

....................

M t

P t

Tr ttC ttTi tt

t Pa t

Vs t

Vrs t

OPOPOCSI

We face an important decision regarding which series to include in the VAR estimation and whether to treat these series as endogenous or exogenous. The vector of endogenous series includes the components of offline buying and online search, defined by the decomposition. The endogenous series relate to their own past and thereby allow for complex, dynamic interactions.

The first vector of exogenous series includes (1) an intercept a (2) four half year dummies (YDy,t) to account for trends of time, (3) two seasonal dummies (high and low) (SDs,t), and (4) a deterministic-trend variable t. The high and low seasonal dummies control for peaks and dips that cannot be explained by the marketing efforts. The marketing efforts, measured as dummies, constitute the second set of exogenous series: (1) offline promotions (OP1,t), (2) online promotions (OP2,t), (3) online communications (OCt), and (4) site introduction (SIt).

In our estimation, we choose the number of lags on the basis of the Schwarz information criterion (SC). We do not interpret the coefficients of cross-channel behavior in the VAR model directly but focus our attention on the impulse response functions (IRF), which simulate the over-time impact of a change (compared with a baseline) in one variable


88

on the full dynamic system. For the cross-channel effect, recall that we consider the cumulative (over 26 weeks) effect of the IRF (Pauwels et al. 2002), whereas for the marketing efforts, we are interested in the immediate (i.e., same week) effects. We interpret the coefficients for the marketing efforts of the VAR model instead of the IRFs.

4.4.3 Model Calibration Steps

We start by inspecting the series graphs plotted against time to obtain general insights into the series’ behavior (graphs not shown here). To assess the temporal behavior (evolution/stationarity) of offline buying and online search, we perform unit root tests (Enders 1995; Maddala & Kim 1996, Dekimpe & Hanssens 1999 for marketing applications). Then, we remove the deterministic component from the series through seasonality dummies. That is, we inspect the series for specific seasonality peaks and dips that the marketing mix instruments cannot explain. Next, we relate all the series to one another with a correlation matrix and remove those series that cause multicollinearity. Finally, we perform Granger causality tests to determine which series of interest in period t causes changes in other series in future periods, which enables us to interpret the results of the subsequent VAR models and IRFs. We make only causal inferences for the impulse series that the Granger test considers to cause the response series.

4.4.4 Unit Root Testing Procedure

Unit root testing determines whether a series is stationary over time, that is, whether it reverts to its stationary mean or trend. If a series is stationary, there are no permanent effects, and we estimate a VAR model in levels.

Generally, the Augmented Dicky Fuller (ADF) method serves to test for a unit root. The ADF contains the null hypothesis that all series have a unit root (i.e., are nonstationary). Other tests are available as well, including the Kwiatkowski, Phillips, Schmidt and Shin (KPSS) test, whose null hypothesis states that the series is stationary. For unit root testing, we must choose whether: • the test specification includes a deterministic time trend, and • on which null hypothesis— unit root (e.g., ADF) versus stationarity

(e.g., KPSS)—it should be based.


89

For the unit root test specification, we follow Enders (1995, p. 257) procedure, which requires that we test in both the presence and the absence of a deterministic time trend. When the first test specification indicates the absence of a unit root, the series is classified as trend stationary. If the second test specification indicates the absence of a unit root, the series is classified as mean stationary. When both specifications indicate a unit root, we classify the series as evolving.

Regarding our second choice, we perform a confirmatory analysis, as suggested by Madalla and Kim (1996, p. 126). With a confirmatory analysis, we can use a test with stationarity as the null hypothesis to confirm the conclusions about the unit root (i.e., compared with the test that uses unit root as the null hypothesis). The confirmatory analyses, rather than just the ADF test, highlight concerns about conventional unit root tests, such as their low power (see e.g., Madalla & Kim 1996). Confirmatory analysis uses two different tests, namely ADF (H0: series is nonstationary) and KPSS (H0: series is stationary). When these test results converge, we obtain greater reliability in terms of the series’ classification. Table 4-2 provides an overview of the unit root testing procedure.

TABLE 4-2 UNIT ROOT TESTING PROCEDURE

Test Specification Result Conclusion 1. ADF With intercept and

trend (t) H0a rejected t significant

Trend stationary

H0 not rejected t significant

Unit root

2. KPSS With intercept and trend (t)

H0b not rejected t significant

Trend stationary

H0 rejected t significant

Unit root

If trend variable t is insignificant 3. ADF With intercept H0 rejected Mean stationary

H0 not rejected Unit root

4. KPSS With intercept H0 not rejected Mean stationary

H0 rejected Unit root a. H0 states that the variable has a unit root b. H0 states that the variable is stationary

In the confirmatory analysis, the level of confirmation between the ADF and KPSS tests depends on the strictness of the classification. Exact confirmation occurs if test 1 and 2 (or 3 and 4) from Table 4-2 result in the same conclusion, namely, trend stationary or mean


90

stationary. We obtain a less strict confirmation if we consider only the evolving versus stationary classification (Enders 1995).

In addition to the unit root test procedure, we test the offline buying series for a structural change with a known breakpoint, namely, the first week after the introduction of the Web site (Madalla & Kim 1996). Specifically, we use the Chow breakpoint test to uncover any significant differences in the estimated equations before and after the site introduction. We do not elaborate on the test results when they confirm the previous unit root test results.

4.4.5 Moderation

In linear regression, moderation can be easily tested with an interaction term. With our VAR model, including 8 endogenous variables, adding an interaction would imply another 8 endogenous variables. Each equation would then be estimated with at least 28 exogenous variables (i.e. in case of one lag). Assuming five observations are needed per parameter, our data would not be sufficient. Moreover, introducing these interactions terms might also cause multicollinearity problems. Hence, to determine the influence of the context characteristics, we follow Lim, Currim and Andrews (2005) by estimating the VAR models for each high versus low group (i.e., product type, experience of online flow, frequency of Web site visits). Comparing the Granger causality and dynamic performance impact through the IRF’s indicates whether there are differences across the groups.

Several studies have classified products as sensory versus nonsensory goods and we draw on these to determine how the products should be classified. Table 4-3 depicts the classification we assigned and in which studies this classification has been used previously.

TABLE 4-3 CLASSIFICATION OF PRODUCT TYPE

Classification Example study Consumer electronics Nonsensory Burke 2002, Citrin et al. 2003 Cd’s, book’s, DVD’s Nonsensory Burke 2002, Citrin et al. 2003 Computer hard- and software

Nonsensory Peterson et al. 1997, Van Baal & Dach 2005

Toys Nonsensory Van Baal & Dach 2005 Clothing Sensory Citrin et al. 2003, Van Baal &

Dach 2005 Shoes & accessories Sensory Peterson et al. 1997, Citrin et al.

2003 Cosmetics Sensory Van Baal & Dach 2005 Furniture Sensory Van Baal & Dach 2005


91


We use the same empirical data as discussed in Section 1.6.

4.5.1 Data

The data used in this chapter pertain to the behavior of 6,594 customers who started using the Web site after its introduction. The data include these customers’ offline buying behavior for 127 weeks. Of the 127 weeks, 60 weeks of buying behavior occur before the introduction of the Web site, and we collect online search behavior for the 67 weeks after the site introduction.

The sample sizes for the analyses differ due to the availability of information about the moderators. We estimate a VAR model at the aggregate level for 6,594 customers. For the product type median split, we use the data from 6,594 customers; for the flow median split, the sample sizes are 2,900 for low flow and 3,481 for high flow; and for the median split of the frequency of site visits, the sample sizes are 3,651 for low frequency and 2,623 for high frequency.

In our median split approach, we use attitudinal and customer behavior data. The attitudinal data, i.e. the experience of online flow, of visitors to the Web site are gathered through two online questionnaires, conducted three months after the introduction of the Web site in May 2001 and one year later in May 2002. Flow is a construct represented by the mean of responses to questions in both questionnaires. All questions use a five-point scale. We measure the frequency of Web visits as the number of site visits a customer makes during the data collection period. On the basis of the median level, we split the customer panel into two segments. Appendix IX provides the items or descriptions of the variables used in the moderation approach, the reliability of these variables, and descriptive statistics13.

4.6 AGGREGATE LEVEL FINDINGS

This section details the results of the VAR model at the aggregate level. We first discuss multichannel behavior, followed by the effects of marketing efforts.

13 Descriptives for product types are not provided, because there is no

median product type; rather products are classified solely as sensory or nonsensory.


92

4.6.1 Multichannel Behavior

Preliminary data inspection. A visual inspection of the series suggests sufficient variability in the data to designate two important seasons: summer and winter. We check the data for seasonality and inspect the residuals after correcting for known marketing efforts. We include a low dummy, a high dummy (the two seasonal dummies), and four half-year dummies in the VAR model.

The correlation between the offline buying series varies from .05 to .30. For online search series, the correlation varies from -.32 to .69. In other words, though the series are correlated, we do not anticipate any important multicollinearity issues.

Unit root testing. The test results in Table 4-4 show that all series are (mean or trend) stationary, according to the ADF test. In the ADF, the test value (tv) must be greater than the critical value (cv) to reject the null hypothesis of a unit root, whereas with KPSS, the test value must be less than the critical value to accept the null hypothesis of stationarity.

We find exact confirmation (i.e., both the AFD and KPSS tests indicate trend or mean stationarity) for five of the eight series. When we consider only whether the series is classified as stationary versus evolving, we confirm that all series are stationary. Compared with Madalla and Kim (p. 128, 1996), who report little convergence for the macroeconomic series they tested, we find more promising results through our confirmatory analysis of these marketing series.

The structural break Chow test indicates a significant structural break in week 61 in the number of trips (F = 6.32 Prob. = .00; Chi-squared log likelihood ratio = 24.45 Prob. = .00) (Madalla & Kim 1996). The unit root tests, of the periods before (weeks 1-60) and after (weeks 62-127) the structural break caused by the Web site introduction in week 61, indicate stationarity (ADF trend significant, t = -8.88 for weeks 1-60, t = -4.93 for weeks 62-127). By allowing for the structural break in the VAR model with a dummy for the site introduction, we estimate the VAR in levels (Kornelis 2002; p. 49).


93

TABLE 4-4 RESULTS FOR THE CONFIRMATORY UNIT ROOT ANALYSIS

ADF KPSS T(rend) tv Cv Result T(rend) tv cv Result

Money 0.00 -4.44 -4.03 mean 0.00 0.11 0.22 mean

-4.47 -3.48 0.15 0.74 Products -0.002 -11.91 -4.03 mean -0.003 0.16 0.22 trend -11.65 -3.48 0.44 0.74 Trips -0.001 -5.22 -4.03 trend -0.001 0.26 0.22 mean -4.77 -3.48 0.52 0.74 Customers 2.50 -9.09 -4.03 trend 2.97 0.14 0.22 trend -8.47 -3.48 0.59 0.74 ADF KPSS t tv Cv Result t tv cv Result

Time -0.03 -7.05 -4.10 mean -0.02 0.10 0.22 mean

-7.11 -3.53 0.11 0.74 Pages -0.08 -6.05 -4.10 trend -0.11 0.24 0.22 mean -4.70 -3.53 0.53 0.74 Visits -0.002 -23.83 -4.10 trend -0.003 0.19 0.22 trend -18.86 -3.53 0.74 0.74 Visitors -6.66 -8.19 -4.10 trend -6.73 0.18 0.22 trend -6.28 -3.53 0.87 0.74

Notes: t(rend) = trend variable, tv = test value, cv = critical value. Coefficients for the trend variable in bold are significant at the .05 level.

Granger causality. To make causal inferences based on the results of the VAR model and IRFs, we test for Granger causality at lags 1-6 between the online search and offline buying behavior and vice versa.14 The results show that at the aggregate level, only a few Granger-caused cross-channel relationships occur. For online search, only “visitors” Granger-cause trips. Thus, if more unique visitors go online in week t, in week t+1, the average number of shopping trips per customer changes. With regard to the offline buying components, “trips” and “customers” Granger-cause an online search component: “visitors”.

VAR Model. We determine the number of lags through the SC and the SC with the lowest value of 37.14 indicated a VAR with one lag. The VAR with one lag shows an acceptable in-sample model fit (the adjusted

14 We do not expect online search to Granger-cause offline buying or vice

versa beyond a period of six weeks.


94

R2 ranges from .29 to .95, and the F-statistic ranges from 3.54 to 132.7). The Lagrange-multiplier (LM) test shows residual correlation at lag order 2 (LM-stat = 94.37; Prob. = .008), but none at lag order 3 (LM-stat = 58.63; Prob. = .66) or lag order 4 (LM-stat = 76.82; Prob. = .13).

Considering the large amount of results a VAR model provides, we interpret only the relations of interest that show Granger causality in at least one test specification. Table 4-5 offers the cumulative IRF results.

TABLE 4-5 CUMULATIVE RESULTS OF THE IRFS FOR MULTICHANNEL BEHAVIOR AT THE AGGREGATE LEVEL

From ↓ to → M P Tr C Ti Pa Vs Vrs A B

Time (Ti) -0.11 0.04 0.00 -64.0 10.76 2.03 0.23 72.0

Page (Pa) 0.19 0.00 0.01 0.0 -2.63 2.20 0.00 0.00

Visits (Vs) 0.14 -0.03 -0.04 -56.6 13.17 8.50 0.62 150.1

Visitors (Vrs) 0.00 -0.07 0.03 55.4 0.53 0.25 -0.02 139.3 C D

Money (M) 1.85 -0.17 -0.03 -23.6 1.97 1.81 0.14 40.9

Products (P) 0.27 0.31 0.00 145.5 -1.89 -1.29 -0.08 -48.6

Trips (Tr) -0.50 0.05 0.06 193.1 0.41 0.19 0.00 22.4

Customers (C) 0.37 0.09 0.04 316.2 -2.24 -1.62 -0.11 -33.7

Notes: Bold parameter estimates indicate Granger causality.

Panel A includes the cross-channel effects from online search in week t on offline buying in weeks t + n, where n can reach a maximum of 26 weeks. Panel B shows the same-channel effects from online search in week t on online search in weeks t + n. Panel C provides the same-channel effects of offline buying in week t on offline buying in weeks t + n. Finally, Panel D includes the cross-channel effects of offline buying in week t on online search in weeks t + n.

Panel A. We find a limited number of cross-channel effects. The only cross-channel behavior effect from online to offline, which is Granger caused, is from visitors to trips. If more customers visit the Web site in week t, the average number of trips to the store by Panel A in weeks t + n increases. The increase in unique visitors online most likely reflects customers with a higher shopping trip baseline.

Panel D. For the cross-channel behavior from offline to online, we find that an increase in trips in week t increases the number of visitors online in weeks t + n. Moreover, an increase in customers in week t,


95

decreases online visitors in weeks t + n. Overall, an increase in offline buying negatively influences online search.

Panel C. Considering the same-channel effects of offline buying, we find among other effects that • If money per product increases, over time, customers buy fewer

products per trip; • If products per trip increase, over time, customers spend more

money per product; and • If trips increase, over time, customers buy more products per trip.

Panel B. For the same-channel effects from online search, we find among other effects that • If time per page increases, over time, customers view more pages per

visit and visit more often; • If pages per visit increase, customers spent less time per page; and • If visits increase, customers spent more time per page and view

more pages per visit. The cumulative cross-channel results indicate that in the long-run the Web site complements the offline store, shown by the increase in the number of shopping trips. However, the offline store retains customers in the offline channel.

4.6.2 Cross-Channel Marketing Efforts

Table 4-6 provides the effects, that is, the VAR coefficients, of marketing efforts on both online and offline behavior. We focus our attention on the immediate effects of online marketing efforts (i.e., promotions and communications) on offline buying and of offline promotions on online search. For benchmark purposes, we also provide the same-channel effects.

The results indicate that offline promotions, online communications, and the site introduction increase one or more components of offline buying. In addition, online promotions, online communications, and site introduction have an effect on online search. The effect of the site introduction on online behavior reflects the average online behavior in the first week of Web site use. We notice that online promotions decreases time spent per page, indicating that online promotions make users browse through the web pages faster.


96

The results for the online and offline promotions confirm previous results summarized by Ansari et al. (2006): Promotions mainly cause same-channel effects. The results pertaining to the online communications and site introduction suggest that these types of marketing efforts have cross-channel effects.

TABLE 4-6 EFFECTS OF MARKETING EFFORTS ON OFFLINE AND ONLINE BEHAVIOR AT THE AGGREGATE LEVEL

Offline Behavior Money Products Trips Customers

Offline promotions -0.03 -0.04 0.07 380.68 Online promotions 0.28 0.13 0.01 67.44

Online communications 0.54 0.54 0.01 199.52

Site introduction 3.04 1.20 -0.06 1218.74 Online Behavior Time Pages Visits Visitors

Offline promotions 0.67 0.3 -0.02 -19.18

Online promotions -3.55 0.75 -0.01 -10.34

Online communications -1.98 -0.37 -0.01 129.21 Site introduction 48.46 21.50 1.27 1571.58

Notes: Bold parameter estimates are significant at the 5%- level.

Table 4-7 summarizes the results and formulated hypotheses. In H1, we posit that online promotions decrease money spent, increase products purchased, decrease trips taken, and increase the number of customers. We cannot confirm this hypothesis because the results indicate insignificant relationships. However, we can confirm, as we suggest in H2, that online communications and the site introduction increase money spent, products purchased, and number of customers, but we cannot support our claim that they decrease the trips taken. We do note that the Web site causes a structural break in the number of store trips. After the introduction the number of store trips decreases significantly. This effect, however, is not found for the online communications.

Overall, we find limited cross-channel effects at the aggregate level and show that marketing efforts mainly trigger behavior in the same channel. Moreover, we find that online marketing efforts are more effective if they communicate brand or product features instead of price.


97

Besides the aggregate VAR, we also performed an exploratory segmentation through cluster analysis. The VAR model is estimated for each segment. The results of this exploratory analysis show differences across segments. However, the segmentation approach does not provide additional insights.

TABLE 4-7 REVIEW OF HYPOTHESES ABOUT MARKETING EFFORTS

From To Lit. Exp. Result Money - - Not confirmed Products + + Not confirmed Trips + - Not confirmed

Online promotions (H1)

Customers + + Not confirmed

Money + + Confirmed Products + + Confirmed Trips + - Not confirmed

Online communications & site introduction (H2)

Customers + + Confirmed Notes: Lit. = literature, Exp. = expectation.

4.7 MEDIAN SPLIT FINDINGS

Our approach to determine the moderating impact of context characteristics (i.e., product type, flow, frequency of site visits) on cross-channel behavior requires us to divide the customer panel by a median split and estimate the persistence modeling framework for each condition (i.e., low versus high). Because not all data are available for all customers, the sample sizes for the median splits vary (see Section 4.5.1). We focus on determining whether context characteristics moderate cross-channel behavior and therefore report only cross-channel effects (Panels A and D in Table 4-6). We focus on the long-term impact that is the total change in behavior over a period of 26 weeks. Moreover, we consider only those relationships that are Granger-caused. For our interpretation of the marketing efforts, we focus on the coefficients from the VAR model.

Preliminary data inspection. For all median splits, we conclude there is sufficient variability in the data and find the same seasons as in the aggregate data.

