University of Groningen Online marketing and advertising ... · Processed on: 23-9-2016...

University of Groningen

Online marketing and advertising researchYudhistira, Titah

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:2016

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):Yudhistira, T. (2016). Online marketing and advertising research: Traditional Theories Revisited Groningen:University of Groningen, SOM research school

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: 11-02-2018

https://www.rug.nl/research/portal/en/publications/online-marketing-and-advertising-research(48502b0b-561c-416e-8275-c015437e3108).html

Processed on: 23-9-2016Processed on: 23-9-2016Processed on: 23-9-2016Processed on: 23-9-2016

505395-L-bw-Yudhistira505395-L-bw-Yudhistira505395-L-bw-Yudhistira505395-L-bw-Yudhistira

Online Marketing and Advertising Research

Traditional Theories Revisited



Publisher: University of Groningen Groningen The Netherlands Printer: Ipskamp Printing, Enschede

ISBN: 978-90-367-9117-5 978-90-367-9116-8 (e-book) © Titah Yudhistira All rights reserved. No part of this publication may be reproduced, stored in a retrieval system of any nature, or transmitted in any form or by any means, electronic, mechanical, now known or hereafter invented, including photocopying or recording, without prior written permission of the publisher.



Online Marketing and Advertising Research

Traditional Theories Revisited

PhD Thesis

to obtain the degree of PhD at the University of Groningen on the authority of the

Rector Magnificus Prof. E. Sterken and in accordance with

the decision by the College of Deans

This thesis will be defended in public on

Monday 10 October 2016 at 11.00 hours

by

Titah Yudhistira

born on 9 May 1976 in Jember, Indonesia



Supervisor Prof. T.H.A. Bijmolt Co-supervisor Dr. K.R.E. Huizingh Assessment committee Prof. M. Clement Prof. H. van Herk Prof. J.E. Wieringa



v

Acknowledgments

Completing a dissertation is a long and very demanding journey. I would never have

finished this work without my colleagues, friends, and family who supported me in

different ways along this journey. I would like to take the opportunity to mention and

thank them.

First and foremost, I am deeply grateful to my promoter and copromoter. I have been

very lucky to have Tammo Bijmolt as my promoter. Not only has his impressive

knowledge and acuity regarding statistical modeling and marketing research always

helped me when I have gotten stuck with my research, but his positive personality that

exudes optimism revived my motivation whenever I felt discouraged during the

process. I also learnt a lot from him on how to be a good researcher and supervisor.

Tammo, thank you so much for your support and confidence on me. It was a great

privilege to be your PhD student.

Eelko Huizingh, my copromoter, has played a very central role since the beginning of

my PhD journey. It was my correspondence with Eelko almost 10 years ago that

paved the way for me to be a PhD student in the Marketing Department at the

University of Groningen. During my studies, Eelko has spent much time and effort in

helping and guiding me with my writing process. He has been very patient and has

always been available to help me develop ideas for structuring and streamlining my

papers. At times, Eelko has been very critical and detail oriented, so it was very

rewarding when he finally said that he had no further feedback for my papers. I have

also enjoyed spending time with Eelko outside of the university, such as dinners

together, barbeque at his house, and especially his recent visit to Bandung. Eelko,

thanks a lot for all of your support from the very beginning of this journey.

Next, I would like to express my gratitude for the members of the reading committee,

Michel Clement, Hester van Herk, and Jaap Wieringa, for their effort and constructive



vi

feedback that improved this dissertation. I especially thank Jaap, who was willing to

discuss some of his comments and feedbacks on Skype. I also would like to thank my

‘internal defense’ reading committee, Bob Fennis, Maarten Gijsenberg, and Lara

Lobschat, for raising some critical issues regarding my research and giving

constructive comments and suggestions to solve those issues. This dissertation is

better because of their feedback.

I also would like to address a special thanks to Daniela Naydenova and Rully Tri

Cahyono for being my paranymphs and helping me in the preparation of my defense

ceremony while I was in Bandung, Indonesia.

Strong empirical research relies on high-quality data. The studies in Chapter 2 and

Chapter 3 are based on massive online banner advertising data provided by Metrixlab;

Lucas Hulsebos and Arjen Groeneveld made the access to these data possible.

Furthermore, Pauline Elema worked on the acquisition and pre-processing of the data

before we could start analyzing the data. The study in Chapter 4 is based on primary

and secondary data collected by Dirk Bloem. I sincerely thank each of them for

making these data available for this research.

I would like also to acknowledge financial support for portions of my studies from

three institutions. First, the University of Groningen granted me the Bernoulli

Scholarship, which covered 2 years at Groningen. Next, the Directorate General of

Higher Education of the Ministry of Education of the Republic of Indonesia funded

the last year of my studies at Groningen. Finally, during my dissertation finishing

stage, Intitut Teknologi Bandung provided the funds to cover my traveling costs to

Groningen.

It was truly a privilege to be a part of the Marketing Department of the University of

Groningen. I have learnt a lot about academic and research ethos from my colleagues

in this great department. I particularly thank Peter Leeflang, Peter Verhoef, Koen

Pauwels, Erjen van Nierop, and Jenny van Doorn, who have opened my eyes on



vii

marketing science with their lectures in the Advanced Marketing Model Building and

Marketing Theory classes. Furthermore, I always enjoyed the social atmosphere of the

department with its birthday treats, unpredictable outings, summer barbeques,

Sinterklaas evening, inspiring (brown bag) seminars, and other activities I now miss

so much. Many thanks to my PhD student colleagues for interesting chats during

coffee breaks and PhD dinners, especially my office mates Adriana, Eline, Shandy,

Tao, Sina, Daniela, and Katrin—thank you for the conversation during the time we

shared an office. For Matilda, Peter van Eck, Ernst, Hans, Auke, Sietske, Umut,

Stanislav, Stefanie, Lisette, Sander, Jacob, Yi-Chun, Niels, Evert, and Alec, thank you

for being such amiable PhD student fellows. Many thanks also to the secretaries of

Marketing Department, Hanneke, Frederika, and Lianne.

I would like also to thank people from the SOM office, especially to Rina Koning,

Ellen Nienhuis, and Astrid Beerta. Special thanks goes to Arthur de Boer, who

patiently provided support for all administrative matters during the preparation of my

defense ceremony. I am also grateful to my PhD coordinators, Martin Land, Linda

Toolsema, and Justin Drupsteen, who have spent their time to ensure that I would

eventually finish my studies.

I also have benefitted from many people outside my academic circle. Firstly, living in

an international student house was a great experience I will never forget. I met and

befriended literally hundreds of people from all over the globe during my stay in

Groningen. The dinners, parties, and outings with my international friends of

Plutolaan, Blue Building, and Diaconessen House were potent stress relievers. In this

regard, I especially mention Anca, Hanna, Chun-ju, Carlos, Berci, Joao, Diana, Luca,

and Ilias, with whom I have had contacts and personal visits long after we left our

international student house. Secondly, being an Indonesian abroad, at times I felt

isolated and alone. My Indonesian friends of PPIG and de Gromiest always provided

friendship and support whenever I needed them. It is impossible to mention them all,

but I would like to especially thank my Indonesian friends Ari, Kadek, Nurul, Poppy,

Teguh, Fean, Naga, Ira, Fegi, Getty, Dodi, Muhsin, Desti, Iging, Bayu, Kang Intan,



viii

Enci, Aree, Cynthia, Tita, Saras, Mira, Tina, Mbak Lia, Mas Yayok, Pak Adjie, Bude

Nanie, Tante Pantja, Oom Basuki, Uwak Asiyah Hofman, Kuswanto, Rully, and Intan

for their help while I was in Groningen.

In the last few years, while finishing my dissertation, I received significant support

from my colleagues at Industrial Engineering Department, Insitut Teknologi Bandung.

I especially mention Pak Senator Nur Bahagia, Bu Lucia Diawati, Pak Suprayogi, Pak

Andi Cakravastia, Bu Yosi Agustina, and Pak Rully Tri Cahyono of Laboratory of

Industrial System Planning and Optimization for their support and concern while I

was in the final stage of finishing my dissertation.

Finally, and most importantly, I would like to thank my family for their support, love,

and care. This dissertation would not be finished without continuous encouragement

from my parents, my mother Tuty Basuki and my late father Masjhud. Ibu dan bapak,

terima kasih sebesar-besarnya atas semua cinta, dukungan, dan doa yang telah

dicurahkan selama ini. My sisters and their families have also been big supporters

and motivators for me to finish this dissertation. Terima kasih Mbak Wahyu, Mbak

Uri, Mas Yan, Mas Yono, dan para keponakanku tercinta semuanya!

Bandung, September 2016



ix

Table of Contents Chapter 1 Introduction 1 1.1 General introduction 1 1.2 Online marketing and advertising research 3 1.3 Aim and objectives 4 1.3.1 Study 1: Effect of banner exposures on consumer memory 4 1.3.2 Study 2: Advertisement and brand metrics for online

banner advertising effectiveness 6

1.3.3 Study 3: Potential of online polls as market research instruments

7

1.4 Outline of the Dissertation 8 Chapter 2 The Effect of Banner Exposures on Consumer Memory 9 2.1 Introduction 9 2.2 Literature Background and Hypotheses 11 2.2.1 Advertisement Scheduling 11 2.2.2 Effect of Number of Exposures 12 2.2.3 Effect of Spacing 13 2.2.4 Effect of Measurement Delay 14 2.2.5 Mediating Effect of Banner Memory 15 2.2.6 Differential Effect of Exposure Variables 17 2.2.7 Other Factors Affecting Consumer Memory of Banner and

Brand

18 2.3 Data Collection and Analysis 19 2.3.1 Data collection and description 19 2.3.2 Model Formulation and Analysis 20 2.4 Results 24 2.5 Discussion 30 2.5.1 Key Findings 30 2.5.2 Practical Implication 33 2.5.3 Research Limitations and Future Research 35 Chapter 3 Ad and Brand Metrics for Banner Advertising Effectiveness 37 3.1 Introduction 37 3.2 Theoretical Background 39



x

3.2.1 Indirect effect of advertising on sales 39 3.2.2 Hypotheses 40 3.3 Empirical Study Design 44 3.3.1 Survey procedure and sample description 44 3.3.2 Banner and brand effectiveness measures 45 3.4 Data analysis 47 3.4.1 Multilevel SEM for clustered observations 48 3.4.2 Categorical variables treatment 48 3.4.3 Missing values treatment 50 3.4.4 Covariates modeling 51 3.5 Results 51 3.5.1 Measurement model 51 3.5.2 Structural model 52 3.5.3 Competing model: Dual mediation 54 3.6 Conclusion 55 3.6.1 Discussion 55 3.6.2 Managerial implication 58 3.6.3 Limitations and further research 59 Chapter 4 Evaluating Website Polls as an Instrument for Customer Surveys

63

4.1 Introduction 63 4.2 Background and Research Question 66 4.3 Study Design 70 4.4 Results 72 4.4.1 Sample Characteristics 72 4.4.2 Response Distribution 73 4.4.3 Sample versus mode effects 73 4.4.4 Weighting results 74 4.5 Discussion 76 4.5.1 Conclusion 76 4.5.2 Limitations and directions for future research 79 4.5.3 Practical Implications 80 Chapter 5 Conclusion and Implication 83 5.1 Overview of Main Findings and Theoretical Implications 84 5.2 Managerial implications 87



xi

5.2.1 Implications for Banner Advertising 87 5.2.2 Implications for Online Marketing Research 89 5.3 Limitations and directions for further research 89 References 93 Samenvatting (Summary in Dutch) 113



xii



Introduction

1

Chapter 1 Introduction

1.1 General Introduction Since the new millennium, the Internet has become increasingly important to

business to consumer (B2C) markets. In 2014, the value of global B2C e-

commerce transactions reached $1.943 trillion and grew 24 percent from the

previous year (Ecommerce Foundation, 2015). Growth is predicted to continue.

Both the value of online transactions and the number of people buying online

are growing rapidly. In 2008, according to the Nielsen Global Online Survey,

the total (global) number of people shopping on the Internet was estimated to

be 875 million (marketingchart, 2008); the number grew to 1.139 billion in

2014 (Ecommerce Foundation, 2015). The proportion of online users who buy

online is also high. According to Mediascope Research (IAB Europe, 2012),

more than 85% of Internet users in Europe shopped online in 2012. With

current global Internet use penetration of three billion people (34 percent of the

world population) and an historical yearly growth average of more than 15

percent in the last 12 years (Internet World Stats, 2015), we can expect the

number of online consumers will continue to increase. As more and more

people buy online, it is not surprising that marketers and marketing academics

are paying more attention to the Internet (e.g. see Kotler and Keller, 2011;

Zikmund and Babin, 2009).

With its technological capabilities and features, the Internet is not only a

transaction channel, but also a medium for relaying marketing communications

and conducting marketing research. In the context of marketing

communications, the Internet has specific technological advantages over other

traditional media (i.e, newspapers, magazines, and television) to relay

marketing messages. Internet advertising can take many forms: text and image



Chapter 1

2

(similar to those in print media), sound and voice (similar to those in radio),

and video and animation (similar to those in television and on billboards).

Internet advertising also enables more interactivity by providing two-way

communication (Ko, Cho, and Roberts, 2005). Furthermore, by using Internet

Protocol (IP) and cookies technology, advertisers can measure audience

responses directly, at the individual level. They can better target audiences and

have more control over the direction of the advertising. Given all these

advantages, it is not surprising that expenditures on online advertising are

significant. In 2014, US non-mobile banner advertising generated $2.26 billion

and accounted for 16% of total Internet advertising (IAB, 2015).

In addition to its role as a marketing communications medium, the

Internet serves as a means of collecting customer information. Compared to

conventional marketing research media, a wider variety of marketing research

practices can be conducted by using the Internet. The Internet can facilitate

surveys (Couper, 2000), in-depth interviews (Pincott and Branthwaite, 2000),

focus group discussions (Montoya-Weiss et al., 1998), and ethnography studies

(Kozinets, 2003). With its ability to transmit text, images, and even videos in

real-time, the Internet can be used as a medium to implement a wide range of

market surveys. More importantly, Internet technologies such as cookies and IP

recognition make it possible to track and record consumer (online) behavior,

which in turn provides market researchers with both attitudinal and real-time

consumer behavior data. This fusion of attitudinal data and behavioral data,

enabled by Internet technology, has created new opportunities to address

marketing research questions and gain insights (Breur, 2011).

Research on how Internet technology affects consumer response to

marketing has been taking place since the introduction of the Internet.

However, both Internet marketing research and advertising instruments—and

the ways in which consumers react to them—are evolving, leading to gaps in

knowledge (Ha, 2008; Schibrowsky, Peltier, and Nill, 2007). This dissertation

aims to fill some of these gaps by developing a greater understanding of how

the Internet impacts online marketing and advertising research.



Introduction

3

1.2 Online Marketing and Advertising Research As a tool for observing consumer online behavior, the Internet has powerful

features such as IP recognition and cookies that record actions, expose

information, and identify devices used. These features have enabled companies

to practice effective customer relationship management (CRM). A number of

studies have been conducted to analyze and fully capitalize on the potential of

the Internet to improve these practices (for a review see, for example, Romano

and Fjermestad, 2003) and make optimal use of information gathered from

online consumer activities.

Traditionally, advertising research has focused on advertising

effectiveness. It has answered questions about how advertising works (ranging

from capturing attention to inducing action) (Vakratsas and Ambler, 2000) and

which metrics best measure the effectiveness of an advertisement. As

advertising research continues to build arguments and propose new theories, it

is becoming more important to study the context of online advertising.

Different types of media may have different effects on cognitive processes

(Sparrow, Liu, and Wegner, 2011). Although previous studies of online

advertising have extensively examined the effects of online advertising on

advertising and brand metrics (e.g., Chatterjee, 2008; Drèze and Hussherr,

2003; Havlena and Graham, 2004; Yoo, 2008), further studies are needed to

build a more comprehensive framework for studying the effectiveness of online

advertising.

Given the ability of Internet surveys to mimic traditional surveys in an

online environment, and the Internet’s abilities to include multimedia and real-

time answer screening, online surveys are used by marketing research agencies

worldwide (Evans and Mathur, 2005). The Internet makes surveying faster and

more economical (Couper, 2000). However, online surveys do suffer from

several weaknesses (e.g.. Fricker and Schonlau, 2002; Furrer and Sudharshan,

2001; Ilieva et al., 2002; Tingling et al., 2003; Wilson and Laskey, 2003),

including problems with validity, representativeness, and generalizability of the

responses. As the Internet spreads into more regions and permeates more social



Chapter 1

4

classes, issues of representativeness and generalizability arguably will become

less serious (Evans and Mathur, 2005). Furthermore, advanced Internet

technologies, such as IP recognition, geo-tagging, and Internet security

(Eysenbach, 2005) make it possible to implement online panels and screen

irrelevant or invalid responses. These technologies, combined with new survey

methodologies such as mixed mode analysis (Blyth, 2008) and complex

weighting procedures (Loosveldt and Sonck, 2008), may alleviate issues and

problems pertaining to online surveys (Callegaro et al., 2014). Further research

on online surveying is needed, particularly with regard to survey types that

have so far received little attention.

1.3 Aim and Objectives This dissertation focuses on three empirical studies, examining substantive

issues related to online marketing and advertising:

(i) Effect of banner exposures on consumer memory.

(ii) Advertisement and brand metrics for online banner advertising

effectiveness.

(iii) Potential of online polls as market research instruments. The main characteristics of these studies are shown in Table 1.1.

The following sections provide a brief discussion of theoretical

background and the contributions of the three studies to marketing literature.

1.3.1 Study 1: Effect of Banner Exposures on Consumer Memory Consumer memory is an important factor in marketing. Consumer decisions are

affected by consumer memory (see Hoyer and MacInnis, 2008, p. 174-176).

Brand memory is an important metric for companies; advertising is one method

of improving it. Many studies have been conducted to study the effectiveness

of advertising in improving brand memory (see e.g., Fennis and Stroebe, 2010,

Ch. 3).



Introduction

5

Table 1.1 Characteristics of the studies in this dissertation Study 1 Study 2 Study 3

Subject Effect of banner exposures on consumer memory

Advertisement and brand metrics for banner advertising effectiveness

Potential of online polls as market research instrument

Data source - Survey data - Internet activity record (cookies data)

- Survey data - Internet activity record

(cookies data)

- Survey data - Company database

Sample n = 8736 21 campaigns

n = 19994 29 campaigns

n = 250 (telephone) n = 1018 (website poll) n = 80000 (company database)

Analysis Method Multilevel Mediation Analysis

Multilevel Structural Equation Modeling

Population means comparison, ANOVA

Recent psychological research (Sparrow, Liu, and Wegner, 2011)

indicates the Internet affects the cognitive processes and memories of Internet

users (see also Carr, 2010). This new understanding demands that we revisit

traditional media theories of how advertising affects consumer memory.

A number of studies have demonstrated that the effectiveness of an

advertising campaign is affected by the campaign’s schedule (e.g.

Zielske, 1959; Heflin and Haygood, 1985; Janiszewski, Noel, and Sawyer,

2003). Campaign schedules are characterized by the number of exposures, time

interval between exposures (spacing), and elapsed time since last exposure

(delay). Several studies have investigated the effects of number of exposures,

spacing, and delay in traditional media, but research on the effects of campaign

schedules in Internet advertising is scant. Previous research (Yoo, 2008; Yoo

and Kim, 2005) has indicated different forms of Internet advertising may have

different effects on consumer memory. Because banner advertising is

ubiquitous, and claims a significant share of the overall online advertising

budget (16% in 2014) (IAB, 2015), Study 1 focuses on banner advertising. The

research question addressed in Study 1 (Chapter 2) is:



Chapter 1

6

“What is the impact of banner advertising number of exposures,

spacing, and delay on consumer memory, in terms of banner and

brand awareness?”

This study makes several contributions to literature. First, in the context of

banner advertising, it is the first to examine the simultaneous effects of number

of exposures, spacing, and delay of banner exposures. Second, it measures

banner and brand memory by recall and recognition. By using these measures,

a more comprehensive hypothesized process can be analyzed. Insights from the

analyses contribute to literature by examining whether and how banner

exposure works to enhance consumer memory.

1.3.2 Study 2: Advertisement and Brand Metrics for Online Banner

Advertising Effectiveness In advertising and marketing communications literature, the hierarchy of

effects is widely used to explain the consumer’s psychological progression

from an initial state of unawareness about a brand to eventual purchase of that

brand (Shimp, 2008). In the context of banner advertising, researchers and

advertisers are interested in the effect of banner exposures on consumer

memory, attitudes, and behavior.

A number of studies of banner advertising have investigated the effects

of banner exposures on banner and brand awareness, banner and brand

attitudes, and purchase intentions (e.g., Chatterjee, 2008; Drèze and Hussherr,

2003; Havlena and Graham, 2004; Yoo, 2008). However, these studies focus

on only one or a few metrics of banner effectiveness. Study 2 aims to

determine the interrelationships among various effect measures to develop a

comprehensive model that links banner memory and banner attitude to brand

memory and brand attitude, and ultimately to purchase intention. Study 2 poses

the following questions:



Introduction

7

“How are advertising and brand metrics interrelated in banner

advertising? How do these interrelationships differ from those in

traditional media advertising?”

By answering these questions, this study makes three key contributions to

advertising literature. First, in contrast to previous banner advertising research

that has investigated separate relations between banner advertising and

individual metrics, this study constructs and tests a comprehensive model

linking banner memory and attitude to brand memory and attitude, and

ultimately to brand purchase intentions. Second, in the framework of structural

equation modeling, the study tests an alternative model of relationships that

have been established for traditional advertising. By analyzing the model

fitness and significant relationships among the metrics in both models, a

comparison can be made of how the advertising ‘hierarchy of effects’ takes

shape in both banner and traditional advertising.

1.3.3 Study 3: Potential of Online Polls as Market Research

Instruments Online surveying began almost as soon as it was apparent that the Internet was

a tool for conducting research (Comley, 1996). Online surveys take many

forms and are administered using different methods. Couper (2000)

summarized varieties of online surveys and concluded that while some forms

of online surveys might produce quality responses, others were deemed to be

unscientific because of their non-probabilistic sampling nature, resulting in

unrepresentative samples of the general population. However, in some

marketing research settings the general population is not relevant; customer

segments are only part of the general population. Furthermore, a number of

marketing research companies use non-probabilistic samples for their online

surveys (Evans and Mathur, 2005). In fact, the panel surveys in Study 1 and

Study 2 of this dissertation are not (technically) random samples from the

population. This situation and the fact that early studies in non-probabilistic



Chapter 1

8

surveys (e.g. Smith, 2001; Yoshimura, 2004) have indicated opportunities to

rectify the problem of non-coverage in non-probabilistic online surveys raise

the question of whether simple surveys—“online polls”—on a company

website have merit as market research instruments. Hence, the main research

questions that will be addressed in Study 3 (Chapter 4) are:

“How different are demographic and response characteristics of

online polls and conventional survey respondents? What are the

sources of and solutions for the differences?”

This study will contribute to literature by being the first to provide results on

the value and usability of self-selected non-probabilistic online surveys, such as

online polls, for marketing research purposes.

1.4 Outline of the Dissertation

This dissertation contributes to existing literature on marketing by providing

new insights on how the Internet impacts marketing communications and

market research practices.

In Chapter 2 (Study 1), we demonstrate how banner exposure

characteristics affect consumer memory. In Chapter 3 (Study 2), we propose a

model relating advertisement and brand metrics for online banner advertising

in a hierarchy of effects context. In Chapter 4 (Study 3), we discuss the

potential of online polls as market research instruments. Finally, in Chapter 5

we summarize the main conclusions of this dissertation, provide a general

discussion of the findings, and discuss further research avenues.



The Effect of Banner Exposures on Consumer Memory

9

Chapter 2 The Effect of Banner Exposures on Consumer Memory

2.1 Introduction

The scheduling of Internet advertising has become an important topic in marketing: in

2103, total spending on Internet advertising finally surpassed total spending on

newspaper and other print media advertising (McKinsey, 2014). Apart from this,

compared to traditional media, Internet advertising, using some Internet technologies,

has more control on to whom, when, and where an advertising to be exposed. In

traditional media, advertising scheduling has long been an important issue and has

been well studied by academics. Although research clearly shows different advertising

schedules lead to different degrees of advertising effectiveness (e.g., recall patterns)

(Zielske and Henry, 1980), there is less clarity about optimal scheduling. For example,

Strong (1977) suggests alternating between high and low exposure rates to achieve

maximum awareness, while Zielske and Henry (1980) suggest massing the exposures

at the beginning of a campaign. Other researchers (e.g., Mahajan and Muller, 1986;

Mesak, 2002; Sasieni, 1989) propose models to analyze various advertising schedule

strategies. These models require input on the effect of the number of exposures, time

interval between exposures (spacing), and elapsed time since last exposure

(measurement delay) to predict the effectiveness of a particular advertising scheduling

strategy.

Many studies of traditional media have investigated the effect of exposure

characteristics (number of exposures, spacing, and delay) but research remains scant

in the area of Internet advertising. For instance, although there are multiple studies on

the effect of Internet advertising number of exposures (e.g., Broussard, 2000; Drèze

and Hussherr, 2003; Yaveroglu and Donthu, 2008), only a few focus on the effect of



Chapter 2

10

spacing and measurement delay in Internet advertising. These include Fang et al.

(2007) on the effect of spacing on brand attitude and Havlena and Graham (2004) on

the effect of measurement delay on brand awareness and attitude. To the best of our

knowledge, no study has investigated the effects of number of exposures, spacing, and

delay on advertising and brand awareness, simultaneously, in a single integrated

Internet advertising model. This gap must be addressed because research in traditional

media has shown the effects of exposure characteristics differ across media due to

differences in the level of attention being paid during advertising exposures (see

Fennis and Stroebe, 2010, p. 42–43). For example, Dijkstra et al. (2005) found

television advertising performed better in evoking cognitive responses than print

advertising. Consumers pay even less attention to Internet advertising than to print

advertising; they consciously avoid it (Drèze and Hussherr, 2003; Dahlen, 2001). This

may impact the effect of number of exposures, spacing, and delay.