The correlations for product type vary between -.25 and .52 for offline buying and between -.37 and .78 for online search. For flow, the correlations with offline buying vary between -.13 and .50 and those with online search between -.04 and .60. With regard to the frequency of Web visits, we find correlations with offline buying that vary between -.16 and .52 and with online search that vary between -.20 and .81.


98

Unit root testing. For all median splits, we perform the confirmatory analysis (Madalla & Kim 1996) and, for the majority of series, find confirmation of the tests, just as we did with the aggregate level tests. A structural break for the site introduction exists for products and trips in most median splits. Both series in all median splits may be classified as stationary, after we allow for the break. As a result of the unit root testing, we consider all variables stationary and therefore include them in the levels for the VAR models.

4.7.1 Product Type

Cross-channel behavior. We may choose a first- or second-order VAR according to the lag length criterion, but on the basis of a comparison of other the fit measures (i.e., adjusted R2, LM, normality, and heteroskedasticity), we use a second-order VAR. The VAR with two lags shows an acceptable in-sample model fit (for sensory products, adjusted R2 ranges from .28 to .99, F-statistic ranges from 3.5 to 1006.85; for nonsensory products, adjusted R2 ranges from .22 to .95, F-statistic ranges from 2.5 to 979.62). The LM test shows no residual correlations for the nonsensory products but some residual correlation for the sensory products.

Table 4-8 provides the cumulative changes in the response series of a one-unit shock in the impulse series for both median splits.

TABLE 4-8 CROSS-CHANNEL IRF RESULTS FOR SENSORY AND NONSENSORY PRODUCTS

Money Products Trips Customers NS S NS S NS S NS S

Time -0.20 0.76 0.00 0.00 0.00 0.00 -6.84 0.00 Pages -0.06 -0.37 0.02 0.01 0.01 0.01 0.00 8.92 Visits 1.54 0.00 0.00 0.00 -0.03 0.00 -70.40 0.00 Visitors -0.42 -0.22 0.06 -0.01 0.06 -0.02 86.09 38.71

Time Pages Visits Visitors NS S NS S NS S NS S

Money 5.28 -1.46 4.07 0.00 0.27 0.00 59.31 0.00 Products 0.00 0.00 -0.14 0.17 -0.01 0.00 -61.63 -14.38 Trips 0.00 0.00 0.39 0.00 0.00 0.00 5.29 0.00 Customers -2.09 0.00 -2.35 0.00 -0.15 0.00 -38.63 0.00

Notes: NS = nonsensory products, S = sensory products. Bold parameter estimates indicate Granger causality.


99

For nonsensory products, shocking online search decreases money spent but increases products purchased, which implies greater consumer awareness of the price due to online search. A shock in offline buying leads to a decrease in online search in terms of pages viewed.

For sensory products, we find that more time per page increases money spent, but an increase in the number of unique visitors online, decreases it. Offline buying does not Granger-cause online search for sensory products. The number of cross-channel effects is similar in both conditions, and the direction of the impact varies across the median splits. For nonsensory products, online search mainly decreases money spent per product, whereas for sensory products, online search mainly increases money spent per product. This indicates that customers search for the cheapest alternative in case of non-sensory products. For sensory products, the findings indicate the customers upgrade to more expensive products.

When reviewing the graphs of offline behavior15 and the R2 adjusted for both sensory and nonsensory products, the differences make more sense. Figure 4-1 shows the money spent per product for sensory and non-sensory products as an example of the difference between both conditions.

We notice that the graphs for the number of customers for both product types roughly show similar patterns. For the other three offline behavior variables, the graphs show very different patterns. Money spent per product has a more volatile pattern for nonsensory products, whereas products and trips are more volatile for sensory products. Second, we see that the variation explained is higher for money spent per product for sensory products (sensory products: money R2a = .59; nonsensory products: money R2a = .25) and products per trip for nonsensory products (sensory products: products R2a = .29; nonsensory products: products R2a = .59).

The cross-channel effects from online to offline are stronger for sensory products. Regarding the direction of the relationship, we find that an informational Web site benefits money spent in case of sensory products. However, for nonsensory products, the cross-channel effects from offline to online are stronger but negative, especially for the number of products per shopping trip.

15 We do not inspect the graphs for online behavior because the online

behavior is equal in the case of this median split.


100

8

10

12

14

16

18

20

25 50 75 100 125

Non-sensory products Sensory products

Mon

ey s

pen

t pe

r pr

odu

ct

t

FIGURE 4-1 COMPARISON OF MONEY SPENT PER PRODUCT FOR SENSORY AND

NON-SENSORY PRODUCTS

4.7.2 Flow Median Split

Cross-channel behavior. According to the lag length criterion, we could choose either a first- or a second-order VAR, but the second-order VAR provides a better in-sample model fit (for low flow, adjusted R2 ranges from .25 to .97, F-statistic ranges from 2.5 to 148.02; for high flow, adjusted R2 ranges from .16 to .99, F-statistic ranges from 1.8 to 676.58). The LM test shows no residual correlations for high flow but some for low flow.

Table 4-9 details the cross-channel IRF results. The number of Granger-caused relationships for low flow (17) is twice as many as for high flow (8). The effect on money spent for low flow is either positive or negative depending on which online search variable represents the shock. We observe that the effect for low flow is stronger than that for high flow. With low flow, an increase in online search increases money spent and decreases trips taken, whereas with high flow, an increase in online search decreases both money and trips.


101

For the high flow condition, we notice that offline behavior only effects online search if money spent per product or the number of products per trip is increased. An increase in money spent, increases the pages seen. An increase in products, decreases time spent per page. For the low flow condition, we notice a lot more effects from offline behavior on online search than in the high flow condition. An increase in money has a positive effect on online search, whereas an increase in products or trips seems to decrease online search.

TABLE 4-9 CROSS-CHANNEL IRF RESULTS FOR LOW AND HIGH FLOW EXPERIENCE

Money Products Trips Customers L H L H L H L H

Time 0.43 0.00 0.00 -0.08 -0.05 -0.02 0.00 -18.63 Pages 1.00 0.24 0.00 0.00 -0.05 0.01 15.37 16.16 Visits 1.58 0.25 0.00 -0.03 -0.10 -0.03 -26.41 0.00 Visitors -0.82 -1.15 0.00 0.06 -0.01 0.01 20.80 18.43

Time Pages Visits Visitors L H L H L H L H

Money 6.81 -0.38 5.38 1.66 0.34 0.03 41.09 13.49 Products 0.00 -0.55 -0.28 0.00 -0.01 0.00 -12.22 -4.96 Trips -0.55 0.00 -1.20 0.44 -0.08 0.00 -1.28 4.22 Customers 0.00 -0.42 0.65 0.22 -0.03 -0.01 0.00 0.00

Notes: L = low flow, H = high flow. Bold parameter estimates indicate Granger causality.

To explain some of the differences between low and high flow, we inspect the graphs (not shown here) and adjusted R2 for offline and online behavior. For the offline behavior variables, we observe that the patterns in the graphs are similar. Considering the explained variance, for low flow the VAR explains more for money spent and trips taken when compared with high flow (for low flow, money R2a = .25, trips = .41; for high flow, money R2a = .16, trips = .32). Reviewing the online behavior graphs and the VAR adjusted R2’s, we find similar patterns. However, the pre-VAR calibration steps (see Section 4.4.3) indicate for high flow that lagged online behavior is a better predictor of online behavior than for low flow. This difference may cause a larger number of Granger-caused relationships for low flow.

In Section 4.2.1 we mentioned that one may expect that the cross-channel effects will be stronger for customers who have experienced low flow. Our findings indicate that in case of an informational Web site, the


102

cross-channel effects are strongest, in terms of the number of Granger-caused relationships and the magnitude of the effects, for low flow. Those customers who experience higher flow integrate the channels less.

4.7.3 Frequency of Site Visits

Cross-channel behavior. The lag length criteria indicates we may choose a first- or second-order VAR, but again, the second-order VAR, shows a better in-sample model fit (for low visit frequency, adjusted R2 ranges from .11 to .96, F-statistic ranges from 1.6 to 127.2; for high visit frequency, adjusted R2 ranges from .34 to .99, F-statistic ranges from 3 to 652.4). Although, the LM test indicates residual correlations for the median split, we estimate a second-order VAR model.

Table 4-10 shows the cross-channel IRF results. Shocking online search increases offline buying, in terms of money spent and trips taken, among low frequency site visitors. For those who visit the site frequently, an increase in online search increases offline buying in terms of trips. For the low frequency of site visits, shocking offline buying mainly decreases online search, whereas for the high frequency of Web visits, shocking offline buying does not influence online search. Furthermore, among low frequency customers, we find more Granger-caused online search.

TABLE 4-10 CROSS-CHANNEL IRF RESULTS FOR LOW AND HIGH FREQUENCY OF SITE VISITS

Money Products Trips Customers L H L H L H L H

Time 0.65 -0.31 -0.01 0.00 -0.02 -0.02 10.53 -20.75

Pages 0.51 0.19 0.03 0.05 0.01 0.00 7.82 3.09

Visits 0.00 0.25 0.05 -0.16 -0.01 -0.06 -11.35 -6.61

Visitors -0.17 -0.40 0.03 0.10 0.01 0.03 30.22 11.24 Time Pages Visits Visitors L H L H L H L H

Money -1.58 -0.69 0.04 0.88 -0.04 0.00 4.03 39.26

Products -2.24 -1.30 -0.76 -2.61 -0.02 -0.17 -2.64 -17.36

Trips -7.06 -1.85 -3.18 -3.49 -0.26 -0.21 -26.94 -15.55

Customers -2.14 -2.72 0.25 -2.97 -0.01 -0.18 5.64 -27.79 Notes: L = low frequency of site visits, H = high frequency of site visits. Bold parameter estimates indicate Granger causality.


103

Based on the offline behavior series, i.e. the graphs and the VAR R2 adjusted, we do not see many differences between the splits. We notice that among customers who frequently visit, lagged money spent is a better predictor of money spent than for those who visit infrequently. With respect to the online behavior series, we observe more volatility in the graphs for low visits, especially for time per page and pages per visit. Besides that, lagged online behavior predicts online behavior better in the case of high visits (R2 adjusted range from .34 to .75 in case of high visits and from .09 to .45 in case of low visits).

The literature (see Section 5.2.1) indicates that that the cross-channel effects are stronger for customers who engage in a low frequency of Web visits, and our findings enable us to confirm this expectation. For customers with a low frequency of visits, we find more cross-channel effects, especially from offline to online.

Investigating the differences across the median splits, we find that

the data aggregation creates varying patterns. Comparing our median split results to the aggregate level results, we find some differences and similarities. At the aggregate level, visitors do not have an effect on money spent but for both low and high flow, we find a negative significant effect. Visitors at the aggregate level do increase trips, which we also find in case of a high frequency of Web site visits. Two possible reasons may cause the differences: aggregation bias over units, and different sample sizes. The differences in the graphs and explanatory behavior of lagged (own series) behavior, indicate a likelihood of parameter heterogeneity across customers for the aggregate level VAR (Leeflang et al. 2000 p. 268).

4.7.4 Cross-Channel Marketing Efforts for Median Splits

Appendix X summarizes the results of online marketing efforts’ effects on offline buying for each of the median splits.

For nonsensory products, only the site introduction significantly increases money spent and the number of customers. A significant structural break in the shopping trips is found for nonsensory products. After the introduction, the number of shopping trips for nonsensory products decreases significantly.

For sensory products, the site introduction increases products purchased, trips taken, and number of customers, and online promotions increase trips. No structural break due to the introduction


104

of the Web site is found in case of sensory products. For sensory products, online promotions influence behavior in the opposite channel.

The effects of the marketing efforts are small and similar for the median split of flow. Only the site introduction has a significant influence on offline buying. For both low and high flow a structural break in the number of shopping trips and products purchased is found. In the case of high flow, the site introduction has a positive immediate influence on both products and customers, whereas with low flow, it only influences the number of customers. For both conditions the long term impact of the site introduction is negative for the number of shopping trips and products bought per shopping trip.

For a low frequency of site visits, the site introduction has a significant impact on offline buying. More frequent visits means the site introduction significantly increases money, products, and customers. Online promotions also significantly increase products purchased among customers who visit more frequently. For both conditions, a structural break caused by the Web site introduction is found for the number of shopping trips and the number of products bought per trip. Both offline buying components diminish after the site introduction.

Compared with the aggregate level results, the median split results tell a different story with respect to the effect on trips taken and the influence of online promotions. At the aggregate level, we find no significant results regarding the number of trips, and for sensory products, we find that online promotions and the site introduction increase the number of shopping trips. The informational Web site does not make customers more efficient but instead causes them shop more often offline in the case of sensory products. In contrast with the aggregate level, for customers who engage in more frequent site visits, we find that online promotions increase the number of products they purchase. The structural breaks found for all conditions, except for sensory products, are in line with the aggregate results. The results indicate that in case of sensory products an informational web site has more positive effects compared to non-sensory products.

4.8 DISCUSSION

In this chapter, we investigate cross-channel effects in an informational Web site/offline store setting. We focus our attention on cross-channel behavior, or how online search affects offline buying and vice versa. We develop an approach to determine cross-channel effects


105

and specify different VAR-models that capture the relationships among different components of offline and online behavior. We also determine how marketing efforts in channel a affect behavior in channel b; and how context characteristics moderate cross-channel behavior and marketing efforts. We discuss our findings in the context of existing multichannel literature.

4.8.1 Cross-Channel Behavior

At the aggregate level, we find limited cross-channel behavior. the results seem to indicate that over time behavior in a particular channel is best explained by past behavior in the same channel. In case of the traditional channel (i.e. the offline channel), offline behavior keeps customers in the offline channel. The online channel complements the offline channel in the long-term.

Compared with these results however, we find more cross-channel effects for the median splits. It appears that aggregation across individuals customers in the panel limits the results at the aggregate level. The differences between the aggregate-level VAR and the median split VARs likely result from heterogeneity in the parameters across customers (Leeflang et al. 2000 p. 268). That is, customers in each median split likely react more similar, which results in parameter homogeneity across customers.

In some of the median-split results, we find that online behavior decreases money spent per product, a result that reinforces our findings from chapter 3 that the majority of individuals, though not all, experience a negative effect. In addition based on the structural break Chow tests, we find that online search generally decreases the number of shopping trips, which may indicate that customers free-ride on the information provided (see e.g. Van Baal & Dach 2005), become more efficient in their decision-making process (see e.g., Alba & Lynch 1997) or become less loyal over time (see e.g., Gensler et al. 2007).

Overall, our results, especially for the median-splits, indicate that behavior in channel a influences behavior in channel b and vice versa. These influences, however, are not necessarily positive. The cross-channel synergies are apparent in the case of money spent per product in some cases, but we cannibalization occurs in the number of shopping trips.


106

4.8.2 Cross-Channel Marketing Efforts

We find that offline marketing efforts mainly drive behavior in the offline channel, in line with Ansari et al. (2006), who show that promotions cause same-channel effects. However, online marketing efforts also create cross-channel effects. The setting of our study —an informational online channel instead of an online transactional channel—may explain this effect. With an online informational channel, buying behavior can only take place offline.

Reviewing our results across the different data aggregations, we find that the effects of marketing efforts are reasonably consistent. The different type of coefficients likely causes this greater consistency, in that we measure marketing efforts as immediate effects on behavior variables, whereas we measure cross-channel effects with Granger-caused IRFs (i.e., cumulative effects).

Overall, our results show that of the online marketing efforts the site introduction has the strongest impact, with positive immediate effects on money spent per product, products bought per shopping trips, and the number of customers in the store. Although the site introduction does not have an immediate effect on the number of shopping trips, the site introduction also entails a structural break for this component. The results of the structural break test show that due to the site introduction, customers reduce their average number of shopping trips per week over time.

4.8.3 Context Characteristics

We investigated moderator effects of product type, flow during Web visits and frequency of Web visits and find that for sensory products, online search improves offline buying. This result is surprising, because previous studies indicate that prepurchase decisions about sensory products rely heavily on consumers’ ability to touch, smell and taste (Degeratu et al. 2000). However, Gupta et al. (2004) also show that the search effort associated with sensory products is significantly higher than that for nonsensory products, and our results confirm this claim. To generate ideas or a consideration set, online information demands less effort than does retrieving the same information offline. Our results confirm the higher search effort required but also show that an informational Web site provides customers with a more efficient way to retrieve initial information about sensory products.


107

We confirm our expectation that a lower level of flow strengthens the cross-channel effects. This finding might be the results of several reasons. First, customers who experience high flow lose their self-consciousness and a sense of their surroundings (e.g., Hoffman & Novak 1996). Therefore, customers with high flow might not perceive the channels as integrated or may consider the online experience satisfactory to fulfill their need to shop. Second, previous research shows that the experience of flow during a site visit improves customers’ attitudes and behavior toward the online channel (see e.g., Hoffman & Novak 1996, Mathwick & Rigdon 2004) but not toward the offline channel. Our results indicate that a high state of flow, though intrinsically enjoyable to the customer, does not enhance offline behavior. Rather, customers who experience low flow tend to exhibit more cross-channel behavior and are more receptive to marketing efforts.

With regard to transactional Web sites, previous research shows that greater visiting frequency improves conversion rates and indicates stronger loyalty or preference for the online channel (see e.g., Shankar et al. 2003; Moe & Fader 2004). Our results confirm that customers who visit the site more frequently prefer the online channel; we find no significant relationships between offline buying and online search. We also show for an informational Web site that customers who visit less frequently exhibit stronger cross-channel behavior and are more receptive to marketing efforts.

4.9 CONCLUSIONS

The objective of this study is to determine to what extent cross-channel behavior takes place when online search is the only option. We estimate a VAR model and thereby capture the relationships among the different types of offline and online customer behavior. We also determine how marketing efforts in channel a influence behavior in channel b. The VAR model estimated for each median split provides insight into cross-channel customer behavior and the effects of marketing efforts given a specific moderator.

Our research also shows that aggregate-level estimations of cross-channel behavior are limited. With median splits, we demonstrate that introducing an informational Web site benefits the organization most with regard to sensory products, low flow during Web visits and a low frequency of Web visits. An increase in online search leads to an


108

increase in offline buying of sensory products. Online search leads to increased money per product, but less shopping trips for those customers who experience less flow during their Web visits. An increase in online search leads to an increase in money and trips for those customers who visit the site less frequently.

Prior research also indicates that marketing efforts influence customers to use a particular channel (Burke 2002; Ansari et al. 2006). It would be reasonable to expect that offline promotions would lead customers to the store and online promotions would lead customers to the Web site. However, in the case of informational Web sites, promotions, whether offline or online, influence customers’ use of the offline channel, in which they may buy products. Customers who experience a high state of flow and engage in frequent Web visits are more receptive to online marketing efforts. The marketing efforts also benefit the organization more in the case of sensory products.

Although our research provides a plethora of new insights, as is any research, it is limited to the variables to which we have access. Most of the variables we collected pertain to individual customer behavior in either channel. For example, with regard to the marketing efforts, we cannot determine how the effects of offline communications influence online search. Our median-split approach shows that customers in our panel do not react in a similar manner. Thus, even though our median-splits allow for deeper insight, more conclusive answers may be possible through latent-class VAR modeling. Furthermore, most insights into the moderators of cross-channel behavior pertain to multiple transactional channels, and because our results differ from prior expectations, further research is needed to uncover the determinants of these differences.

Any organization is subject to competitors, about which we unfortunately have no information. Additional research might improve our insights by determining how cross-channel behavior varies, given that consumers use multiple providers to search for and purchase particular products. However, obtaining actual individual consumer search and buying behavior in multiple channels for multiple organizations represents a great challenge.

In conclusion, our study provides some initial insights into cross-channel behavior for online search and offline buying. Online search positively may influence the money spent per product but lowers the average number of shopping trips customers make. An increase in


109

offline behavior mainly translates into an increase in online, which indicate increased overall behavioral loyalty to the store. Cross-channel behavior also is moderated by context characteristics in a manner that conflicts with previous research. Among customers who experience low flow online, visit the site less frequently, and are interested in sensory products, cross-channel behavior is stronger. Therefore, for an informational Web site, our results show that online marketing efforts drive offline buying.

111

5 Discussion and Conclusions In this chapter, we provide an overview of the studies presented in Chapters 2–4, including the main conclusions, general similarities, and differences across studies. We discuss the main theoretical and managerial implications of these findings and conclude by identifying some limitations that additional research might address.

5.1 INTRODUCTION

In this dissertation, we provide various insights into the knowledge gaps identified in Chapter 1 (see Table 1-2). We focus specifically on how an informational Web site affects the offline channel at the level of individual customers. Chapter 2 deals with determining the effects at the attitudinal and behavioral levels, Chapter 3 addresses the effects of the introduction of a Web site at the customer level, and Chapter 4 studies how online and offline behavior affect each other over time. Table 5-1 lists the characteristics of the studies presented in Chapters 2–4.

TABLE 5-1 MAIN CHARACTERISTICS OF THE RESEARCH IN THIS DISSERTATION

Chapter 2 Chapter 3 Chapter 4 Subject Customer attitudes

and behavior Individual customer behavior

Cross-channel effects

Description Influence of online attitudes on offline attitudes & behavior

Effects of online behavior on trips & category spending

Sequential process of online search & offline buying

Model Structural equation model

Poisson model, multivariate Probit model

Vector Autoregressive model

Range of data

Cross-sectional, survey May 2002 behavior aggregate March ’01–May ‘02

January 2000–May 2002, a total of 29 months

January 2000–May 2002, a total of 127 weeks

Sample size

Overall = 2,877 Moderators = 2,816 Longitudinal = 422

Five random samples of approx. 215 customers per sample.

Aggregate = 6,594 Flow = 3,233 and 3,822 Product type = 6,594 Web visits = 3,651 and 2,623

Aggregation level

Customer Yearly (March ’01-May ’02)

Customer Monthly

Panel, weekly Median split, weekly


112

5.2 SUMMARY

5.2.1 Attitudinal Framework (Chapter 2)

Most multichannel research has focused on the effects of an additional transactional channel on customer attitudes toward the incumbent channel or the organization as a whole. At the aggregate level, informational Web sites can have a positive effect, depending on the position of the additional channel (e.g.. Deleersnyder et al. 2002; Lee & Grewal 2004). At the individual customer level, the effects of an additional informational channel remain unknown. Considering that 10 years after the introduction of the Internet for commercial purposes, online purchases remain minimal, the effects of informational Web sites represent an intriguing area for study.

In Chapter 2, we focus on the effects of an additional online informational channel and specifically investigate the effects of online attitudes and behavior on offline attitudes and behavior through structural equation modeling. We also study how these relationships hold up given a longitudinal design and whether and how these relationships are moderated.

We find that customers with a positive site attitude also have a positive store attitude. This relationship is stronger for certain customer segments, such as female and less educated customers. We offer both a general and a more Internet-related explanation for the differences caused by sociodemographics. The relationship between site and store attitude is also stronger for customers who perceive higher channel integration.

We find a negative effect between site attitude and actual behavior, which is unexpected given previous results. We indicate that the difference between our findings and previous studies may be a result from the setting of the study and the type of data used. Lastly, we find a positive effect between site behavior and store behavior. However, the association is relatively small and instable.

Overall, this chapter indicates that “good” customers—those with positive attitudes and higher spending—tend to be good customers in both channels. Nevertheless, these good customers gain efficiency, which is noticeable in the informational Web site’s negative effect on the spending levels

Chapter 5: Discussion & Conclusions

113

5.2.2 Individual Customer Behavior (Chapter 3)

In Chapter 3, we provide insight into the effects of the use of an informational Web site on individual offline buying behavior. Within our decomposition, we model two components to determine whether the use of the informational Web site changes (1) the frequency of offline shopping trips and (2) the average amount of money spent per shopping trip in six different categories. Moreover, we determine whether category-specific site pages affect the money spent per category. Because individual customers do not shop at the department store on a daily or even weekly base, we aggregate the data into monthly periods. We focus on the effect of using the Web site on offline buying behavior in the same month.

Our results show that the majority of customers decrease their buying behavior, both in terms of both shopping trips and money spent per category, as a result of visiting the Web site. Ansari et al. (2006) and Gensler et al. (2007) obtain similar negative effects on individual customer behavior in the case of an added transactional Web site, and our results also confirm Van Baal and Dach’s (2005) finding, which reveal 10% customer retention within the same organization across online and offline channels. The remaining 90% of customers use the firm’s online channel to collect specific information and then go elsewhere to purchase the products; that is, they free-ride on the provided information. Hence, multiple channels provide customers with the benefits but do not necessarily offer similar benefits to the firm that introduces a Web site.

The results from studies focusing on a more general (non-firm specific) level indicate that consumers benefit from using multiple channels (e.g., Burke 2002). These studies also seem to indicate that firms’ offline channels can benefit from this behavior. The findings from our study and Gensler et al. 2007, Ansari et al. 2006 and Van Baal and Dach 2005 indicate that this might not be the case. Firms may experience negative effects because customers achieve (1) more efficient decision making, (2) fewer impulse-driven purchases, and/or (3) lower switching costs when they use an informational Web site.

For a small percentage of customers, visiting the Web site has a positive impact on offline behavior. Approximately 20% of customers visit the offline store more often, and 10% buy more products. Our post hoc comparison shows that customers with positive coefficients on average spend more at the department store. Therefore, an


114

informational Web site can be beneficial with regard to the company’s top customers. Overall, the use of an informational Web site influences the purchase patterns of customers even when the site itself does not provide customers an opportunity to buy online. Our research also demonstrates that the implementation of an informational Web site should be considered with great care, because customers benefit more than the firm from the provision of online information.

5.2.3 Cross-Channel Effects (Chapter 4)

In this chapter, we investigate the long-term cross-channel effects that take place in an informational Web site/offline store setting at an aggregate level. We focus our attention on (1) cross-channel behavior, or how online search affects offline buying and vice versa over a period of 26 weeks; (2) cross-channel marketing efforts, or how marketing efforts in channel a immediately affect behavior in channel b; and (3) how context characteristics moderate cross-channel behavior and marketing efforts.

We estimate a vector autoregression (VAR) model to capture the relationships between the different types of offline and online customer behavior. We also determine how marketing efforts in channel a influence behavior in channel b. The VAR model estimated for each median split provides additional insights into cross-channel customer behavior and the effects of marketing efforts.

At the aggregate level, we find limited long-term cross-channel behavior, but find more such effects at the median-split level. We discover that the introduction of the Web site decreases the number of shopping trips through a structural break, which indicates that customers free-ride on the information provided (Van Baal & Dach 2005), become more efficient in their decision-making process (Alba & Lynch 1997) or become less loyal over time (Gensler et al. 2007).

We demonstrate the moderator effects of product type, experience of flow during Web visits and frequency of Web visits. Our results confirm a greater search effort associated with sensory products but also show that an informational Web site provides customers with a more efficient way to retrieve initial information about sensory products. A high state of flow, though intrinsically enjoyable to the customer, does not enhance offline behavior. On the contrary, customers who experience low flow, demonstrate more cross-channel behavior and are more receptive to marketing efforts. Finally, our results confirm that


115

customers with a high visiting frequency prefer the online channel; we find no significant relationships between offline buying and online search.

5.3 INSIGHTS

Table 5-2 lists the main findings from various selected studies on cross-channel effects. Table 5-3 lists our main findings (Chapters 2–4).

TABLE 5-2 SELECTED STUDIES ON CROSS-CHANNEL EFFECTS

S?a Channels Cross-Channel Relationships Result Ansari et al. 2006

N Web site, catalog

Internet use → buying behavior -

Gensler et al. 2007

N Web site, tv-station

Online purchase → customer retention -

Kushwaha & Shankar 2006

Web site, catalog

Use of multiple channels → buying behavior

+

Y Online attitudes → online use + Online attitudes → overall satisfaction +

Montoya-Weiss et al. 2003

Web site, store

Offline attitudes → online channel use -

Y Online search → offline buying behavior + Store search → online/catalog buying +

Nicholson et al. 2002

Web site, catalog, store

Catalog buying → online buying +

Y Online search firm a → offline purchase firm b

+ Van Baal & Dach 2005

Web site, store

Offline search firm a → online purchase firm b

+

Y Online search → offline buying behavior + (n.s.)

Catalog search → online buying behavior +

Verhoef et al. 2007

Web site, catalog, storeb

Store search → catalog buying behavior +

Y Usage of multiple channels → channel perceptions

+ Wallace et al. 2004

Web site, catalog, store

Channel perceptions → satisfaction + a. S? indicates whether the data are collected through a survey (Y) or represent

actual purchase behavior (N). b. The researchers focus on both the informational and transactional purposes

of channels.


116

TABLE 5-3 CROSS-CHANNEL EFFECTS FOUND IN THIS DISSERTATION

S?a Channels Cross-Channel Relationships Result Y Online attitudes → offline attitudes + Chapter 2 Online attitudes → offline buying

behavior -

Online search behavior → offline buying behavior (short-term)

• Majority of customer -

Chapter 3 N

• Top customers (10%) + Online search behavior → offline buying behavior (long-term)

• Aggregate level +

• Sensory products, low online visits +

• Structural break for store trips - Offline buying behavior → online search behavior (long-term)

• Aggregate level -

• Nonsensory products -

• Low online flow experience +

Chapter 4

N Web siteb,

store

• Low online visits -

a. S? indicates whether the data are collected through a survey (Y) or represent actual purchase behavior (N).

b. The channel can be used only as an informational channel.

Reviewing the findings, we notice that studies that use survey data mainly find positive cross-channel effects, with the exception of Van Baal and Dach (2005), who show that the majority of consumers tend to switch to the offline channel from another firm after searching online. In Chapter 2, we provide the following insights compared with these studies: • An informational Web site improves customer attitudes toward the

offline transactional channel. Web sites do not have to offer a transaction function for the firm to benefit at the attitudinal level.

• Customers with a positive site attitude decrease their offline store spending, which indicates that free-riding behavior takes place not only for transactional Web sites, but also, as we demonstrate for informational Web sites.

Previous studies using behavioral customer data find that an

additional transactional channel can have positive or negative effects. Gensler et al. (2007) and Ansari et al. (2006) both argue that customers


117

using multiple channels decrease their buying behavior over time and become less loyal (lower customer retention). Chapters 3 and 4 provide the following insights compared with these studies: • Using an informational Web site decreases the number of shopping

trips in the offline transactional channel in the same month for the majority of customers. Customers experiencing a positive effect from visiting the informational Web site can be categorized as more loyal, because they use both channels more than customers with a negative coefficient do. Therefore, firms can increase customer buying behavior without necessarily implementing an online transactional channel.

• The majority of customers, who are less loyal, benefit through more efficiency, less impulse buying and better decision making.

• In the long-run, the introduction of the Web site causes a structural break in the number of shopping trips. The introduction of the Web site decreases the number of shopping trips in the short and long-run.

• In addition to the decrease in the number of shopping trips, most customers spent less in the offline channel. Again, for the better customers (i.e. those who spend more), we find a positive effect on money and products.

• Offline and online marketing efforts, in the context of an informational Web site, stimulate behavior in the channel that allows the customers to purchase products.

Overall, the three studies presented in this dissertation lead to the

following conclusions: • A positive experience online benefits attitude toward the store.

Customers clearly appreciate access to and the use of multiple channels.

• Customers decrease their offline buying behavior in response to an informational Web site. The majority of customers free-ride on the information provided online. They also may become more rational and improve their decision-making process because of online information.

• Customers become more efficient in their shopping behavior after the introduction of an informational Web site. That is, customers


118

decrease the number of shopping trips they take, but also may increase spending per shopping trip.

• The firm’s top customers with regard to purchases, improve their offline buying behavior because of the informational Web site.

• Higher perceived channel integration increases cross-channel synergies. Customers who indicate that they use online information to go offline to purchase items experience a stronger relationship in their channel attitudes.

• Sensory products benefit more from an informational Web site than do nonsensory products.

• A strong experience of online flow and frequent online visiting decrease the number of cross-channel relationships, signaling lower perceived channel integration and/or a preference for the online channel.

• Online marketing efforts, in the context of an informational Web site, may stimulate offline buying behavior. Specifically, the introduction of an informational web sites increases offline buying behavior in the week of the promotion.

• The models developed in this research project can determine the effects of an added (informational) channel. Moreover, using multiple models allows insight into the differences across (1) types of data (survey versus actual behavior), (2) unit aggregation levels (individual versus entire customer panel), and (3) temporal aggregation levels (weekly versus monthly or yearly). In conclusion, we note that the different models used in Chapter 2-

4 show both consistency in the results and additional insights due to the differences. Table 5-1 already shows several reasons for some of these insights, such as variation in the sample size. In addition, the aggregation across units and time provide additional insights into the results (see e.g., Tellis & Franses 2006 on a discussion of temporal aggregation). The aggregation across units fails to take customer heterogeneity into account. Hence, the models in Chapter 3 provide a deeper insight into the differences across customers.

Additional insights are gained through the differences in the type of effects, i.e., same-period effects (Chapter 3) versus next-period effects (Chapter 4) or immediate effects (Chapter 3) versus cumulative (26 weeks) effects (Chapter 4). Lastly, in Chapter 3 we focus on money


119

spent in a particular category, while in Chapter 4 we focus on money spent per product in general, that is, not with a specific category. A negative impact on money spent in a product category does not need to imply that money spent per product cannot increase. If customers buy the more expensive items but also buy less items from a particular category, then combining the results from Chapters 3 and 4 provides additional insight into the buying behavior of customers.

5.4 MANAGERIAL IMPLICATIONS

Understanding the possible functions and benefits of an informational Web site is necessary to determine how to use the Internet as part of a multichannel strategy. This dissertation provides new insights into the effects of informational Web sites that managers can use in three main areas, namely, (1) to strategically implement (informational) Internet activities, (2) to use the methodology to gain firm-specific insights and (3) to manage the effects of informational Web sites.

First, managers already likely realize that ignoring the Internet is no longer an option. Consumers use the Internet during several stages of their decision-making process to improve their knowledge of the marketplace, increase their efficiency and minimize their impulse buying. Consumers appreciate and often expect multiple channels, so the question becomes how managers can implement Internet activities that benefit overall firm performance. The first question they confront is whether to implement an informational or a transactional Web site. This depends on the incumbent channels; for example catalog retailers could easily implement a transactional channel because the logistical processes needed to deal with remote orders are already available. Other factors to take into consideration include competitors, the type of products, purchase frequency and previous experiences with implementing additional channels. In terms of the type of product and purchase frequency, Van Baal and Dach (2005) show that retention across transactional channels is higher for infrequently purchased products. Moreover, we show that customer retention, in the context of an informational Web site, is greater for sensory products. More important, managers should carefully consider if their online activities should be available to all customers. Previous research shows that customer retention diminishes as a result of an Internet application (e.g. Gensler et al. 2007), and our findings further imply that Internet


120

activities benefit the firm only when it comes to the firm’s top customers. Therefore, firms may differentiate themselves by offering their best customers, for instance through exclusive access to, an (informational) Web site.

Second, for firms already using multiple channels, we provide several models they might use to determine cross-channel effects for their unique situations. Aggregate-level analyses, such as firm sales, do not indicate what happens at an individual customer level. Although sales models provide valuable insights into the overall performance of a firm (possibly compared with competitors), the contribution of channels, and the general effectiveness of marketing efforts, they cannot show whether individual customers become more loyal. Improved sales levels might indicate simply that additional channels have expanded the customer base. However, models at the customer level might tell a different story. Our research shows that models at the individual customer level provide valuable insights into customer heterogeneity and clarify that customers react differently to the use of multiple channels, as well as to the marketing efforts used across these channels. Therefore, managers should combine aggregate-level models with customer-level models to determine how their channels are performing. Managers can gain further insights by investigating attitudes versus behavior and effects over time or by combining our proposed models with cost analyses for each channel. Finally, we stress that collecting individual customer data related to search and buying behavior can greatly enhance managers’ understanding of customers, though the collection of such data remains problematic.

Third, we provide some implications in terms of managing the effects of informational Web sites, such as free-riding behavior. Managers can minimize free-riding by increasing channel integration, because customers enjoy using multiple channels, and beneficial cross-channel effects increase when customers perceive channels to be integrated. Managers can stimulate channel integration by (1) providing consistent messages across channels; (2) allowing customers to integrate the use of channels, such as ordering online but picking up the product in the offline store; and (3) stimulating customers to use multiple channels. In addition to improving channel integration, managers should focus their online activities on their best customers; informational Web sites can reinforce relationships with these customers. Personalizing the content of Web sites to the best customers


121

might even reinforce the relationship further. Finally, we stress that the “new” multichannel environment makes “old” marketing practices, such as differentiation, even more crucial. In conclusion, this dissertation provides the following managerial implications. • Carefully consider what type of Web site is suitable given the

organization’s incumbent channels, competitors, type of product, purchase frequency and previous experience.

• Carefully consider whether the Web site should be accessible to all customers or a selection of customers.

• Combine aggregate-level models with customer-level models to determine channel performance.

• Minimize free-riding behavior through channel integration. • Improve customer perceived channel integration through consistent

messages, the possibility of integrated channel use, and stimulation of using multiple channels.

• The ‘new’ multichannel environment makes ‘old’ marketing practices even more crucial.

5.5 LIMITATIONS AND FUTURE RESEARCH

This research provides several insights into the effects of informational Web sites in a multichannel environment. As with any research, however, it contains limitations, and we recognize that we fail to address several issues. We review some of the issues related to cross-channel effects, multichannel empirical generalizations, and omitted variables in the following sections.

5.5.1 Cross-Channel Effects

This dissertation focuses on the cross-channel effects of an online informational channel and an offline transactional channel for a specific firm. Neslin et al. (2006) refer to some interesting initial research about multichannel approaches, but various aspects pertaining to cross-channel effects also warrant further research.

First, our study is limited to one Web site for one organization, which makes it hard to clarify to what extent these results are generalizable to other organizations or Web sites. To generalize our insights about cross-channel effects, more research is needed to elaborate on these effects for other firms, industries/branches, and other (online) channels, such as brand communities (e.g., Muniz &


122

O’Guinn 2001). For example, an experimental setting could be used to investigate the effects of Web site design on customer behavior. Moreover, further research could investigate whether these effects differ across Web site type, i.e. comparing transactional and informational Web sites.