Banner advertising accounts for a significant share of Internet advertising

spending. In 2014, U.S. non-mobile banner advertising generated $2.26 billion and

accounted for 16% of total Internet advertising (IAB, 2015). In addition to

encouraging click-through, the purpose of banner advertising is to improve brand

performance metrics such as brand awareness. When brand awareness levels are

higher, consumers exposed to banner advertisements will be able to recall or

recognize the advertised brand when making a purchase decision. Along with click-

through rate, enhanced consumer memory is an important measure of banner

effectiveness (Briggs and Hollis, 1997).

The overall objective of our study is to reveal how the various characteristics of

banner exposures (number of exposures, spacing, and measurement delay) affect

consumer memory of the banner and the brand. By determining these effects,

advertisers can estimate the effectiveness of various banner-based advertising

schedules. Using data from 21 actual banner campaigns involving almost 9,000

consumers, we find the effect of number of exposures on consumer memory for both

the banner and the brand is initially positive, but becomes less so as the number

increases. The positive effect of number of exposures is greater if there is a moderate

interval between exposures, but enhanced memory resulting from banner exposure




11

eventually decays. This finding suggests continuous advertising with even spacing is

the best advertising scheduling strategy for banner advertising. Furthermore, although

we find banner exposures have similar effects on consumer memory as advertising

exposures in traditional media, there is an important difference in how the effect is

achieved. In banner advertising, the effect of exposures on brand memory is indirect

and mediated by banner memory. This finding implies it is crucial that consumers

remember the actual banner advertisements.

In the following section, we present a literature background and our research

hypotheses. Next, we describe the research design, including data collection, model

formulation, and data analysis. We follow with results and conclude with a discussion,

practical implications, limitations, and future research.

2.2 Literature Background and Hypotheses

2.2.1 Advertisement Scheduling Advertisers use several strategies for scheduling advertisements (Mahajan and Muller,

1986). Three commonly applied strategies are:

1. Blitz, in which most or all advertisements are concentrated in a short period at the

beginning of a campaign;

2. Pulsing, in which advertisements are broadcast or published by alternating very

high intensity with and low or zero intensity during the campaign;

3. Even, in which advertisement exposures are held almost constant during the

campaign.

Within a particular campaign time period, these strategies can be characterized

by three variables: (i) number of exposures, (ii) average time interval between

exposures (spacing), and (iii) time between last exposure and a consumer action (e.g.,

related to the purchase cycle). For instance, a blitz campaign is characterized by a high

frequency at the beginning and a very low or even zero frequency after a short period

of time, a very low average time interval between exposures, and a large distance

between the last exposure and the end of the current campaign time (because most of

the advertising budget is spent at the beginning of the campaign). In contrast, an even



Chapter 2

12

campaign is characterized by a moderate interval time between exposures and

continuous exposure during a particular campaign period. Knowledge of the effect of

the various exposure variables is needed to predict the consequences of various banner

scheduling strategies on banner effectiveness.

In the next sections, we discuss findings from psychology, marketing, and

advertising literature on the effects of number of exposures, spacing, and

measurement delay on consumer memory for both banner and brand.

2.2.2 Effect of Number of Exposures Through repeated exposures to advertisements, consumers can rehearse information

about the advertisements. Rehearsal may improve information storage to, and recall

from, long-term memory (Dark and Loftus, 1976). Advertisement repetition may

enhance advertising and brand awareness; experimental and field advertising studies

have shown the number of exposures affects advertising and brand awareness

(Pechmann and Stewart, 1989). When attention to advertising is low, the first few

exposures may not improve awareness. Advertising effects build over time, following

a wear-in process (Blair, 2000). However, the build-up effect of repetition decreases

and may become negative after a certain number of exposures, a phenomenon known

as the wear-out effect (see Pechmann and Stewart, 1989). These possible wear-in and

wear-out effects underscore the importance of understanding the relationship between

the number of exposures and measures of advertising effectiveness.

The issue of the number of exposures effectiveness is even more complicated

for banner advertising because banners are likely to be overlooked by website visitors.

In an eye-tracking study, Drèze and Hussherr (2003) show Internet users avoid

looking at banners. Hoffman and Novak (1996) find experienced Internet users,

compared with novice Internet users, focus more on the main content of a webpage

and are less susceptible to the influence of banners. Dahlen (2001) confirms banners

are more effective for novice Internet users.

Given this situation, it is not surprising there are mixed findings in studies on

the effect of exposures for banners. Although Drèze and Hussherr (2003) and Havlena

and Graham (2004) find a positive effect of exposures on banner and brand awareness,




13

Yoo (2008) does not; however, he does find positive effects on brand attitude and

brand consideration. He attributes these effects to implicit memory resulting from the

exposures. In his experiment, Yoo (2008) used only one exposure rather than multiple

exposures; if there are positive effects of exposures to banners, a single exposure may

not be strong enough to be explicitly recorded in memory. With higher frequencies of

exposures, the wear-in effect (as discussed above) may be strong enough to cause

banners and advertised brands to be remembered. Therefore, our first hypothesis is:

Hypothesis 1: The number of banner exposures has a positive effect on a) banner

memory and b) brand memory.

2.2.3 Effect of Spacing A given number of exposures can be spread in several ways during the campaign

period. Different time intervals are expected to have different memory effects. For

example, extensive research in verbal learning has found providing a longer time

interval between exposures to a stimulus can lead to memory enhancement

(Janiszewski et al., 2003), an effect known as spacing (Noel and Vallen, 2009; Sawyer

et al., 2009).

Several theories have been proposed to explain the cognitive mechanism

behind the spacing effect (Appleton-Knapp et al., 2005; Janiszewski et al, 2003; Noel

and Vallen, 2009). Two of the most frequently cited mechanisms are encoding

variability (D’Agostino & DeRemer, 1973) and the study-phase retrieval process

(Thios & D’Agostino, 1976). Encoding variability theory posits that subsequent recall

can be enhanced if there is variation in how information is encoded from multiple

exposures to the stimulus. Differences in encoded information provide manifold

routes for recall. Because of the change of context for the information, spacing

ensures the stimulus encountered in the subsequent exposure will be encoded

differently from that of the previous exposure. This mechanism implies the greater the

time interval between exposures, the more impact on memory gained from the

subsequent exposure (Appleton-Knapp et al., 2005).



Chapter 2

14

Alternatively, the study-phase retrieval process assumes that subsequent

exposure to a stimulus enables information retrieval about that stimulus recorded in

memory from the previous exposure. The retrieval itself is another learning process, in

that the information being retrieved is more easily recalled in the future. Furthermore,

future recall is greater when the level of difficulty in retrieving the information from

the previous exposure is higher. Because longer spacing between two exposures

makes it more difficult to retrieve information from the first exposure, longer spacing

results in enhanced memory recall. However, this longer interval is limited to a point

where such retrieval starts to fail (Appleton-Knapp et al., 2005). Optimal spacing

occurs when the interval between two exposures is long enough to make the retrieval

process not too easy and not too difficult. Therefore, the study-phase retrieval process

predicts an inverted U-shape effect of exposure spacing on recall.

Early research (Singh et al., 1994; Unnava and Burnkrant, 1991) proposed that

encoding variability is the dominant mechanism of the spacing effect in advertising.

However, more recent research (Appleton-Knapp et al., 2005; Janiszewski et al.,

2003) provides more support for the study-phase retrieval process. Heflin and

Haygood (1985) found inverted U-shape of recall and recognition response curves

against interval length between exposures. These findings seem to support the study-

phase retrieval process. Hence, we predict the effect of spacing will follow an inverted

U-shape, with spacing has the greatest effect when the interval between exposures is

moderate.

Hypothesis 2: The effect of spacing on a) banner memory and b) brand memory

follows an inverted U-shape.

2.2.4 Effect of Measurement Delay We do not expect the effect of an advertising exposure will be everlasting. Consumers

will forget advertising over time (Hutchinson and Moore, 1984). Therefore, the time

interval between exposures and advertising effectiveness measurement (i.e., the

measurement delay), directly influences the estimated advertising effectiveness. It is




15

crucial to take this effect into account in situations in which the measurement delay

varies between consumers, as is the case with banner advertising.

Consumer forgetting occurs when associations between pieces of information

in memory decay over time (Warlop et al., 2005). Several studies of print and

television advertising have demonstrated that consumer forgetting in print and

television advertising is a function of time (e.g., Gregan-Paxton and Loken, 1997).

The studies suggest the decline of memory of an advertisement and its message is

inevitable, due to the time effect and the interference effect of advertising clutter

(Burke and Srull, 1988). Only one study of banner advertising has specifically

investigated the effect of measurement delay on advertising and brand recognition.

Havlena and Graham (2004) analyzed online advertising in three different industries;

they found no significant negative effect of measurement delay on advertising and

brand recognition. Hence, in contrast to findings in print and television advertising,

Havlena and Graham (2004) conclude “the short duration of time since last exposure

to brand measurement is an issue, but not a serious one” (p. 331).

Despite the surprising finding of Havlena and Graham (2004), we expect a

negative effect of measurement delay. We base our prediction on both the extensive

evidence of this effect in traditional media and its theoretical logic. We formulate our

hypothesis as:

Hypothesis 3:Measurement delay has a negative impact on a) banner memory and b)

brand memory.

2.2.5 Mediating Effect of Banner Memory Human memory can be represented as a complex associative network. Memories for

components of an advertisement (e.g., logo, images, music, or endorsing figures),

information from the advertising (e.g., brand claims, or brand positioning in a product

category), and the brand itself are interrelated. Information about the advertising and

the brand is encoded in memory as a pattern of links between concept nodes (Burke

and Srull, 1988). When a consumer is exposed to a stimulus, parts of the associative

network are activated. The more frequently a node is activated, the more salient other



Chapter 2

16

nodes within the associative network in the consumer mind, connected with the

activated node, become (du Plessis, 2008). The subsequently activated node in the

associative network could be brand name. When an advertisement contains a brand

name, logo, and product image representing the product category, all of the

components and their relationships become more accessible for future reference.

When a consumer is asked to name a soft drink brand and when the name of the brand

or the product (for example, the beverage bottle) is absent, the consumer will search in

memory for a soft drink brand name. During this process, a recollection of personal

experience or others’ experiences (through advertising) triggers the memory of the

brand name. In this way, memory of the product advertisement mediates memory of

the brand.

Many laboratory studies that measure the effect of advertising exposures on

brand memory require subjects to view advertising. Pechman and Stewart (1989)

characterize these studies as massed and forced exposures. With massed and forced

exposure in a controlled environment, any changes in brand memory can be attributed

to the advertising exposures. If there is a significant difference in awareness of a

brand, particularly a hypothetical brand, this difference must be caused by the

advertising exposures. However, the situation is more complicated in banner

advertising studies based on actual campaigns. A person may be presented with a

banner on a webpage but may not see the banner; perhaps the visitor does not scroll

down the webpage or is too absorbed with the article or other stimuli on the webpage

to look at the advertisement (Drèze and Hussherr, 2003; Dahlen, 2001).

When there is no information available about whether a respondent has actually

looked at the banner, memory for the banner can serve as a cue that the respondent has

previously looked at and, to some degree, consciously processed the banner. When

respondents recognize they have seen a banner (perhaps several times), the effect of

banner exposure on brand memory is also more likely to occur. Hence, we formulate

our hypothesis as:

Hypothesis 4: The effects of (a) number of exposures, ((b) spacing, and (c) measurement delay on brand memory are mediated by banner memory.




17

2.2.6 Differential Effect of Exposure Variables Measuring memory traces of information from advertising is a complex process (see,

for example Baddeley, 1997); the information is not always easily retrievable. There is

no universal way of measuring consumer memories of advertising. Detection of

memory traces depends on using a memory test with external cues (recognition) rather

than a memory test without external cues (recall), because the presence of cues may

help the information retrieval process (Lynch and Srull, 1982; Keller, 1987). It is

important to know the differential effect of exposure variables on recall and

recognition, because both may be relevant, depending on the situation faced by a

consumer when making a buying decision (Singh et al., 1988). In a situation where a

stimulus is presented (for example, in a supermarket) during the decision-making,

recognition is more relevant. When the decision is made at home in the absence of the

stimulus, recall is more relevant. In this context, a pertinent research question is to

what extent advertising exposure variables differ in affecting recall and recognition.

In the context of brand awareness, Laurent et al. (1995) demonstrated that

recognition and recall measure the same underlying construct: brand salience; the

score difference was the result of task difficulty differences. The threshold for brand

salience is higher for a person to recall than to recognize the brand. As a result,

recognition scores tend to be higher than recall scores (du Plessis, 1994; Laurent et al.,

1995).

With regard to the number of exposures, according to the two-stage learning

model (Pechmann and Stewart, 1989), multiple exposures are needed to allow

detection of advertising traces in consumer memory through either recall or

recognition tasks. Since recall tasks are more difficult than recognition tasks, we

expect the number of exposures needed to perform a recall task is greater than the

number needed for a recognition task. In other words, individuals subject to the same

number of exposures are more likely to recognize advertisements or brands than they

are to recall them.

With respect to measurement delay, both recognition and recall suffer from the

memory decay of brand salience. However, because recognition is easier, it has a



Chapter 2

18

lower threshold than recall for successful performance. As time progresses and brand

salience decays, the salience will drop below the threshold for recall, but remain

above the threshold level for recognition. With regard to spacing effect, it is not easily

determined whether it is stronger for recall or for recognition; the outcome depends on

the underlying mechanism creating the spacing effect.

We do not formulate explicit hypotheses for the differential effects of number

of exposures, spacing, and measurement delay on recall versus recognition because

the effect sizes will depend on whether certain thresholds are reached. Instead, we

empirically explore the differences between recall and recognition effect measures.

All hypothesized relationships above are depicted in Figure 2.1. 2.2.7 Other Factors Affecting Consumer Memory of Banner and Brand

In addition to being affected by exposure characteristics, consumer memory for

advertising or a brand is also affected by individual characteristics such as age and

gender (Krishnan and Chakravarti, 1999). Older consumers have more difficulty

accessing their memory (Burke and Light, 1981). Females may have an advantage in

performing recall and recognition tasks, especially for stimuli with lower exposure

levels, due to their lower encoding threshold (Meyers-Levy and Sternthal, 1991).

Previous experience with the brand is also an influencing factor: such

experience conveys brand familiarity, which in turn affects memory of the brand

(Campbell and Keller, 2003). Users of the brand are likely better able to recall or

recognize the brand and the advertising of the brand. We do not formulate specific

hypotheses regarding these factors. Instead, we use them as control variables and to

check face validity of the model results.

In addition to consumer factors, banner memory and brand memory also

depend on characteristics of the banner and the brand. However, most of these factors

are not observable; because of their influence, each observation within a campaign is

not independent from other observations of the same campaign. We therefore use

multilevel analysis to cope with the issue of non-independent observations. We




19

Figure 2.1. Diagrammatic Representation of Hypotheses

include ‘brand popularity’ and ‘product category’ as control variables for recall and

recognition in our model at the campaign level, but we do not propose any specific

hypotheses.

2.3 Data Collection and Analysis 2.3.1 Data Collection and Description The dataset used for this study is part of a database of Advertising Campaign

Evaluation (ACE) studies conducted by MetrixLab, a global online market research

company with operations covering more than 44 countries worldwide. ACE evaluates

the impact of several aspects of a campaign on branding. Our study includes data from

21 different banner campaigns conducted in seven countries in Europe and Latin

America; respondents were recruited randomly from websites featuring campaign

banners. There is a total of 8,736 respondents, with sample sizes per campaign ranging

from 47 to 1,272. Each respondent was exposed at least twice to the banners under

study, meeting the minimum number of exposures required to measure exposure

characteristics.

Using proprietary tagging technology, MetrixLab collects two types of data

during an ACE study. The first type is exposure-related data, a data series that records

- Usage- Gender- Age- Brand Popularity- Product Category

BANNER MEMORY(Recall/Recognition)

BRAND MEMORY(Recall/Recognition)

EXPOSURECHARACTERISTICS- Frequency- Spacing- Delay

- Usage- Gender- Age- Brand Popularity- Product Category



Chapter 2

20

the time at which the respondent is exposed to a particular banner. The data are

recorded by means of a cookie and extracted at the time of the survey. From these

data, we can derive the variables of number of exposures, average time between

exposures, and time between the survey and last banner exposure. The second type of

data is drawn from survey questions, completed by respondents, that relate to the

campaign and the brand; they gather information about banner and brand awareness as

well as demographics.

In the survey, respondents are asked about one brand in a product category.

They are then questioned about their awareness of all other brands in the category.

Next, they are asked to identify media in which they have seen the brand advertised.

This procedure measures whether respondents are able to recall the banner as a form

of advertising (in addition to television or printed advertisements, for example) of the

brand. They are then shown the banner to which (according to the cookie) they have

been exposed, and asked whether they recognize it. Finally, respondents answer

questions related to their usage of the brand and their demographics. Table 2.1

provides a summary of the descriptive statistics and survey questions.

2.3.2 Model Formulation and Analysis We employ a structural equation model to test the relationships proposed in

Hypotheses 1–4 of our research framework (Figure 2.1). Iaobucci et al. (2007) shows

that a structural equation model (SEM) approach performs better in analyzing

mediated relationships than the classical Baron and Kenny (1986) approach, which

uses a set of independent regression models. In addition, because banner memory and

brand memory are measured as binary variables, the test of significance of the indirect

effects using the classical mediation analysis is more complicated than if the variables

are measured using metric scales (MacKinnon and Dwyer, 1993). This complexity is

amplified by the hierarchical nature of the data. In recent years, new developments in

the application of SEM methodology have enabled researchers to assess mediation in

multilevel data (Preacher et al., 2010) by using supporting software (Muthén and

Muthén, 2010). Hence, based on these theoretical and practical considerations we use

the SEM approach.



T

able

2.1

. Mea

sure

men

ts a

nd D

escr

iptiv

e St

atis

tics

V

aria

ble

N

Mea

n St

d. D

ev.

Skew

ness

K

urto

sis

Mea

sure

s/Q

uest

ions

Dep

ende

nt v

aria

bles

B

anne

r rec

all (

bina

ry)

8682

.1

9 .3

9 -

- W

here

, if a

t all,

hav

e yo

u se

en, h

eard

, or r

ead

any

adve

rtisi

ng fo

r [br

and]

rece

ntly

? B

rand

reca

ll (b

inar

y)

8615

.6

4 .4

8 -

- W

hich

bra

nds o

f [pr

oduc

t cat

egor

y] y

ou c

an y

ou

thin

k of

? B

anne

r rec

ogni

tion

(bin

ary)

80

75

.29

.45

- -

Plea

se h

ave

a lo

ok a

t the

follo

win

g ad

verti

sing

[a

bann

er is

show

n]. D

o yo

u re

call

seei

ng a

n ad

verti

sem

ent s

imila

r to

the

one

show

n ab

ove

on

the

Inte

rnet

? B

rand

reco

gniti

on (b

inar

y)

6916

.8

0 .4

0 -

- H

ave

you

hear

d of

any

of t

he fo

llow

ing

[pro

duct

ca

tego

ry] b

rand

s?

In

depe

nden

t var

iabl

es

Expo

sure

s (co

unt)

8736

8.

22

9.22

3.

120

13.2

21

Num

ber o

f ban

ner e

xpos

ures

, rec

orde

d in

the

cook

ie fi

le

Spac

ing

inte

rval

(min

utes

) 87

36

2848

.10

6542

.67

12.7

42

329.

259

Ave

rage

inte

rval

bet

wee

n ex

posu

res,

reco

rded

in

the

cook

ie fi

le

Mea

sure

men

t del

ay

(min

utes

) 87

36

6724

.78

1321

5.92

3.

238

13.9

03

Tim

e el

apse

d be

twee

n la

st e

xpos

ure

and

the

surv

ey, r

ecor

ded

in th

e co

okie

file

Pa

st u

sage

(bin

ary,

1 =

use

r)

8736

.3

1 0.

46

- -

Hav

e yo

u ev

er u

sed

[bra

nd]?

A

ge (y

ears

) 87

36

39.8

0 12

.75

.036

-.7

93

How

old

are

you

? G

ende

r (bi

nary

, 1=

fem

ale)

87

36

0.59

0.

49

- -

Wha

t is y

our g

ende

r?



Chapter 2

22

We estimate two models. The first model is designed to study the effect of

banner exposure variables on consumer memory, measured by banner and brand

recall. The second model determines memory by banner and brand recognition. Our

structural models can be represented by the following equations:

(1)

(2)

(3)

(4)

yadx : banner recall or banner recognition (x is to be replaced by either recall or

recognition)

ybrx : brand recall or brand recognition (x is to be replaced by either recall or

recognition)

expos : number of banner exposures (in log)

delay : delay between last exposure and survey time (in log of minutes)

spacing : average time between exposures (in log of minutes)

usage : dummy variable indicating whether the respondent has experience with the

brand

gender : dummy variable indicating whether the respondent is female (1) or male (0)

age : age of the respondent

brandpop : brand popularity, measured by average recognition of the brand

cat1 : dummy variable in which cat1 = 1, indicating the brand is a nondurable

product

cat2 : dummy variable in which cat2 = 1, indicating the brand is a durable

product, hence cat1 =0 and cat2 = 0, indicating the brand is of a service

i,j : subscripts for individual i in campaign j




23

Equations (1) and (2) capture the effect of the exposure variables and the covariates

on banner memory; equations (3) and (4) capture those effects for brand memory.

Parameters β1 and β2 capture the direct effects of number of exposures and of

measurement delay on banner and brand memory, and β3 and β4 together capture the

spacing effect: if β4 is negative, the hypothesized inverted U-shape from Hypothesis 4

is supported. Parameter β8 captures the direct effect of banner memory on brand

memory. Hence, the indirect effects of the exposure variables on brand memory (the

effect mediated by banner memory) are β1,2,3,4 x β8. Furthermore, β0j depicts the base

of banner memory and brand memory for campaign j, which are affected by brand

popularity and brand product category. The heterogeneity of the bases for each

campaign is captured by u0j. The model is estimated using Bayesian methods with

Mplus software (Muthén and Muthén, 2010).

2.4 Results Results from the model estimations are presented in Table 2.2. A summary of the

results is depicted in Figure 2.2.

Effect of number of exposures. We find highly significant direct effects of number of

banner exposures on banner recall ( = 0.078, p < 0.001) and banner recognition ( =

0.035, p < 0.040). Because the model contains the log of number of exposures, the

implication is that the effect diminishes as the number increases. Nevertheless, the

relationship curve is monotonic; we do not observe the phenomenon of wear-out for

banner advertising. Figure 2.3 depicts the effects of varying number of exposures on

the probability to recall and recognize the banner, assuming all other variables to be

constant.

The direct effects of number of exposures are not significant on brand recall

( = -0.018, p = 0.187) and brand recognition ( = 0.013, p = 0.314). Hence, with

regard to Hypothesis 1, we conclude banner exposures have a direct and positive

impact on banner memory (H1a), but a direct effect on brand memory is not observed

(H1b).



Chapter 2

24

(a) Recall Model

(b) Recognition Model

Note: Figures in parentheses are posterior standard deviations

Figure 2.2 Summary of Model Estimation Result

The absence of a direct effect of number of exposures on brand memory leads

to the question of whether the effect could be indirect and mediated by banner

memory (Hypothesis 4a). In Table 2.2, we see the effect of number of exposures on

brand memory is mediated by banner memory ( 1 x 8) and is significantly positive

(with p < 0.001 for recall and p < 0.05 for recognition). This significant indirect effect

and the non-significant direct effect of number of exposures on brand memory shows

the effect of exposures on brand memory is fully mediated by banner memory,

thereby confirming Hypothesis 4.

Banner Recall

Brand Recall

Exposure Variables- Frequency (F)- Spacing (S)- Spacing squared (S2)- Measurement delay (D)

- F : .078 (.022)- S : .026 (.021)- S2: -.003 (.002)- D : -.030 (.006)

.135 (.024)

- F : .078 (.022)- S : .026 (.021)- S2: -.003 (.002)- D : -.030 (.006)

Banner Recognition

Brand Recognition

Exposure Variables- Frequency (F)- Spacing (S)- Spacing squared (S2)- Measurement delay (D)

- F : .035 (.020)- S : .033 (.019)- S2: -.004 (.002)- D : -.022 (.006)

.079 (.029)

- F : .013 (.026)- S : -.020 (.024)- S2: -.002 (.003)- D : -.006 (.007)




25

Figure 2.3. Effect of Number of Exposures on Probability to recall and Recognize the Banner

Effect of spacing. The spacing effect is captured by the parameters β3 and β4, where

β4 is the parameter for the squared time interval between exposures. A significant

positive β3 and a significant negative β4 imply the effect of spacing has an inverted-U

shape, as proposed by Hypothesis 2. For banner recognition, both β3 ( 3 = 0.033, p =

0.044) and β4 ( 4 = -0.004, p = 0.036) are significant at the 0.05 level. Figure 2.4

depicts the effect of varying spacing intervals on probability to recognize the banner

when the effect of other variables is held constant. We see there is a peak on the

curve. This peak is reached when spacing is about 62 minutes.

The direct effect of spacing is insignificant for brand recognition. However,

further analysis shows Hypothesis 4a is supported by the data; there is a significant

indirect effect of spacing on brand recognition ( 3 x 8= 0.003, p = 0.047 and 4 x

8= -0.0004, p = 0.039). These findings indicate banner spacing also affects brand

recognition, but in an indirect manner.

However, the spacing effect seems to be lacking for recall. Both β3 and β4 are

insignificant for banner recall as well as for brand recall, implying Hypothesis 2 is

rejected for recall. As a consequence of a lack of direct effect of spacing on banner

recall, the hypothesis that the spacing effect on brand recall is mediated by banner

recall is also rejected (Hypothesis 4b).