Some initial studies (e.g., Kushwaha & Shankar 2005) offer consumer behavior findings across multiple firms, but must studies have employed a single-firm perspective when they use behavioral data and a more general perspective when they use survey data. The majority of multichannel studies refer to a retail setting, which leaves limited insights into the multichannel effects in the service industry or business-to-business channels. In addition, most current studies focus on the following channels: • Catalog, • Store (offline), and/or • The Internet (transactional Web sites).

Admittedly, collecting customer behavior data for transactional

channels is easier than collecting data from customers who use informational channels, but our research shows that the effects of informational channels are not necessarily the same as those of other channels. Therefore, further investigation into the effects of channels other than “classic” transactional channels is warranted.

Second, we note that knowledge is limited about several specific cross-channel effects. For example, relatively little is known about how channel attitudes influence each other and behavior over time. Our research shows relatively consistent effects of attitudes on behavior, but our survey data are limited to two observations over time (2001 and 2002). To determine how channel attitudes affect each other and behavior over time, further research might consider, in addition to longitudinal effects, cross-category effects, channel-switching behavior, and other stages of the decision-making process.

Although we cannot demonstrate that information search in category a influences purchases in category b across different channels, prior research shows that sales levels and marketing activities for different categories can influence one another (e.g., Manchanda et al. 1999). The question therefore becomes whether searching for a particular product category in a channel influences purchase behavior in another product category in a different channel. If it does, how


123

should managers use their marketing instruments? Finally, further research should attempt to determine how customers switch between channels during the different stages (e.g., information collection, purchase, after sales) in the decision-making process. Another interesting avenue of research is to determine how consumers their technology preferences influence the cross-channel effects. More specifically, what factors cause customers to switch channels within or across firms? To what extent is across-firm channel switching different from just switching firms?

5.5.2 Generalizations

Multichannel behavior as a field of research has emerged because of the increase in the number of shopping channels available to consumers. Although the field remains relatively new, some studies offer overviews of the state of knowledge (see e.g., Rangaswamy & Van Bruggen 2005; Neslin et al. 2006). Because multichannel research comprises various topics that have received extensive attention, generalizations are valuable and could create synthesis for the field..

For example, as Neslin et al. (2006) show, several researchers have focused on the determinants of channel choice and found ample evidence of six basic determinants of channel selection (Neslin et al. 2006). The extents to which these determinants are conclusive and apply in various circumstances represent relevant research avenues for a meta-analysis. Such a study should also take into account pre-and post-hype effects. Porter (2001) indicates that the effects of Internet after the hype could be substantially different. These effects may also differ considering the faster Internet connections and a more diverse Internet population.

Other topics that need further investigation to achieve empirical generalization include (1) the effects of multiple channels on customer buying behavior, (2) cross-channel synergies and channel cannibalization, (3) the contribution of an additional channel, or (4) the effects of channel integration versus channel separation. Within each of these topics, further research is needed to determine under which circumstances certain effects occur.

5.5.3 Omitted Variables

We were able to collect only a limited number of variables, and most of these pertain to customer behavior offline or online. Hardly any


124

variables related to marketing instruments other than the Web site, such as prices and advertising expenses, were available. Similarly, most current multichannel research focuses on how customer behavior varies across channels and how different channels affect one another. To obtain conclusive insights, further research should include the effects of marketing instruments other than channels on customer behavior.

Any organization is subject to competition, on which we also unfortunately have no data. Because we deal with many product categories sold in the department store, the potential competitors are many. Furthermore, the competitive profiles of the 58 outlets of the department store differ, which makes it impossible to collect data about all potential competitors. Therefore, additional research should improve our insights by determining how cross-channel behavior varies when customers use multiple providers to search for and purchase particular products. Obtaining actual individual consumer search and buying behavior in multiple channels for multiple organizations will be a great challenge but also a great advance of existing research.

125

Appendix I. Multichannel Studies

OVERVIEW OF SELECTED MULTICHANNEL BEHAVIOR STUDIES

Study Type of channels

Empirical setting Key findings

Bendoly et al. 2005

Stores

Internet Survey, 3 US retailers, n = 1598.

Higher levels of channel integration are associated with greater firm loyalty. Channel integration does not predict channel choice.

Knox 2005 Catalog Internet

Purchase data, US retailer, n = 2000, 139 weeks.

Channel preferences evolve over time and in response to firm marketing. Customer behavior and response to marketing varies across segments.

Cha

nnel

cho

ice

Kushwaha & Shankar 2005

Catalog Internet

Purchase data, 750 firms, n = 1 mln., 4 years.

Multichannel customers are more valuable. Demographics, shopping traits and product associations affect channel choice.

Ansari et al. 2006 Catalog Internet

Purchase data, durable retailer, n = 500, 4 years.

Increased use of the Internet leads to lower purchases. Marketing plays a pivotal roll in channel usage.

Gensler et al. 2007

Call center Internet

Purchase data, home-shopping TV station, n = 15 mln., 15 months.

The incumbent channel is still the dominant channel. Customers who increasingly use the Internet have lower behavioral loyalty.

Gupta et al. 2004

Stores Internet

Survey, n = 337. Consumer switch online for search products. Price-search intention and evaluation effort drive switching online.

Cha

nnel

mig

ratio

n

Sullivan & Thomas 2004

Stores Catalogs Internet

Purchase data, US retailer, 37,000 orders, 1997 – 2001.

The combination of channels used by the customer does not automatically signal profitability to the firm.


126



Thomas & Sullivan 2005


Purchase data, US retailer, n = 4,1000, 1 year.

Two distinct segments are catalog & Internet and bricks and mortar loyal

Dholakia et al. 2005


Purchase data, US retailer, n = 530,000, 24 month.

Customers use the same channel as their original entry channel, rather add a new channel than replace the old and switch between similar channels.

Montoya-Weiss et al. 2003

Call center Internet

Survey, financial provider and US university, n = 1,137 and 493.

Overall satisfaction with firm is determined by service quality provided through both channels.

Schoenbachler & Gordon 2002

Multiple channels

Conceptual. Not applicable.

Mul

ti-ch

anne

l buy

ing

beha

vior

Shankar et al. 2003

Stores Internet

Survey, lodging industry, survey, n = 350.

The online medium can be used to reinforce overall loyalty to the firm.

Nicholson et al. 2002


Case studies, UK fashion retailer, n = 48.

Consumers combine available purchase channels due to lifestyle.

Balasubramanian et al. 2005


Conceptual. Not applicable.

Burke 2002 Multiple channels

Survey, n = 200. The majority of consumers use multiple channels. Internet use is preferred for product information and comparison.

Mul

ti-ch

anne

l sea

rch

and

buyi

ng b

ehav

ior

Van Baal & Dach 2005

Stores Internet

Survey, n = 1,094.

Multichannel firms lose more customers across channels than they retain. Retailers are not compensated for the information service outputs provided.

Appendices

127



Verhoef et al. 2007


Survey, n = 396. There are both within and between channel cross-over effects across different shopping tasks. Channel attributes can be manipulated to mitigate the 'research-shopper' phenomenon.

Verhoef & Donkers 2005

Multiple channels

Purchase data, financial provider, n = 3,317.

Customer loyalty differs among acquisition channels. Cross-buying is affected by marketing efforts.

Acq

uisi

tion

chan

nel

Villanueva et al. 2003

Multiple channels

Purchase data, Internet firm providing free Web hosting services, 70 weeks.

Firm performance differ among acquisition channels.

Notes: Stores means offline outlets.


128

Appendix II. Survey Chapter 2

DESCRIPTIVE STATISTICS FOR THE ATTITUDES AND BEHAVIOR VARIABLES

Mean Std. Deviation Personnel 3.5 .73 Store interior 3.42 .80 Price 3.48 .69 Merchandise 3.5 .68 Content 3.2 .79 Design 3.52 .73 Store attitude 3.85 .74 Site attitude 3.45 .81 Store behavior (ln) 3.15 (.81) 3.43 (.76) Site behavior (ln) 12.91 (2.23) 11.42 (.84) Channel involvement 3.29 .74 Involvement 4.49 3.89

RANGE OF THE INTER-ITEM AND ITEM-TO-TOTAL CORRELATIONS FOR THE ATTITUDE VARIABLES

Inter-Item Item-to-Total Personnel .53 - .70 .58 - .71

Store interior .54 - .74 .64 - .79 Price .61 - .75 .66 - .76 Merchandise .39 - .56 .44 - .57 Content .61 - .73 .72 - .82 Design .54 - .80 .67 - .84 Store attitude .61 - .78 .64 - .78 Site attitude .62 - .73 .68 - .77 Channel involvement .39 - .44 .48 - .52 Notes: All correlations are significant at the .01 level (2-tailed)

Appendices

129

Appendix III. Full Conditional Posterior Distributions

In this appendix we specify the full conditional posterior distributions of the parameters of interest in our multivariate Tobit model. Individuals are indexed by i , and they each make iT purchases.

Yes/no decision Sampling of α To obtain the full conditional posterior distribution of α we rewrite (5) from Section 3.3.1 as

(1) 1 1 1 1

*2 2 2 2= ,it it i it itZ H Hα α ε− − − −

Σ − Σ Σ + Σ

where 1= ( , , )'it i t iCtH H H… , for individuals =1, ,i I… and purchase

occasions =1, , it T… . We can interpret (15) as C regression equations

with regression coefficient α and uncorrelated normal distributed error terms with unit variance. Hence, the full conditional posterior

distribution of α given *Z and Σ , is normal. The mean and variance result from the OLS estimator of α in (1), see Zellner (1971 Chapter VIII).

Sampling of iα To sample iα we can follow a similar approach as for α . We rewrite (5)

from section 3.3.1 as

(4) 1 1 1 1

*2 2 2 2=it it it i itZ H Hα α ε− − − −

Σ − Σ Σ + Σ ,

for =1, ,j C… , =1, ,i I… and =1, , it T… . This represents =1TitC∑

regression equations with regression coefficient iα and uncorrelated

normal distributed error terms with unit variance. Hence, the full

conditional posterior distribution of iα given α , αΣ , Σ , and *Z is

normal. The mean and variance result from the OLS estimator of iα in

(2).


130

Sampling of αΣ For αΣ it holds that

(5) 11( | ) exp( ),2

'i i ip α αα α α−Σ ∝ − Σ

hence αΣ can be sampled from an inverted Wishart distribution, see

Zellner (1971 Chapter VIII).

Sampling of Σ To sample Σ we note that

(6)

12 1=1

=1 =2

1( | , ) ( ) =| | exp( ).2

ITi I i 'i

it iti t

Tp Zα π ε ε

−∗ −

∑Σ ∝ Σ Σ − Σ∑ ∑

where,

(7) *= ( ) for =1, , ,it it it i iZ V t Tε α α− + …

for =1, ,i I… . As Σ is not a free covariance matrix (the diagonal elements are 1), the full conditional distribution is not inverted Wishart. In fact, the full conditional posterior distribution of Σ is not standard. To sample Σ we propose a sampler based on Basag and Green (1993) and Damien, Wakefield and Walker (1999). Loosely speaking, this sampler interchanges the two steps in the Metropolis-Hastings sampler. A possible Metropolis-Hastings sampler for Σ is: • Step 1. Draw the elements of the matrix Σ from a uniform

distribution on the interval ( 1,1)− under the restriction of positive

definiteness, resulting in newΣ . • Step 2. Draw u from a uniform distribution on the interval (0,1)

and accept newΣ if new old( )/ ( ) > uπ πΣ Σ otherwise take new old=Σ Σ .

For the sampler used here we interchange these two steps. We first draw u from a uniform distribution on the interval (0,1) . In the second

step we keep sampling candidate draws of the elements of Σ from a

uniform distribution on the interval ( 1,1)− until newΣ is positive definite

Appendices

131

and new old( )/ ( ) > uπ πΣ Σ . The advantage of the latter approach is that it

always results in a new draw, which is not the case for the Metropolis-Hastings sampler, see Damien et al. (1999) for details. The disadvantage is that the sampler is slower as one has to draw new candidates until acceptance. Another possibility to generate Σ based on the Metropolis-Hastings sampler is given in Chib and Greenberg (1998) or the hit-and-run algorithm in Manchanda et al. (1999).

Sampling of *Z

To sample *itZ , =1, ,i I… , =1, , ,it T… we consider

(8) * = ( ) ,it it i itZ H α α η+ +

hence *itZ is normal distributed with mean ( )it iH α α+ and variance Σ .

The full conditional posterior distributions of the elements of *itZ are of

course also normal. Hence, *ijtZ can be sampled from truncated normal

distributions in the following way

(9) ( )( )

* * ijt, ,

ijt

normal on ,0 if Z| normal on 0, if Z 1ijt i j t

oZ Z −

⎧ −∞ =⎨ ∞ =⎩

∼ ,

where * *, , = ( for all )i j t iktZ Z k j− ≠ , see Geweke (1991) for details.

Purchase amount decision Sampling of β

To obtain the full conditional posterior distribution of β we rewrite (7)

from Section 3.3.1 as

(10) 1 1 1 1

*2 2 2 2= ,it it i it itY G Gβ β η− − − −

Ω − Ω Ω + Ω

where 1= ( , , )'it i t iCtG G G… , for =1, ,i I… , =1, , it T… . We can interpret (8)

as C regression equations with regression coefficient β and

uncorrelated normal distributed error terms with unit variance. Hence,

the full conditional posterior distribution of β given iβ , *Y and Ω , is


132

normal. The mean and variance result from the OLS estimator of β in

(22), see Zellner (1971 Chapter VIII).

Sampling of iβ

To sample iβ we can follow a similar approach as for iα . We rewrite (7)

from Section 3.3.1 as

(11) 1 1 1 1

*2 2 2 2=it it it i itY G Gβ β η− − − −

Ω − Ω Ω + Ω ,

for =1, ,j C… , =1, ,i I… and =1, , it T… . This represents =1TitC∑

regression equations with regression coefficient iβ and uncorrelated

normal distributed error terms with unit variance. Hence, the full

conditional posterior distribution of iβ given β , βΩ , Ω , and *Y is

normal. The mean and variance result from the OLS estimator of iβ in

(9).

Sampling of βΣ

For βΣ it holds that

(12) 11( | ) exp( ),2

'i i ip β ββ β β−Σ ∝ − Σ

hence βΣ can be sampled from an inverted Wishart distribution, see

Zellner (1971 Chapter VIII). In the application, for identification purposes, βΣ is set to identity.

Sampling of Ω The covariance matrix Ω is drawn from an inverted Wishart

distribution with =1= Iii Tν ∑ degrees of freedom.

Sampling of *Y

*ijtY is equal to the observed ijtY if > 0ijtY , otherwise we sample *

ijtY

from a normal distribution, truncated above at 0. In this case, consider

Appendices

133

(13) * = ( ) ,it it i itY G β β η+ +

This shows that *ijtY is distributed normal with mean ( )it iG β β+ and

variance Ω . The full conditional posterior distributions of the elements

of *itY are of course also normal. Hence, *

ijtY can be sampled from

(truncated) normal distributions in the following way

(14) ( )* *, ,

draw normal on ,0 if 0| = if 0

ijtijt i j t

ijt ijt

YY Y Y Y−⎧ −∞ =⎨ >⎩

,

where * *, , = ( for all )i j t iktY Y k j− ≠ , again see Geweke (1991) for details.


134

Appendix IV. Descriptives Chapter 3

DESCRIPTIVE STATISTICS ENTIRE DATASET AND ESTIMATION SAMPLE

Before site implementation After site implementation Entire dataset Mean St.

dev. Range Mean St.

dev. Range

# Shopping Trips 2.6 1.8 1 - 13 2.1 1.8 0 – 16

€ Ladies Fashion 15.8 28.2 0 -286 13.2 28.5 0 – 365

€ Men’s Fashion 8.0 26.0 0 – 332 7.4 24.9 0 – 329

€ Children 11.2 28.1 0 – 256 8.5 29.4 0 – 897

€ Accessories 9.7 19.9 0 – 300 9.3 23.3 0 – 352

€ Interior Design 10.0 35.0 0 – 403 8.4 38.5 0 – 1144

€ Sports 4.9 15.6 0 – 190 5.0 31.5 0 – 1300

# Pages online 0.0 0.0 0 – 0 2.5 9.7 0 – 216

# Site visits 0.0 0.0 0 – 0 0.2 0.6 0 – 14

Estimation sample Mean St.

dev. Range Mean St.

dev. Range

# Shopping Trips 2.4 1.9 0 – 15 1.8 1.7 0 – 14

€ Ladies Fashion 13.8 30.6 0 – 330 14.1 44.8 0 – 1096

€ Men’s Fashion 6.7 22.5 0 – 276 6.3 24.1 0 – 293

€ Children's 9.0 22.8 0 – 268 8.6 26.7 0 – 379

€ Accessories 8.9 19.2 0 – 153 7.0 18.4 0 – 176

€ Interior Design 10.4 66.3 0 – 1547 7.9 31.7 0 – 624

€ Sports 5.8 20.4 0 – 280 5.6 32.2 0 – 936

# Pages online 0.0 0.0 0 – 0 6.7 26.2 0 – 696

# Site visits 0.0 0.0 0 – 0 0.4 0.8 0 – 11

Appendices

135

Appendix V. T-Test Comparison

COMPARISON NUMBER OF SHOPPING TRIPS OF VISITORS & NON-VISITORS

Sample size Average shopping trips Month Non-visitors Visitors Non-visitors Visitors T-value

1 438 2867 2.57 2.47 0.95

2 392 2589 2.49 2.47 0.23

3 466 3447 2.74 2.64 0.95

4 413 2829 2.27 2.39 -1.19

5 491 3116 2.45 2.52 -0.65

6 443 3092 2.63 2.45 1.81

7 428 3051 2.54 2.44 1.03

8 355 2664 2.45 2.44 0.05

9 648 3819 2.52 2.6 -1.00

10 529 3749 2.37 2.15 2.51 11 461 3911 2.73 2.4 3.26 12 532 4221 2.86 2.89 -0.32

13 397 2961 2.61 2.52 0.78

14 354 2688 2.34 2.37 -0.22

15 471 5418 2.6 1.73 8.00 16 380 3738 2.31 1.79 5.59 17 460 4229 2.41 1.92 4.95 18 402 4149 2.44 1.85 6.28 19 372 3368 2.35 1.99 3.70 20 307 3263 2.27 1.87 3.75 21 461 4304 2.51 2.27 2.53 22 369 3553 2.4 1.93 4.84 23 416 3915 2.48 2.08 3.89 24 520 4377 2.8 2.64 1.55

25 451 4061 2.04 1.89 1.78

26 433 3651 1.81 1.79 0.18

27 523 4712 2.03 2.08 -0.57

28 444 3694 1.81 1.82 -0.12

29 483 4262 1.97 1.93 0.55 Notes: Bold t-values are significant at the 5%- level.