0,15

0,17

0,19

0,21

0,23

0,25

0,27

0,29

0,31

0,33

1 6 11 16 21 26 31 36 41 46

Prob

abili

ty

Frequency

Recognition

Recall



Tab

le 2

.2. P

aram

eter

Est

imat

es fo

r M

odel

s

Dep

ende

nt v

aria

ble

Inde

pend

ent v

aria

ble

Ban

ner r

ecal

l

Bra

nd re

call

Ban

ner r

ecog

nitio

n B

rand

reco

gniti

on

Dir

ect e

ffect

s

� 1 (n

umbe

r of e

xpos

ures

) .0

78

(.022

)***

-.018

(.0

33)

.035

(.0

20)*

* .0

13

(.026

) � 2

(mea

sure

men

t del

ay)

-.030

(.0

06)*

**

-.0

02

(.002

) -.0

22

(.006

)***

-.0

06

(.007

) � 3

(spa

cing

- lin

ear)

.0

26

(.020

)*

-.0

03

(.003

) .0

33

(.018

)**

-.020

(.0

24)

� 4 (s

paci

ng -

quad

ratic

) -.0

03

(.002

)*

.0

01

(.100

) -.0

04

(.002

)**

.002

(.0

03)

� 5 (u

sage

) .4

16

(.037

)***

.832

(.0

96)*

**

.191

(.0

35)*

**

1.05

5 (.0

59)*

* � 6

(gen

der)

-.1

57

(.055

)***

-.261

(.0

04)*

**

-.252

(.0

55)*

**

-.072

(.0

72)

� 7 (a

ge)

-.016

(.0

01)*

**

-.0

02

(.001

)*

-.007

(.0

01)*

**

-.008

(.0

02)*

**

� 8 (b

anne

r rec

all)

--

.135

(.0

24)*

**

--

.0

79

(.012

)***

� 00 (in

terc

ept)

.099

(.3

21)

-1

.257

(.3

94)*

**

-.309

(.1

98)*

* -.2

45

(.717

) � 0

1 (b

rand

pop

ular

ity)

.7

78

(.430

)*

.2

32

(.969

) .4

66

(.356

)*

2.48

3 (.9

63)*

**

� 02 (d

umm

y –

non-

dura

ble

) -.4

30

(.281

)**

-.4

90

(.623

) -.5

71

(.218

)***

.6

68

(.552

) � 0

3 (d

umm

y - d

urab

le)

-.050

(.3

38)

-1

.501

(.6

60)*

**

-.397

(.2

40)*

* -.0

17

(.779

) u 0

j .1

73

(.088

)***

.888

(.0

88)*

**

.148

(.0

74)*

* .7

30

(.685

)

Indi

rect

effe

cts

� 1x � 8

(num

ber o

f exp

osur

es)

--

.010

(.0

04)*

**

--

.0

03

(.002

)**

� 2 x

�8 (m

easu

rem

ent d

elay

)

-

-

-.0

04

(.001

)***

--

-.0

02

(.001

)***

� 3 x

�8 (s

paci

ng -

linea

r)

- -

-.003

(.0

03)

--

-.002

6 (.0

015)

**

� 4 x

�8 (s

paci

ng -

quad

ratic

)

--

-.0

01

(.001

)

--

-.0

004

(.000

2)**

N

ote:

O

ne-ta

iled

test

ing

* p-

valu

e <

0.1

** p

-val

ue <

0.0

5 **

* p-

valu

e <

0.01




27

Note: spacing has only marginal effect on banner recall, therefore not depicted in the graph. Figure 2.4. Effect of Spacing on Probability to Recognize the Banner

Effect of delay in measurement. We find a significant negative direct effect of

measurement delay on banner memory ( 2 = -0.030, p = 0.006 for banner

recall; 2 = -0.022, p = 0.006 for banner recognition), but not on brand memory

( 2 = -0.002, p = 0.002 for brand recall; 2 = -0.006, p = 0.007 for brand

recognition). Hence, Hypothesis 3 is supported only for banner memory (H3a).

As expected, the delay between the last banner exposure and the measurement

of memory for the banner has a significant negative effect on the probability of

recalling or recognizing the banner.

We also find indirect effects of measurement delay on brand recall ( 2 x

8 = - 0.004, p < 0.001) and brand recognition ( 2 x 8 = - 0.002, p = 0.007) to

be significant. Hence, with regard to Hypothesis 4c, the effect of measurement

delay on brand memory is mediated by banner memory. This result shows

decay in brand memory is affected by the inability to remember the banner as a

cue for recalling or recognizing the brand.

Differential effects of exposure variables on recall and recognition. We find

no major differences in the effect of exposure variables between recall and

0,27

0,275

0,28

0,285

0,29

0,295

0,3

0,305

0,31

0,315

0,32

1 25 50 75 100 125 150 175 200 225 250 275 300 325 350 375

Prob

abili

ty

Spacing time (in minutes)



Chapter 2

28

recognition. Specifically, the effect of measurement delay on banner recall (β =

- 0.030, p < 0.001) is comparable to that on banner recognition (β = - 0.022, p <

0.001). Figure 2.5 depicts the effects of varying levels of measurement delay

on the probability to recall and recognize the banner, holding the effects of

other variables constant. Although the point estimates of the effect of exposure

variables on recall are different from those of recognition (see Table 2.2), these

differences are not statistically significant. Because both recall and recognition

models have the same set of independent variables and both the dependent

variables on recall and recognition models have the same scales, testing the

differential effects of each exposure variable on recall and recognition is

straightforward. By holding all other independent variables constant, we use a

t-test to determine whether one unit increase of a particular exposure variable

will result in a different increase on the recall and the recognition odds ratio.

The results of these t-tests are insignificant, with p > 0.05. Hence, the

magnitudes of the effects of exposure variables on recall and recognition are

not statistically different.

Figure 2.5. Effect of Measurement Delay on Probability to Recall and Recognize the Banner

0

0,05

0,1

0,15

0,2

0,25

0,3

0,35

Prob

abili

ty

Measurement Delay (in minutes)

Recognition

Recall




29

Relationship between banner memory and brand memory. There is a

significant relationship between banner memory and brand memory measured

by recall (β = 0.135, p < 0.001), showing the probability of recalling a brand is

significantly higher when a consumer can recall the banner for the brand. A

similar relationship is found when memory is measured by recognition (β =

0.079, p < 0.001). Furthermore, the β for recall is higher than the corresponding

β for recognition, showing the relationship between banner memory and brand

memory is stronger when memory is measured by recall, though it is

marginally significant (p = 0.068).

Control variable (e.g., age, gender, product category) results. Users are

expected to have better recall and recognition of the brand than non-users. Our

analyses confirm this prediction by revealing brand usage is a positive and

significant predictor of both brand recall (t = 8.67, p < 0.001) and brand

recognition (t = 17.88, p < 0.001). In line with Campbell & Keller (2003), our

analyses also show experience with the brand significantly increases banner

recall (t = 11.24, p < 0.01) and banner recognition (t = 5.45, p < 0.001).

Furthermore, our analyses suggest male consumers have better memory

for both banner advertisements and brands. All coefficients for the dummy

variable ‘gender,’ except for brand recognition, are significantly negative

(Table 2.2). One possible explanation for this result is that males tend to use

cognition more than emotion when processing an advertisement (Meyers-Levy

and Maheswaran, 1991; Meyers-Levy and Sternthal, 1991). However, because

we do not have information about banner content in our study, we cannot make

conclusions about the causes of this gender difference.

With regard to age, it is widely accepted that the ability to memorize

decreases with age. Some studies in advertising literature (for example, Cole

and Houston, 1987) document that elderly people have deficits in learning

advertising messages and consequently perform worse on advertising memory

tests. Our analyses show age has a strongly significant negative effect on brand

and banner memory (see Table 2.2).



Chapter 2

30

In summary, the parameters for the control variables add to the face

validity of our model results.

2.5 Discussion 2.5.1 Key Findings Advertising scheduling is an important issue for both marketing researchers

and practitioners. With the advent of the Internet, this issue has become even

more important because the Internet provides sophisticated technologies to

control when, which, and how often customers are exposed to advertisements

such as banners. This study offers new insights on the effects of banner

exposures and provides guidance for formulating effective banner scheduling

strategies. We examine the effects of various aspects of banner exposures,

including the number of exposures, spacing (interval between exposures), and

measurement delay, on consumer memory of banner and brand. Our analyses

lead to five major findings.

First, the number of banner exposures has a direct positive effect on

consumer memory of the banner, which in turn enhances consumer memory for

the advertised brand. This effect is found for both recognition and recall, and

the magnitude of the effect is similar for both. A possible explanation for this

finding is that the correlation between recognition and recall is especially high

on stimuli (brands) with high recall scores (Laurent et al., 1985). The average

brand recall in our study is 51%. When consumers recognize a stimulus

(brand), they are also likely to recall the stimulus (brand). Furthermore, the

positive linear effect of the log of number of exposures on memory implies the

effect does not wear out completely, although the effect diminishes as the

number of exposures increases. Previous research on advertising response

(especially in laboratory experiments, where subjects are required to process

advertising) suggests wear-out occurs when consumers get tired of being

exposed to advertising, although this wear-out effect is less apparent in actual

advertising campaigns with voluntary attention to advertising (Pechmann and

Stewart, 1989). Our findings confirm wear-out does not occur in actual banner




31

campaigns. We also confirm results from earlier experimental banner

advertising studies (Drèze and Hussherr, 2003; Havlena and Graham, 2004)

that show banner exposures increase brand awareness. This finding is

important because it challenges the prediction of a decade ago that when

consumers became more adept at using the Internet, they would become more

skilled in filtering out unimportant information such as banner advertisements

(Dahlen, 2001). Our findings demonstrate that ten years later, in an age of

Internet adeptness, the effect of banner advertising on consumer memory is still

significant and positive.

Second, we also find there is a spacing effect for banner advertising; it is

likely the result of the study-phase retrieval mechanism. The spacing effect is

greatest for enhancing consumer memory for the banner and the brand, when

the time interval is not too short or too long. However, we find the spacing

effect for recognition only, not recall. This is an unexpected finding, because

previous research that found the spacing effect in advertising (e.g., Appleton-

Knapp et al., 2005; Heflin and Haygood, 1985) also used recall as a memory

measure. Our finding is perhaps due to the study setting: both Appleton-Knapp

et al. (2005) and Heflin and Haygood (1985) conducted studies in laboratory

settings in which the subjects were required to process advertising, whereas in

our study processing was voluntary. In real life, consumers often pay little

attention to advertising, which may result in weaker effects of advertising on

memory (Pechmann and Stewart, 1989). Our finding also shows the spacing

effect is not strong enough to provide recall benefits from spacing. Recall is

more difficult than recognition; it requires more memory traces than can be

provided by the spacing effect of banner exposures.

Third, the effect of memory enhancement from banner exposures decays

significantly over time for both recall and recognition, in contrast to the

common belief that recognition does not decay (Lucas, 1960). Our finding is

consistent with the Singh et al. (1988) study of television advertising.

Furthermore, in the context of banner advertising, this finding contrasts with

work of Havlena and Graham (2004), who found no significant negative effect



Chapter 2

32

of measurement delay on banner and brand awareness. The small effect of

measurement delay on banner memory might explain their findings: with β = -

0.030, after a one-day delay the decrease in recall probability is only about

three percent (see Figure 2.5).

Fourth, with regard to the mediating role of banner memory, our

analyses show (1) significant direct effects of banner exposure variables on

banner recall and recognition (2) insignificant direct effects of banner exposure

variables on brand recall and recognition, and (3) significant indirect effects of

banner exposure variables on brand recall and recognition (except spacing on

brand recall). Following the typology of mediations proposed by Zhao et al.

(2010), these results suggest an indirect-only mediation, or a full mediation

according to the terminology of Baron and Kenny (1986). In indirect-only

mediation, the presence of omitted mediators is unlikely. This indicates that

instilling consumer memory of the banner advertisement itself is vital to the

success of campaigns that seek to enhance consumer memory for the brand: to

create stimulate brand memory, the banner must be processed by the consumer,

with sufficient cognitive resources. Banners that fail to attract attention, or are

only minimally processed, do not improve the recall and recognition of the

brand they advertise. We also find the relationship between banner and brand

memory is stronger when memory is measured through recall. A possible

explanation for this finding is that memory tests show recall is more difficult

than recognition (Krishnan and Chakravarti, 1999; Laurent et al., 1995; Singh

and Rothschild, 1983). Recall of a stimulus requires deeper memory than

recognition of the stimulus. When consumers profoundly remember

advertisements and their features (for example, product images, logo), they are

more likely to recall the associated advertising brand when asked to name one

brand in the product category.

Fifth, the effects of the exposure variables are similar for recall and

recognition. Singh et al. (1988) asserted recognition is more able than recall to

access the tinier traces of memory, implying (i) less number of exposures is

needed for recognition than for recall and (ii) recognition decay is less than




33

recall decay. We take a cautious stand in this discussion because the brands in

our study have a relatively high level of familiarity. Because such brands

already have strong memory traces in consumer minds, the differential impact

of their banner exposures on recall and recognition might not be observed.

In summary, the exposure effects we find are mostly similar to those

found in previous studies of both banner advertising and traditional media. The

main difference in our findings is that contrary to studies that show banners

directly affect brand memory, our study does not find a direct effect of banner

exposures on brand memory. Instead, we find this effect is indirect and

mediated by banner memory. Our study shows the success of banner

advertising depends on consumer memory of the banner itself, in addition to

memory of specific advertising messages, such as brand name and brand

claims, contained within the banner.

2.5.2 Practical Implication Our findings also have practical implications. First, recall and recognition are

both important for measuring the impact of advertising on consumer brand

choice. A consumer’s brand choice decision is affected by brand memory

(Fennis and Stroebe, 2010). On one hand, recall is more relevant when brand

decision-making happens in the absence of the stimulus. On the other hand,

when the brand choice decision must be made in the presence of the stimulus

(for example, when choosing a brand of canned soup in a supermarket),

recognition is more important. Our study finds both brand recall and

recognition are enhanced by banner exposures; product choices made in either

memory-based (for example, durable products) or stimulus-based (for example

nondurable products) decision-making situations can be influenced by banner

advertisements. Furthermore, our finding with regard to the exposures response

curve implies brands will not suffer from too many banner exposures. The

impact of overexposure is limited to budget efficiency because the response

curve shows diminishing returns.



Chapter 2

34

Second, the interval between subsequent banner exposures must not be

too long or too short; banner exposures should be moderately spaced. Our

analyses provide initial interval estimates that suggest a one-hour interval

between exposures is optimal. This estimate can be used as a basis for

determining the optimal interval in a specific situation. Companies may also

take into account the effect of other factors that determine optimal spacing,

such as advertising clutter from similar products and demographics of the

target audience. Once the optimal spacing target has been determined using

available Internet advertising technologies (e.g., Plummer et al., 2007), targeted

optimal spacing can be attained without major difficulty.

Third, we examined the impact of the time interval between last

exposure and response measurement. In practice, this interval may refer to the

time between last exposure and the moment when the actual buying decision is

made. Our findings clearly show the delay between the last banner exposure

and the buying decision time should be short enough to ensure the effect gained

from banner exposures has not decayed too much. For example, by using IP

recognition technology and cookies information, a company can plan the next

exposure before further decay occurs. The finding that memory of banner

exposure decays also implies that the current practice of some advertising

research agencies— the testing of banner effectiveness by measuring

effectiveness directly after exposure—may overestimate the impact of banners

on purchase decisions by failing to take into account the effect of memory

decay.

By relating these three implications to existing advertising scheduling

strategies from marketing literature, we can conclude that even, rather than

massed (blitz), banner scheduling, is the best scheduling strategy for banner

advertising. More specifically, our model and findings can be used as a starting

point to build a mathematical model to optimize banner advertising spending

(i.e., number of exposures and when to deliver), given the purchase cycle and

advertising cost per exposure.




35

Another practical implication of our findings is that because the effect of

exposures is mediated by banner memory, it is important to have banners that

capture consumer attention. Banners that that do not stand out are not

remembered and will have no effect in enhancing brand memory.

2.5.3 Research Limitations and Future Research Our study has several limitations. First, our dataset, although extensive with

almost 9,000 consumers from seven countries, is drawn mainly from banner

advertising campaigns for established brands. This may limit the applicability

of our findings with regard to banner advertising that promotes new brands. For

example, established brands may limit the effect of exposures variables on

brand memory due to ceiling effects (Singh et al., 1988). Future research could

address this issue by determining whether brand familiarity moderates the

effects of banner exposures found in this study.

Another limitation is that we did not include banner design factors, such

as size or shape, position on a webpage, and interactivity (i.e., static or

animated) because this information was not available. Previous studies on

banner advertising have found design factors may impact banner effectiveness.

For example, a recent study by Draganska, Hartmann, and Stanglein (2014)

showed that different online advertisements had different effects on lifting

consumers recall of brand message. It would be interesting to uncover which

design factors moderate the relationships we found in our study.

Both our limitations and our findings suggest avenues for future

research. First, having demonstrated that banner exposures improve brand

awareness, we question how to achieve maximum improvement. Based on our

findings, we recommend an even, continuous advertising schedule as the best

banner advertising scheduling strategy. It would be interesting to study not

only the factors that influence optimal spacing, but also the optimal number of

exposures, the factors that influence decay rate, and ways in which firms can

minimize the impact of memory decay by tying banner exposures more closely

to consumer decision-making.



Chapter 2

36

Second, although our study has shown how various aspects of banner

exposures affect brand recall and recognition, the question of how these aspects

affect brand attitudes remains. Earlier research indicates banner exposures may

improve brand attitudes, but more specific effects of exposure characteristics,

such as the number of exposures, spacing, and measurement delay are still

unknown. This issue is important to address; if brand attitudes wear out, gains

in enhanced brand awareness from, for example, a high number of banner

exposures might be offset by declines in brand attitude due to boredom.

Finally, future research could address the effect of banner exposure

characteristics on cross-media synergy (Naik and Peters, 2009). Different

banner exposure characteristics may have different cross-media effects.

However, future research could examine advertising effectiveness

consequences of exposure scheduling within and across multiple media.

Empirical study of differential effects of exposure characteristics on media

synergy will help companies with their integrated marketing communication

strategies, especially with scheduling of multi-media advertising campaigns.

Findings from our study can serve as starting point for this empirical research

avenue.



Ad and Brand Metrics for Banner Advertising Effectiveness

37

Chapter 3 Ad and Brand Metrics for Banner Advertising Effectiveness

3.1 Introduction In the early days of the Internet, banner advertising studies focused on direct

responses from consumers, in the form of the clickthrough rate. When consumers

became more experienced users, they no longer perceived banners as novel or

interesting and stopped clicking them, such that clickthrough rates declined from 7 %

in 1997 (Drèze and Hussherr, 2003) to just .06 % in 2014 (Chaffey, 2015). Yet banner

advertising spending continues to increase. In 2014, U.S. banner advertising generated

$ 8.0 billion (IAB, 2015), a four-fold increase from $ 1.9 billion in 2010 (IAB, 2011).

Banner advertising accounted for 16% of total Internet advertising in 2014 (IAB,

2015). Therefore, banner effectiveness studies need to focus on a broader set of

effects, including brand metrics (Hollis, 2005). Drèze and Hussherr (2003)

recommend traditional measures, such as brand awareness and advertising recall.

Comparable to offline advertising, banner advertising could have indirect effects on

consumer purchasing, through improved brand awareness and brand attitude. To

uncover these potential indirect effects of banner advertising, we seek to determine the

interrelated changes in a range of advertising and brand metrics.

Previous research links banner exposures to various effectiveness measures,

including consumer attitudes toward the ad and the brand (Briggs and Hollis, 1997;,

Yoo, 2008); advertising recall, brand recall, and brand recognition (Drèze and

Hussherr, 2003; Yaveroglu and Donthu, 2008); and repeat purchase intentions

(Manchanda et al., 2006). Most studies examine only one or a few banner

effectiveness metrics though, so we lack understanding of how banner advertising,

through various intermediate and interrelated advertising and brand metrics, affects



Chapter 3

38

purchase intention. Instead, these interrelationships have been studied extensively for

traditional media, such as television and print advertising (Brown and Stayman, 1992;

MacKenzie et al., 1986; Vakratsas and Ambler, 1999). From both academic and

practical points of view, it is important to understand the interrelationships among

effect measures for banner advertising, because of the differences in consumer

information processing in different media (Chaudhuri and Buck, 1995; De Pelsmacker

et al., 2002).

The goal of this study is to determine interrelationships among advertising and

brand metrics in banner advertising. In pursuit of this goal, this study makes three key

contributions to advertising literature. First, we build on previous research that has

investigated separate relations between banner advertising and individual metrics to

develop and test a comprehensive model linking banner memory and attitude to brand

memory and attitude, and ultimately to brand purchase intentions. We find that banner

advertising —through banner memory, banner attitude, brand memory, and brand

attitude— indirectly affects purchase intentions positively. Second, we compare these

relationships with those established for traditional advertising (Brown and Stayman,

1992), as in the dual mediation model. The indirect effect of banner advertising is

similar but not identical to that established for advertising in traditional media.

Futhermore, by examining alternative paths for the indirect relationships, we find that,

whereas in traditional media advertising attitude has a stronger effect on brand attitude

than does brand cognition/memory, in banner advertising the effects of banner attitude

and brand memory on brand attitude are similar.

Beyond these theoretical contributions, our findings have practical relevance.

Knowledge of the influential processes that arise between banner exposures and

consumer actions help managers design banner advertising strategies. For example, by

realizing the extent to which banner memory affects brand attitudes through both

banner attitudes and brand awareness, managers can weigh the beneficial and

detrimental effects of “annoying” banner designs on users’ attitudes toward the brand.

In the next section, we discuss banner and brand effectiveness measures from

previous studies and their interrelations in traditional media advertising. On the basis

of our discussion of differences in processing banner versus traditional advertising and




39

findings from previous studies on banner advertising, we propose a model of the

relationships between banner and brand constructs for banner advertising. In the

following section, we provide the details of our field study and data collection from

almost 20,000 consumers. Following the empirical analyses, we discuss the findings

and their implication for theory and practice. This article concludes with presenting

limitations and directions for further research.

3.2 Theoretical Background 3.2.1 Indirect Effect of Advertising on Sales

Many advertising and brand metrics have been proposed and studied in advertising

and marketing literature. These metrics generally constitute five major categories:

advertising cognition/memory, advertising attitude, brand cognition/memory, brand

attitude, and purchase intentions (Brown and Stayman, 1992; Vakratsas and Ambler,

1999). Many comparable measures also have appeared in banner advertising studies,

such as banner recall, banner recognition, banner attitude, and brand awareness

(Chatterjee, 2008; Yaveroglu and Donthu, 2009; Yoo, 2008).

Studies of traditional print and television advertising describe complex

interdependent relationships among these metrics. In general, advertising and brand

attitude depend on advertising and brand cognitive responses, such as recall of the

advertising or the brand (Brown and Stayman, 1992, MacKenzie et al., 1986). In

addition, brand attitude may be influenced by advertising attitude. According to the

elaboration likelihood model (ELM), when the motivation to process the brand is low,

consumers use peripheral cues about the ad to form their brand attitude (Petty et al.,

1983). In turn, brand attitude affects purchase intentions and, ultimately, actual

behavior. For example, recent research (Bruce, Peters, and Naik, 2012) shows that

advertising can increase sales through intermediate effects of mind-set metrics. This

structuring of relationships among advertising and brand metrics into sequential stages

that lead to actual behavior is known as a hierarchy-of-effects model. Although

subject to some criticisms (Barry and Howard, 1990; Vakratsas and Ambler, 1999;

Weilbacher, 2001), it has been used widely to explicate the advertising

communication process from the consumer perspective, as a series of psychological



Chapter 3

40

steps (Shimp, 2008 p.147). We examine a general version of this hierarchy-of-effects

model for banner advertising (Figure 3.1) and elaborate on each of the relationships

depicted in the model, in the context of banner advertising.

3.2.2. Hypotheses

3.2.2.1. Effect of Banner Memory on Banner Attitude

Banner advertisements may attract attention from consumers, who then are

stimulated to process the information, which leads to stronger banner memory.

Remembering the banner means that consumers can access their own cognitive

reactions to the banner, which enhances their banner attitudes. In traditonal research,

Mackenzie and Lutz (1989) demonstrate that ad cognitions are antecedents of ad

attitude. However, empirical research on banner advertising offers equivocal results

regarding the relationship between banner memory and banner attitude. Yoo (2008)

finds that banner exposures can improve attitudes toward the advertisement, even

when the banner cannot be recalled, sugesting that there is no explicit relationship

BannerMemoryBanner

Memory

BrandMemoryBrand

Memory

BannerAttitudeBanner Attitude

BrandAttitudeBrand

AttitudePurchaseIntentionPurchase Intention

H1

H4

H2 H3

H5

ovariatesCovariates CovariatesCovariates

CovariatesCovariates CovariatesCovariates CovariatesCovariates

Covariates:- Usage- Gender-Age

Covariates:- Usage- Gender-Age

Figure 3.1. Structural model of the hypotheses




41

between banner memory and banner attitude. Yoo and Kim (2005) indicate that the

relationship is complex and moderated by the level of arousal ignited by the banner,

such that too much arousal (e.g. from animation) may initiate unpleasant feelings and

negatively affect attitudes. As we examine static banners, which induce cognitive

processes similar to that prompted by print advertising, we propose the following

relationship:

Hypothesis 1: Banner memory has a positive relationship with banner attitude

3.2.2.2. Effect of Banner Memory on Brand Memory

Although banner advertising research indicates that banner exposures increase both

banner and brand awareness (Drèze and Hussherr, 2003, Havlena and Graham, 2004),

prior studies have not investigated the relationship between these metrics. For

traditional media, Schmitt et al. (1993) show that pictorials have better mnemonic

values and are easier to remember. Using eye tracking methods, Wedel and Pieters

(2000) find that the more attention a consumer pays to pictorials and other

advertisement elements related to the brand, the better he or she remembers the brand.

Because banners mostly consist of pictorials, such as logos, symbols, and pictures or

photos of endorsers, banner exposures should strengthen consumers’ memory of the

pictorials. Familiar brands may benefit from an already established associative

network between the brand and its brand-related pictorials too (du Plessis, 2008), so

banner exposures should make such brands even more salient. The activation of

memory nodes for the pictorials subsequently will activate memory for the brand in

the associative network. Therefore, we hypothesize:

Hypothesis 2: Banner memory has a positive relationship with brand memory.

3.2.2.3. Effects of Brand Memory on Brand Attitude

Consumers evaluate brands based not only on information about the brand itself but

also on how they feel when processing information about the brand. The ease of

information processing might serve as a signal for a positive evaluation of the brand



Chapter 3

42

(Barry and Howard, 1990, Mantonakis et al., 2008). Brands that are easily retrieved

from memory evoke better evaluations than brands that do not readily come to mind,

especially in low level processing situations (Nordhielm, 2002). However, prior

negative associations might interfere with this fluency effect (Lee and Labroo, 2004).

Because we focus on established brands, we assume they generally do not have

substantially negative associations. Furthermore, the processing level of brand

information in a banner is mostly low, so we hypothesize:

Hypothesis 3: Brand memory has a positive relationship with brand attitude.

3.2.2.4. Effects of Banner Attitude on Brand Attitude

In traditional media, brand attitude generally is influenced by advertising attitude. A

mechanism that seeks to explain this relationship is the affect transfer hypothesis

(MacKenzie et al., 1986), which states that affective responses to advertising transfer

to the brand, following a peripheral route in the ELM. Consumers usually process

banners at a pre-attentive level (Chatterjee, 2008; Yoo, 2008), so peripheral brand

processing likely takes place when consumers try to form attitudes using knowledge

about the brand that they learned from banner advertising. The peripheral cues might

include banner attractiveness, likeability, or believeability, which in general determine

the consumers’ attitude toward the banner. We thus predict a positive effect.

Hypothesis 4: Banner attitude has a positive relationship with brand attitude.

3.2.2.5. Effects of Brand Attitude on Purchase Intention

The role of brand attitude in predicting consumer behavior has been long

acknowledged in marketing literature (see e.g. Mitchell & Olson, 1981; Bian &

Moutinho, 2011; Zarantonello & Schmitt, 2010). Therefore, advertising research on

traditional media frequently includes the relationship between attitude toward the

brand and brand purchase intention (e.g., Brown & Stayman, 1992; Keller & Aaker,

1992; Morwitz et al., 1993). In traditional advertising consumer motivation to process

advertising is generally high, which is in contrast to banner advertising. However, the




43

effect of brand attitude on purchase intention is not affected by how the consumer

processed the advertising, although when the motivation to evaluate the brand is

higher, the effect of brand attitude on purchase intention becomes even higher

(MacKenzie & Spreng, 1992). Therefore, though we do not expect the relationship of

brand attitude and purchase intention to be as strong as it would be in advertising

processed with high motivation, here brand attitude still have a positive effect on

purchase intentions.

Hypothesis 5: Brand attitude has a positive relationship with purchase intentions.

3.2.2.6. Covariates

In addition to these hypothesized effects, banner effectiveness metrics may be

influenced by other factors related to the brand and the consumer. Campbell and

Keller (2003) show that the effect of advertising on attitude toward the advertising is

greater for familiar than for less familiar brands. For familiar brands, brand attitude

depends more on brand memory, rather than advertising attitude, whereas for less

familiar brands, advertising attitude has a greater impact on brand attitude.

Furthermore, users or past users of a brand may remember and like the brand better

than nonusers. They may also have different goals in processing the banners, which in

turn can affect the effectiveness of the banners (Wang et al., 2009). Therefore, we

include experience with the brand to control for the effect of brand familiarity on

advertising and brand metrics.

Previous research also shows that consumer memory for advertising or brands

is affected by individual characteristics such as age and gender (Krishnan and

Chakravarti, 1999). Specifically, younger consumers have better memory for

advertising (Cole and Houston, 1997), and consumers of different genders engage in

unique cognitive advertising processing (Meyers-Levy and Maheswaran, 1991;

Meyers-Levy and Sternthal, 1991). Age and gender also may affect banner and brand

attitudes, because some campaigns and brands target specific market segments.

Younger consumers thus may have a different attitude toward the advertising or the



Chapter 3

44

brand than older consumers, as might consumers of different genders. Therefore, we

include gender and age as covariates in our model.

3.3. Empirical Study Design 3.3.1. Survey Procedure and Sample Description The data we used to test our hypotheses consist of surveys of 29 banner advertising

campaigns that involved mostly well-known brands in 10 European and Latin

America countries: Belgium, Finland, France, Germany, Italy, Mexico, The

Netherlands, Norway, Spain, and the United Kingdom. For each campaign, potential

respondents were invited randomly from the company panel of respondents who were

visiting websites on which the banners for a particular brand had been placed during

the brand campaign. The profile of invited respondents was matched to the defined

profile of the brand target customers. Upon a potential respondent accepted the

invitation, he or she was presented with a set of questions measuring the brand and ad

metrics used in this study. Furthermore, using proprietary tagging technology, we

could identify if respondents had been exposed to the particular banner or not. For this

study, we only used data from respondents who had been exposed to the banner. The

survey was conducted by MetrixLab, a leading global online market research

company.

In total, the data set contains 19,994 respondents. The smallest sample size for

any individual campaign is 72 respondents, and the largest is 2,281 respondents.

Almost 73% of the respondents are women, largely because some of the focal

campaigns targeted women only (e.g., beauty and personal care brands). The average

age of the respondents is 36.5 years. Though the demographic characterics of the

sample are different from those of the general population, for each campaign they are

similar to the demographic characteristics of the targeted brand. This is because the

respondents were invited selectively to match with the defined profile of the brand

target customers.




45

3.3.2. Banner and Brand Effectiveness Measures

The campaign surveys used online, semi-standardized questionnaires that contained

various advertising effect measures, demographic questions, and items pertaining to

the product category, brand, and banner. The questions relevant for our study were

standardized across all campaigns. Table 3.1 contains the list of brand and advertising

indicators; we explain the measures for each construct, as well as the rationale for

including these indicators, next.

To measure banner memory and brand memory, we used recall and recognition

measures (Table 3.1). Both recall and recognition have been used widely to measure

the effect of advertising on consumer memory (du Plessis, 1994). The basic difference

is whether the stimulus, as the retrieval target, appears during the test or not: The

stimulus appears in a recognition test but not in a recall test, so these measures may

pertain to different dimensions of memory (Finn, 1992). However, Bagozzi and Silk

(1983) argue that recall and recognition represent a single memory state, if we control

for the effect of variation in reader interest. Therefore, we use recall and recognition

as indicators of memory construct for a particular banner or brand. In the context of

banner advertising studies, banner recall (Yaveroglu and Donthu, 2008) and banner

recognition (Drèze and Hussherr, 2003) similarly can measure the effect of banner

exposures on banner memory (Chatterjee, 2008; Havlena and Graham, 2004).

Previous studies have used many indicators to measure advertising attitude,

such as advertising favorability, attractiveness, and likeability (MacKenzie and Lutz,

1989; MacKenzie et al., 1986; Yoo, 2008). Therefore, we use multiple indicators to

measure the various aspects of banner attitude, including questions about banner

likeability and the extent to which the banner is striking to measure the affective

aspect, as well as banner relevance, believability, and curiosity to measure the

cognitive aspect. Our scales are adapted from scales widely used to measure general

attitude toward the ad (Bruner, 2009 p.97-98). With a Cronbach-alpha of 0.695, the

reliability of our adapted scales is within an acceptable range (Garson, 2013),



Tab

le 3

.1. M

odel

con

stru

cts,

indi

cato

rs, s

cale

s, an

d de

scri

ptiv

e st

atis

tics

Con

stru

ct

Indi

cato

r Su

rvey

Que

stio

n

Scal

e an

d C

odin

g M

ean

SD

Bra

nd

mem

ory

Bra

nd re

call

W

hich

bra

nds o

f [pr

oduc

t cat

egor

y] th

at y

ou c

an th

ink

of?

Bin

ary

0: T

arge

t bra

nd is

not

in th

e lis

t 1:

Tar

get b

rand

is in

the

list

0.51

0.

50

Bra

nd

reco

gniti

on

Hav

e yo

u he

ard

of a

ny o

f the

follo

win

g [p

rodu

ct c

ateg

ory]

bra

nds?

B

inar

y 0:

No

1

: Yes

0.

79

0.41

Bra

nd

attit

ude

B

rand

fa

vora

bilit

y

How

wou

ld y

ou d

escr

ibe

your

ove

rall

opin

ion

of th

e fo

llow

ing

[pro

duct

cat

egor

y] b

rand

s?

Plea

se p

rovi

de u

s with

you

r ove

rall

opin

ion,

eve

n if

you

have

nev

er u

sed

this

pro

duct

bef

ore

Like

rt (1

-5)

1: V

ery

unfa

vora

ble

5: V

ery

favo

rabl

e

3.41

1.

15

Bra

nd

cons

ider

atio

n

If yo

u w

ere

look

ing

to p

urch

ase

[pro

duct

cat

egor

y], h

ow li

kely

wou

ld y

ou b

e to

con

sider

ea

ch o

f the

follo

win

g br

ands

B

inar

y 0:

Wou

ld n

ot c

onsid

er p

urch

asin

g 1:

Wou

ld c

onsi

der p

urch

asin

g

0.48

0.

50

Ban

ner

mem

ory

Ban

ner r

ecal

l W

here

, if a

t all,

hav

e yo

u se

en, h

eard

, or r

ead

any

adve

rtisi

ng fo

r [br

and]

rece

ntly

? B

inar

y 0:

The

Inte

rnet

is n

ot in

the

list

1: T

he In

tern

et is

in th

e lis

t

0.22

0.

41

Ban

ner

reco

gniti

on

Plea

se h

ave

a lo

ok a

t the

follo

win

g ad

verti

sing

[a b

anne

r is s

how

n]. D

o yo

u re

call

seei

ng a

n ad

verti

sing

sim

ilar t

o th

e on

e sh

own

abov

e on

the

inte

rnet

? B

inar

y 0:

No

1

: Yes

0.

31

0.46

Ban

ner

attit

ude

Ban

ner

likea

bilit

y

Ove

rall,

I lik

e th

e ad

verti

sing

Li

kert

(1-5

) 1:

Stro

ngly

dis

agre

e 5:

Stro

ngly

agr

ee

3.25

1.

03

Ban

ner

rele

vanc

y Th

e ba

nner

was

rele

vant

to m

y ne

eds

Like

rt (1

-5)

1: S

trong

ly d

isag

ree

5: S

trong

ly a

gree

2.81

1.

03

Ban

ner

strik

ingn

ess

The

bann

er w

as d

iffer

ent f

rom

oth

er [p

rodu

ct c

ateg

ory]

ads

Li

kert

(1-5

) 1:

Stro

ngly

dis

agre

e 5:

Stro

ngly

agr

ee

3.22

1.

05

Ban

ner

belie

veab

ility

Th

e ba

nner

was

bel

ieva

ble

Like

rt (1

-5)

1: S

trong

ly d

isag

ree

5: S

trong

ly a

gree

2.67

1.

20

Ban

ner

curio

sity

Th

e ba

nner

mak

es m

e w

ant t

o fin

d ou

t mor

e ab

out [

bran

d]

Like

rt (1

-5)

1: S

trong

ly d

isag

ree

5: S

trong

ly a

gree

3.10

1.

01

Purc

hase

in

tent

ion

Purc

hase

in

tent

ion

Con

side

ring

ever

ythi

ng y

ou k

now

abo

ut [b

rand

], ho

w li

kely

wou

ld y

ou b

e in

pur

chas

ing

this

pro

duct

, if y

ou w

ere

look

ing

for [

prod

uct c

ateg

ory]

?

Bin

ary:

0:

Wou

ld n

ot p

urch

ase

one

1: W

ould

pur

chas

e on

e

0.37

0.

48

Cova

riate

s

Use

Do

you

have

exp

erie

nces

with

[bra

nd]?

B

inar

y:

0: N

o, I

do n

ot

1:

Yes

, I d

o 0.

31

0.46

Age

How

old

are

you

? In

yea

rs

36.5

4 13

.40

Gen

der

W

hat i

s you

r gen

der?

B

inar

y:

0: M

ale

1: F

emal

e 0.

72

0.

44




47

Similar to the banner attitude construct, brand attitude comprises two aspects: cognitive

and affective (Percy and Elliott, 2005). Following this line of research, we use brand

consideration to measure the cognitive aspect and brand favorability to measure the

affective aspect of brand attitude (see also Putrevu & Lord, 1994). Finally, we measured

purchase intention with a question about whether the respondent would buy the brand on

the next purchase occasion. There are several items to measure behavioral intention used

in marketing literature (Bruner, James, and Hensel, 2001), but most of the study include

the question similar to ‘how likely will you purchase the brand/product’. Therefore, we

believe the our single item here has the predictive validity of purchase intention (see also

Bergkvist & Rossiter, 2007). Since we borrow the scales from various advertising studies

on traditional media, we test the unidimensionality of these scales later in our

measurement model analysis.

As we show in Table 3.1, an average of 22%(31%) of the respondents recalled

(recognized) the banner to which they had been exposed. With respect to banner attitude,

the average scores ranged from 2.67 for banner believeability to 3.25 for banner

likeability. These scores varied strongly across campaigns; for example the lowest

average score for banner curiosity was 1.78, whereas the highest campaign average was

3.57. On average, 51% of the respondents could recall and 79% recognized the brands we

studied. The brands scored 3.41 on brand favorability on average, and 48% of

respondents reported that they would consider and 37% that they intended to buy the

brand.

3.4 Data Analysis To test the relationships among constructs, measured using indicator variables (Figure

3.1), we applied structural equation modeling (SEM). We used SEM because our model

consists of latent constructs and we want to test the relationships simultaneously.

Previous research testing relationships among advertising and brand constructs also used

SEM (for example: MacKenzie and Lutz, 1989; Brown and Stayman, 1992). However,

three characteristics of the data demand special consideration: (1) clustered observations



Chapter 3

48

in campaigns, (2) variables that are categorical instead of continuous, and (3) the

prevalence of missing values.

3.4.1 Multilevel SEM for Clustered Observations Our database revealed a hierarchical structure: The observations of individual

respondents are clustered in campaigns. We therefore expect that unobserved campaign-

level factors influence responses, such that observations within a campaign are not

independent of one another. In other words, there could be a correlation between

observations within a campaign, which then would violate the assumption that all

observations are independent. Ignoring the clustered structure of the data would lead to

erroneous conclusions (Snijders and Bosker, 2003), due to biases in the structural model

parameter estimates and their standard errors (Muthén, 1989; Muthén and Satorra, 1995).

Muthén (1994, 1997) suggests data should be analyzed in a multilevel setting if the intra-

cluster (in our case, within-campaign) correlations exhibit values greater than .10. In our

study, more than 50% of the intra-cluster correlations (see Table 3.2, below the diagonal)

indicate such values, which strongly suggests the need to use multilevel SEM.

Our theoretical background and hypotheses address effects at the individual level.

For example, consumers with better banner memory likely score better on the brand

memory measure. We do not have clear expectations about relationships at the campaign

level, but we specify one general latent construct at the between-campaign model level.

This approach is in line with (Hox, 2002) and (Muthén, 1989) when the theoretical

foundation for the model at the higher (between-campaign) level is lacking.

3.4.2 Categorical Variables Treatment

The database contains both dichotomous and categorical variables, which violates the

assumption of a multivariate normal distribution of the variables, so a maximum

likelihood estimator may produce biased parameter estimates and standard errors (Bollen,

1989). The development of continunous/categorical variables methodology (CVM)

(Muthen, 1984; Muthen & Asparouhov, 2002) and robust weighted least square



Tab

le 3

.2. W

ithin

- and

bet

wee

n-ca

mpa

ign

corr

elat

ions

of b

rand

and

ban

ner

met

ric

indi

cato

rs

1

2 3

4 5

6 7

8 9

10

11

12

1. B

rand

reca

ll 1

0.68

**

0.04

0.

64**

0.

57**

-0

.26

-0.0

8 0.

12

0.28

0.

48**

-0

.09

0.26

2. B

rand

reco

gniti

on

0.46

**

1 -0

.29

0.60

**

0.62

**

-0.2

8 0.

07

0.23

0.

11

0.52

**

-0.2

2 0.

25

3. B

rand

favo

rabi

lity

0.14

**

0.15

**

1 0.

15

0.18

-0

.20

-0.1

1 0.

34

0.19

0.

22

0.30

0.

11

4. B

rand

con

side

ratio

n 0.

30**

0.

41**

0.

42**

1

0.71

**

0.21

0.

12

0.44

**

0.61

**

0.55

**

0.02

0.

62**

5. B

rand

inte

ntio

n 0.

21**

0.

27**

0.

43**

0.

65**

1

0.22

0.

17

0.35

0.

68**

0.

60**

0.

03

0.71

**

6. B

anne

r rec

all

0.18

**

0.21

**

0.10

**

-0.3

2**

-0.3

3**

1 0.

22

-0.2

5 -0

.24

-0.2

7 -0

.04

-0.3

9*

7. B

anne

r rec

ogni

tion

0.12

**

0.09

**

0.06

**

-0.1

7**

0.17

**

0.35

**

1 -0

.10

0.14

-0

.18

0.12

0.

17

8. B

anne

r lik

eabi

lity

0.03

**

0.10

**

0.24

**

0.15

**

0.25

**

0.14

**

0.09

**

1 0.

34

0.39

* 0.

28

0.58

**

9. B

anne

r rel

evan

ce

0.04

**

0.07

**

0.27

**

0.15

**

0.29

**

0.09

**

0.08

**

0.54

**

1 0.

37*

0.50

**

0.85

**

10. B

anne

r stri

king

ness

0.

02**

0.

06**

0.

21**

0.

25**

0.

42**

0.

11**

0.

06**

0.

45**

0.

40**

1

0.01

0.

44**

11. B

anne

r bel

ieva

bilit

y 0.

03**

0.

07**

0.

26**

0.

64**

0.

57**

0.

11**

0.

08**

0.

55**

0.

56**

0.

41**

1

0.44

**

12. B

anne

r cur

iosi

ty

0.04

**

0.10

**

0.30

**

0.60

**

0.62

**

0.16

**

0.10

**

0.62

**

0.64

**

0.46

**

0.59

**

1 N

otes

: Cor

rela

tions

bel

ow th

e di

agon

al a

re w

ithin

-leve

l cor

rela

tions

(i.e

., w

ithin

cam

paig

ns);

corr

elat

ions

abo

ve th

e di

agon

al a

re

betw

een-

leve

l cor

rela

tions

(i.e

., be

twee

n ca

mpa

igns

). *

p-va

lue

< 0.

05

** p

-val

ue <

0.0

1



Chapter 3

50

(WLS) estimation procedure have provided the correct test statistics for models

with any combination of dichotomous, ordinal, or continuous outcome

variables (Kline, 2010). We use the robust WLSMV estimator available in the

Mplus software (Muthén and Muthén, 2010) to estimate the multi-level SEM

model.

3.4.3 Missing Values Treatment The overall percentage of missing values for the indicators is 9.3 %, with a

maximum of 21.5 % for purchase intentions. In the covariates, 11.2 % of the

values are missing, with a maximum of 33.5 % for brand usage. We decided

not to delete observations with missing values, because doing so would lead to

10,872 deleted observations and increase the risk of biases, if the complete

observations did not adequately represent the overall sample (Schafer and

Olsen, 1998).

To overcome the missing value problem, we applied multiple imputation

(Rubin, 1996). Multiple imputation uses simulation to generate multiple

complete data sets, with missing values in the original data set replaced by

plausible values generated randomly from their predictive distribution. Then,

by conducting statistical procedures, we can provide “averaged” values for the

missing data. In the context of SEM studies, multiple imputation can be

performed using a restricted or unrestricted model (Asparouhov and Muthén,

2010). For multiple imputation using a restricted model, the data get imputed

using a factor structural model. For unrestricted model multiple imputation, in

contrast, the models are more general, to avoid misspecification bias. The

downside of an unrestricted model multiple imputation is the convergence

problem that arises due to the large number of parameters in the model. In two-

level data with many categorical variables, an analysis using multiple

imputation with an unrestricted model results in substantially better

performance than an analysis using unimputed data (Asparouhov and Muthén,

2010). Furthermore, the performance of multiple imputation is similar for

unrestricted or restricted models (Asparouhov and Muthén, 2010). Therefore,




51

we use unrestricted model multiple imputation in our analysis. Specifically, we

adopt variance-covariance model multiple imputation and include all indicators

and covariates in the data set in the model. We include all indicators for our

multiple imputation, to avoid the risk of bias in the subsequent analyses that

might result from the omission of variables that could have significant

relationships with imputed variables when performing the imputation (Schafer

and Olsen, 1998).

3.4.4 Covariates Modeling We included the covariates as predictors for each latent construct, rather than

as predictors for each indicator (Rabe-Hesketh, et al., 2007), with one

exception because, the brand purchase intention indicator had no underlying

latent construct.

3.5 Results 3.5.1 Measurement Model As we explained in the previous section, we tested our measurement model

with the factor indicator structure described in Table 3.1 for the within-level

model and one general factor for the between-level model. The estimation of

the measurement model showed good fit with the data. The �2-value of 122.9

(103 degrees of freedom, p = .089) indicated reasonable fit, considering that the

�2 statistic is sensitive to huge sample sizes. Other approximate fit indices

confirmed the fit of the model (root mean square error of approximation

[RSMEA] = .003, confirmatory fit index [CFI] = .999, square root mean

residual [SRMRwithin] = .042). All loadings in the measurement model were

statistically significant (p < .001). Table 3.3 contains the loading parameters,

standard errors, and t-values. The reasonable fit of the measurement model led

us to estimate and interpret the structural model.



Chapter 3

52

3.5.2 Structural Model

The structural model (Figure 3.2) showed good fit (�2 = 145.91, 125 df, p =

.0973, CFI = .999, RSMEA = .003, SRMRwithin = .052). As the standardized

parameter estimates for this model in Table 3.3 reveal, we found support for all

our hypotheses. In particular, banner attitude was associated positively with

banner memory (β = .245, p < .001), in support of H1. The path between

banner memory and brand memory was statistically significant (β = .517, p <

.001), indicating a positive relationship between banner and brand memory, in

line with H2. Brand attitude also showed a significant positive effect of banner

attitude (β = .321, p < .001), in support of H3. The parameter for the direct

effect of brand memory on brand attitude was positive (β = .293) and

statistically significant (p < .001), which revealed support for H4. We also

confirmed H5, because brand attitude related positively to purchase intentions

(β = 2.223, p < .001). Finally, the model accounted for a significant proportion

of within-level variation in each endogenous banner and brand construct. The

variance accounted (R-square) values were as follows: banner memory .092

(SE = .004), brand memory .537 (SE = .048), banner attitude .100 (SE = .007),

brand attitude .698 (SE = .021), and purchase intention .758 (SE = .078).

Regarding the covariate effects, almost all significant effects were as expected.

Experience with the brand related positively to most constructs, with the

notable exception of a negative (direct) effect on brand intention, which is

counterintuitive. However, considering that brand intention appears at the end

of the structural model, and experience was included as a covariate for other

constructs in the previous stages, the total effect may be positive, after we take

all indirect effects into account. The total effect of experience on brand

intention was positive (β’ = 1.92). Furthermore, as expected, age showed a

negative association with banner memory, but we did not find any significant

relationship with brand memory. For gender, the analysis revealed that female

consumers exhibited significantly better brand memory and more positive

banner attitudes than male consumers.




53

Table 3.3 Within-level parameter estimates (standardized) for hypothesized and competing dual-mediation model Hypothesized Model Competing Model Parameter Value SE t-Statistic Value SE t-Statistic Measurement model

Brand memory � Brand recall

0.863 0.051 16.922 0.873 0.052 16.788

Brand memory � Brand recognition

1.467 0.173 8.480 1.484 0.185 8.022

Brand attitude � Brand favorability

0.890 0.036 24.722 0.889 0.037 24.027

Brand attitude � Brand consideration

1.046 0.050 20.920 1.045 0.051 20.490

Banner memory �Banner recall

1.425 0.080 17.813 1.491 0.088 16.943

Banner memory �Banner recognition

0.497 0.017 29.235 0.498 0.016 31.125

Banner attitude �Banner likeability

1.077 0.030 35.900 1.071 0.03 35.700

Banner attitude �Banner relevance

1.248 0.039 32.000 1.248 0.039 32.000

Banner attitude � Banner strikingness

0.682 0.020 34.100 0.683 0.02 34.150

Banner attitude � Banner believability

1.083 0.029 37.345 1.084 0.029 37.379

Banner attitude �Banner curiosity

1.626 0.050 32.520 1.63 0.05 32.600

Structural Model

Banner memory �Brand memory

0.517 0.039 13.256 n.a n.a n.a

Banner attitude � Brand memory

n.a n.a n.a 0.141 0.017 8.294

Banner memory � Banner attitude

0.245 0.015 16.333 0.236 0.015 15.733

Brand memory � Brand attitude

0.293 0.029 10.103 0.281 0.03 9.367

Banner attitude � Brand attitude

0.321 0.039 8.231 0.317 0.04 7.925

Brand attitude �Purchase intention

2.223 0.561 3.963 2.235 0.575 3.887



Chapter 3

54

BannerMemoryBanner

Memory

BrandMemoryBrand

Memory


BrandAttitudeBrand


0.245(0.015)

0.293(0.029)

0.517(0.039)

0.321(0.039

)

2.223(0.561)

Notes: Values are standardized parameter estimates, and values in parentheses are their standard errors.

Figure 3.2. Structural model results, hypothesized model

3.5.3 Competing Model: Dual Mediation Although our model received statistical support from the data, we cannot

conclude the data exclusively confirm the proposed model. Therefore, we

considered the dual mediation model (MacKenzie et al, 1986), which has

gained considerable empirical support in traditional media contexts (Brown and

Stayman, 1992), as an alternative model. In the dual mediation model, the

effect of advertising attitude on brand attitude is mediated by brand cognition.