136

Appendix VI. MVP Model Selection

FIT CRITERIA MODEL SELECTION

Model

1 Model

2 Model

3 Model

4 Model

5 Model

6 Hit rate Stage 1 Ladies 0.71 0.65 0.71 0.71 0.71 0.71 Men’s 0.85 0.84 0.85 0.85 0.85 0.85 Children 0.81 0.76 0.82 0.82 0.82 0.82 Accessories 0.71 0.63 0.71 0.71 0.71 0.71 Living 0.78 0.78 0.78 0.78 0.78 0.78 Sports 0.87 0.86 0.87 0.87 0.87 0.87 RMSE Stage 2 Ladies 20.77 22.97 21.01 20.84 21.04 20.82 Men’s 18.88 19.50 18.73 18.92 18.73 18.94 Children 16.81 18.82 17.13 17.03 17.12 17.03 Accessories 15.93 17.69 16.65 15.93 16.66 15.94 Living 21.50 21.90 21.52 21.48 21.52 21.49 Sports 12.16 12.19 12.21 12.16 12.30 12.17 MAPE Stage 2 Ladies 0.72 0.82 0.73 0.72 0.73 0.72 Men’s 0.84 0.88 0.85 0.84 0.85 0.84 Children 0.72 0.84 0.73 0.72 0.74 0.72 Accessories 0.75 0.76 0.77 0.76 0.77 0.76 Living 0.78 0.80 0.78 0.78 0.78 0.78 Sports 0.86 0.86 0.87 0.86 0.87 0.86

Appendices

137

RESULTS CHOW TEST FINAL MVP MODEL SELECTION

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

LRSS 9672163 11016439 9875302 9727224 9881545 9731082

Df 120 20 30 40 40 50

Df unused 2867 2967 2957 2947 2947 2937 Chow test results

1 2 3 4 5 6

1 3.98 0.67 0.20 0.78 0.25

2

3 34.17

4 19.53 4.49

5 16.92 0.19

6 12.93 2.18 0.12 4.54 Critical F-values

1 2 3 4 5 6

1 1.69 1.69 1.69 1.87 1.69

2

3 2.99

4 2.30 2.99

5 2.30 2.99

6 2.02 2.30 2.99 2.99


138

Appendix VII. Post-Hoc Comparison

COMPARISON OF CUSTOMERS WITH POSITIVE AND NEGATIVE EFFECTS OF VISITING THE WEB SITE

Notes: Bold test values are significant at the 5%. Test values with an * are concern a Pearson Chi-square. All other test values concern a T-test.

Customers with

positive site use effect

negative site use effect Test value

Individual level

Age 39.75 38.98 -1.166

Number of children 1.3 1.2 -1.102

Number of adults 2.3 2.2 -1.101

At most high school education 75.0% 25.0% 4.636*

College education 70.6% 29.4% Distance to closest store 5.62 6.56 2.643

Gender: male 27.6% 72.3% 1.819*

Gender: female 29.1% 70.9%

Zip code area

Households social class A 18.37% 17.43% -1.836

Single households 31.31% 29.23% -2.576 Buying through a catalog 57.50% 58.26% 2.048

Behavior

Number of shopping trips 2.17 1.78 -4.968 Total money spent 67.55 52.42 -5.156

Total products bought 9.29 7.19 -4.503 Money spent in Ladies 13.22 11.19 -2.386 Money spent in Men’s 7.22 5.02 -3.463

Money spent in Children 8.08 6.18 -2.741 Money spent in Accessories 6.05 7.86 -3.501

Money spent in Living 9.94 7.46 -1.927

Money spent in Sport 5.39 4.92 -0.891 Number of site visits .24 .22 -0.686

Number of Web site pages 4.52 3.45 -2.084 N 7634.0 951.0

Appendices

139

Appendix VIII. Description Variables Chapter 4

t

t

MP

= the mean monetary value spent per product a in period t. The

series indicates if customers over time spend more, less or roughly the same amount of money on average per product.

t

t

PTr

= the mean number of products purchased per shopping trip in

period t (basket size). The series indicates if customers over time change the number of products they buy on average per trip per week.

t

t

TrC

= the mean number of shopping trips per customer in period t (store

traffic). The series indicates if customers change their number of shopping trips in a week.

tC = total number of customers in period t (total store traffic). The series

shows how many unique customers are visiting the store each week.

t

t

TiPa

= the mean amount of time per page in period t (depth of search).

The series indicates the mean intensity or per page duration that customers spend online.

t

t

PaVs

= the mean number of pages seen per visit in period t (width of

search). The series shows how the depth of online search changes. That is, do customers on average increase or decrease the number of pages they visit during a Web visit.

t

t

VsVrs

= the mean number of online visits per visitor in period t (site

traffic). The series shows how the frequency of the online visits changes over time.

tVrs = total number of Web visitors in period t (total site traffic). The

series shows how many unique customers are visiting the Web site each week.


140

Appendix IX. Moderation Variables

OVERVIEW MEDIAN SPLIT VARIABLES

Items/description Reliability Flow During my visit, I often forget my immediate

surroundings. During my visit, I often do not realize the duration of my Web visit. During my visit, I lose self-consciousness. During my visit, time seems to fly by.

0.93 in 2001 0.94 in 2002

Site visits Number of Web visits made by the customer during the period of data collection (March 2001- May 2002)

DESCRIPTIVES MEDIAN SPLIT VARIABLES

Mean S.D. Median Mode Flow 2.52 0.96 2.50 3 Site visits 3.23 4.53 2 1

Appendices

141

Appendix X. Moderation Results PRODUCT TYPE: EFFECTS ONLINE MARKETING ON OFFLINE BUYING

Nonsensory products Money Products Trips Customers Online promotions -0.18 0.05 0.03 24.92 Online communications -0.59 -0.09 0.03 -56.28 Site introduction 8.38 -0.22 0.14 675.85

Sensory products Money Products Trips Customers Online promotions 0.15 0.00 0.26 168.44 Online communications 0.16 0.06 0.05 -51.86 Site introduction 0.78 0.21 0.67 958.66

Notes: Bold parameter estimates are significant at the 5% level.

FLOW: EFFECTS ONLINE MARKETING ON OFFLINE BUYING

Low experience of flow Money Products Trips Customers Online promotions 0.71 0.28 0.01 48.25 Online communications -1.04 -0.15 -0.02 -67.36 Site introduction 3.79 1.15 -0.08 337.41

High experience of flow Money Products Trips Customers Online promotions 0.61 0.38 -0.02 42.99 Online communications -0.29 -0.11 0.02 -28.48 Site introduction 2.70 1.55 -0.03 564.14


FREQUENCY SITE VISITS: EFFECTS ONLINE MARKETING ON OFFLINE BUYING

Low frequency of visits Money Products Trips Customers Online promotions 0.14 0.28 -0.01 46.29 Online communications -0.04 0.03 -0.01 -32.30 Site introduction 2.40 1.34 -0.04 519.13

High frequency of visits Money Products Trips Customers Online promotions 0.38 0.52 0.00 54.85 Online communications 0.80 0.28 0.01 -24.69 Site introduction 6.35 1.49 -0.05 375.22


143

References

A Alba, J. & J. Lynch. 1997. Interactive Home Shopping: Consumer, Retailer, and

Manufacturer Incentives to Participate in Electronic Marketplaces. Journal of Marketing. 61(3) 38-54.

Allenby, Greg M. & Peter E. Rossi. 1999. Marketing models of consumer heterogeneity. Journal of Econometrics. 89 57—78.

Alpar, P., M. Porembski & S. Pickerodt. 2001. Measuring the Efficiency of Web Site Traffic Generation. International Journal of Electronic Commerce. 6(1) 53-74.

Amemiya, T. 1985. Advanced Econometrics. Oxford, UK: B Blackwell. Anderson, J.C. & D.W. Gerbing. 1988. Structural Equation Modeling in

Practice: A Review and Recommended Two-Step Approach. Psychological Bulletin. 103(3) 411-423.

Andrews, Rick L. & Imran S. Currim. 2004. Behavioral Differences between Consumers Attracted to Shopping Online versus Traditional Supermarkets: Implications for Enterprise Design and Marketing Strategy. International Journal Internet Marketing and Advertising. 1 38-61.

Ansari, Asim, Carl Mela & Scott Neslin. 2006. Customer Channel Migration. Working Paper Teradata Center. Duke University. Durham NC. Paper Series No.13. Conditionally accepted by Journal of Marketing Research.

B Bahn, David L. & Patrick P. Fisher. 2003. Clicks and Mortar: Balancing Brick

and Mortar Business Strategy and Operations with Auxiliary Electronic Commerce. Information Technology and Management. 4 319-334.

Baker, Julia, A. Parasuraman & Glenn B. Voss. 2002. The Influence of Multiple Store Environment Cues on Perceived Merchandise Value and Patronage Intentions. Journal of Marketing. 66(April) 120-141.

Bakos, Yannis & Erik Brynjolfsson. 2000. Bundling and Competition on the Internet. Marketing Science. 19(1) 63-82.

Balasubramanian, Sridhar, Rajagopal Raghunathan & Vijay Mahajan. 2005. Consumers in a Multichannel Environment: Product Utility, Process Utility and Channel Choice. Journal of Interactive Marketing. 19(2) 12 – 29.

Basag, J. & P.J. Green. 1993. Spatial Statistics and Bayesian Computation. Journal of the Royal Statistical Society. 55 25-37.

Beales, Howard, Michael B. Mazis, Steven C. Salop, Richard Staelin. 1981, Consumer Search and Public Policy. Journal of Consumer Research. 8(June) 11-22.


144

Belk, Russel W. 1975. Situational Variables and Consumer Behavior. Journal of Consumer Behavior. 2 157-164.

Bell, David R., Jeongwen Chiang & V. Padmanabhan. 1999. The Decomposition of Promotional Response: An Empirical Generalization. Marketing Science. 18(4) 504-526.

Bendoly, Elliot, James D. Blocher, Kurt M. Bretthauer, Shankar Krishnan & M.A. Venkataramanan. 2005. Online/In-Store Integration and Customer Retention. Journal of Service Research. 7(4) 313-327.

Biyalogorsky, Eyal & Prasad Naik. 2003. Clicks and Mortar: The Effect of Online Activities on Offline Sales. Marketing Letters. 14 21-32.

Blattberg, Robert C., Richard Briesch & Edward J. Fox. 1995. How Promotions Work. Marketing Science. 14(3) G122 – G131.

Broekhuizen, Thijs L.J.. 2006. Understanding Channel Purchase Intentions: Measuring Online and Offline Value Perceptions. Ph.D. Dissertation University of Groningen. Labyrint Publication, Netherlands.

Burke, Raymond R. 2002. Technology and the Customer Interface: What Consumers Want in the Physical and Virtual Store. Journal of the Academy of Marketing Science. 30(4) 411-432.

Byrne, Barbara M. 1998. Structural Equation Modeling with LISREL. PRELIS and SIMPLIS: Basic concepts, Applications and Programming. Lawrence Erlbaum Associates Inc. New Yersey. United States of America.

C Carroll, Conor. 2002. Web Based Marketing Communications - A Content

Analysis of the Top 500 Irish Companies Corporate Websites. Presented paper at European Marketing Academy Conference. University of Minho, Braga, Portugal. May 29, Book of Abstracts p 76. Full paper on CD Rom. ISBN 972-8755-00-7.

Casella, G. & E.I. George. 1992. Explaining the Gibbs Sampler. American Statistician. 46 167-174.

CBS. 2006. De Digitale Economie. Centraal Bureau voor de Statistiek (Statistics Netherlands). Voorburg/Heerlen. The Netherlands. www.cbs.nl.

Chandon, Pierre, Vicki G. Morwitz & Werner J. Reinartz. 2005. Do Intentions Really Predict Behavior? Self-Generated Validity Effects in Survey Research. Journal of Marketing. 69(April) 1-14.

Chandon, Pierre, Brian Wansink & Gilles Laurent. 2000. A Benefit Congruency Framework of Sales Promotion Effectiveness. Journal of Marketing. 64(October) 65-81.

Chen Q. & W.D. Wells. 1999. Attitude toward the Site. Journal of Advertising Research. September - October 27-37.

Chib, Siddhartha. 1992. Bayes inference in the Tobit censored regression model. Journal of Econometrics. 51(1-2) 79-99.

http://www.cbs.nl/

References

145

Chib, Siddhartha & Edward Greenberg. 1998. Analysis of Multivariate Probit Models. Biometrika.

Churchill, Gilbert A. Jr. 1979. A Paradigm for Developing Better Measures of Marketing Constructs. Journal of Marketing Research. 16(February) 64-73.

Citrin, Alka Varma, Donald E. Stem, Eric R. Spangenberg & Michael J. Clark. 2003. Consumer Need for Tactile Input. An Internet Retailing Challenge. Journal of Business Research. 56 915-922.

Coelho, Filipe, Chris Easingwood & Arnaldo Coelho. 2003. Exploratory Evidence of Channel Performance in Single vs. Multiple Channel Strategies. International Journal of Retail and Distribution Management. 31(11) 561-573.

CrossMedia Services. 2006. Outlook for Web-to-Store Shopping Seen as Stronger than Online E-Commerce. Available at http://www.crossmediaservices.com/pr_17.html.

D Dabholkar, Pratibha A. & Richard P. Bagozzi. 2002. An Attitudinal Model of

Technology-Based Self-Service: Moderating Effects of Consumer Traits and Situational Factors. Journal of the Academy of Marketing Science. 30(3) 184-201.

Damien, Paul, Jon Wakefield & Stephen Walker. 1999. Gibbs Sampling for Bayesian Non-Conjugate and Hierarchical Models by using Auxiliary Variables. Journal of the Royal Statistical Society. 61 331-344.

Danaher, Peter J., Isaac W. Wilson & Robert A. Davis. 2003. A Comparison of Online and Offline Consumer Brand Loyalty. Marketing Science. 22 461-476.

Dann, Susan & Stephen Dann. 2001. Strategic Internet Marketing. John Wiley & Sons Milton Australia, Ltd.

Davis, Fred D. 1989. Perceived Usefulness, Perceived Ease of Use and User Acceptance of Information Technology. MIS Quarterly. September 318-340.

Degeratu, A.M., A. Rangaswamy & J. Wu. 2000. Consumer Choice Behavior in Online and Traditional Supermarkets: The Effect of Brand Name, Price, and Other Search Attributes. International Journal of Research in Marketing. 17(1) 55-78.

Dekimpe, Marnik G. & Dominique M. Hanssens. 1999. Sustained Spending and Persistent Response: A New Look at Long-term Marketing Profitability. Journal of Marketing Research. 36(November) 397-412.

Deleersnyder, Barbara, Inge Geyskens, Katrijn Gielens & Marnik Dekimpe. 2002. How Cannibalistic is the Internet Channel?. International Journal of Research in Marketing. 19(4) 337-348.

Dholakia, Ruby Roy, Miao Zhao & Nikhilesh Dholakia. 2005. Multichannel Retailing: A Case Study of Early Experiences. Journal of Interactive Marketing. 19(2) 62 - 74

http://www.crossmediaservices.com/pr_17.html


146

Donovan, Robert J., John R. Rossiter, Gilian Marcoolyn & Andrew Nesdale. 1994. Store Atmosphere and Purchasing Behavior. Journal of Retailing. 70(3) 283-294.

DoubleClick. 2004. Retail Details: Best Practices in Multi-Channel Integration. March. New York: DoubleClick. Inc.

Ducoffe, Robert H. 1996. Advertising Value and Advertising on the Web. Journal of Advertising Research. September – October 21-35.

Dutta, Soumitra & Arie Segev. 1999. Business Transformation on the Internet. European Management Journal. 17(5) 466-476.

E Ehrenberg, A.S.C. 1959. The Pattern of Consumer Purchases. Applied Statistics.

8 26-41. Enders, Walter. 1995. Applied Econometric Time Series. John Wiley & Sons,

New York, USA. Engel, James F., Roger D. Blackwell & Paul W. Miniard. 1993. Consumer

Behavior. The Dryden Press. 7th International Edition. Orlando, USA.

F Fornell, Claes & David F. Larcker. 1981. Evaluating Structural Equation

Models with Unobservable Variables and Measurement Error. Journal of Marketing Research. 18(February) 39-50.

Forrester Research. 2005. Forrester Research US eCommerce Forecast: Online Retail Sales To Reach $329 Billion By 2010. Available at http://www.forrester.com/ER/Press/Release/0,1769, 1033 ,00.html

Forsythe, Sandra M. & Bo Shi. 2003. Consumer Patronage and Risk Perceptions in Internet Shopping. Journal of Business Research. 56 867-875.

Fox, E.J., A.L. Montgomery & L.M. Lodish. 2004. Consumer Shopping and Spending across Retail Formats. Journal of Business. 77(2) 25-60.

Fox, Edward J. & Stephen J. Hoch. 2005. Cherry-Picking. Journal of Marketing. 69(January) 46-62.

Franses, P.H. & R. Paap. 2001. Quantitative Models in Marketing Research. Cambridge, UK: Cambridge University Press.

G Gallagher, K., J. Parsons & K.D. Foster. 2001. A Tale of Two Studies:

Replicating Advertising Effectiveness and Content Evaluation in Print and on the Web. Journal of Advertising Research. (July – August) 71-81.

Gallagher, K.D. Foster & J. Parsons. 2001. The Medium is Not the Message: Advertising Effectiveness and Content Evaluation in Print and on the Web. Journal of Advertising Research. (July – August) 57-70.

References

147

Geman, S. & D. Geman. 1984. Stochastic Relaxations, Gibbs Distributions, and the Bayesian Restoration of Images. IEEE Transaction on Pattern Analysis and Machine Intelligence. 6 721-741.

Gensler, Sonja, Marnik G. Dekimpe & Bernd Skiera. 2007. Evaluating Channel Performance in Multi-Channel Environments. Journal of Retailing and Consumer Services. 14(1) 17-23.

Geweke, John F. 1991. Efficient Simulation from the Multivariate Normal and Student-t Distributions Subject to Linear Constraints. Computer Science and Statistics: Proceedings of the 23rd Symposium on Interface. American Statistical Association, Alexandria, VA. 571-578.

Geyskens, Inge, Katrijn Gielens & Marnik G. Dekimpe. 2002. The Market Valuation of Internet Channel Additions. Journal of Marketing. 66 102-119.

Greene, William. 2003. Econometric Analysis. New Yersey, USA: Pearson Education International, Fifth Edition.

Gulati, Ranjay & Jason Garino. 2000. Get the right Mix of Bricks and Clicks. Harvard Business Review. May-June 107-114.

Gupta, Alok, Bo-chiuan Su & Zhiping Walter. 2004. An Empirical Study of Consumer Switching from Traditional to Electronic Channels: A Purchase-Decision Process Perspective. International Journal of Electronic Commerce. 8(3) 131-161.

Gupta, Sunil. 1988. Impact of Sales Promotions on When, What, and How Much to Buy. Journal of Marketing Research. 25(4) 342-355.

H Hair, Joseph F., Rolph E. Anderson, Ronald L. Tatham & William C. Black.

1998. Multivariate Data Analysis. 5th edition. Prentice Hall. New Yersey. United States of America.

Hoffman, Donna L., Thomas P. Novak, & P. Chatterjee. 1995. Commercial Scenarios for the Web: Opportunities and Challenges. Journal of Computer-Mediated Communications. 1(3) 1-20.

Hoffman, Donna L., Thomas P. Novak, & Marcos Peralta. 1999. Building Consumer Trust Online. Communications of the ACM. 42(4) 80-85.