Although we measure brand memory, instead of the more general brand

cognition concept to which the dual mediation model refers, it is interesting to

determine whether including a mediating role for brand memory can improve

model performance. Furthermore, unlike our model, the dual mediation model

lacks a direct path between advertising cognition and brand cognition. To

estimate a dual mediation model, we add a path from banner attitude to brand

memory and delete the path between banner and brand memory (Figure 3.3).

The competing model offered comparable model fit indices (�2(125) =

148.80, p = .072, CFI = 0.999, RSMEA = .003, SRMR = .062). Although all

these indices indicated reasonable model fit, the chi-square value and SRMR

index in this competing model were slightly worse than those in the

hypothesized model (Figure 3.1). A comparison of the individual parameter




55

BannerMemoryBanner

Memory

BrandMemoryBrand

Memory


BrandAttitudeBrand


0.236(0.015)

0.281(0.030)

0.141(0.017) 0.317

(0.040)

2.235(0.575

)

Notes: Values are standardized parameter estimates, and values in parentheses are their standard errors. Figure 3.3. Structural model results, dual mediation model

estimates in Table 3.3 showed that most of the parameters were similar to those

of the original model and statistically significant, which provided support for

all the hypothesized relationships between banner and brand metrics. A notable

difference arose in how the two models explained brand memory though. In

our model, brand memory was directly affected by banner memory, whereas in

the competing dual mediation model, it was affected only indirectly by banner

memory, through banner attitude. As hypothesized in our model, the

relationship between banner memory and brand memory was positive and

significant (β = .517). In the dual mediation model, the relationship between

banner attitude and brand memory also was positive and significant, but it had

a much lower β of .141. Furthermore, in our hypothesized model, the R-square

for brand memory was .537, whereas in the dual mediation model it was only

.310. On the basis of these values, we concluded that the competing model

offered slightly worse fit than our hypothesized model, so we use our model

presented in Figures 3.1 and 3.2 as the basis for our concluding discussion.

3.6 Conclusion 3.6.1 Discussion This study has determined the interrelationships among advertising and brand

metrics in banner advertising. In general, we find that the effects of banner

advertising on advertising and brand metrics are similar to those in traditional



Chapter 3

56

advertising contexts. Our model supports the notion that banner advertising has

an ultimately positive effect on purchase intentions, through intermediate

effects on brand memory and brand attitude. All the hypotheses within our

structural model, representing the series of intermediate effects from banner

memory to purchase intention (through banner attitude, brand memory, and

brand attitude), received support from our data.

Not all effects were equivalent to the effects established in traditional

advertising studies though. We found a stronger positive relationship between

banner memory and brand memory than between banner attitude and brand

memory, which highlights an important difference between our model and the

dual mediation model. In the latter, a widely used model to assess advertising

in traditional media, there is no direct influence of advertising cognition on

brand cognition. We consider several explanations for this difference. First, it

may be caused by the differential nature of banner advertising, compared with

television or print advertising. In the dual mediation model, the relationship

between advertising attitude and brand cognition builds on the premise that

advertising affect may facilitate the acceptance of brand claims (MacKenzie et

al., 1986). Because most banners are very simple, the effect of banner attitude

on brand memory, through banner affect, might not exist. Second, even if some

banners eventually generate affection, which enhances brand memory, this

effect might be low compared with the enhancement of brand memory

generated by the stronger links between banner and brand memory in a

consumer associative network that has been influenced by banner exposures.

Third, we use banner and brand memory constructs, instead of the advertising

and brand cognitions that appear in the dual mediation model for traditional

media. Although memory is part of the cognitive system, advertising/brand

memory is not the same as advertising/brand cognition. More complex

processes might define the relationships between advertising and brand

cognition, as well as between advertising/brand cognition and

advertising/brand attitudes.




57

Another important difference between our model and the dual mediation

model entails the size of the effect of advertising attitude on brand attitude,

compared with that of brand cognition/memory. In the dual mediation model

(Brown and Stayman, 1992), the effect of advertising attitude on brand attitude

is more than twice as great as the effect of brand cognition on brand attitude

(standardized estimates: .57 versus .20). MacKenzie et al. (1986) even

considered the effect of brand cognition on brand attitude so low that it was

insignificant. In our model, both the effects of banner attitude on brand attitude

and of brand memory on brand attitude were significant and similar in size,

though the former were slightly larger (standardized estimates: .32 versus .29).

This finding highlights the lessened effect of advertising attitude in banner

advertising, compared with that in traditional advertising (standardized

estimates: .57 versus .32). The effect of brand memory on brand attitude even

appeared stronger (standardized estimates: .20 versus .29) in banner

advertising. There are two possible explanations. First, consumers might

process banners using minimum resources (Chatterjee, 2008; Yoo, 2008), such

that the banner attitude that results from banner advertising messages and

features has a weaker effect on brand attitude than would be the case in

traditional media advertising. Yet, a dominant brand logo and other pictorials

in banners may make the brand more salient in consumer memory, which

would then be perceived as more familiar to the consumer. This familiarity

would be translated subsequently into an emotional response that serves as a

positive cue for attitude toward the brand (Nordhielm, 2002). Second, the

stronger effect of brand memory on brand attitude may be due to the

confounding effect of brand familiarity, built over time. Unfortunately, we

cannot disentangle this effect, because our data set did not include new or

unfamiliar brands.

Regarding the measurement model, the relationships between

advertising/brand metric constructs (i.e., banner memory, brand memory,

banner attitude, and brand attitude) and their indicators in traditional media

apply similarly to banner advertising. This finding may serve as an



Chapter 3

58

endorsement of the adoption of advertising and brand metrics from traditional

media advertising research into banner advertising research. Regarding the

structural model, we find that basically (cf. the differences we discussed

previously), the relationships between advertising and brand metrics in

traditional media advertising appear in banner advertising. This finding aligns

with the suggestion that advertising on the Internet communicates in a way

similar to advertising in traditional media (Percy and Elliott, 2005), such that

“with advertising and the Internet we are essentially dealing with a print

advert” (Percy and Elliott, 2005, p. 11). However, this similarity might

disappear when we address other kinds of Internet advertising, such as search

engine advertising, pop-ups, or games, because of their lesser similarity with

traditional media advertising.

Finally, with regard to the covariate effects, the finding that banner

memory correlates negatively with consumers’ age offers face validation for

the results. The insignificant link between brand memory and age instead might

arise because in our study, most of the brands were prominent and easy to

remember for both younger and older consumers. Female consumers expressed

significantly better brand memory and more positive banner attitudes than their

male counterparts, probably because more brands in our data set targeted

female consumers.

3.6.2 Managerial Implications This study has several implications for advertisers and marketing managers. In

line with the idea that the Internet is both an information and a transaction

medium, we confirm that, above and beyond the direct click-through and sales

effects of banners, banner exposures affect brand metrics and thus have

positive, indirect effects on purchase intentions. The indirect effect is similar to

the one established for traditional media advertising. This finding justifies

developments in online advertising practice that has started to use traditional

advertising and brand metrics, such as advertising and brand recall,

recognition, or attitude. It also constitutes an important observation, in that




59

banners have lost much of the novelty that stimulated direct responses to them

in the 1990s and early 2000s.

If we view our model as one variation of the hierarchy-of-effect model,

our findings can translate into strategies to improve brand performance. In

banner advertising, brand attitude can be enhanced through (1) the transfer of

affection for the banner to the brand and/or (2) heightened awareness of or

familiarity with the brand, which leads to a misattribution of positive attitude to

the brand. The indirect effect of advertising memory on brand attitude, through

brand memory, also is stronger than the one through advertising attitude.

Therefore, banners need to facilitate stronger links to the brand, rather than

attempt to provoke stronger affect. In practice, advertisers might benefit from a

brand logo or slogan as the main element of their banner advertisements.

However, intrusive and annoying banners, such as pop-ups, may increase

awareness but also exert negative impacts on banner attitudes. This negative

impact will decrease positive attitudes toward the brand as well.

3.6.3 Limitations and Further Research Similar to other empirical studies, our project has limitations, which suggest

directions for further research. First, our data set contains only established

brands. Previous research (Campbell and Keller, 2003; Dahlén, 2001;

Draganska et al., 2014) show that brand familiarity influences the effectiveness

of advertising exposures. We did not include less familiar brands in our study,

so we cannot speak to the effect of banners on unfamiliar or new brands. This

question remains for further research.

Second, the banners in our data set were similar, in that they were all

static. However, they may have differed in size or location on the webpage; this

information was not recorded in the database. Therefore, we cannot include

banner characteristics as explanatory variables, even though the interactivity of

the banner (e.g., animation, gamification, video), its size, and its location on the

page likely influence the relationships between advertising and brand metrics

in banner advertising. For example, animation in a banner can increase both



Chapter 3

60

awareness and irritation (Chatterjee, 2008), so the relation between memory

constructs and an attitude construct could be negative for animated banners.

Furthermore, brands advertised in complex banners are harder to recognize

(Lee and Ahn, 2012). Future research should feature other types of banner

advertising, such as animated, video, or pop-up banners. It also would be

interesting to study the relationships among advertising and brand metrics for

other types of Internet advertising, such as interstitial, take-over, and paid

search advertising, to determine if the similarities we find between traditional

and banner advertising extend to these realms.

Third, our model excluded any effects of advertising in other media.

Integrated marketing communication theory recommends careful planning of

promotional activities, because one promotional activity likely interacts with

other, related activities. The objective is to achieve synergy among these

promotional activities. In an advertising context, the effect of advertising in one

medium may enhance the effect of advertising in other media (Naik and

Raman, 2003). For example, Zenetti, Bijmolt, Leeflang, and Klapper (2014)

find substantial interaction effects of different online and television advertising.

We lack data on advertising in other media for the brands in our study though,

so we could not control for such effects. Rather, we assume that the intensity of

advertising in other media is comparable across campaigns. A natural extension

of our study would be to investigate how banner advertising synergizes with

other media. Such insights should help advertisers develop more effective

media mixes.

Fourth, we used SEM and cross-sectional data to examine relationships

between advertising and brand constructs in a causal manner. In this condition,

no single statistical analysis, regardless of how sophisticated it is, can confirm

or disconfirm causality (Bullock et al., 1994). Our findings therefore should be

interpreted as a possible means to model the interdependent relationships

among banner and brand metrics. Single directionality in our model

relationships implies causal relationships only in a steady-state condition

(Percy and Elliott, 2005). However, our finding is in line with the interpretation




61

of an advertising hierarchy-of-effects (Barry and Howard, 1990) and the

general aim of advertising to build the brand over the long term (Percy and

Elliott, 2005).

Internet advertising has become increasingly important for virtually

every business in the past decade, and our study contributes to a better

understanding of the multifaceted impact of banner advertising. We encourage

other researchers to use these insights to further enhance the effectiveness of

banner advertising.



Chapter 3

62



Evaluating Website Polls as an Instrument for Customer Surveys

63

Chapter 4 Evaluating Website Polls as an Instrument for Customer Surveys 4.1. Introduction Imagine this situation faced by the marketing manager of a small national

restaurant chain: the company has received information that within a month, a

main competitor will launch a new dessert menu, promoted by a massive

advertising campaign. The chain may well lose customers to the competitor.

Unfortunately, the manager has no data about customer preferences for

desserts, and there is no time or budget to conduct a nationwide customer

survey. However, the company does have a website that attracts more than ten

thousand visitors a day. The desperate marketing manager wonders: “What if

we post some simple questions: ‘What is your most favorite dessert?’ ‘How

often do you have dessert at lunch?’ Do you like rice pudding?’ If we put the

questions on our website, and invite our customers to respond, can we use their

answers to guide our decisions?” In other words, can companies treat online

responses to quick online polls as if they come from reliable, well-designed

surveys?

This study explores the usefulness of website polls as quick customer

survey instruments to provide market information to companies. In the study, a

website poll is defined as a form of survey with a limited number of closed

questions—often just a single question—posted on a website, with all site

visitors invited to submit their responses or votes. We focus on website polls

posted on company websites and directed to customers. Such polls are usually

placed in a prominent area of the site (e.g., the home page). Respondents can

access a summary of all current responses, often in the form of bar graphs or

pie charts. (See Figure 4.1 for an example of a website poll, related to the



Chapter 4

64

newly-designed Honda CR-Z, as posted on the Honda website.) Website polls

add a playful element to websites and increase their perceived interactivity

(Song and Zinkhan, 2008). Visitors may also be interested in the opinions of

fellow site visitors, thus increasing the perceived value of site visits. As

customer survey instruments, website polls may represent fast, cheap, and

easy-to-use methods of collecting information from (prospective) customers.

However, because website poll samples are by nature non-probabilistic and

self-selected, their representativeness is questionable. Participation in the polls

is limited to Internet users; therefore, respondent samples may not represent the

general population. For reviews of the limitation of Internet surveys, see

Couper (2000), Evans and Mathur (2005), Eysenbach (2005), and Fricker and

Schonlau (2002).

Despite these drawbacks, there are several reasons that website polls

may have potential as customer survey instruments. First, as is often the case in

social surveys, the target population of a customer survey may be a specific

group of (potential) customers rather than the general population. The selection

mechanism of website polls could be an advantage, because participants are

likely to be the company’s (potential) customers. However, as in other survey

forms, there may be selection bias caused by the decision to participate or not

in a website poll. Second, website polls gather a large number of participants in

a short time period at low cost (McDonald and Adam, 2003), making them one

of the most popular forms of web surveys (Couper, 2000). For example, the

Honda poll depicted in Figure 4.1 attracted more than ten thousand

participants, without any need to contact individual respondents or offer

incentives for participation. Third, Internet technologies control access to

website polls, preventing multiple submissions or the inclusion of irrelevant

subjects (Eysenbach, 2005). Fourth, with regard to missing data, the quality of

web surveys equals that of conventional surveys (such as mail surveys) and—

in the case of open-ended questions—exceeds them (Barrios et al., 2011). As a

result of these advantages, early research has shown the potential of website

polls to produce realistic estimates (Yoshimura ,2004).




65

Previous studies of web surveys have used the general population as

their research context (e.g., Curtice and Sparrow, 2010; Fricker et al., 2005;

Roster et al,. 2004; Schillewaert and Meulemeester, 2005; Strabac and Aalberg,

2011). Bethlehem (2009), for example, compared self-selection-based online

surveys and probability-based conventional surveys in predicting the

composition of the Dutch parliament and concluded that probability-based

surveys perform better when the target population is the general population.

Our study focuses on situations in which the target population differs from the

general population; it evaluates website polls as customer survey instruments

for companies. To our knowledge, we are the first to study website polls in the

context of a company’s customer base or market segment. We contribute to

existing literature by studying (i) sample characteristics of website polls

compared to telephone surveys and their representativeness with regard to

company customer databases, (ii) response differences between both survey

modes and the sources of these differences, and (iii) ways to correct the biases

observed. Knowledge of these issues will help companies make more informed

decisions about whether and how to use website polls for customer surveys.

Figure 4.1. Examples of website poll questions



Chapter 4

66

In the next section, we provide background information on Internet surveys,

particularly website polls, and a detailed description of our research questions.

We follow with a description of the research design and survey methodology.

Next, we present our empirical findings. We conclude with a discussion of the

results and suggestions for future research directions to further explore the

usefulness of website polls as customer survey instruments.

4.2. Background and Research Question Since Internet surveys were introduced about twenty years ago, they have

evolved into multiple forms, ranging from email surveys to interactive web

surveys. Couper (2000) provides a thorough discussion of web surveys and

suggests a classification based on their sampling methods. Web surveys can be

categorized into non-probability and probability-based methods. Website polls

belong to the non-probability category. They have limited coverage, are self-

selected, and often have little or no control over who responds to the polls.

Therefore, website poll samples are often not representative of the general

population. So far, website polls are popular, but are generally used for

entertainment purposes and not as surveys in the scientific sense (Couper,

2000).

In contrast, probability-based web surveys aim to achieve

representativeness by improving the sampling process using probability-based

sampling methods. The most popular survey types in this category use pre-

recruited panels. Panel members are recruited using probability-sampling

methods such as random digit dialing (RDD), to be representative of the

general population. However, empirical studies involving the general

population (e.g., Fricker et al., 2005; Loosveld and Sonck, 2008; Lugtig et al.,

2011; Strabac and Aalberg, 2011, Vicente and Reis, 2012) find significant

sample differences between pre-recruited panel web surveys and traditional

surveys.

Although support for the effectiveness of web surveys in representing

the general population is still lacking, it is interesting to review this issue when




67

the population of interest differs from the general population. For example, in a

survey of Internet banking, the population of interest is not the general

population but users of Internet banking facilities. The demographics of

banking users may differ from those of the general population; they may be

younger, more educated, and more technology-oriented. Hence, in this setting,

it is likely that web surveys can achieve equal or better sample

representativeness compared with traditional survey methods. We can expect

that web site polling samples are representative of the target population (i.e.,

the company’s customers) because the poll is posted on the company website.

Visitors to a particular company website are likely to be similar to, or even

belong to, the target segment of the company and the population of interest for

the survey.

A notable development is that response rates for telephone surveys are

declining. Curtin, Presser, and Singer (2005) show the telephone survey

response rate in the United States declined at an average rate of 1.5% per year

in the period 1997–2003. A similar situation is found in Europe (Blyth, 2008).

Many young people have never had a fixed telephone line; their mobile phones

are their only telephone connections. Since mobile phone numbers are

generally not listed, this growing segment will not be reached by RDD surveys.

Given these developments, it is not clear whether traditional probability-

sampling procedures such as those used for telephone surveys are superior to

self-selected web surveys with targeted respondents, such as website polls.

Accordingly, we test the following hypotheses:

H1a: The demographics of website poll respondents are comparable to the

demographics of telephone survey respondents.

H1b: The demographics of website poll respondents are comparable to the

demographics of the company’s customers.

H1c: The demographics of telephone interview respondents are comparable to

the demographics of the company’s customers.



Chapter 4

68

Another important issue related to web surveys is the mode effect. Mode

effects occur when respondents report different responses under different

survey modes (Voogt and Saris, 2005), (in our case, when answering a website

poll or a telephone interview). A number of studies have reported the existence

of mode effects (e.g., De Leeuw, 2005), especially when surveys contain

sensitive questions (Tourangeau and Smith, 1996). Several factors vary with

survey modes and may be the source of response differences. These include

media-related factors, information transmission differences, and interviewer

effects (De Leeuw, 1992). Although recent studies show web surveys can

deliver responses with similar quality as conventional surveys (Barrios et al.,

2011; Holland and Christian, 2009), a number of empirical studies show

responses to web surveys are different from responses to traditional surveys

(Deutskens et al., 2004). Smith (2001) compared face-to-face and web surveys

based on probability sampling using a panel constituted by RDD and found

systematic differences (especially regarding ‘do not know’ responses).

Subsequently, Grandcolas, Rettie, and Marusenko (2003) found response

distributions differ between web and paper-based questionnaires. In a

comparison between a telephone survey and a web survey with panel

respondents, Roster et al. (2004) concluded that the web survey garnered a

lower response rate, more item omissions, and more negative or neutral

evaluations than the telephone survey. In another study, Fricker et al. (2005)

found web respondents tended to give less differentiated responses and less

item non-response. Furthermore, Chang and Krosnick (2009) found web

surveys produced less socially desirable response bias than telephone surveys.

Because other kinds of web surveys tend to show mode effects, it is

interesting to study whether the responses to website polls differ from the

responses to telephone interviews. In practice, response differences can be

caused by both the mode of interviewing and differences in the samples.

Therefore, we formulate three hypotheses regarding these effects. The first

hypothesis includes a simple comparison of the answers in both surveys, the

second hypothesis incorporates possible differences in respondent




69

demographics that may influence responses, and the third hypothesis explores

the possible effects of the interviewing mode.

H2a: Responses to website polls and telephone interviews are different.

H2b: Differences in responses are related to the mode of interviewing, while

controlling for demographics.

H2c: Responses are related to demographics, while controlling for mode

effects.

When mode effects are absent, response differences among survey

modes may solely reflect non-coverage bias due to limited coverage of a single

mode of survey. There are several methods to cope with this problem. One

weighting method for reducing the bias of web surveys is propensity score

weighting (PSW). In experiments involving non-randomized subject

assignments, a direct comparison between treatments may be misleading

because there may be systematic differences (i.e., covariates) between subjects

exposed to each treatment. The propensity score—the conditional probability

of assignment to a particular treatment given a vector of observed covariates

(Rosenbaum and Rubin 1983) —can be used as an adjustment to remove bias

due to these systematic differences. In particular, PSW can be applied to reduce

biases caused by the inclination of certain groups of the population to

participate in web surveys. However, Loosveldt and Sonck (2008) show that

applying PSW does not always solve the problem, and the bias sometimes

becomes even greater. Yoshimura (2004) concludes PSW works only if the

covariates of the variables are observed and identified.

Another popular method of reducing bias is demographic weighting.

Taylor (2000) applied demographic weighting to several web survey data sets

and concluded it could solve the problem of non-coverage bias. However, not

all researchers have been successful in applying weighting methods for web

survey studies. For example, Vehovar, Manfreda, and Batagelj (1999) found

that weighting could not solve the non-coverage bias in their web surveys.



Chapter 4

70

They also found weighting could be problematic because responses can be

highly diverse, making it an inappropriate solution to the problem of non-

response bias.

Based on these extant empirical studies, we conclude it is not clear

whether some form of weighting can solve the problem of response biases

caused by non-coverage and non-response effects for a web survey. Previous

studies suggest that whether weighting works is contingent on several factors.

In this study we explore whether demographic weighting can be applied to

reduce response biases for website polls. This rather simple method may be

effective because website polls usually pose only simple questions and

demographic characteristics may be sufficient to explain and correct

differences in the actual responses. We formulate our last hypothesis as

follows:

H3: After applying demographic weighting, the responses to the website polls

and the telephone interviews are comparable.

4.3. Study Design We administered a telephone interview survey and a series of website polls in

collaboration with a Dutch insurance company. The questions in the poll were

classified into two groups: demographics and opinions. The demographics

include gender, age, and education (see Table 4.1). For the opinions, using

five-point scales, the respondents answered questions on the following four

issues:

1. The importance of comparing offers from various health insurance

companies (1 = very important and 5 = very unimportant).

2. The monthly premium difference that would make the respondent switch to

another insurance company (0-5 euros, 6-10 euros, 11-15 euros, more than

15 euros, do not switch based on premium difference).




71

3. Whether an insurance company should request a higher premium from

persons with serious diseases (1 = strongly agree and 5 = strongly disagree).

4. Whether an insurance company is allowed to reject someone who has

committed an insurance claim fraud (1 = strongly agree and 5 = strongly

disagree).

The formulation of the questions was the same for the telephone survey and the

website poll.

Respondents for the telephone interview survey were chosen randomly

from the list of customers in the company database. In total, 250 respondents

answered three demographic questions and four opinion questions during

telephone interviews. In the website polls, data were gathered from

questionnaires posted on the company website. In contrast to the telephone

survey, website poll respondents answered only four questions: three

demographic questions and one opinion question. Each week, we posted a

website poll with a different opinion question on the company’s website. In

total, we obtained 1,018 responses from website polls, consisting of 265, 250,

253, and 250 responses for opinion questions 1–4, respectively.

Table 4.1 Sample and customer characteristics Sample Target Population

Website Poll (n = 1,018)

Telephone Interview (n = 250)

Company Database

(N =+/- 80,000) Gender female 50.2% 50.0% 48.1%

male 49.8% 50.0% 51.9%

Age 18-25 years 41.3% 8.4% 21.6%

26-35 years 25.8% 18.4% 24.3%

36-45 years 17.1% 16.8% 18.5%

46-55 years 8.4% 22.4% 16.5% > 55 years 7.4% 34.0% 19.1%

Education Level

Primary school 1.5% 4.4% 9.0% Low secondary 13.0% 31.2% 24.0% High secondary 25.6% 23.2% 41.0%

Bachelor 39.9% 29.2% 16.0% Master-Doctoral 20.0% 12.0%

10.0%



Chapter 4

72

4.4 Results 4.4.1 Sample Characteristics We first compared the characteristics of both the interview and poll samples

with customers in the company database. The data are summarized in Table

4.1. In terms of gender, samples from both survey modes show an even

proportion of male and female respondents, which closely reflects the

customers’ gender distribution in the database. However, with regard to age

and education, distributions in both samples are different from the distributions

in the company database. Website poll respondents tend to be younger, with

about 40% younger than 26 years. They also tend to be more highly educated,

with 60% having a university degree (bachelor’s or master’s). In contrast,

telephone respondents tend to be older and less educated, with 56% older than

45 years and only 40% with a university degree. The distributions of age and

education of the company’s customers are in between the distributions of the

interview and poll samples.

We conducted Chi-square tests to examine whether the differences for

gender, age, and education were statistically significant. For gender, none of

the tests are significant (all p-values are larger than 0.05), whereas all three

comparisons—website versus customer database, telephone versus customer

database, and website versus telephone –—show highly significant differences

for age and education (all p-values are less than 0.001). Based on these results,

except for gender, we reject hypotheses H1a, H1b, and H1c for each

demographic.

Our first three hypotheses address the representativeness of both

samples. With two out of three demographics showing significant differences,

we conclude neither of the two samples, website poll or telephone interview, is

representative of the target population. Also, both samples differ in terms of

respondent demographics (for two of the three demographics).




73

4.4.2 Response Distribution Our next analyses addresses answers to the four opinion questions. Table 4.2

summarizes the responses in terms of means and standard deviations (see rows

labeled ‘unweighted’). We conducted a series of t-tests to examine the

differences in the means for both survey modes and found only the response

mean for Question 1 was significantly different (p = .004). Next, we examined

the distribution of the responses. We applied Chi-square tests to test for

differences in the proportion of responses in each category of answers between

the website polls and the telephone interviews. We found significant

differences for three of the four questions: Question 1 (p < .001), Question 2 (p

<. 001), and Question 4 (p =. 013). The difference was insignificant only for

Question 3 ( = .291). Therefore, although the means are only marginally

different, the responses are spread differently across the categories of answers.

Based on these two series of tests we accept H2a and conclude the responses to

the website polls and the telephone interviews are different.

4.4.3 Sample versus Mode Effects Hypotheses H2b and H2c focus on the source of response differences: are they

related to the sample characteristics, the modes of survey, or both? We employ

an analysis of variance (ANOVA) with four factors: mode (two levels), gender

(two levels), age (five levels), and education (five levels). The model also

includes interaction effects between the survey mode and each demographic

variable (i.e., mode x gender, mode x age, and mode x education). Table 4.3

displays a summary of the ANOVA results. Of the four questions, none shows

a significant main effect of the survey mode (all p > .05), leading to the

rejection of H2b. The only significant effect involving survey mode is related

to the interaction effect of mode and gender for Question 1. However, this

effect is small, reflected by a squared partial eta of only 0.9%. There are

significant main effects for all demographic variables, although each concerns

only one question. Education has a significant main effect for Question 2 (p =

.007), age for Question 3 (p = .023), and gender for Question 4 (p = .001).



Chapter 4

74

Table 4.2. Unweighted and weighted response mean estimates (on a 1-5 scale) Question 1 Question 2 Question 3 Question 4 Mean S.E Mean S.E Mean S.E Mean S.E

Website Poll

Unweighted 2.48 .062 3.37 .082 4.10 .048 2.22 .062 Weighted by age 2.41 .066 3.44 .144 4.17 .032 2.13 .037 Weighted by education 2.35 .079 3.49 .117 4.14 .185 2.12 .082

Telephone Interview

Unweighted 2.22 .065 3.39 .087 4.02 .056 2.26 .053 Weighted by age 2.22 .071 3.27 .136 3.95 .057 2.22 .040 Weighted by education 2.28 .077 3.47 .139 4.00 .059 2.35 .052

Therefore, we accept H2c and conclude that the responses are significantly

related to the sample characteristics.

By further examining the data, we found the response differences related

to the demographics have high face validity. Higher-educated respondents

regarded comparing offers from different insurers (Question 1) as more

important than lower-educated respondents, older respondents tended to

disagree on imposing higher premiums for persons with more serious diseases,

and females were more lenient on people with a fraud record. Again, we could

not find any particular response patterns related to the survey mode. With

regard to our second set of hypotheses we conclude the differences in

responses between both survey modes are minor and are more related to

differences in sample demographics than to the survey mode.

4.4.4 Weighting Results The final research question proposes demographic weighting as a potential

solution to correct biases caused by non-coverage. Because the ANOVA results

show significant effects of sample characteristics on responses and because we

also found differences in sample characteristics between both survey modes,

we cannot use website polls just only to replace the telephone interview.

Although t-tests detected marginal differences only between the response

means of both survey modes, the responses were distributed differently. Thus,

weighting to adjust for biases related to differences in sample characteristics




75

Table 4.3. ANOVA summary

p-values Source of Variation Question 1 Question 2 Question 3 Question 4 Survey Mode 0.538 0.957 0.143 0.067 Gender 0.351 0.785 0.972 0.001 Age 0.126 0.489 0.023 0.533 Education 0.146 0.007 0.720 0.728 Mode * Gender 0.034 0.888 0.951 0.199 Mode * Age 0.371 0.267 0.233 0.133 Mode * Education 0.212 0.424 0.238 0.317

could make the results of each survey mode more similar. Demographic

weighting could lower the bias problems because we found no significant mode

effect in the ANOVA results.

We have estimated the population response means using simple

demographic weightings based on age and education. We did not use gender

for the weighting because there are no significant differences in gender

distribution between both samples and the customer data in the company

database. The result summary is shown in Table 4.2 (see results weighted by

age and by education)

When we compare results across the questions and the weighting

variables, there is no clear pattern. For Question 1, the weightings reduce the

mean differences between both modes from .26 to .07 (see Table 4.4) without

much inflation of the standard errors. In contrast, weighting widens the

differences for Question 4 from .04 to .23. In the case of Question 2, weighted

by education, the mean differences are the same before and after weighting. In

general, weighting also tends to result in larger standard errors, although in

most cases the inflations are not substantial. However, we expected the means

estimated from weighted responses of both modes to be similar because we

found no significant mode effects in the ANOVA. We conducted a series of t-

tests to test the hypothesis that the difference between the mean estimated

using weighted telephone responses and the mean estimated using weighted

website polls responses is equal to zero (i.e., H0: �weighted telephone - �weighted website

polls = 0). The p-values of these tests are displayed in Table 4.4 (in parentheses).



Chapter 4

76

Because none of the eight p-values is less than 0.1, we can accept H3 and

confidently conclude there is no statistical difference between mean estimates

derived from weighted responses of the website polls and the telephone

interviews.

4.5 Discussion 4.5.1 Conclusion The ubiquity of the Internet has made it an attractive alternative technology for

conducting surveys (Tingling, Parent, and Wade, 2003). With an ever-

increasing number of Internet users (more than 3 billion in 2014) (Internet

World Stat, 2015), the Internet has become a powerful tool for collecting

information on individuals. Most of the world’s largest research firms provide

web surveys as part of their services (Evans and Mathur, 2005). Although

several methodologies for web surveys, each with their own strengths and

weaknesses, have been proposed (Couper, 2000; Evans and Mathur, 2003;

Eysenbach, 2005; Fricker and Schonlau, 2002), the web survey arena is still

developing (see, for example, Heinze et al., 2013; Jin, 2011; Van Selm and

Jankowski 2006; Williams, 2012). This study focuses on a less-studied variant

of web surveys, the website poll, which invites website visitors to submit quick

online responses to simple questions.

By conducting a series of website polls and a telephone interview

survey, we have explored the usefulness of website polls as customer survey

Table 4.4. Difference between website polls and telephone interview mean estimates (p-values from corresponding t-tests are in parentheses)

Mean estimates difference (D)

Data Question 1 Question 2 Question 3 Question 4

Unweighted -0.26 (0.004) 0.02 (0.867) -0.08 (0.259) 0.04 (0.589)

Weighted by age -0.19 (0.194) -0.17 (0.465) -0.22 (0.301) 0.09 (0.486)

Weighted by education -0.07 (0.468) -0.02 (0.740) -0.14 (0.562) 0.23 (0.381)




77

instruments. The results lead to three main conclusions. First, we find

differences in respondent characteristics between each of the survey modes and

the company’s customer database. Second, response differences between both

surveys are mainly attributable to sample effects, rather than mode effects.

Third, we find demographic weighting can be used to reduce the bias resulting

from sample effects, (in our study case, self-selection effects). Therefore, our

overall conclusion is that polls on a company website can be a fast, cheap, and

easy-to-use method of collecting information from (prospective) customers,

and they merit further customer research.

With regard to our first set of hypotheses, our analyses show sample

characteristics of the website polls and the telephone interview survey are

different from those of the target population (the company’s customer

database). In line with previous studies (e.g., Roster et al., 2004; Schillewaert

and Meulemeester, 2005), we find our website poll respondents are younger

and have higher education levels than both telephone survey sample and the

customer database. A possible explanation for this finding is that Internet users

are not spread homogeneously over the target population; there are still

demographic groups with a low Internet penetration. Furthermore, with regard

to age and education, telephone interview respondents’ characteristics are not

similar to those of the company’s customers, although our samples were

randomly drawn from the customer database. This finding suggests we should

be more aware of non-response problems in telephone interviews, as indicated

by Curtin et al. (2005), Tourangeau (2004), and Voogt and Saris (2005). As

younger people are less likely to have fixed telephone lines, they tend to be

under-represented in traditional telephone samples (Strabac and Aalberg 2011).

For the website polls we find the opposite effect: younger and more educated

individuals are over-represented in the sample.

Our study also shows differences in response distributions between the

website polls and the telephone interviews, which confirms previous findings

by Roster et al. (2004) and Schillewaert and Meeulemester (2005). In this case,

there could be confounding effects of sample characteristics due to self-



Chapter 4

78

selection and effects of the modes of survey. Based on ANOVA results, we

conclude response differences are much more affected by sample

characteristics than by mode effects. This finding is in line with the suggestion

of Vehovar et al. (2001) that the differences between web survey and other

survey modes becomes negligible in some research settings. The lack of mode

effects in our study is a propitious finding. Aside from the issue of non-

response bias, the absence of mode effects means the biases in website polls

and in telephone interviews are attributable to non-coverage due to limitations

of both survey modes in reaching the target population. Consequently,

demographic weighting can correct the bias caused by non-coverage in both the

website polls and the telephone interviews.

Given that some population groups are harder to reach using website

polls and that responses are affected by sample characteristics, weighting

becomes necessary to derive valid inferences from the website polls data. We

applied a simple weighting procedure based on demographic variables to show

the estimates based on weighted website polls data and weighted telephone

interview data are not statistically different. This finding is in line with Witte,

Amoroso, and Howard (2000), who argue that voluntary Internet-based survey

research can yield meaningfully comparable data about both Internet users and

larger populations. We also find estimates based on weighted data tend to have

larger standard error of estimates, which is not unexpected (see for example,

Vehovar et al., 2001). However, the inflation of the standard errors (32% on

average in our study) is modest.

There is a final issue of non-response bias. As mentioned, website polls,

as examples of open web surveys, are criticized for having samples based on

convenience sampling It is widely documented that web surveys result in lower

response rates (for a thorough meta-analysis on response rates comparison see

Manfreda et al., 2008 and Shih and Fan, 2008). However, non-response does

not necessarily lead to non-response errors, since non-response errors occur

only if the non-respondents would have provided different responses compared

to those who did respond (Manfreda et al., 2008; Voogt and Saris, 2005).




79

Groves (2006) presents empirical findings with regard to this statement.

Because this convenience sampling problem seems unresolved, other authors

argue that oversampling could add more credibility to a non-random web

survey, although oversampling cannot automatically solve the problems related

to convenience sampling (Couper, 2000). Due to their low administration cost

and speed of data gathering, website polls can provide sufficient responses to

qualify for oversampling.

4.5.2 Limitations and Directions for Future Research Our study has several limitations that suggest directions for future research.

First, we applied demographic weighting instead of more complex

weighting procedures such as propensity score weighting (PSW), widely used

for web survey studies (e.g Loosveldt and Sonck, 2008; Schonlau et al., 2009).

In practice, PSW is not applicable to website polls because the probability of a

respondent’s participation cannot be estimated (knowledge of background

characteristics requires participation). In online panel surveys, researchers can

derive this information from sampling frames. Future research might explore

whether complex weighting procedures result in smaller biases, thereby

warranting investment in collecting covariate information on the respondents.

Second, this study did not include conventional survey modes other than

telephone, such as postal mail or face-to-face interviews. Future research could

compare website polls with a broader range of survey modes.

Third, we did not address the design of the website poll, including

factors such as presentation of the polls, web page format, and incentive

scheme. We consider these issues to be promising avenues for future research.

By understanding the effects of types of questions, poll presentations, and

incentive schemes, models can be developed to reduce potential biases

resulting from this single mode of survey. Since website polls are voluntary by

nature, it is crucial to develop poll designs that are attractive to participants.

Fourth, our study was conducted with samples taken from one company

customer base and website (an insurance company). Different companies or



Chapter 4

80

industries may show different customer responses with regard to telephone or

online polls. For example, an insurance company website may attract only

visitors with (demographic) characteristics similar to the insurance company

customers. However, a fancy car company website may attract a wider range of

visitors who are not necessarily similar to the company customers or target

market. Replication of this study with a variety of companies or industries is

needed to understand to what extent the findings of this study can be

generalized.

Finally, we did not investigate the survey non-response. Although non-

response does not necessarily lead to non-response bias (Groves 2006), more

knowledge on this phenomenon in the specific case of website polls is needed

to better understand the non-response bias. A number of recent studies have

been conducted to examine the number of non-responses in online surveys

(e.g., Bruggen et al., 2011; Downes-Le Guin et al., 2012; Tehranian and

Bremer, 2012). These and other future studies on website polls will provide an

even richer understanding of how and when to apply website polls for customer

survey purposes. They will increase the usefulness of website polls as new and

promising customer survey instruments.

4.5.3 Practical Implications Interactivity is an important element of successful websites (Voorveld, Neijens,

and Smit 2009). Website polls can be used to make a site more interactive.

Most website polls or quick votes deal with trivial questions and merely serve

to entertain visitors. This study shows for simple questions, and after applying

demographic weighting, responses to website polls and telephone interviews

are not statistically different. This finding should encourage firms to further

explore the potential of website polls for customer research. Website polls are

attractive because they are cheap, fast, and easy to use. Website polls are

especially useful for exploratory purposes, such as capturing (prospective)

customers’ experiences and opinions. In terms of the costs and benefits of

web-based survey methods (Dahan and Hauser, 2000), development research




81

projects can benefit from the cost and time advantages of website polls. Further

research on the design elements of website polls should increase our insight

into the usability and value of website polls for more complex market research

studies.

The preconception of academic literature is that website polls are

unscientific (e.g., Couper, 2000; Simsek and Vega, 2001). Although we cannot

completely remove this limitation, we can minimize it. IP recognition

technology or cookies, for example, can trace participants, limit participation to

only certain regions, and prevent multiple submissions (Ball, 2013; Eysenbach,

2005; Gosling et al., 2004; Tingling et al., 2003). To minimize self-selection

bias, questions used in website polls should not be sensitive. By placing

website polls on company websites, no just general web portals such as Yahoo

or news websites, companies can reduce discrepancies between website poll

participants and their target population. Website polls are a promising and

fruitful tool, not only for increasing website interactivity and entertainment, but

also for collecting useful customer data.



Chapter 4

82



Conclusion and Implication

83

Chapter 5 Conclusion and Implications Internet technology is evolving and new technologies are finding their way into

commercial tools. As new forms of online marketing instruments emerge,

Internet marketing has become a dynamic process. Recent research findings

(Huang, Yen, and Hun, 2012; Sparrow, Liu, and Wegner, 2011) suggest the

Internet may impact consumers’ cognitive processes and cognitive styles,

including their thoughts, perceptions and memories. In the context of

marketing communications, these findings suggest the effect of advertising on

the Internet is different from its effect on traditional media, such as television

and print. Some studies have shown these differences. For example, Dijkstra,

Buijtels, and Van Raaij (2005) found Internet advertising is less effective in

evoking cognitive responses than print and television advertising. Dahlen,

Murray, and Nordenstam (2004) found Internet advertising outperformed print

advertising in transferring implicit meaning. These results lead us to question

which aspects of advertising’s effect on traditional media hold true for Internet

adverting. Indeed, this question has been one of the main research agendas of

Internet marketing (Ha, 2008; Schibrowsky, Peltier, and Nill, 2007).

The main goal of this dissertation is to contribute to Internet marketing

literature by providing a greater understanding of how the Internet impacts

marketing and advertising research. To pursue this goal, we conducted three

studies, as described in Chapters 2–4. In Chapter 2, we examine the effect of

banner exposures characteristics (number of exposures, spacing, and delay) on

consumer memory measured by banner and brand awareness. Chapter 3

focuses on how banner exposure effect measures are interrelated. Finally,

Chapter 4 addresses the value of online polls, a form of self-selected Internet

survey, as a marketing research instrument for companies.



Chapter 5

84

The next section provides an overview and discussion of the findings

from the three studies. This overview is followed by a discussion of their

managerial implications. Finally, the dissertation concludes with some

limitations of the studies presented in this dissertation, as well as suggestions

for further research directions.

5.1 Overview of Main Findings and Theoretical Implications The main research question of Chapter 2 (Study 1) is how banner advertising

number of exposure, spacing, and delay impact consumer memory, in terms of

banner and brand awareness. By analyzing data from 21 banner campaigns in a

multilevel analysis setting, the study reveals that the effect of number of

exposures on consumer memory for both the banner and the brand is initially

positive but becomes less so as the number increases. It further shows the

effect of banner spacing on consumer memory follows an inverted U-shape. In

other words, the positive effect of banner exposures is greater if the time

interval between exposures is not too short or too long; our analyses reveal that

the optimal interval is around 60 minutes. However, enhanced memory

resulting from banner exposures eventually decays over time. Furthermore,

mediation analysis suggests banner exposures affect consumer brand memory

only when the exposures improve consumer memory for the banner itself.

The results of this study are mostly in line with previous research on

effects of number of exposures on consumer memory of traditional advertising

media (e.g., Appleton-Knapp, Bjork, and Wickens, 2005; Heflin and Haygood,

1985; Singh et al., 1988) and banner advertising (Drèze and Hussherr, 2003;

Havlena and Graham, 2004). In addition to these findings, further analyses

reveal new insights on how banner exposures effect banner and brand

awareness, thereby extending our knowledge of how banner advertising works.

First, it appears that awareness wear-out—when more repetitions have no

significant effect or even have a negative effect—does not occur in banner

advertising. This finding contrasts with general findings of field studies on




85

brand information recall with regard to conventional advertising (e.g.,

Pechmann and Stewart, 1989, p. 295). Second, the observed spacing effect in

banner advertising follows a study-phase retrieval mechanism (Appleton-

Knapp et al., 2005), suggesting moderate spacing of banner exposures makes

consumers retrieve information from previous banner exposures, which in turn

enhances memory for the banner and information in the banner. Hence, the

consumer learning process associated with the brand is more effective if there

is just enough spacing time between exposures. Third, in contrast to previous

research on banner advertising (e.g., Drèze and Hussherr, 2002; Yaveroglu and

Donthu, 2008), our findings identify the importance of banner memory, and

highlight its crucial role.

Our goal in Chapter 3 (Study 2) is to determine the interrelationships

among banner effect variables to develop a comprehensive model that links

banner memory and attitude to brand memory and attitude, and ultimately to

purchase intention. In pursuing this goal, and in contrast to previous studies on

banner advertising effects, we examine the relationships between various effect

measures simultaneously. We analyze online survey data from respondents

who had been exposed to actual banner campaigns (29 campaigns in total),

using a multilevel structural equation modeling methodology. The main finding

of this study is that banner advertising—through banner memory, banner

attitude, brand memory, and brand attitude—exerts an indirect and positive

effect on purchase intentions. Although these effects of banner advertising are

mostly similar to those in traditional advertising, some notable differences are

revealed. First, we find the banner–brand memory relationship is stronger than

the banner attitude–brand memory relationship. This finding corroborates the

finding in Chapter 2 (Study 1) that memory for brands is mediated by memory

for banners. Second, we also find the strength of the banner–brand attitude

relationship is similar to that of the brand memory–brand attitude relationship.

These two findings indicate that the dual-mediation hypothesis, postulating that

the attitude towards the advertising is channeled through brand cognitive



Chapter 5

86

response to improve brand attitude, that fits well traditional media advertising,

may not fit banner advertising.

With regard to the use of survey data from self-selected online

respondents, Chapter 4 (Study 3) investigates whether companies can use

online surveys for their marketing research purposes. To answer this research

question, we conducted an experiment involving a telephone survey and a

series of online polls on a company website. Our analysis of the data from this

experiment provides several insights. First, not only do the respondent

characteristics differ between the two survey modes, they also differ from those

recorded in the company’s customer database. This reflects that conventional

communication media, such as land telephones, are used less and less, and the

superiority of probability surveys using this media cannot be warranted

because of increasing unreachability of part of the population (see also Blyth,

2008). Second, we find that response differences between both survey modes

are attributable to sample effects, not to survey mode effects. Third, in contrast

with previous studies, we find that demographic weighting can reduce the

response bias resulting from the sample effects between website polls and

telephone surveys. Compared to previous research on online surveys (e.g.,

Fricker et al., 2005; Roster et al., 2004; Schillewaert and Meulemeester, 2005;

Strabac and Aalberg, 2011), the findings of our study suggest the problem of

differences between a self-selected (online) survey and a simple random survey

can be alleviated. This finding may be related to the fact that our study

population is a company customer base and the online polls were placed on the

company website. Hence, there were no extremely under-represented (in terms

of general population) demographic groups that led to problematic weighting

(see Vehovar, Manfreda, Batagelj, 1999). Our finding that the response bias

can be alleviated suggests that in self-selected Internet surveys, the population

of reference and the location of online polls are important factors.

Some forms of Internet marketing instruments resemble their traditional

counterparts. For example, banners resemble print advertising and online polls

resemble paper- based polls. Nevertheless, this study shows the results, as well




87

as the underlying processes, of these comparable forms of marketing

communication and research are not necessarily the same. Many expected

relationships that are built on theories from traditional marketing still hold, but

some notable differences are found. For example, how the eventual effect of

advertising exposures eventually affects purchase intention is not the same for

online banner and traditional advertising. The difference lies in the path taken

for the effect to work in the context of the hierarchy of effects. Given that in

Internet marketing, the effect of contextual factors of consumer information

processing (e.g., type and scheduling exposure pattern of the online

advertising) and their interactions are more complex, the effects of Internet

marketing efforts cannot be predicted only from currently available marketing

theories. Marketing theories may require refinement or adjustment to be

applied successfully in the Internet setting.

5.2 Managerial Implications The results of our three studies have important implications for how managers

can use the Internet as an effective marketing communication and marketing

research medium. We distinguish between implications for banner advertising

and implications for Internet marketing research.

5.2.1 Implications for Banner Advertising Advertising media planning is an aspect of marketing communication that

managers need to address. Media planning requires decisions about the best

media schedule for advertising exposures, given a certain budget and campaign

duration (Naik, Mantrala, and Sawyer, 1998). The findings of our research that

the number of banner exposures has a diminishing positive effect, and that

spacing has an inverted-U shape effect on consumer memory, imply the best

banner campaign scheduling strategy is even or continuous scheduling, rather

than massed scheduling. For best results, consumers should be periodically and

continuously exposed to the banner during the campaign period. Our empirical



Chapter 5

88

model suggests an optimal period of about one hour (i.e., a period

corresponding to the top of the inverted-U shaped response curve). In practice,

cookie technology makes this kind of scheduling manageable. The diminishing

effect of number of exposures also implies overexposure will impact budget

efficiency, suggesting budget spent on banner advertising will be no longer be

optimally effective after a number of exposures to the same consumer. In our

study, the first ten exposures increased the probability of banner recognition by

3 percent, but the next 30 exposures increased the probability only by 1

percent. To avoid overexposure, advertisers can also monitor the number of

banner exposures to individual consumers by using cookie technology.

Furthermore, when managers evaluate the effectiveness of the banner

campaign, the timing of the measurement is important, because we find

memory of banner exposures decays over time. Measuring banner effectiveness

directly after the exposure may overestimate the effects (e.g., Havlena and

Graham, 2004). Finally, since memory for the banners mediates memory for

the brands from banner exposures, it is important to have simple banners with a

clear message (e.g., brand names, logos, endorsers) that is easy to process and

requires minimum cognitive resources. However, simple banners must be

capable of attracting consumer attention.

Our research also suggests the effect of banner exposures goes beyond

banner memory. Study 2 shows that although both banner memory and effect

have a positive relationship with brand memory, the relationship is stronger for

banner memory. This implies that when banners are used for advertising, they

need to facilitate stronger links to the brand than to the banner. Therefore,

banners should not distract from their main function of increasing brand

awareness. Lovely figures, cute animals, or beautiful nature images in banner

design may attract attention, but they should not distract already-low consumer

attention from the brand name, logo, or claims. Given that banner exposures

increase brand awareness in the consumer’s mind, an important implication of

our study is that spending on banner advertising may be justified, despite the

evidence of recent data that show click-through rates for banner advertising are




89

very low. Banner exposures affect banner memory; further, following a

hierarchy of effects, they improve brand attitude and increase brand purchase

intention.

5.2.2 Implications for Online Marketing Research The results of our research suggest a positive prospect for self-selected online

surveys posted on company websites, such as online polls. As more and more

people move to wireless communication devices, conventional surveys such as

random digit dialing are no longer more representative than online surveys,

because some population segments can no longer be reached through

traditional media. In Study 3, we find consumer characteristics in surveys are

highly affected by the characteristics of the users of that particular survey

medium, suggesting managers should consider mixed- mode surveys when

they need more representative responses. However, for a simple survey with a

specific target population—a common scenario in marketing research—a

company can use online polls as survey instruments to gather data for further

processing using statistical methods (e.g., weighting) to adjust possible biases.

Hence, in addition to including study questions, the polls should also include

background questions; such information is needed to correct bias, using

sample weighting methods. However, despite the potential of online polls,

some cautions still apply. Managers should ensure that ‘robots’ or automatic

submissions by software, multiple entries, or other attempts to bias the polls are

avoided by using appropriate Internet technology.

5.3 Limitations and Directions for Further Research There are several limitations to the findings of Chapter 2 (Study 1) and Chapter

3 (Study 2). First, due to limitations of our data, we were not able to study

effects of banner design (e.g., size, color). Previous studies (e.g., Chen, Ross,

Yen, and Akhapon, 2009; Li and Bukovac, 1999) indicate banner designs also

have effects on banner effectiveness measures. Studies 1 and 2 could be

extended to include banner design factors to examine their moderating effects



Chapter 5

90

on the relationships established in the studies. Second, the data used in our

studies are related to established brands only. Hence, the findings in both

chapters may not be generalizable to brands that are new or less popular. It

would be interesting and relevant to extend our research to the effectiveness of

banner advertising for newer and less popular brands. Third, we used cross-

sectional data to test the hypothesized relationships in a causal context (i.e.,

advertising hierarchy of effects). Although this research is based on a large

amount of data from a wide range of actual banner campaigns, thereby

increasing generalizability, experimental data could provide stronger evidence

for the suggested causality of our findings.

Our research also does not take into account the effect of other media

advertising the same brands concurrent to their banner advertising campaign.

This opens the door to future research that examines cross-media effects of

banner and other media advertising. By analyzing the advertising effect of

budget allocation among the media, Naik and Peters (2009) show synergies

between online and offline advertising. Zenetti, Bijmolt, Leeflang, and Klapper

(2014) find substantial interaction effects of various types of online and

television advertising. However, it is still a challenge to determine how optimal

synergies can be built from scheduling exposures across multiple media.