Hoffman, Donna L. & Thomas P. Novak. 1996. Marketing in Hypermedia Computer-Mediated Environments: Conceptual Foundations. Journal of Marketing. 60 50-68.

Hoque, A. Y. & Gerald L. Lohse. 1999. An Information Search Cost Perspective for Designing Interfaces for Electronic Commerce. Journal of Marketing Research. 36(August) 387-394.

Huff, D. 1964. Defining and Estimating a Trade Area. Journal of Marketing. 28 34-8

Huizingh, Eelko K. R. E. 2000. The Content and Design of Web Sites: An Empirical Study. Information & Management. 37 123-134.


148

I Inman, Jeffery J., Venkatesh Shankar & Rosellina Ferraro. 2004. The Roles of

Channel-Category Associations and Geodemographics in Channel Patronage Decision. Journal of Marketing. 68 (April) 51-71.

J Järveläinen, Jonna & Jussi Puhakainen. 2004. Distrust of One's Own Web

Skills: A Reason for Offline Booking after Online Information Search. Electronic Markets. 14(4) 333-343.

Johnson, Eric J., Wendy Moe, Peter S. Fader, Steven Bellman & Gerald L. Lohse. 2004. On the Depth and Dynamic of Online Search Behavior. Management Science. 50(March) 299-308.

Jöreskog, Karl G. & Dag Sörbom. 1993. LISREL 8: Structural Equation Modeling with the SIMPLIS Command Language. Scientific Software International. Chicago. United States of America.

K Kaul, Anil & Dick Wittink. 1995. Empirical Generalizations about the Impact of

Advertising on Price Sensitivity and Price. Marketing Science Special Issue on Empirical Generalizations in Marketing. 14(3) (part 2) 151-160.

Keen, Cherie, Martin Wetzels, Ko De Ruyter & Richard Feinberg. 2004. E-Tailers versus Retailers. Which Factors Determine Consumers Preferences. Journal of Business Research. 57 685-95.

Kelley, Christopher M., Kate Delhagen & Alanna Denton. 2002. Capturing Cross-Channel Dollars. Forrester Research. May.

Kelly, J. Patrick & Scott M. Smith. 1983. Determinants of Retail Patronage: An Examination of Alternative Models. Advances in Consumer Research. 10(1). 345-351.

Knox, George. 2005. Modeling and Managing Customers in a Multichannel Setting. Extended Doctoral Proposal. Marketing Department Wharton School. University of Pennsylvania. Philadelphia, USA.

Kornelis, Marcel. 2002. Modeling Market Shake-Ups Using Time Series Data. An Application to the Advertising Market of the Netherlands. Ph.D. Dissertation University of Groningen. Labyrint Publication, Netherlands.

Kumar V. & Rajkumar Venkatesan. 2005. Who Are Multichannel Shoppers and How Do they Perform?: Correlates of Multichannel Shopping Behavior. Journal of Interactive Marketing. 19(Spring) 44-61.

Kushwaha, Tarun L. & Venkatesh Shankar. 2006. Optimal Multichannel Allocation of Marketing Efforts. Working Paper. Texas A&M University. College Station. TX 7785.

References

149

L Lam, Shun Yin, Mark Vandenbosch, John Hulland & Michael Pearce. 2001.

Evaluating Promotions in Shopping Environments: Decomposing Sales Response into Attraction, Conversion, and Spending Effects. Marketing Science. 20(2) 194-216.

Lattin, J., J.D. Carroll & P.E. Green. 2003. Analyzing Multivariate Data. Thomson Learning Inc, Australia.

Lee, Ruby P. & Rajdeep Grewal. 2004. Strategic Responses to New Technologies and Their Impact on Firm Performance. Journal of Marketing. 68(October) 157-171.

Leeflang, Peter S.H., Dick R. Wittink, Michel Wedel & Philippe A. Naert (2000), Building Models for Marketing Decisions. Dordrecht, The Netherlands: Kluwer Academic Publishers.

Lim, J., I.S. Currim & R.L. Andrews. 2005. Consumer Heterogeneity in the Longer-Term Effects of Price Promotions. International Journal of Research in Marketing. 22(4) 441-457.

M Macintosh, Gerrard & Lawrence S. Lockshin. 1997. Retail Relationships, Store

Loyalty: A Multi-Level Perspective. International Journal of Research in Marketing. 14 487-497.

Madalla, G.S. & I.M. Kim. 1996. Unit Roots, Cointegration, and Structural Change. Cambridge University Press, Cambridge.

Maddox, R. Neil. 1982. The Structure of Consumers' Satisfaction: Cross-product Comparisons. Journal of the Academy of Marketing Science. 10(1). 37-17.

Manchanda, P., A. Ansari & S. Gupta. 1999. The Shopping Basket: A Model for Multicategory Purchase Incidence Decisions. Marketing Science. 18(02) 95-115.

Mathwick, Charla, Naresh K. Malhotra & Edward Rigdon. 2002. The Effect of Dynamic Retail Experiences on Experiential Perceptions of Value: An Internet and Catalog Comparison. Journal of Retailing. 78 51-60.

Mathwick, Charla & Edward Rigdon. 2004. Play, Flow and the Online Search Experience. Journal of Consumer Research. 31(September), 324-32.

Mendelsohn, Tamara, Carrie A. Johnson & Sean Meyer. 2006. Understanding US Cross-Channel Shoppers. Forrester Research. April 19.

Meuter, Matthew L., Amy L. Ostrom, Mary Jo Bitner & Robert I. Roundtree. 2003. The Influence of Technology Anxiety on Consumer Use and Experiences with Self-Service Technologies. Journal of Business Research. 56 899-906.

Mick, David Glen & Susan Fournier. 1998. Paradoxes of Technology: Consumer Cognizance, Emotions and Coping Strategies. The Journal of Consumer Research. 25(2) 123-143.


150

Mittal, Vikas & Wagner A. Kamakura. 2001. Satisfaction. Repurchase Intent, Repurchase Behavior: Investigating the Moderating Effect of Customer Characteristics. Journal of Marketing Research. 38(1) 131-142.

Moe, Wendy W. & Peter S. Fader. 2004. Capturing evolving visit behavior in clickstream data. Journal of Interactive Marketing. 18(1), 5-19.

Montoya-Weiss, Mitzi M., Glenn B. Voss & Dhruv Grewal. 2003. Determinants of Online Channel Use and Overall Satisfaction with a Relational Multichannel Service Provider. Journal of the Academy of Marketing Science. 31(4) 448-458.

Morgan, C. Adapted from Sandor, G. 1994. Attitude (Not Age) Defines the Mature Market. American Demographics. 18-21.

Morrison, Donald G. 1969. On the Interpretation of Discriminant Analysis. Journal of Marketing Research. 6(May) 156-163.

Muniz, Albert M. Jr. & Thomas C. O’Guinn. 2001. Brand Community. Journal of Consumer Research. 27(March) 412-432.

N Neslin, Scott A., Dhruv Grewal, Robert Leghorn, Venkatesh Shankar, Marije L.

Teerling, Jacquelyn S. Thomas & Peter C. Verhoef. 2006. Challenges and Opportunities in Multichannel Customer Management. Journal of Services Research Special Issue on Customer Relationship Management. 9(2) 1-18.

Newman, Joseph W. & Richard Staelin. 1972. Prepurchase Information Seeking for New Cars and Major Household Appliances. Journal of Marketing Research. 9(August) 249-257.

Nicholson, Michael, Ian Clarke, & Michael Blakemore. 2002. One brand, three ways to shop: situational variables and multichannel consumer behavior. The International Review of Retail, Distribution and Consumer Research. 12(2) 131-148.

Nijs, Vincent R., Marnik G. Dekimpe, Jan Benedict E.M. Steenkamp & Dominique M. Hanssens. 2001. The Category Demand Effects of Price Promotions. Marketing Science. 20 1-22.

Noble, Stephanie M., David A. Griffith, Marc G. Weinberger. 2005. Consumer Derived Utilitarian Value and Channel Utilization in a Multi-Channel Retail Context. Journal of Business Research. 58 1643 – 1651.

Novak, Thomas P., Donna L. Hoffman & Yiu-Fai Yung. 2000. Measuring the Customer Experience in Online Environments: A Structural Modeling Approach. Marketing Science. 19(1) 22-42.

O Oliver, Richard L. 1999. Whence Consumer Loyalty? Journal of Marketing.

63(Special Issue 1999) 33-44.

References

151

Olson, James M. & Mark P. Zanna. 1993. Attitudes and Attitude Change. Annual Review of Psychology. 44 117-154.

Otto, James R. & Q.B. Chung. 2000. A Framework for Cyber-Enhanced Retailing: Integrating E-Commerce Retailing with Brick-and-Mortar Retailing. Electronic Markets. 10 185-191.

P Parasuraman, A. 2000. Technology Readiness Index (TRI) A Multiple-Item Scale

to Measure Readiness to Embrace New Technologies. Journal of Service Research. 2(4) 307-320.

Pauwels, Koen H., Dominique M. Hanssens & S. Siddarth. 2002. The Long-Term Effects of Price Promotions on Category Incidence, Brand Choice and Purchase Quantity, Journal of Marketing Research. 39(November) 421-439.

Pavlou, Paul A. & Mendel Fygenson. 2006. Understanding and Predicting Electronic Commerce Adoption: An Extension of the Theory of Planned Behavior. MIS Quarterly. 30(1) 115-143.

Peck, Joann & Terry L. Childers. 2003. To Have and To Hold: The Influence of Haptic Information on Product Judgements. Journal of Marketing. 67(April) 35-48.

Peterson, R.A., A. Balasubramanian & B.J. Bronnenberg. 1997. Exploring the Implications of the Internet for Consumer Marketing. Journal of the Academy of Marketing Science. 25(4) 329-346.

Podsakoff, Philip M., Nathan P. Podsakoff & Jeong-Yeon Lee. 2003. Common Method Bias in Behavioral Research: A Critical Review of the Literature and Recommended Remedies. Journal of Applied Psychology. 88(5) 879-903.

Porter, M. 2001. Strategy and the Internet. Harvard Business Review. March 63-78.

Punj, Girish N. & Richard Staelin. 1983. A Model of Consumer Information Search Behavior for New Automobiles. Journal of Consumer Research. 9(March) 366- 380.

R Rangaswamy, Arvind & Gerrit H. van Bruggen. 2005. Opportunities and

Challenges in Multichannel Marketing: An Introduction to the Special Issue. Journal of Interactive Marketing. 19(2) 5-11.

Rasbash, J., F. Steele, W. Browne & B. Prosser. 2004. A User’s Guide to MlwiN. Centre for Multilevel Modelling, Institute for Education, University of London.

Rasch, Stefan & Alexander Lintner. 2001. The Multichannel Consumer, the Need to Integrate Online and Offline Channels in Europe. The Boston Consulting Group. Boston, USA. Available via: http://www.bcg.com/publications/files/MultichannelConsumer_ summary.pdf

http://www.bcg.com/publications/files/MultichannelConsumer_ summary.pdf

http://www.bcg.com/publications/files/MultichannelConsumer_ summary.pdf


152

Ratchford, Brian T., Myung-Soo Lee & Debabrata Talukdar. 2003. The Impact of the Internet on Information Search for Automobiles, Journal of Marketing Research. 40(2) 193-209.

Reynolds, Kristy E. & Sharon E. Beatty. 1999. A Relationship Customer Typology. Journal of Retailing. 75(4) 509-523.

Rindfleisch, Aric., Alan J. Malter, Shankar Ganesan & Christine Moorman. 2006. Cross-Sectional Versus Longitudinal Survey Research: Concepts, Findings and Guidelines. Working Paper University of Wisconsin-Madison. Under review at Journal of Marketing Research.

Roy, Subroto & Sanjoy Ghose. 2006. Internet Adoption as a Two-Stage Transition. Converting Internet Non-Users to Internet Users and to Online Buyers. International Journal of Market Research. 48(3) 321-349.

Rust, R.T. & N. Donthu. 1995. Capturing Geographically Localized Misspecification Error in Retail Store Choice Model. Journal of Marketing Research. 32(February) 103-110.

S Schlosser, Anne E., Tiffany Barnett White & Susan M. Lloyd. 2006. Converting

Web Site Visitors into Buyers: How Web Site Investment Increases Consumer Trusting Beliefs and Online Purchase Intentions. Journal of Marketing. 70(April) 133-148.

Schoenbachler, Dennis D. & Geoffrey L. Gordon. 2002. Multi-Channel Shopping: Understanding what Drives Channel Choice. Journal of Consumer Marketing. 19(1) 42-53.

Seiders, Kathleen, Glenn B. Voss, Druh Grewal & Andre L. Godfrey. 2005. Do Satisfied Customer Buy More? Examining Moderating Influences in a Retailing Context. Journal of Marketing. 69(October) 26-43.

Shankar, Venkatesh, Amy K. Smith & Arvind Rangaswamy. 2003. The Relationship between Customer Satisfaction and Loyalty in Online and Offline Environments. International Journal of Research in Marketing. 20(2). 153-175.

Shiv, B. & A. Fedorikhin. 1999. Heart and Mind in Conflict: The Interplay of Affect and Cognition in Consumer Decision Making. Journal of Consumer Research. 12 214-224.

Sims, Christopher A. 1980. Macroeconomics and Reality. Econometrics. 48 1-48.

Sirgy, M. Joseph, Dhruv Grewal & Tamara Mangleburg. 2000. Retail environment. Self-Congruity. and Retail Patronage: An Integrative Model and a Research Agenda. Journal of Business Research. 49 127-138.

Sismeiro, Catarina & Randolph E. Bucklin. 2004. Modeling Purchase Behavior at an E-Commerce Web Site: A Task-Completion Approach. Journal of Marketing Research. 41(3) 306-323.

References

153

Srinivasan, Srini S., Rolph Anderson & Kishore Ponnavolu. 2002. Customer Loyalty in E-commerce: an Exploration of its Antecedents and Consequences. Journal of Retailing. 78 41-50.

Steenkamp, Jan Benedict E.M. & H.C.M. van Trijp. 1991. The Use of LISREL in Validating Marketing Constructs. International Journal of Research in Marketing. 8(4) 283-299.

Steenkamp, Jan-Benedict E. M. & Hans Baumgartner. 2000. On the Use of Structural Equation Models for Marketing Modeling. International Journal of Research in Marketing. 17 195-202.

Stewart, David W. & Paul A. Pavlou. 2002. From Consumer Response to Active Consumer: Measuring the Effectiveness of Interactive Media. Journal of the Academy of Marketing Science. 30(4) 376-396.

Sullivan, Ursula & Jacquelyn S. Thomas. 2004. Customer Migration: An Empirical Investigation Across Multiple Channels. http:/www.business.uiuc/working_papers/papers/04-0112.pdf

T Tang, Christopher S., David R. Bell & Teck-Hua Ho. 2001. Store Choice and

Shopping Behavior: How Price Format Works. California Management Review. 43(2) 56-74.

Tanner, M.A. & W.H. Wong. 1987. The Calculation of Posterior Distributions by Data Augmentation. Journal of the American Statistical Association. 82(398) 528-540.

Taylor, Steven A. & Gary L. Hunter. 2002. The Impact of Loyalty with e-CRM Software and e-Services. International Journal of Service Industry Management. 13(5) 452 – 474.

Tellis, Gerard J. 1988. Advertising Exposure, Loyalty, and Brand Purchase: A Two-Stage Model of Choice. Journal of Marketing Research. 25(May) 134-144.

Tellis, Gerard J. & Philip Hans Franses. 2006. Optimal Data Interval for Estimating Advertising Response. Marketing Science. 25(3) 217-229.

Theo, Thompson, S.H. & Yujun Pian. 2004. A Model for Web Adoption. Information and Management. 41(4) 457-468.

Thomas, Jacquelyn S. & Ursula Sullivan. 2005. Managing Marketing Communications with Multichannel Customers. Journal of Marketing. 69(October) 239-251.

Tierney, L. 1994. Markov Chains for Exploring Posterior Distributions. Annals of Statistics. 22 1701-1762.

U US Census Bureau. 2006. http://www.census.gov/mrts/www/data/html/

06Q3.html


154

V Van Baal, Sebastian & Christain Dach. 2005. Free Riding and Customer

Retention Across Retailers' Channels. Journal of Interactive Marketing. 19(2) 75-85.

Van der Heijden, Hans. 2003. Factors Influencing the Usage of Web sites: the Case of a Generic Portal in The Netherlands. Information and Management. 40(6) 541-549.

Van Heerde, Harald J., Peter S.H. Leeflang & Dick R. Wittink. 2001. The Estimation of Pre- and Postpromotion Dips with Store-Level Scanner Data. Journal of Marketing Research. 37(3) 383-395.

Van Heerde, Harald J., Peter S.H. Leeflang & Dick R. Wittink. 2004. Decomposing the Sales Promotion Bump with Store Data. Marketing Science. 23(3) 317-334.

Van Heerde, Harald J. & Tammo H.A. Bijmolt. 2005. Decomposing the Promotional Revenu Bump for Loyalty Program Members versus Non-Members. Journal of Marketing Research. 42(4) 443-457.

Van Kenhove, Patrick, Kristof De Wulf & Walter van Waterschoot. 1999. The Impact of Task Definition on Store-Attribute Saliences and Store Choice. Journal of Retailing. 75(1) 125-137.

Venkatesh, Viswanath & Fred D. Davis. 2000. A Theoretical Extension of the Technology Acceptance Model: Four Longitudinal Field Studies. Management Science. 46(2) 186-204.

Verhoef, Peter C. & Bas Donkers. 2005. The Effect of Acquisition Channels on Customer Loyalty and Cross-Buying. Journal of Interactive Marketing. 19(2) 31-43.

Verhoef, Peter C., Scott A. Neslin & Bjorn Vroomen. 2007. Multichannel Customer Management: Understanding the Research Shopper Phenomenon. Forthcoming International Journal of Research in Marketing.

Villanueva, Julian, Shijin Yoo & Dominique Hanssens. 2003. The Impact of Acquisition Channels on Customer Equity. IESE Working Paper No. D/516. Available at SSRN: http://ssrn.com/abstract=459200

W Wallace, David W., Joan L. Giese & Jean L. Johnson. 2004. Customer Retailer

Loyalty in the Context of Multiple Channel Strategies. Journal of Retailing. 80 249 - 263.

Wu, Jianan & Arvind Rangaswamy. 2003. A Fuzzy Set of Search and Consideration with an Application to an Online Market. Marketing Science. 22(3) 441-434.

http://www.rug.nl/economie/faculteit/medewerkers/VerhoefPC/Multichannel_IJRM.pdf

http://www.rug.nl/economie/faculteit/medewerkers/VerhoefPC/Multichannel_IJRM.pdf

http://ssrn.com/abstract=459200

References

155

Z Zellner, Arnold. 1971. An Introduction to Bayesian Inference in Econometrics.

Wiley, New York. Zhang, J. & L. Krishnamurthi. 2004. Customizing Promotions in Online Stores:

Methodology and a Comparison to Brick-and-Mortar Stores. Marketing Science. 23(4) 561-578.