Managers are faced with questions about how many exposures, which order of

media exposures, and what delays between exposures within and across media

are most effective.

In Study 1, we do not test the effect of number of exposures, spacing,

and delay on brand attitude and purchase intention. By relating the findings of

Study 2 to the findings of Study 1, we can hypothesize that effects of banner

exposure characteristics on brand memory would also be observed on brand

attitude and purchase intention. Future research can be conducted to test this

hypothesis.

A major limitation in Chapter 5 (Study 3) is the sole use of telephone

survey as the traditional survey in our experiment. Previous research (e.g., De

Leeuw, 1992; Dillman and Christian, 2005) found differences between surveys




91

using different traditional media. Future research could include other types of

traditional surveys in comparing the results with online polls. Another

limitation of this study is the format of online polls used in the experiment. We

use only one type of poll. Literature on online surveys indicates differences

between online survey formats, both in terms of response rates (Bethlehem,

2010) and results (Healey and Gendall, 2008). This limitation suggests our

study could be extended to include different online poll design aspects and

question types, for example, text only, decoratively visual, functionally visual,

and gamified (see Downes-Le Guin, et. al., 2012) in future experiments.

Further, given the limitations of the experimental setting, our study did not

elaborate on the effect of non-response bias. In some surveys, non-response can

be a significant source of survey response bias (Groves, 2006). Future research

could combine aspects of online poll design, types of questions, and non-

response effect on the quality of self-selected online surveys. By conducting a

more complex experiment involving these factors, the main effects as well as

interaction effects of these factors could be understood and the appropriate

action needed to minimize the bias of online surveys could be identified and

implemented.

Finally, online polls generate a massive amount of big data—often tens

of thousands of responses. When online polls are equipped with Internet

technology that collects even more data (such as IP recognition and geo-

tagging technology, with area demographics info and access times), they

provide market researchers with the opportunity to construct more

representative samples by employing statistical methods such as resampling

(Yu, 2003) and sample matching (Dickson, Deutsch, Wang, and Dube, 2006).

Future research could explore these techniques to overcome the potential bias

problem of online polls. Another avenue for future research is to use data

fusion methodology (e.g., Van Hattum and Hoijtink, 2008) to better understand

consumers. Users are increasingly accessing the Internet using their mobile

devices. Unlike Internet access through PCs, with issues related to cookies and

multiple users, the fact that one mobile telephone number represents one user



Chapter 5

92

allows consumer Internet behavior to be tracked more closely and collected

more completely. With this kind of data, a wide range of research questions can

be addressed and answered. We believe that findings from the studies in this

dissertation can serve as a useful foundation for future research in Internet

marketing.



References

93

References Appleton-Knapp, S.L., R.A. Bjork, and T.D. Wickens (2005). Examining the

Spacing Effect in Advertising: Encoding Variability, Retrieval Processes,

and Their Interactions. Journal of Consumer Research, 32(2), 266-276.

Asparouhov, T. and B.O. Muthén (2010). Multiple imputation with Mplus.

Retrieved from http://www.statmodel.com/download/Imputations7.pdf.

(February 10, 2012).

Baddeley, A.D. (1997). Human Memory: Theory and Practice (Rev. ed., pp. 9-

27). Hove, UK: Psychology Press.

Bagozzi, R.P. and A.J. Silk (1983). Recall, recognition, and the measurement

of memory for print advertisement. Marketing Science, 2(2), 95-134.

Ball, A. (2013). The cookie is still crumbling: the challenges facing cookie

tracking research. International Journal of Market Research, 55(1) 34-41.

Baron, R.M and D.A. Kenny (1986). The Moderator-Mediator Variable

Distinction in Social Psychological Research: Conceptual, Strategic and

Statistical Considerations. Journal of Personality and Social Psychology,

51 (6), 1173–1182.

Barrios, M., A. Villarroya, A. Borrego and C. Olle (2011). Response Rates and

Data Quality in Web and Mail Surveys Administered to PhD Holders.

Social Science Computer Review, 29 (2), 208-220.

Barry, T.E. and D.J. Howard (1990). A review and critique of the hierarchy of

effects in advertising. International Journal of Advertising, 9, 121-135.

Bergkvist, L., & J.R. Rossiter (2007). The predictive validity of multiple-item

versus single-item measures of the same constructs. Journal of Marketing

Research, 44(2), 175–184.

Bethlehem, J.K. (2009). Applied survey methods: a statistical perspective.

Hoboken, NJ : Wiley.



References

94

Bian, X. and L. Moutinho (2011). The Role of Brand Image, Product

Involvement, and Knowledge in Explaining Consumer Purchase

Behaviour of Counterfeits: Direct and Indirect Effects. European Journal

of Marketing, 45(1/2), 191-216.

Blair, M.H. (2000). An Empirical Investigation of Advertising Wearin and

Wearout. Journal of Advertising Research, 40(6), 95-100.

Blyth, B. (2008). Mixed Mode: the only ‘fitness’ regime? International Journal

of Market Research, 50 (2), 241-266.

Bollen, K.A. (1989). Structural Equations with Latent Variables. New York:

Wiley.

Breur, T. (2011). Data Analysis Across Various Media: Data Fusion, Direct

Marketing, Clickstream Data and Social Media. Journal of Direct, Data

and Digital Marketing Practice, 13(2), 95–105.

Briggs, R. and N. Hollis (1997). Advertising on the web: Is there response

before click-through? Journal of Advertising Research, 37(2), 33-45.

Broussard, G. (2000). How Advertising Frequency Can Work to Build Online

Advertising Effectiveness. International Journal of Market Research, 42

(4), 439–457.

Brown, S.P. and D.M. Stayman (1992) Antecedents and consequences of

attitude toward the ad: a meta-analysis. Journal of Consumer Research,

19(1), 34-51.

Bruce, N.I., K. Peters, and P. Naik (2012) Discovering how advertising grows

sales and builds brands. Journal of Marketing Research, 49 (6), 793-806.

Bruggen, E., M. Wetzels, K. de Ruyter, and N. Schillewaert (2011). Individual

differences in motivation to participate in online panels. International

Journal of Market Research 53 (3): 369-390

Bruner, G.C. (2009). Marketing Scales Handbook (Vol. 5). Carbondale,

Illinois: GCBII Production.

Bruner, G.C., K.E. James, and P.J. Hensel (2001). Marketing Scales Handbook

(Vol. 3). Chicago: American Marketing Association



References

95

Bullock, H.E., L.L. Harlow, and S.A. Mulaik (1994). Causation issues in

structural equation modeling. Structural Equation Modeling, 1 (3), 253-

267.

Burke, D.M and L.L. Light (1981). Memory and Aging: The Role of Retrieval

Processes. Psychological Bulletin, 90 (3), 513–546.

Burke, R.R. and T.K. Srull (1988). Competitive Interference and Consumer

Memory for Advertising. Journal of Consumer Research, 15, 56–68.

Callegaro, M., R. Baker, J. Bethlehem, A.S. Goritz, J.A Krosnick, and P.J.

Lavrakas, eds. (2014). Online Panel Research: A Data Quality

Perspective. Sussex, UK: Wiley.

Campbell, M.C. and K.L. Keller (2003). Brand Familiarity and Advertising

Repetition Effects. Journal of Consumer Research, 30 (2), 292–304.

Carr, N. (2010). The Shallows: What the Internet is Doing to Our Brains. New

York, NY: W.W. Norton and Company.

Chaffey, D. (2015). Display Advertising Clickthrough Rates. Retrieved from:

http://www.smartinsights.com/archive/internet-advertising/internet-

advertising-analytics/ (July 4, 2015).

Chang, L. and J.A. Krosnick. (2009). National Surveys via RDD Telephone

Interviewing versus the Internet: Comparing Sample Representativeness

and Response Quality. Public Opinion Quarterly, 73 (4), 641-678.

Chatterjee, P. (2008). Are unclicked ads wasted? Enduring effects of banner

and pop-up ad exposures on brand memory and attitudes. Journal of

Electronic Commerce Research, 9(1), 51-61.

Chaudhuri, A. and R. Buck (1995). Media differences in rational and emotional

responses to advertising. Journal of Broadcasting & Electronic Media,

39(1), 109-125

Chen, V., W.H. Ross, D.C. Yen, and L. Akhapon (2009). The effect of types of

banner ad, web localization, and customer involvement on Internet users’

attitudes. Cyber Psychology and Behaviour, 12(1) 71-73.



References

96

Cole, C.A. and M.J. Houston (1997). Encoding and Media Effects on

Consumer Learning Defficiencs in the Elderly. Journal of Marketing

Research, 24 (1), 55–64.

Comley, P. (1996). Internet Surveys: The Use of the Internet as a Data

Collection Method. ESOMAR/EMAC: Research Methodologies for “The

New Marketing” Symposium, ESOMAR Publication Services, 204, 335–

346.

Couper, M.P. (2000). Web Survey: A review of issues and approaches. Public

Opinion Quarterly, 64 (4), 464-494.

Curtice, J. and N. Sparrow (2010). Too clever by half? The impact of

weighting and adjustments on the accuracy of the polls. International

Journal of Market Research 52 (5), 678-686.

Curtin, R., S. Presser, and E. Singer (2005). Changes in Telephone Survey

Non-response Over the Past Quarter Century. Public Opinion Quarterly,

69 (1): 87-98.

D’Agostino, P. R. and P. DeRemer (1973). Repetition Effects as a Function of

Rehearsal and Encoding Variability. Journal of Verbal Learning and

Verbal Behavior, 12 (1), 108–13.

Dahan, E. and J.R. Hauser (2002). The Virtual Customer. Journal of Product

Innovation Management, 19 (5): 332-353.

Dahlen, M. (2001). Banner Advertisements Through a New Lens. Journal of

Advertising Research, 41 (4), 23–30.

Dahlen, M., M. Murray, and S. Nordenstam (2004). An empirical study of

perceptions of implicit meanings in World Wide Web advertisements

versus print advertisements. Journal of Marketing Communication, 10(1),

35-47.

Dark, V.J. and G.R. Loftus (1976). The Role of Rehearsal in Long-Term

Memory Performance. Journal of Verbal Learning and Verbal Behavior,

15, 479-490.

De Leeuw, E.D. (1992). Data Quality in Mail, Telephone, and Face-to-face

Surveys. Amsterdam, The Netherlands: TT-Publicaties.



References

97

De Leeuw, E.D. (2005). To Mix or Not to Mix Data Collection Modes in

Surveys. Journal of Official Statistics 21 (2), 233-255.

De Pelsmacker, P., M. Geuens, and P. Anckaert (2002). Media context and

advertising effectiveness: The role of context appreciation and context/ad

similarity. Journal of Advertising, 31(2), 49-61.

Deutskens, E., K. de Ruyter, M. Wetzels, and P. Oosterveld (2004). Response

rate and response quality of Internet-based surveys: an experimental

study. Marketing Letters, 15 (1), 21-36.

Dickson, C. A., C.K. Deutsch, S.S. Wang, & W.V. Dube (2006). Matching-to-

sample assessment of stimulus overselectivity in students with intellectual

disabilities. American Journal on Mental Retardation, 111, 447–453

Dijkstra, M., H.E.J.J.M Buijtels, and W.F. van Raaij (2005). Separate and Joint

Effects of Medium Type on Consumer Responses: A Comparison of

Television, Print, and the Internet. Journal of Business Research, 58(3),

377–386.

Dillman, D.A. and L.M. Christian (2005). Survey mode as a source of

instability in responses across surveys. Field methods, 17(1), 30-52

Downes-Le Guin, T., R. Baker, J. Mechling, and E. Ruyle (2012). Myths and

realities of respondent engagement in online surveys. International

Journal of Market Research, 54 (5), 613-633.

Draganska, M., W. R. Hartmann, and G. Stanglein (2014). Internet vs

Television Advertising: A Brand Building Comparison. Journal of

Marketing Research, 51 (5), 578-590.

Drèze, X. and F. Hussherr (2003). Internet Advertising: Is Anybody Watching?

Journal of Interactive Marketing, 17 (4), 8–23.

duPlessis, E. (1994). Recognition Versus Recall. Journal of Advertising

Research, 34 (3), 75–91.

duPlessis, E. (2008). The Advertised Mind: Ground-Breaking Insights into

How Our Brains Respond to Advertising. London: Kogan Page.



References

98

Ecommerce Foundation (2015). Global B2C Ecommerce Report 2015 Light.

Retrieved from: http://www.ecommerce-europe.eu/stream/global-b2c-

ecommerce-report-2015-light.pdf (December 26, 2015).

Evans, J.R and A. Mathur (2005). The value of online surveys. Internet

Research, 15 (2), 195-219.

Eysenbach, G. (2005). Using the Internet for Surveys and Research. In: K.J

Hannah and M.J Ball (Eds), Evaluating the Organizational Impact of

Healthcare Information Systems (pp.129-143). New York, NY: Springer.

Fang, X., S. Singh, and R. Ahluwalia (2007). An Examination of Different

Explanations of Mere Exposure Effect. Journal of Consumer Research,

34(1), 97-103.

Fennis, B.M. and W. Stroebe (2010). The Psychology of Advertising. East

Sussex: Psychology Press.

Finn, A. (1992). Recall, recognition, and the measurement of memory for print

advertisements: A reassessment. Marketing Science, 11(1), 95-100.

Fricker, R.D. and M. Schonlau (2002). Advantages and Disadvantages of

Internet Research Surveys: Evidence from the Literature. Field Methods,

14 (4), 347-367

Fricker, R.D., M. Galesic, R. Tourangeau, and T. Yan (2005). An

experimental comparison of Web and telephone surveys. Public Opinion

Quarterly, 69 (3), 370-392.

Furrer, O. and D. Sudharshan (2001). Internet Marketing Research:

Opportunities and Problems. Qualitative Market Research, 4 (3), 123–

129.

Garson, G.D. (2013). Validity and Reliability. Asheboro, NC: Statistical

Association Publishing.

Gosling, S.D., S. Vazire, S. Srivastava, and O.P John. (2004). Should we trust

web-based studies? A comparative analysis of six preconceptions about

Internet questionnaires. American Psychologist, 59 (2), 93-104.

Grandcolas, U., R. Rettie and K. Marusenko. (2003). Web survey bias: sample

or mode effect? Journal of Marketing Management, 19 (5-6), 541-561.



References

99

Gregan-Paxton, J. and B. Loken (1997). Understanding Consumer Memory for

Ads: A Process View in Measuring Advertising Effectiveness, eds. W.D.

Wells, Mahwah, NJ: Lawrence Erlbaum Associates, pp. 183-202.

Groves, R.M. (2006). Non-response Rates and Non-response Bias in

Household Surveys. Public Opinion Quarterly, 70 (5), 646-675.

Ha, L. (2008). Online Advertising Research in Advertising Journals: A

Review. Journal of Current Issues and Research in Advertising, 30(1),

31-48.

Havlena, W.J. and J. Graham (2004). Decay Effects in Online Advertising:

Quantifying the Impact of Time Since Last Exposure on Branding

Effectiveness. Journal of Advertising Research, 44 (4), 327–332.

Healey, B. and P. Gendall (2008). Asking the age question in mail and online

surveys. International Journal of Market Research, 50(3), 309-317.

Heflin, D.T.A and R.C. Haygood (1985). Effects of Scheduling on Retention of

Advertising Messages. Journal of Advertising, 14 (2), 41–64.

Heinze, A., E. Ferneley, and P. Child (2013). Ideal participatns in online

market research. International Journal of Market Research 55 (6): 769-

789.

Hoffman, D.L. and T.P. Novak (1996). Marketing in Hypermedia Computer-

Mediated Environments: Conceptual Foundations. Journal of Marketing,

60 (3), 50–68.

Holland, J.L. and L.M. Christian (2009). The Influence of Topic Interest and

Interactive Probing on Responses to Open-Ended Questions in Web

Surveys. Social Science Computer Review, 27 (2), 196-212.

Hollis, N. (2005). Ten years of learning on how online advertising builds

brands. Journal of Advertising Research, 45(2), 255 – 268.

Hox, J.J. (2002). Multilevel analysis: Techniques and applications. Mahwah,

NJ: Erlbaum.

Hoyer, W.D. and D.J. Macinnis (2008). Consumer Behavior (5th ed.). Mason,

Ohio: South-Western Cengage Learning.

Huang, A.H., D.C. Yen, and S-Y Hung (2012). Research issues of Internet-



References

100

integrated cognitive style. Computers in Human Behavior, 28(5), 1547-

1554.

Hutchinson, J.W. and D.L. More (1984). Issues Surrounding the Examination

of Delay Effects in Advertising. Advances in Consumer Research, 11,

650-655.

IAB (2011). IAB Internet Advertising Revenue Report 2010 Full Year Report.

Retrieved from:

http://www.iab.net/media/file/IAB_Full_year_2010_0413_Final.pdf.

Retrieved on March 3rd, 2013

IAB (2015). IAB Internet Advertising Revenue Report 2014 Full Year Report.

Retrieved from:

http://www.iab.net/media/file/IAB_Internet_Advertising_Revenue_FY_2

014.pdf. Retrieved on June 25th, 2015.

IAB Europe (2012). Mediascope Europe 2012 Pan-European Launch

Summary. Retrieved from: http://www.iabeurope.eu/research-and-

papers/mediascope-2012-pan-european-launch-presentation-summary.

Retrieved on June 25, 2015.

Iaobucci, D., N. Saldanha and X. Deng (2007). A Meditation on Mediation:

Evidence That Structural Equations Models Perform Better Than

Regressions. Journal of Consumer Psychology, 17 (2), 140-154.

Illieva, J., S. Baron, and N. Healey (2002). Online Surveys in Marketing

Research: Pros and Cons. International Journal of Market Research,

44(3), 361-382.

Internet World Stat. (2015). Internet Usage Statistics. Retrieved from

http://www.Internetworldstats.com/stats.htm. Retrieved on on July 2,

2015.

Janiszewski, C., H. Noel, and A.G. Sawyer (2003). A Meta-Analysis of the

Spacing Effect in Verbal Learning: Implications for Research on

Advertising Repetition and Consumer Memory. Journal of Consumer

Research, 30 (1), 138–149.



References

101

Jin, L. (2011). Improving response rates in web surveys with default setting.

International Journal of Market Research 53 (1): 75-92.

Keller, K.L. (1987). Memory Factors in Advertising: The Effect of Advertising

Retrieval Cues on Brand Evaluations. Journal of Consumer Research, 14

(3), 316-333.

Keller, K.L. and D.A. Aaker (1992). The Effects of Sequential Introduction of

Brand Extensions. Journal of Marketing Research, 29(1), 35-50.

Kline, R.B. (2010). Principle and practice of structural equation modeling.

New York: Guilford Press.

Ko, H., C-H. Cho, and M.S. Roberts (2005) Internet Uses and Gratifications: A

Structural Equation Model of Interactive Advertising. Journal of

Advertising, 34 (2), 57–70.

Kotler, P. and K.L. Keller (2011). Marketing Management (14th ed.). Upper-

Saddle River, New Jersey: Prentice Hall.

Kozinets, R.V. (2002). The field Behind the screen: Using Netnography For

Marketing Research In Online Communities. Journal of Marketing

Research, 39 (1), 61–72.

Krishnan, H.S. and D. Chakravarti (1999). Memory Measures for Pretesting

Advertisements: An Integrative Conceptual Framework and a Diagnostic

Template. Journal of Consumer Psychology, 8 (1), 1–37.

Laurent, G., J-N. Kapferer, and F. Roussel (1995). The Underlying Structure of

Brand Awareness Scores. Marketing Science, 14 (3), G170–G179.

Lee, A.Y. and A.A. Labroo (2004). The effect of conceptual and perceptual

fluency on brand evaluation. Journal of Marketing Research, 41(2), 151-

165.

Lee, J.W. and J-H. Ahn (2012). Attention to banner ads and their effectiveness:

an eye-tracking approach. International Journal of Electronic Commerce,

17(1), 119-137.

Li, H. and J.L. Bukovac, J.L. (1999). Cognitive Impact of Banner Ad

Characteristics: An Experimental Study. Journal of Mass Communication

Quarterly, 76(2), 341–53.



References

102

Loosveldt, G. and N. Sonck. (2008). An Evaluation of the weighting

procedures for an online access panel survey. Survey Research Methods,

2 (2), 93-105.

Lucas, D.B. (1960). ABCs of ARF’s PARM. Journal of Marketing, 25 (1), 9–

20.

Lugtig, P., Lensvelt-Mulders, G.J.L.M., Frerichs, R., Greven, A. 2011.

Estimating non-response bias and mode effects in a mixed-mode survey.

International Journal of Market Research, 53 (5), 669-686.

Lynch, J.G. and T.K. Srull (1982). Memory and Attentional Factors in

Consumer Choice: Concepts and Research Methods. Journal of

Consumer Research, 9 (1), 18–36.

MacKenzie, S.B. and R.J. Lutz (1989). An empirical examination of the

structural antecedents of attitude toward the ad in an advertising

pretesting context. Journal of Marketing, 53(2), 48-65.

MacKenzie, S.B, R.J. Lutz, and G.E. Belch (1986). The role of attitude toward

the ad as a mediator of advertising effectiveness. Journal of Marketing

Research, 23(2), 130-143.

MacKenzie, S.B. and R.A. Spreng (1992). How does motivation moderate the

impact of central and peripheral processing on brand attitudes and

intentions? Journal of Consumer Research, 18(4), 519-529.

MacKinnon, D.P and J.H. Dwyer (1993). Estimating Mediated Effect in

Prevention Studies. Evaluation Review, 17 (2), 144–158.

Mahajan, V. and E. Muller (1986). Advertising Pulsing Policies for Generating

Awareness for New Products. Marketing Science, 5 (2), 89–106.

Manchanda, P., J.-P. Dubé, K.Y. Goh, and P.K. Chintagunta (2006). The effect

of banner advertising on Internet purchasing. Journal of Marketing

Research, 43(1), 98-108.

Manfreda, K.L., M. Bosnjak, J. Berzelak, I. Haas, and V. Vehovar (2008). Web

surveys versus Other Survey Modes: A Meta Analysis Comparing

Response Rates. International Journal of Market Research, 50 (1), 79-

104.



References

103

Mantonakis, A., B.W.A. Whittlesea and C. Yoon (2008). Consumer memory,

fluency, and familiarity. In C. Haugtvedt, P.M. Herr, and F.R. Kardes

(Eds.), Handbook of Consumer Psychology New York: Taylor & Francis,

pp. 77-102.

MarketingChart (2008). 875MM consumers have shopped online – up 40% in

two years. Available at: http://www.marketingcharts.com/online/875mm-

consumers-have-shopped-online-up-40-in-two-years-3225/. Retrieved on

June 25, 2015.

McDonald, H. and S. Adam (2003). A comparison of online and postal data

collection methods in marketing research. Marketing Intelligence and

Planning, 21 (2), 85-95.

McKinsey (2014). Global Media Report 2014. McKinsey & Company.

Available at:

http://www.mckinsey.com/client_service/media_and_entertainment/latest

_thinking/global_media_report_2014. Retrieved on: June 25, 2015.

Mesak, H.I. (2002). On the Impact of Initial Performance on the Effectiveness

of Advertising Pulasation Policies. Journal of the Operational Research

Society, 53 (11), 1247–1255

Meyers-Levy, J. and D. Maheswaran (1991). Exploring differences in males’

and females’ processing strategies. Journal of Consumer Research, 18(1),

63-70.

Meyers-Levy, J. and B. Sternthal (1991). Gender differences in the use of

message cues and judgments. Journal of Marketing Research, 28 (1), 84-

96

Mitchell, A.A. and J.C. Olson (1981). Are Product Attribute Beliefs the Only

Mediator of Advertising Effects on Brand Attitude? Journal of Marketing

Research, 18, 318-332.

Montoya-Weiss, M.M, A.P. Massey, and D.L. Clapper (1998). On-Line Focus

Groups: Conceptual Issues and a Research Tool. European Journal of

Marketing, 32 (7/8), 713–723.



References

104

Morwitz, V. G., E. Johnson, and D. Schmittlein (1993). Does measuring intent

change behavior? Journal of Consumer Research, 20 (1), 46–61.

Muthén, B. O. (1984). A general structural equation model with dichotomous,

ordered categorical, and continuous latent variable indicators.

Psychometrika, 49(1), 115-132

Muthén, B. O. (1989). Latent variable modeling in heterogeneous populations.

Psychometrika, 54(4), 557-585.

Muthén, B. O. (1994). Multilevel covariance structure analysis. Sociological

Methods and Research, 22(3), 376–398.

Muthén, B. O. (1997). Latent variable modeling of longitudinal and multilevel

data. In A.E. Raftery (Ed.), Sociological Methodology. Cambridge, MA:

Blackwell, pp. 453-480.

Muthén, B. O. and T. Asparouhov (2002) Latent variable analysis with

categorical outcomes: Multiple-group and growth modeling in Mplus.

Mplus Web Note No. 4, at http://

www.statmodel.com/mplus/examples/webnote.html

Muthén, B. O. and A. Satorra (1985). Complex sample data in structural

equations modeling. Sociological Methodology, 25, 267-316.

Muthén, L. K., and B.O. Muthén (2010). Mplus User’s Guide: Statistical

Analysis with Latent Variables (6th ed.). Los Angeles, CA: Muthén and

Muthén.

Naik, P.A and K. Peters (2009). A Hierarchical Marketing Communications

Model of Online and Offline Media Synergies. Journal of Interactive

Marketing, 23 (4), 288–299.

Naik, P.A. and K. Raman (2003). Understanding the impact of synergy in

multimedia communications. Journal of Marketing Research, 40(4), 375-

388.

Naik, P.A., M.K. Mantrala, and A.G. Sawyer (1998). Planning Media

Schedules in the Presence of Dynamic Advertising Quality. Marketing

Science, 17(3), 214-235.



References

105

Noel, H. and B. Vallen (2009). The Spacing Effect in Marketing: A Review of

Extant Findings and Directions for Future Research. Psychology and

Marketing, 26 (11), 951–969.