157

Author Index

A Alba J., 48, 51, 108, 116 Allenby G.M., 55 Alpar P., 84 Amemiya T., 54 Anderson J.C., 26, 29, 30 Anderson R.E., 24, 26 Andrews R.L., 19, 92 Ansari A., 6, 17, 19, 21, 22, 39, 48, 49, 50,

51, 54, 73, 75, 82, 86, 98, 108, 110, 115, 117, 118, 127

B Bagozzi R.P., 35, 39, 41 Bahn D.L., 19 Baker J., 25 Bakos Y., 50 Balasubramanian A., 2 Balasubramanian S., 22, 44, 73, 128 Barnett White T., 1, 4 Basag J., 132 Baumgartner H., 26 Beales H., 7 Beatty S.E., 28 Belk R.W., 58 Bell D.R., 25, 85 Bellman S., 85 Bendoly E., 6, 19, 20, 23, 34, 127 Bijmolt T.H.A., 85 Bitner M.J., 1, 4 Biyalogorsky E., 17, 19, 48, 49, 50 Black W.C., 26 Blackwell, 24 Blakemore M., 2, 4, 117 Blattberg R.C., 86

Blocher J.D., 6 Bretthauer K.M., 6 Briesch R., 86 Broekhuizen T.L.J., 4 Bronnenberg B.J., 2 Browne W., 56 Brynjolfsson E., 50 Bucklin R.E., 47 Burke R.R., 6, 21, 22, 39, 43, 61, 75, 83,

92, 110, 115, 128 Byrne B.M., 26, 27, 31

C Carroll C., 8, 79 Casella G., 56 CBS, 2 Chandon P., 37, 85, 87 Chatterjee P., 2 Chen Q., 84 Chiang J., 85 Chib S., 54, 133 Childers T.L., 83 Chung Q.B., 19 Churchill G.A. Jr., 29 Citrin A.V., 5, 48, 83, 92 Clark M.J., 5 Clarke I., 2, 4 Coelho A., 50 Coelho F., 49, 50 CrossMedia Services, 80 Currim I.S., 19, 92

D Dabholkar P.A., 35, 39, 41 Dach C., 6, 8, 18, 22, 38, 39, 73, 75, 82, 92,

108, 115, 116, 117, 118, 121, 128


158

Damien P., 132, 133 Danaher P.J., 19, 49 Dann S., 3 Davis F.D., 25, 28 Davis R.A., 19 De Ruyter K., 4 De Wulf K., 24 Degeratu A.M., 82, 83, 109 Dekimpe M.G., 2, 6, 19, 85, 88, 90 Deleersnyder B., 48, 49, 50, 114 Delhagen K., 47 Denton A., 47 Dholakia N., 6 Dholakia R.R., 6, 19, 21, 39, 49, 50, 128 Donkers B., 129 Donovan R.J., 25 Donthu N., 52 DoubleClick, 21 Ducoffe R.H., 84 Dutta S., 8

E Easingwood C., 50 Ehrenberg A.S.C., 53 Enders W., 89, 90, 91 Engel, 24

F Fader P.S., 47, 75, 83, 85, 109 Fedorikhin A., 51, 52, 74 Feinberg R., 4 Ferraro R., 4 Fischer P.P., 19 Fornell C., 29 Forrester Research, 47, 80 Forsythe S.M., 4 Foster K.D., 84 Fournier S., 4, 51, 86, 87 Fox E.J., 54, 56, 85, 86 Franses P.H., 54, 120 Fygenson M., 25

G Gallagher K., 84 Ganesan S., 32 Garino J., 18, 23 Geman D., 56 Geman S., 56 Gensler S., 2, 6, 10, 17, 19, 21, 48, 49, 50,

51, 73, 75, 82, 108, 115, 116, 117, 118, 121, 127

George E.I., 56 Gerbing D.W., 26, 29, 30 Geweke J.F., 133, 135 Geyskens I., 19, 48, 49, 50 Ghose S., 4, 48 Gielens K., 19 Giese J.L., 20 Godfrey A.L., 22 Gordon G.L., 4, 128 Green P.J., 132 Greenberg E., 133 Greene W., 56 Grewal D., 5, 6, 22, 25, 117 Grewal R., 1, 3, 10, 48, 49, 50, 114 Griffith D.A., 5 Gulati R., 18, 23 Gupta A., 1, 7, 109, 127 Gupta S., 52, 54, 85

H Hair J.F., 26, 27, 29, 30, 36 Hanssens D.M., 85, 88, 90 Ho T.H., 25 Hoch S.J., 85 Hoffman D.L., 2, 3, 4, 7, 48, 74, 83, 109 Hoque A.Y., 52 Huff D.L., 58 Huizingh K.R.E., 2, 8 Hulland J., 52 Hunter G.L., 24

I Inman J.J., 4

Author Index

159

J Järveläinen J., 4, 20 Johnson C.A., 47 Johnson E.J., 85 Johnson J.L., 20 Jöreskog K.G., 27, 34, 35

K Kamakura W.A., 22, 24, 37, 43, 44 Kaul A., 86, 87 Keen C., 4, 20 Kelley C.M., 47 Kelly J.P., 25 Kim I.M., 90, 91, 92, 95, 100 Knox G., 6, 48, 49, 50, 127 Kornelis M., 95 Krishnamurthi L., 52 Krishnan S., 6 Kumar V., 51 Kushwaha T.L., 6, 21, 39, 44, 48, 49, 51,

117, 124, 127

L Lam S.Y., 52, 85 Larcker D.F., 29 Laurent G., 85 Lee J.Y., 45 Lee R.P., 1, 3, 10, 48, 49, 50, 114 Lee M.S., 7 Leeflang P.S.H., 52, 53, 86, 106, 108 Leghorn R., 5 Lim J., 92 Lintner A., 47 Lloyd S.M., 1, 4 Lockshin L.S., 24 Lodish L.M., 54 Lohse G.L., 52, 85 Lynch J., 48, 51, 108, 116

M Macintosh G., 24

Madalla G.S., 91, 92, 95, 100 Maddox R.N., 28 Mahajan V., 22 Malhotra N.K., 84 Malter A.J., 32 Manchanda P., 54, 69, 124, 133 Mangleburg T., 25 Marcoolyn G., 25 Mathwick C., 73, 83, 84, 109 Mazis M.B., 7 Mela C., 6 Mendelsohn T., 47 Meuter M.L., 1, 4, 48 Meyer S., 47 Mick D.G., 4, 51, 86, 87 Miniard, 24 Mittal V., 22, 24, 37, 43, 44 Moe W.W., 47, 75, 83, 85, 109 Montgomery A.L., 54 Montoya-Weiss M.M., 6, 17, 18, 19, 20, 21,

25, 38, 39, 42, 117, 128 Moorman C., 32 Morrison D.G., 68 Morwitz V.G., 37 Muniz A.M., 123

N Naert A., 53 Naik P., 17, 19, 48, 49, 50 Nesdale A., 25 Neslin S.A., 2, 5, 6, 10, 48, 51, 52, 74, 82,

117, 123, 125 Newman J.W., 7 Nicholson M., 2, 4, 6, 75, 117, 128 Nijs V.R., 85, 86 Noble S.M., 5 Novak T.P., 2, 3, 7, 4, 74, 83, 109

O O’Guinn T.C., 124 Oliver R.L., 24 Olson J.M., 23, 24 Ostrom A.L., 1, 4


160

Otto J.R., 19

P Paap R., 54 Padmanabhan V., 85 Parasuraman A., 25, 76 Parsons J., 84 Pauwels K.H., 85, 89 Pavlou P.A., 25, 52 Pearce M., 52 Peck J., 83 Peralta M., 4 Peterson R.A., 2, 4, 17, 83, 92 Pian Y., 3 Pickerodt S., 84 Podsakoff N.P., 45 Podsakoff P.M., 45 Ponnavolu K., 24 Porembski M., 84 Porter M., 52, 74, 125 Prosser B., 56 Puhakainen J., 4, 20 Punj G.N., 7

R Raghunathan R., 22 Rangaswamy A., 7, 17, 51, 82, 125 Rasbash J., 56 Rasch S., 47 Ratchford B.T., 7, 22, 38, 39, 51, 84 Reinartz W.J., 37 Reynolds K.E., 28 Rigdon E., 73, 83, 85, 109 Rindfleisch A., 32, 34, 46 Rossi P.E., 55 Rossiter J.R., 25 Roundtree R.I., 1, 4 Roy S., 4, 48 Rust R.T., 52

S Salop S.C., 7

Schlosser A.E., 1, 4 Schoenbachler D.D., 4, 128 Segev A., 8 Seiders K., 22, 23, 39, 40 Shankar V., 4, 5, 6, 17, 19, 20, 21, 38, 39,

44, 48, 49, 51, 83, 109, 117, 124, 127, 128

Shi B., 4 Shiv B., 51, 52, 74 Siddarth S., 85 Sims C.A., 89 Sirgy M.J., 24 Sismeiro C., 47 Skiera B., 2, 6, 117 Smith A.K., 17 Smith S.M., 25 Sörbom D., 27, 34, 35 Spangenberg E.R., 5 Srinivasan S.S., 24 Staelin R., 7 Steele F., 56 Steenkamp J.B., 26, 29, 30, 85 Stem D.E., 5 Stewart D.W., 52 Su B., 1, 7 Sullivan U., 6, 18, 19, 50, 80, 82, 127, 128

T Talukdar, 7 Tang C.S., 25 Tanner M.A., 56 Tatham R.L., 26 Taylor S.A., 24 Teerling M.L., 5 Tellis G.J., 44, 46, 120 Teo T.S.H., 3 Thomas J.S., 5, 6, 18, 19, 50, 80, 82, 127,

128 Tierney L., 56 Tobin’s Q, 50

U US Census Bureau, 1

Author Index

161

V Van Baal S., 6, 8, 18, 22, 38, 39, 73, 75, 82,

92, 108, 115, 116, 117, 118, 121, 128 Van Bruggen G.H., 18, 125 Van Der Heijden H., 25, 28 Van Heerde H.J., 52, 85, 86 Van Kenhove P., 24, 28 Van Trijp H.C.M., 26, 29, 30 Van Waterschoot W., 24 Vandenbosch M., 52 Venkataramanan M.A., 6 Venkatesan R., 51 Venkatesh V., 25 Verhoef P.C., 2, 5, 7, 21, 22, 23, 38, 39, 43,

44, 51, 80, 82, 117, 129 Villanueva J., 129 Voss G.B., 6, 22, 25 Vroomen B., 2, 5, 117

W Wakefield J., 132

Walker S., 132 Wallace D.W., 19, 20, 21, 23, 39, 51, 117 Walter Z., 1, 7 Wansink B., 85 Wedel M., 53 Weinberger M.G., 5 Wells W.D., 84 Wetzels M., 4 Wilson I.W., 19 Wittink D., 52, 53, 86, 87 Wong W.H., 56 Wu J., 7, 51, 82

Y Yung Y.F., 83

Z Zanna M.P., 23, 24 Zellner A., 131, 132, 134 Zhang J., 52 Zhao M., 6

163

Subject Index

A advertising

classical, 72 mass, 9 nonintrusive, 3

aggregation bias, 103 data, 94, 98 entity, 97, 110 level, 111 temporal, 103 unit, 94, 110

analyses descriptive, 23 exploratory, 56 post hoc, 65

antecedents, 15, 20, 22, 23, 28 affective, 19 behavioral, 19 cognitive, 19

attitudes, 5, 19, 104, 109 offline, 9 online, 9 site, 9 store, 9

B behavior

buying, 9 free-riding, 71, 109 impulse buying, 39, 40, 48, 66 offline, 9 online, 9 search, 9 site, 9 store, 9

bias common method, 41 interpretation, 18, 41

C channel(s), 5, 8

acquisition, 5 addition, 45 cannibalization, 115 choice, 5 incumbent, 104, 117 informational, 13 integration, 5, 19 lock-in, 8 migration, 5 selection, 115 switching, 19 transactional, 13 utility, 71

cluster analysis, 89 common method variance, 41 conditional posterior distribution, 52, 120 construct, 22, 23 co-occurrence, 63 correlation, 25, 50, 63, 79, 83

contemporaneous, 51 cross-category, 59 inter-item, 24 item-to-total, 24 pairwise, 25 residual, 85, 88

cross-channel behavior, 40, 69, 71, 97 effects, 6, 7, 8, 10, 70, 103, 107, 113 marketing efforts, 98 synergy, 8, 19, 98, 115

customer(s) attitudes. See attitudes contact point, 5, 8 demographics, 14 efficiency, 40 Internet savvy, 40 panel, 10, 44 retention, 65, 105, 107, 109, 111, 112


164

segment, 104

D decision-making process, 47, 65, 97 decomposition, 48, 73 deterministic trend, 78

E effects

complementary, 37 cross-category, 59, 63 cross-channel. See cross-channel moderation, 34 same-channel, 85, 98 substitution, 40

error multivariate distribution, 51 variances, 24, 33

F factor analysis

exploratory, 23 fit indices

CFI, 25 Chi-square index, 25 Cronbach’s alpha, 23 hitrate, 58 lag length criterium, 99 Lagrange multiplier, 93 MAPE, 64 NNFI, 25 R2 adjusted, 85 RMSEA, 25 SC, 84

flow, 74, 91, 107, 110, 131

G Gibbs sampling, 52 Granger causality, 79, 82

H heterogeneity, 49, 51, 63, 64 heteroskedasticity, 99

I impulse response functions, 79 information

dissemination, 48, 72 persuision, 72 search, 43, 69, 114

involvement, 18

L likelihood

estimation, 52 maximum, 52

loyalty attitudinal, 20 behavioral, 20, 41, 71, 88

M marketing

efforts, 7, 70, 74, 86, 100, 110 instruments, 67

Markov chain, 52 MCMC methodology, 52 model(s)

factor-analytical, 22 identification, 50 longitudinal, 33 multi-level, 52 MVP, 50 Poisson, 49 sales, 112 SEM, 24 structural, 26 type-II Tobit, 49 VAR, 70

moderation, 30, 82 multi-channel

behavior, 13

Subject Index

165

buying behavior, 5 consumer behavior, 5 customers, 5, 16, 70 environment, 4, 113 retailers, 69 search behavior, 5 strategy, 111

multichannelers, 48 multicollinearity, 54, 83

P parameter homogeneity, 97 pooling, 60 price sensitivity, 74, 75 product(s)

category, 44, 52,67 non-sensory, 110 sensory, 99, 110 type, 72, 82, 89, 99, 107

promotion non-price, 74 price, 74 supported, 74

psychological bonds, 66, 71 purchase

behavior. See behavior incidence, 48, 50, 62 intention, 16, 38

R relationship(s)

attitude-behavior, 32 cross-channel, 16 partial, 58

research shopper, 71

S sample

convenience, 23 estimation, 53, 54 validation, 53

satisfaction, 20

search attractiveness, 16 behavior. See behavior

segmentation, 83 self-selection, 30, 40, 55 stationarity, 79 store

interior, 20 merchandise, 20 personnel, 20

switching costs, 48, 65, 71, 106

T technology

acceptance model (TAM), 23 adoption, 20 anxiety, 1, 4, 45 self-service, 4

tests Augmented Dicky-Fuller, 79 Chi-square difference, 33 Chow breakpoint, 89 Chow pooling test, 60 KPSS, 79 pre-test, 23 unit root, 79 unit root confirmatory, 80 validation. See validation

Tobin’s Q, 46

U unidimensionality, 24

V validation, 36, 58, 64 validity

construct, 23 convergent, 24 discriminant, 25

variable(s) dummy, 49, 78 endogenous, 22


166

exogenous, 22 latent, 22, 26 time-varying, 50

W web site(s)

content, 20

conversion rate, 43 design, 20 effectiveness, 72 informational, 2 introduction, 70 transactional, 43

web-to-store shopping, 69

167

Samenvatting Recente ontwikkelingen bieden consumenten, naast de ‘traditionele

winkel’, de mogelijkheid om bij het doen van aankopen, aanvullende kanalen te gebruiken. Met een kanaal bedoelen we een middel waarmee klanten en bedrijven contact met elkaar kunnen hebben. Bedrijven gebruiken kanalen om met klanten te communiceren, om aankopen te faciliteren (transactiemogelijkheid) of om producten of diensten te distribueren. Het kanaal dat het afgelopen decennium de grootste impact heeft gehad op het consumentengedrag is het internet. Nu het internet langzaam maar zeker volwassen begint te worden, wordt duidelijk dat voor internetactiviteiten strategische planning nodig is. Daarvoor is inzicht nodig in de verschillende functies van het internet.

Op basis van de communicatie- en transactiefunctie van een website, zijn er twee typen commerciële websites te onderscheiden. Het eerste type is de website die voornamelijk communiceert met de klant en informatie biedt. Dit is een informatieve website. Het tweede type is een transactionele website. Op een dergelijke site krijgt de bezoeker informatie en kan hij producten en/of diensten kopen.

De praktijk wijst uit dat de meeste bedrijven sneller kiezen voor een informatieve website, aangezien voor transactionele websites onder andere meer investeringen en interne aanpassingen nodig zijn. Daarnaast zijn er bij consumenten nog obstakels voor het doen van aankopen via internet, bijvoorbeeld angst, privacy en vertrouwen (Meuter et al. 2003; Schlosser et al. 2006). Dit blijkt ook in het percentage aankopen via internet ten opzichte van de totale consumentenaankopen. In Nederland was dat in 2004 slechts 1.7% (CBS 2006).

Toch is er sinds 1998 voornamelijk wetenschappelijk marketingonderzoek gepubliceerd op het gebied van de transactiemogelijkheden van het internet. Mede omdat het relatief eenvoudig is te meten wat een transactionele website bijdraagt aan het bedrijfsresultaat. De opbrengsten van informatieve websites, die door de meeste organisaties gebruikt worden, zijn minder eenvoudig te bepalen. Via dit type website kun je immers geen producten of diensten aankopen.

In mijn onderzoek staat juist de commerciële bijdrage van informatieve websites centraal. Een informatieve website geeft


168

informatie over het bedrijf en de producten, draagt het imago van het bedrijf uit, en/of probeert het opbouwen van een langdurige relatie met klanten te ondersteunen. Een informatieve website biedt geen mogelijkheden om producten te bestellen of te kopen. Dit type website is vooral geschikt voor producten die zich minder lenen voor verkoop via internet, respectievelijk voor consumenten internet gebruiken om zich te orienteren alvorens in de traditionele winkel overgaan tot een aankoop.

Het blijkt dat consumenten eerder geneigd zijn om meerdere kanalen, zoals internet en de winkel in combinatie met elkaar te gebruiken dan het ene kanaal in te wisselen voor het andere (Nicholson et al. 2002; Verhoef et al. 2005). In veel gevallen zoekt de consument via het internet informatie op om vervolgens het product of de dienst in de winkel te kopen. Deze ontwikkeling in consumentengedrag, van het gebruik van een enkel kanaal naar het gebruik van meerdere kanalen naast elkaar, heeft onderzoek op het gebied van ‘multichannel consumer behavior’ in gang gezet.

Het onderzoek naar multichannel consumer behavior bekijkt hoe consumenten meerdere kanalen gebruiken in de verschillende fasen van het aankoopproces en hoe de kanaalkeuze hun aankoopgedrag beïnvloedt. De consument kan eerst via een informatieve website uitgebreid informatie zoeken voordat hij het product via een ander kanaal koopt. Informatieve websits hebben derhalve de potentie om het aankoopgedrag van de consument te beïnvloeden.

Uit multichannel onderzoek blijkt dat de keuze van de consument voor een bepaald kanaal gestuurd wordt door demografische kenmerken, in hoeverre kanalen geïntegreerd zijn en marketingactiviteiten (Bendoly et al. 2005). Een aantal onderzoeken toont aan dat klanten die meerdere kanalen van dezelfde aanbieder gebruiken meer besteden dan klanten die uitsluitend via hetzelfde kanaal aankopen (Kushwaha & Shankar 2005). Echter, er zijn ook onderzoeken die aantonen dat het gebruik van internet leidt tot minder aankopen (Ansari et al. 2006; Gensler et al. 2007). Tot slot tonen Dholakia et al. (2005) aan dat consumenten liever het internetkanaal gebruiken naast de winkel dan deze de winkel te laten vervangen.

Ook is er onderzoek gedaan naar het effect van een additioneel (extra) kanaal op de manier waarop klanten tegen een organisatie aankijken, de klantperceptie. Op een geaggregeerd niveau blijkt dat informatieve websites een positief effect hebben op klantpercepties.

Nederlandse Samenvatting

169

Op individueel klantniveau blijkt dat de effecten van een additioneel informatief kanaal grotendeels onbekend zijn.

Zoals ik al eerder heb aangegeven is er vooral met betrekking tot het effect van informatieve websites op consumentengedrag nog relatief weinig bekend. Hierdoor is er nog een aantal belangrijke vragen waarop antwoord gegeven moet worden. Ten eerste is het onduidelijk hoe consumenten zich daadwerkelijk gedragen in de verschillende kanalen, met name op informatieve websites, en wat voor effect dit gedrag heeft. Ten tweede is het van belang om aan te tonen hoe het zoekgedrag in het ene kanaal het aankoopgedrag in het andere kanaal beïnvloedt. Ten derde is het de vraag of er verschillen zijn in dit gedrag voor bepaalde groepen consumenten en producttypen. Tot slot, welke achtereenvolgende kanaaleffecten vinden plaats en wat betekent dit gedrag voor de organisatie? Mijn onderzoek geeft een antwoord op deze vragen aan de hand van de volgende onderzoeksvragen: • Hoe en in welke mate zijn er relaties tussen de percepties ten

opzichte van een informatieve website en de percepties en gedrag in een traditionele winkel en hoe worden deze relaties beïnvloed door consumenteneigenschappen?

• Hoe en in welke mate heeft het gedrag op een informatieve website invloed op het gedrag in een traditionele winkel?

• Welke achtereenvolgende kanaaleffecten vinden plaats voor een informatieve website en een traditionele winkel en hoe beïnvloeden marketingactiviteiten de kanaaleffecten? Verschillen deze effecten voor bepaalde segmenten en producttypen? Dit onderzoek draagt bij aan de bestaande literatuur door inzicht te

geven in (1) het effect van informatieve websites op klantengedrag, (2) het sequentiële zoeken koopproces van klanten, (3) specifieke effecten van informatieve websites voor segmenten en productcategorieën, (4) gebruik van informatie verkregen via de website door klanten, en (5) de methodologie die gebruikt kan worden om de effecten aan te tonen.

ONDERZOEKSGEGEVENS

Het onderzoek heeft plaatsgevonden binnen de detailhandel. De data om de onderzoeksvragen te beantwoorden, zijn verzameld in samenwerking met een Nederlandse detailist. Deze retailer [idem] heeft 58 warenhuizen in de grootstedelijke gebieden. Elke winkel heeft over


170

het algemeen dertien verschillende afdelingen, zoals vrouwenkleding, mannenkleding, accessoires, interieurartikelen en persoonlijke verzorgingsproducten.

In maart 2001 heeft de detailist een informatieve website geïntroduceerd voor haar klanten. Het doel van de website is om de winkelactiviteiten te ondersteunen en het verhogen van de verkopen in de winkels. De site biedt klanten informatie over lifestyle (bijvoorbeeld mode en vakantie), producten die in winkels worden aangeboden, promoties en de onderneming zelf. Bovendien biedt de site mogelijkheden voor vermaak, zoals het versturen van een e-mailkaart naar vrienden. De site heeft voor de klant meerwaarde omdat het een betere kijk op de producten biedt die in de winkel te koop zijn, tevens kan de klant via de website op ideeën komen.

Om de onderzoeksvragen te beantwoorden zijn er gegevens verzameld met betrekking tot de aankopen in de winkel, het zoekgedrag op de informatieve website en de percepties van individuele klanten. Het aankoopgedrag in de winkel, verzameld voor 8.847 klanten, beslaat de periode januari 2000 tot en met mei 2002. Van de 8.847 klanten, hebben er 6.594 van maart 2001 tot en met mei 2002 gebruik gemaakt van de site. De overige 2.253 klanten hebben de website niet bezocht. Om de klantpercepties ten opzichte van de winkel en de website te verzamelen is er in mei 2001 en mei 2002 een enquête uitgevoerd via internet. In mei 2001 hebben 3.128 klanten de enquête ingevuld en in mei 2002 4.865 klanten. In totaal hebben 749 klanten de enquête in beide jaren ingevuld. Naast de percepties zijn via de website demografische gegevens verzameld op individueel klantenniveau en via Acxiom op postcodeniveau. Bij de aankopen in de winkel, het bezoek aan de website en het invullen van de enquête wordt gebruikt gemaakt van de klantenkaart, daardoor zijn de verschillende gegevensbronnen op individueel klantniveau aan elkaar te koppelen.

HOOFDSTUK 2

In hoofdstuk 2 ligt de nadruk op het bepalen van de effecten van een informatieve website op de percepties en het gedrag van de klant in de winkel. Dit gebeurt door middel van een structureel model (structural equation model). De vraag is in hoeverre de houding (attitude) ten aanzien van de informatieve website een effect heeft op de houding ten aanzien van de winkel en hoe het aankoopgedrag in de winkel erdoor beïnvloed wordt. We onderzoeken ook hoe de verbanden


171

zich verhouden over de tijd en hoe de verbanden beïnvloed worden door consumentenkenmerken zoals leeftijd.

De resultaten tonen aan dat klanten met een positieve houding ten aanzien van de website ook een positieve houding hebben ten aanzien van de winkel. Deze verbanden zijn sterker voor een bepaalde groep klanten, namelijk vrouwen en lager opgeleide klanten. Voor deze bevindingen geven we zowel een algemene als een meer internetgerelateerde verklaring. Het verband tussen website- en winkelperceptie is ook sterker voor klanten die een sterkere integratie tussen de kanalen ervaren.

Naast dit positieve verband vinden we ook een negatief verband tussen websiteperceptie en het aankoopgedrag in de winkel. Dit is een onverwacht verband gezien de voorgaande onderzoeken. Dit komt doordat ons onderzoek zich richt op één specifieke organisatie. In het algemeen verzamelen consumenten graag via internet informatie om vervolgens de aankoop in een winkel te doen. Echter deze positieve relatie is minder logisch in het geval van twee kanalen voor één organisatie. Immers in dit geval kunnen factoren zoals loyaliteit en concurrentie de relatie beïnvloeden en zelfs een negatief verband veroorzaken. Tot slot blijkt dat er een positief verband is tussen het zoekgedrag op de informatieve website en het aankoopgedrag in de winkel, echter deze relatie is zwak en instabiel.

Het hoofdstuk geeft aan dat de ‘beste’ klanten – klanten met een positieve perceptie en gedrag – in beide kanalen de ‘beste’ klanten zijn, echter het onderzoek toont ook aan dat de informatieve website een negatief effect heeft op het aankoopgedrag in de winkel. Dit resultaat is een sterke indicatie dat klanten efficiënter worden in hun aankoopgedrag door het gebruik van de informatieve website.

HOOFDSTUK 3

Het onderzoek dat besproken wordt in hoofdstuk 3 bepaalt de effecten van het gebruik van de informatieve website op het individuele aankoopgedrag in de winkel. Een decompositie van het aankoopgedrag wordt gehanteerd om te bepalen hoe het gebruik van de informatieve website het aankoopgedrag in de winkel beïnvloedt. Decompositie van gedrag houdt in dat de totale aankopen van een klant worden opgedeeld in een aantal elementen. In dit onderzoek betreft het de frequentie van koopbezoeken en het gemiddelde aankoopgedrag per koopbezoek in zes


172

verschillende productcategorieën. Het effect van de website op deze elementen wordt bepaald door een tweetal modellen: 1. Een Poisson model voor de frequentie van het aantal koopbezoeken

in de winkel. 2. Een multivariaat probit model voor het gemiddelde aankoopbedrag

per koopbezoek in zes productcategorieën. In het multivariaat probit model bepalen we ook in welke mate

webpagina’s die gerelateerd zijn aan specifieke categorieën, invloed hebben op het aankoopbedrag in een bepaalde productcategorie. Omdat individuele klanten niet met hoge regelmaat aankopen doen in een warenhuis, schatten we de modellen op basis van maandelijkse aankoopgegevens. We bepalen het effect van het gebruik van de informatieve website in dezelfde maand als het aankoopgedrag in de winkel.

De resultaten tonen aan dat door het bezoeken van de informatieve website de meerderheid van de klanten minder vaak in de winkel komt en gemiddeld ook minder per productcategorie besteedt. In het geval van een transactionele website tonen Ansari et al. (2006) en Gensler et al. (2007) vergelijkbare negatieve effecten aan op het aankoopgedrag van klanten. Daarnaast zijn onze bevindingen ook in overeenstemming met de studie van Van Baal en Dach (2005) waarin zij aantonen dat slechts 10 procent van de klanten bij dezelfde organisatie blijft als zij meerdere kanalen gebruiken. De overige 90 procent van de klanten gebruikt de website van een bedrijf voor het zoeken naar informatie, om vervolgens bij een ander bedrijf de aankoop te doen. Gezien deze bevindingen lijkt het erop dat vooral consumenten voordeel halen uit het aanbod van meerdere kanalen.

Naast de onderzoeken betreffende de effecten voor een specifiek bedrijf zijn er tevens verschillende onderzoeken die de effecten van de multichannel omgeving in het algemeen tot onderwerp hebben (o.a. Burke 2002; Verhoef et al. 2005). Hieruit blijkt dat bedrijven wel voordeel kunnen halen uit het de multichannel omgeving. Echter, gezien de onderzoeken van Ansari et al. (2006), Gensler et al. (2007) en onze bevindingen, lijkt het er op dat bedrijven eerder negatieve effecten kunnen ondervinden van de multichannel omgeving. Deze negatieve effecten zijn waarschijnlijk gevolg van (1) een efficiënter aankoopproces van consumenten, (2) minder impulsaankopen en (3) gemakkelijker veranderen van aanbieder.


173

Bij een klein percentage klanten leidt het bezoeken van de informatieve website tot meer aankopen in de winkel. Circa 20 procent van de klanten heeft meer koopbezoeken en circa 10 procent van de klanten besteed een hoger bedrag in de desbetreffende productcategorie. Een aanvullende analyse toont aan dat deze klanten, dus de klanten voor wie een positief effect van de informatieve website gevonden is, gemiddeld genomen meer besteden in de winkel. Zodoende kan een informatieve website voordeel opleveren voor het bedrijf, maar alleen in het geval van de ‘beste’ klanten.

Concluderend: het gebruik van een informatieve website heeft invloed op het aankoopgedrag van klanten in de traditionele winkel. Ons onderzoek toont ook aan dat bedrijven de implementatie van een informatieve website zorgvuldig moeten overwegen, gezien de te verwachten positieve en negatieve effecten.

HOOFDSTUK 4

In hoofdstuk 4 ligt de nadruk op het bepalen van de langetermijneffecten, namelijk (1) hoe gedrag in het ene kanaal het gedrag in het andere kanaal beïnvloedt over een periode van 26 weken, (2) hoe marketingactiviteiten in het ene kanaal het gedrag in het andere kanaal beïnvloeden en (3) hoe omgevingskenmerken dit gedrag en het effect van de marketingactiviteiten beïnvloeden.

We schatten een vector autoregressie model (VAR) om de verbanden tussen de verschillende componenten van het aankoopgedrag in de winkel en het zoekgedrag op de website te bepalen. Daarnaast bepalen we met het VAR het effect van de marketingactiviteiten in het ene kanaal op het gedrag in het andere kanaal. We schatten het VAR op een geaggregeerd niveau en op het niveau van de omgevingskenmerken.

Op het geaggregeerde niveau vinden we weinig langetermijneffecten van het gedrag in het ene kanaal op het gedrag in het andere kanaal. Echter op het niveau van de omgevingskenmerken komen deze effecten wel naar voren. Uit de resultaten blijkt dat de introductie van een informatieve website een structurele afname veroorzaakt in het aantal koopbezoeken. Dit resultaat bevestigt dat consumenten de informatie, verstrekt via de website, gebruiken om de beste aanbieder te selecteren (van Baal & Dach 2005). De klanten worden na introductie van een informatieve website efficiënter in hun aankoopproces en minder trouw aan de organisatie.


174

Verder kijken we naar het effect van een aantal omgevingskenmerken, om precies te zijn het effect van het producttype, de mate waarin de consument zichzelf ‘verliest’ tijdens het bezoek aan de website (flow), en de frequentie van het aantal websitebezoeken. De resultaten bevestigen dat voor zintuiglijke producten, zoals kleding, consumenten meer zoekgedrag vertonen. Echter, ons onderzoek toont ook aan dat voor deze producten een informatieve website klanten een efficiëntere wijze van informatie zoeken biedt. Ondanks dat een hoge mate van ‘flow’ tijdens het websitebezoek erg plezierig is voor de klant, heeft dit geen positief effect op het aankoopgedrag in de winkel. Het tegenovergestelde wordt zelfs waargenomen. Klanten die weinig flow ervaren, vertonen sterkere verbanden tussen het gedrag in de verschillende kanalen en zijn ontvankelijker voor marketingactiviteiten. Tot slot zijn er geen significante verbanden tussen het aankoopgedrag in de winkel en het zoekgedrag op de website voor klanten die de website vaak bezoeken.

BELANGRIJKSTE RESULTATEN

In hoofdstuk 5 (tabel 5-2 en 5-3) worden de uitkomsten van de verschillende hoofdstukken vergeleken met voorgaande onderzoeken op dit gebied. De belangrijkste inzichten die ons onderzoek ten opzichte van deze onderzoeken geeft zijn de volgende: • Een informatieve website verbetert de percepties van klanten

betreffende de traditionele winkel. Voor een dergelijke verbetering van percepties zijn transacties (aankopen) via de website niet noodzakelijk.

• Klanten met een positieve perceptie betreffende de website spenderen minder in de traditionele winkel. Gebruik maken van informatie die via het internet aangeboden wordt om het beste aanbod te vinden, gebeurt niet alleen bij transactionele websites, maar ook bij informatieve websites.

• Het bezoekgedrag op de website leidt tot een afname van het aantal aankoopbezoeken in de winkel in dezelfde maand voor de meerderheid van de klanten. Klanten met een toename van het aantal aankoopbezoeken als gevolg van het websitebezoekgedrag, gebruiken beide kanalen meer. Verbetering van het klantgedrag is mogelijk door de implementatie van een informatieve website, maar dit geldt alleen voor de ‘beste’ klanten. De meerderheid van de klanten ondervindt voordeel van de informatieve website door een


175

efficiënter aankoopproces, minder impulsieve aankopen en betere beslissingen.

• Op de lange termijn heeft de introductie van de informatieve website een structurele afname veroorzaakt in het aantal koopbezoeken.

• Naast de afname van het aantal koopbezoeken, veroorzaakt de informatieve website voor de meerderheid van de klanten een afname in het bestede bedrag.

• Marketingactiviteiten in beide kanalen, via de website en via traditionele media, stimuleren het klantgedrag in het kanaal dat het mogelijk maakt om aankopen te doen.

CONCLUSIE

De onderzoeken die in dit proefschrift gepresenteerd worden, tonen aan dat klanten het gebruik van meerdere kanalen waarderen. Echter, deze waardering leidt niet automatisch tot een toename in het aankoopgedrag. Voor de meerderheid van de klanten leidt het bezoeken van het additionele informatieve kanaal (website) tot een afname in het aankoopgedrag. De meest waarschijnlijke redenen hebben te maken met de efficiëntie die de klant ondervindt van de informatieve website.

Door het bezoeken van de informatieve site hoeven klanten niet meer naar de winkel voor informatie. Voorheen gingen dit soort informatiebezoeken aan de winkel meestal gepaard met impulsieve aankopen. Met de introductie van een informatieve website hebben klanten de mogelijkheid informatie te vinden zonder in de verleiding te komen om impulsief producten te kopen. Daarnaast zijn klanten dankzij de site beter in staat om beslissingen te nemen en hun aankoopproces efficiënter te doen.

Bij een klein percentage van de klanten heeft een informatieve website een positief effect op het aankoopgedrag in de winkel. Deze klanten zijn te kenmerken als de ‘beste’ klanten van de organisatie. Ze kopen frequenter, besteden meer geld, bezoeken de website vaker en bekijken meer pagina’s tijdens een website bezoek. Daarnaast laat ons onderzoek zien dat de integratie van kanalen van belang is voor het bereiken van een verbetering van het aankoopgedrag in het traditionele kanaal.

IMPLICATIES VOOR MANAGERS

Naast de wetenschappelijke bijdrage, geeft dit proefschrift een aantal inzichten voor managers over de effecten van informatieve


176

websites. Managers kunnen deze inzichten allereerst gebruiken voor het strategisch implementeren van (informatieve) internetactiviteiten. Immers voor de meeste bedrijven is internet geen keuze maar eerder een verplichting. De moderne consument waardeert niet alleen het gebruik van meerdere kanalen maar verwacht het ook. De vraag is niet óf een bedrijf internet moet gebruiken, maar hóe zij internet het beste kan inzetten. De eerste afweging is wat voor soort website een bedrijf het beste kan implementeren. Deze afweging is afhankelijk van de huidige kanalen, de concurrentie, de frequentie van aankoop, en voorgaande ervaringen met het implementeren van additionele kanalen. Daarnaast moeten bedrijven zich goed afvragen of de internetactiviteiten voor alle klanten zijn bedoeld of dat men zich op een bepaald segment wil richten.

Ten tweede kunnen managers de inzichten gebruiken om de effecten van informatieve websites te sturen. Het negatieve effect op het aankoopgedrag kan geminimaliseerd worden door de integratie van de kanalen te maximaliseren. Deze integratie kan gestimuleerd worden door een consistente boodschap via de kanalen te communiceren. Daarnaast door ook klanten de mogelijkheid te bieden om de kanalen geïntegreerd te gebruiken. Bijvoorbeeld door klanten een product online te laten bestellen en het in de traditionele winkel op te laten halen.

Documents

· Published by: Labyrinth Publications Pottenbakkerstraat 15 – 17 2984 AX Ridderkerk The Netherlands Print: Offsetdrukkerij Ridderprint B.V., Ridderkerk ISBN 90-5335-106-X 978-90-