Nordhielm, C.L. (2002). The influence of level of processing on advertising

repetition effects. Journal of Consumer Research, 29(3), 371-382

Pechmann, C. and D.W. Stewart (1989). Advertising Repetition: A Critical

Review of Wearin and Wearout, in Current Issues and Research in

Advertising 1988, James H. Leigh and Claude R. Martin Jr., eds., Ann

Arbor, MI: University of Michigan, 285–329.

Percy, L. and R. Elliott (2005). Strategic advertising management (2d. edition).

Oxford: Oxford University Press,

Petty, R.E., J.T. Cacioppo and D. Schumann (1983). Central and peripheral

routes to advertising effectiveness: the moderating role of involvement.

Journal of Consumer Research, 10(2), 135-146.

Pincott, G. and A. Branthwaite (2000). Nothing New Under the Sun?

International Journal of Market Research, 42 (2), 137–155.

Plummer, J., S. Rappaport, T. Hall, and R. Barocci (2007). The Online

Advertising Playbook: Proven Strategies and Tested Tactics from The

Advertising Research Foundation. Hoboken, New Jersey: John Wiley and

Sons.

Preacher, K.J., M.J. Zyphur, and Z. Zhang (2010). A General Multilevel SEM

Framework for Assessing Multilevel Mediation. Psychological Methods,

15 (3), 209–233.

Putrevu, S. and K.R. Lord (1994). Comparative and Noncomparative

Advertising: Attitudinal Effects under Cognitive and Affective

Involvement Conditions. Journal of Advertising, 23, 77-90.

Rabe-Hesketh, S., A. Skrondal, and X. Zheng (2007). Multilevel structural

equation modeling. In: S.Y. Lee (Ed.), Handbook of latent variable and

related models. Amsterdam: Elsevier, pp. 209-227.

Rivers, D. (2006). Sample matching: representative sampling from Internet

panels. Palo Alto, CA: Polimetrix, Inc.



References

106

Romano, N.C. and J. Fjermestad(2003). Electronic Commerce Customer

Relationship Management: A Research Agenda. Information Technology

and Management, 4, 233–258.

Rosenbaum, P.R. and D.B. Rubin. (1983). The central role of the propensity

score in observational studies for causal effects. Biometrika, 70 (1), 41-

55.

Roster, C.A., R.D. Rogers, G. Albaum and D. Klein (2004). A comparison of

response characteristics from web and telephone surveys. International


Rubin, D.B. (1996). Multiple imputation after 18+ years. Journal of the

American Statistical Association, 91(434), 473-489.

Sasieni, M.W. (1989). Optimal Advertising Strategies. Marketing Science, 8

(4), 358–370.

Sawyer, A.G., H. Noel, and C. Janiszewski (2009) The Spacing Effects of

Multiple Exposures on Memory: Implications for Advertising Scheduling.

Journal of Advertising Research, 49 (2), 193–197.

Schafer, J.L. and M.K. Olsen (1998). Multiple imputation for multivariate

missing-data problems: a data analyst’s perspective. Multivariate

Behavioral Research, 33(4), 545-571.

Schibrowsky, J.A., J.W. Peltier, and A. Nill (2007). The state of Internet

marketing research: a review of the literature and future research

direcitons. European Journal of Marketing, 41(7/8), 722-733.

Schillewaert, N. and P. Meulemeester (2005). Comparing response

distributions of offline and online data collection methods. International


Schmitt, B.H., N.T. Tavassoli and R.T. Millard (1993). Memory for print ads:

understanding relations among brand name, copy, and picture. Journal of

Consumer Psychology, 2 (11), 55-81.

Schonlau, M., A. van Soest, A. Kapteyn, and M.P. Couper (2009). Selection

Bias in Web Surveys and the Use of Propensity Scores. Sociological

Research and Methods, 37 (3), 291-318.



References

107

Shih, T.H and X. Fan (2008). Comparing Response Rates from Web and Mail

Surveys: A Meta-Analysis. Field Methods, 20(3), 249-271.

Shimp, T.A. (2008). Advertising promotion and other aspects of integrated

marketing communications (8th edition). Mason, OH: South-Western

Cengage Learning.

Simsek, Z. and J.F. Veiga (2001). A primer on internet organizational surveys.

Organizational Research Methods, 4(3), 218-235.

Singh, S.N and M.L. Rothschild (1983). Recognition as a Measure of Learning

from Television Commercials. Journal of Marketing Research, 20(3),

235–248.

Singh, S.N., S.M. Mishra, N. Bendapudi, and D. Linville (1994). Enhancing

Memory of Television Commercials through Message Spacing. Journal

of Marketing Research, 31(2), 72–80.

Singh, S.N., M.L Rothschild, and G.A. Churchill Jr. (1988). Recognition

versus Recall as Measures of Television Commercial Forgetting. Journal

of Marketing Research, 25(1), 72–80.

Smith, T.W. (2001). Are representative Internet surveys possible? In

Proceedings of Statistics Canada Symposium. Ottawa, Canada: Statistics

Canada.

Snijders, T.A.B. and R.J. Bosker (2003). Multilevel analysis: an introduction to

basic and advance multilevel modeling. London: Sage.

Song, J.H and G.M Zinkhan (2008). Determinants of Perceived Web Site

Interactivity. Journal of Marketing, 72, 99-113.

Sparrow, B., J. Liu, and D.M. Wegner (2011). Google effects on memory:

cognitive consequences of having ifnormation at our fingertips. Science,

333, 776-778.

Strabac, Z. and T. Aalberg (2011). Measuring Political Knowledge in

Telephone and Web Surveys: A Cross-National Comparison. Social

Science Computer Review, 29 (2), 175-192.

Strong, E.C. (1977). The Spacing and Timing of Advertising. Journal of

Advertising Research, 17 (6), 25-31.



References

108

Taylor, H. (2000). Does Internet research work? Comparing online survey

results with telephone survey. International Journal of Market Research,

42 (1), 51–63.

Terhanian, G. and J. Bremer (2012). A smarter way to select respondents for

surveys? International Journal of Market Research, 54 (6), 751-780.

Thios, S. J. and P.R. D’Agostino (1976), Effects of Repetition as a Function of

Study-Phase Retrieval. Journal of Verbal Learning and Verbal Behavior,

15(5), 529–36.

Tingling, P., M. Parent, and M. Wade (2003). Extending the capabilities of

Internet-based research: lessons from the field. Internet Research, 13 (3),

223-235.

Tourangeau, R. (2004). Survey research and societal change. Annual Review of

Psychology, 55, 775-801.

Tourangeau, R. and T.W. Smith (1996). Asking sensitive questions: The

impact of data collection mode, question format, and question context.

Public Opinion Quarterly, 60 (2), 275-304.

Unnava, H.R. and R.E. Burnkrant (1991). Effects of Repeating Varied Ad

Executions on Brand Name Memory. Journal of Marketing Research,

28(4), 406–416.

Vakratsas, D. and T. Ambler (1999) How advertising works: what do we really

know? Journal of Marketing, 63(1), 26-43.

Van Hattum, P. and H. Hoijtink (2008). The proof of the pudding is in the

eating Data fusion: an application in marketing. Journal of Database

Marketing and Customer Strategy Management, 15(4), 267-284.

Van Selm, M. and N.W. Jankowski (2006). Conducting Online Surveys.

Quality and Quantity, 40 (3), 435-456.

Vehovar, V., Z. Batagelj, Manfreda, and M. Zaletel (2001). Non-response in

Web Surveys. In R.M. Groves, D.A. Dillman, J.L. Eltinge, and R.J.A.

Little (Eds.), Survey Nonrespons (pp. 229-242). New York, NY: John

Wiley and Sons.



References

109

Vehovar, V., K.L. Manfreda, and Z. Batagelj (1999). Web Surveys: Can the

Weighting Solve the Problem? In Proceedings of the Section on Survey

Research Methodology (pp. 962-967). Alexandria,VA: American

Statistical Association.

Vicente, P. and E. Reis (2012). Coverage error in Internet surveys.

International Journal of Market Research, 54 (3), 323-344.

Voogt, R.J.J and W.E. Saris (2005). Mixed Mode Design: Finding the Balance

Between Non-response Bias and Mode Effects. Journal of Official

Statistics 21 (3), 367-387.

Voorveld, H.A.M., P.C. Neijens, and E.G. Smit (2009). Consumers’ responses

to brand websites: an interdisciplinary review. Internet Research, 19(5),

535-565.

Wang, K., E.T.G. Wang, and C-K. Farn (2009). Influence of Web advertising

strategies, consumer goal-directedness, and consumer involvement on

Web advertising effectiveness. International Journal of Electronic

Commerce, 13(4), 67-96.

Wang, Y., D. Liu, and Y. Wang (2003). Discovering the Capacity of Human

Memory. Brain and Mind, 4(2), 188–198.

Warlop, L., S. Ratneshwar, and S.M.J. van Osselaer (2005). Distinctive Brand

Cues and Memory for Product Consumption Experiences. International

Journal of Research in Marketing, 22(1), 27–44.

Wedel, M. and R. Pieters. (2000). Eye fixations on advertisements and memory

for brands: A model and findings. Marketing Science, 19(4), 297-312.

Weilbacher, W.M. (2001) Point of view: does advertising cause a ‘hierarchy of

effects’? Journal of Advertising Research, 41(6), 19-26.

Williams, J. (2012). Survey methods in an age of austerity. International


Wilson, A. and Laskey, N. (2003). Internet Based Marketing Research: A

Serious Alternative to Traditional Research Methods? Marketing

Intelligence and Planning, 21 (2), 79–84.



References

110

Witte, J.C., L.M. Amoroso, and P.E.N. Howard (2000). Method and

Representation in Internet-Based Survey Tools – Mobility, Community,

and Cultural Identity in Survey2000. Social Science Computer Review,

18 (2), 179-195.

Yaveroglu, I. and N. Donthu (2008). Advertising Repetition and Placement

Issues in On-Line Environments. Journal of Advertising, 37(2), 31–44.

Yoo, C.Y. (2008). Unconscious processing of web advertising: effects on

implicit memory, attitude toward the brand and consideration set. Journal

of Interactive Marketing, 22(2), 2-18.

Yoo, C.Y. and K. Kim (2005). Processing of animation in online banner

advertising: the role of cognitive and emotional responses. Journal of

Interactive Marketing, 19(4), 18-34.

Yoshimura, O. (2004). Adjusting Responses in a Non-probability Web Panel

Survey by the Propensity Score Weighting. In Proceedings of Joint

Statistical Meeting (pp. 4660-4665). Toronto, Canada: American

Statitstical Association.

Yu, C.H. (2003). Resampling methods: concepts, applications, and

justification. Practical Assessment, Research, and Evaluation, 8(19).

Retrieved July 13, 2015 from

http://PAREonline.net/getvn.asp?v=8andn=19.

Zarantonello, L. and B.H. Schmitt (2010). Using the Brand Experience Scale to

Profile Consumers and Predict Consumer Behaviour. Journal of Brand

Management, 17, 532-540.

Zenetti, G., T.H.A. Bijmolt, P.S.H. Leeflang, and D. Klapper (2014). Search

Engine Advertising Effectiveness in a Multimedia Campaign.

International Journal of Electronic Commerce, 18(3), 7–38.

Zhao, X., J.G. Lynch Jr., and Q. Chen (2010). Reconsidering Baron and

Kenny: Myths and Truths about Mediation Analysis. Journal of

Consumer Research, 37(4), 197–206.

Zielske, H.A. (1959). The Remembering and Forgetting of Advertising.

Journal of Marketing, 23 (3), 239–243.



References

111

Zielske, H. A. and W.A. Henry (1980). Remembering and Forgetting

Television Ads. Journal of Advertising Research, 20(2), 7–13.

Zikmund, W.G. and B.J. Babin (2009). Essentials Of Marketing Research (4th

ed.). Mason, Ohio: South-Western Cengage Learning.



Samenvatting (Summary in Dutch)

112




113

Samenvatting (Summary in Dutch) Inleiding

Internet is steeds belangrijker geworden voor business-to-consumer-markten

(B2C). Zowel de waarde van online transacties als het aantal mensen dat online

aankopen doet neemt snel toe, en voorspellingen duiden erop dat deze groei zal

doorzetten. Maar dankzij de technische mogelijkheden is internet meer dan

alleen een transactiekanaal. Het is ook een medium voor het overbrengen van

marketingboodschappen (marketingcommunicatie) en het doen van

marktonderzoek. Binnen de context van marketingcommunicatie heeft internet

specifieke technologische voordelen boven andere, traditionele media als het

gaat om het doorgeven van marketingboodschappen. Daarom is het niet

verrassend dat er flinke bedragen worden uitgegeven aan online reclame. Als

medium voor het doen van marktonderzoek kan internet een veelheid aan

marketing-onderzoeksmethoden faciliteren, tegen lagere kosten en met snellere

resultaten.

Al sinds de opkomst van internet wordt er onderzoek gedaan naar de invloed

van internettechnologie op consumentengedrag. Echter, zowel het

marktonderzoek als de reclame-instrumenten op internet zijn aan verandering

onderhevig, en daarmee ook de manier waarop consumenten op reclame-

instrumenten reageren. Hierdoor ontstaan er hiaten in onze kennis. Dit

proefschrift heeft tot doel enkele van deze hiaten te vullen door meer te weten

te komen over de invloed van internet op onderzoek naar online marketing en

online reclame. In Hoofdstuk 2 laten we zien hoe de kenmerken van

bannerblootstelling van invloed zijn op het consumentengeheugen. In

Hoofdstuk 3 stellen we een model voor, waarin we de advertentie- en

merkmeetwaarden voor online bannerreclame in de context van een hiërarchie

van effecten plaatsen. In Hoofdstuk 4 bespreken we het potentieel van online

polls als marktonderzoeksinstrument. Ten slotte vatten we in Hoofdstuk 5 de




114

belangrijkste conclusies van dit proefschrift samen, bespreken we de

bevindingen in het algemeen, en geven we mogelijkheden aan voor verder

onderzoek.

Hoofdstuk 2

De belangrijkste onderzoeksvraag van Hoofdstuk 2 is hoe het

consumentengeheugen wordt beïnvloed door het aantal blootstellingen, de

spreiding (spacing) en vertraging, met betrekking tot merken

bannerbekendheid. Uit onze studie blijkt dat het aantal blootstellingen in eerste

instantie een positief effect heeft op het consumentengeheugen, zowel voor de

banner zelf als het merk, maar dat dit effect minder wordt naarmate de

hoeveelheid blootstellingen toeneemt. Verder toont de studie aan dat het effect

van de spreiding tussen het laten zien van de banners (banner spacing) een

omgekeerde U-vorm volgt. Met andere woorden, het positieve effect van

blootstellingen aan banners is groter wanneer er niet te veel maar ook niet te

weinig tijd tussen de blootstellingen zit. Het verbeterde merkgeheugen ten

gevolge van het aantal blootstellingen neemt echter na enige tijd weer af.

Verder opperen mediation-specialisten dat het merkgeheugen van consumenten

alleen wordt beïnvloed door bannerblootstellingen indien de blootstellingen

ook het geheugen van de consument voor de banner zelf vergroten.

De bevindingen van deze studie komen goeddeels overeen met eerder

onderzoek naar de invloed van bannerblootstellingen op het

consumentengeheugen, zowel in de traditionele advertentiemedia (bijv.

Appleton-Knapp, Bjork en Wickens, 2005; Heflin en Haygood, 1985; Singh et

al., 1988) als bij bannerreclame (Drèze en Hussherr, 2003; Havlena en Graham,

2004). Naast deze bevindingen biedt deze studie nieuwe inzichten in de wijze

waarop merken bannerbekendheid worden beïnvloed door

bannerblootstellingen. Dit vergroot onze kennis van hoe bannerreclame werkt.

Allereerst lijkt het zo te zijn dat het gaan vervelen van de boodschap – ook

wear-out genoemd, waarbij vaker herhalen geen effect meer heeft, of zelfs een

negatief effect heeft – niet optreedt bij bannerreclame. Dit is in contrast met




115

wat er gewoonlijk wordt gevonden in praktijkstudies over het onthouden en

reproduceren van merkinformatie met betrekking tot conventionele

advertentiemethoden (bijv. Pechmann en Stewart, 1989, p. 295). Ten tweede

verloopt het gevonden spreidingseffect bij bannerreclame volgens een

studiefase-retrievalmechanisme (Appleton-Knapp et al., 2005), wat suggereert

dat een matige spreiding van bannerblootstellingen ervoor zorgt dat

consumenten zich de informatie van eerdere blootstellingen herinneren, wat er

op zijn beurt weer voor zorgt dat de banner en de informatie uit de banner beter

onthouden worden. Dit betekent dat het leerproces van consumenten met

betrekking tot het merk effectiever is wanneer er precies genoeg tijd tussen de

blootstellingen zit. Ten derde, en in tegenstelling tot eerder onderzoek naar

bannerreclame (bijv. Drèze en Hussherr, 2002; Yaveroglu en Donthu, 2008),

hebben onze bevindingen het belang van bannergeheugen aangetoond; het

merkgeheugen wordt indirect beïnvloed door bannerblootstellingen. Banners

die geen aandacht trekken of slechts in beperkte mate door de consument in

zich worden opgenomen, dragen niet bij aan verbetering en herinnering van het

geadverteerde merk.

Hoofdstuk 3

Het doel van onze studie in Hoofdstuk 3 is het bepalen van de interrelaties

onder banner-effectvariabelen, met als doel de ontwikkeling van een

allesomvattend model dat bannergeheugen en -attitude koppelt aan

merkgeheugen en -attitude, en uiteindelijk aan koopintentie. Om dit doel te

bereiken onderzoeken we tegelijkertijd de relaties tussen verschillende

effectmetingen; dit in tegenstelling tot eerder uitgevoerde studies over de

effecten van bannerreclame. We analyseren online enquêtegegevens van

respondenten die waren blootgesteld aan echte bannercampagnes (29

campagnes in totaal). Hierbij maken we gebruik van een methodologie met

meerlagige structurele vergelijkingsmodellen. De belangrijkste bevinding van

deze studie is het feit dat bannerreclame een indirect en positief effect heeft op




116

de koopintentie, door middel van bannergeheugen, bannerattitude,

merkgeheugen en merkattitude.

Maar hoewel deze effecten van bannerreclame voor het grootste deel

overeenkomen met de effecten van traditionele advertentiemethoden, dat wil

zeggen, de hiërarchie van effecten van adverteren, zijn er enkele opvallende

verschillen waargenomen. Ten eerste is gebleken dat de relatie tussen

bannergeheugen en merkgeheugen sterker is dan die tussen bannerattitude en

merkgeheugen. Deze bevinding bevestigt de bevinding uit Hoofdstuk 2, dat het

geheugen voor banners een indirecte bijdrage levert aan het geheugen voor

merken. Ten tweede is gebleken dat de relatie tussen bannerattitude en

merkattitude ongeveer even sterk is als die tussen merkgeheugen en

merkattitude. Deze twee bevindingen duiden erop dat de dual-

mediationhypothese, die goed aansluit bij adverteren in traditionele media en

die poneert dat de attitude ten opzichte van de advertenties door cognitieve

merkrespons wordt gestuurd om de merkattitude te verbeteren, mogelijk niet

passend is voor bannerreclame.

Hoofdstuk 4

Hoofdstuk 4 onderzoekt of bedrijven online enquêtes kunnen gebruiken voor

hun marktonderzoeksdoelen. Om deze onderzoeksvraag te beantwoorden,

hebben we een experiment uitgevoerd. Hiervoor hebben we een telefonische

enquête en een reeks online enquêtes op een bedrijfswebsite uitgevoerd. Onze

analyse van de data uit dit experiment heeft diverse inzichten opgeleverd. Ten

eerste is er niet alleen een verschil in respondentkenmerken tussen de twee

enquêtesoorten onderling, maar verschillen de kenmerken bovendien van de

kenmerken die zijn vastgelegd in de klantendatabase van het bedrijf. Ten

tweede is gebleken dat het verschil in de antwoorden tussen beide soorten kan

worden toegeschreven aan de steekproeftrekking, en niet aan het effect van de

enquêtesoort. Ten derde is gebleken dat een demografische weging de

responsvertekening die voortkomt uit de steekproeftrekking tussen de online

poll en de telefonische enquête kan verminderen. Dit komt niet overeen met




117

bevindingen uit eerdere studies. In vergelijking met eerder onderzoek naar

online enquêtes, wijzen de bevindingen uit onze studie erop dat het probleem

van verschillen tussen een (online) enquête op basis van zelfselectie en een

eenvoudige willekeurige steekproef kan worden verminderd indien er geen

sterk ondervertegenwoordigde demografische groepen zijn (vergeleken met de

algemene bevolking) die tot een problematische wegingsfactor leiden. Onze

bevinding dat de responsvertekening kan worden verminderd, wijst erop dat de

referentiepopulatie en de locatie van de online polls belangrijke factoren zijn

bij internet-enquêtes op basis van zelfselectie.

Gevolgen voor marketingtheorie

Sommige vormen van internetmarketing lijken op hun traditionele

tegenhangers. Zo lijken banners bijvoorbeeld op gedrukte reclame en lijken

online polls op papieren enquêtes. Desondanks blijkt uit deze studie dat de

resultaten, evenals de onderliggende processen, van deze vergelijkbare vormen

van marketingcommunicatie en marktonderzoek niet noodzakelijkerwijs

hetzelfde zijn. Veel verwachte relaties die zijn gebaseerd op theorieën uit de

traditionele marketing houden stand, maar er zijn enkele opvallende verschillen

gevonden. Zo is bijvoorbeeld de manier waarop het uiteindelijke effect van

blootstelling aan advertenties van invloed is op de koopintentie niet hetzelfde

voor bannerreclame als voor traditionele reclame-uitingen. Het verschil zit in

het pad dat moet worden afgelegd zodat het effect werkt binnen de context van

de hiërarchie van effecten. Aangezien het effect van de contextuele factoren

van informatieverwerking door consumenten (bijv. type en planning van het

blootstellingspatroon van de online reclame) en hun interacties complexer zijn

in het geval van internetmarketing, kan de uitwerking van de

internetmarketing-inspanningen niet alleen worden voorspeld aan de hand van

de marketingtheorieën die op dit moment beschikbaar zijn. Mogelijk moeten

marketingtheorieën aangepast of verfijnd worden voordat ze succesvol kunnen

worden toegepast in een internetomgeving.




118

Gevolgen voor managers

De bevindingen van ons onderzoek dat het aantal bannerblootstellingen een

afnemend positief effect heeft, en dat spreiding een omgekeerd u-vormig effect

heeft op het consumentengeheugen, wijzen erop dat een gelijkmatig of

doorlopend schema een betere strategie is voor een bannercampagne dan

massed scheduling. Voor het beste resultaat moeten consumenten periodiek en

doorlopend aan de banner worden blootgesteld tijdens de campagneperiode.

Uit ons empirische model blijkt een optimale periode van ongeveer een uur

(d.w.z., een periode die overeenkomt met de bovenkant van de omgekeerde u-

vorm van de responscurve). In de praktijk is dit type planning uitvoerbaar

dankzij cookie-technologie. Verder is de timing van de meting van belang

wanneer managers de effectiviteit van de bannercampagne evalueren,

aangezien uit ons onderzoek is gebleken dat de herinnering aan

bannerblootstelling in de loop der tijd afneemt. Wanneer de bannereffectiviteit

direct na de blootstelling wordt gemeten, worden de effecten mogelijk

overschat (bijv. Havlena en Graham, 2004). Ten slotte, aangezien het geheugen

voor banners door middel van bannerblootstelling een indirecte bijdrage levert

aan het geheugen voor merken, is het belangrijk dat de banners eenvoudig zijn

en een heldere boodschap hebben (bijv. merknamen, logo's, endorsers) die

eenvoudig verwerkt kan worden en een minimum aan cognitieve belasting

kost. Eenvoudige banners moeten echter wel in staat zijn de aandacht van

consumenten te trekken.

Ons onderzoek lijkt er ook op te wijzen dat het effect van bannerblootstelling

het bannergeheugen overtreft. Uit Studie 2 blijkt dat zowel bannergeheugen als

-effect weliswaar een positieve relatie hebben met merkgeheugen, maar dat de

relatie sterker is voor bannergeheugen. Dit duidt erop dat wanneer banners

onderdeel zijn van een reclamecampagne, die banners moeten zorgen voor een

sterkere koppeling naar het merk dan naar de banner. Banners moeten derhalve

niet afleiden van hun belangrijkste doel: het vergroten van de merkbekendheid.

Mooie figuurtjes, schattige beestjes of fraaie natuurplaatjes in een

bannerontwerp kunnen weliswaar de aandacht trekken, maar mogen de toch al




119

beperkte aandacht van de consument niet afleiden van de merknaam, het logo

of de argumenten. Aangezien bannerblootstelling de merkbekendheid bij de

consument vergroot, is een belangrijke gevolgtrekking van onze studie dat het

gerechtvaardigd kan zijn om geld uit te geven aan bannerreclame, hoewel

recente gegevens hebben aangetoond dat het doorklikpercentage zeer laag is.

Bannerblootstelling is van invloed op bannergeheugen; bovendien zorgt

bannerblootstelling er via een hiërarchie van effecten voor dat zowel de

merkattitude als de koopintentie voor het merk wordt verbeterd.

De resultaten van ons onderzoek wijzen op een positief vooruitzicht voor

online enquêtes op basis van zelfselectie die worden gepubliceerd op websites

van bedrijven, zoals online polls. Naarmate er meer mensen gebruik maken van

draadloze communicatie-apparaten, zijn gebruikelijke enquêtevormen zoals

Random Digit Dialing niet meer representatiever dan online enquêtes, omdat

bepaalde bevolkingsgroepen niet meer via traditionele media bereikt kunnen

worden. Uit Studie 3 is gebleken dat de consumentenkenmerken binnen

enquêtes sterk worden beïnvloed door de kenmerken van de gebruikers van het

desbetreffende medium, wat erop wijst dat managers gemengde enquêtevormen

moeten inzetten wanneer ze een representatievere respons nodig hebben.

Echter, voor een eenvoudige enquête met een specifieke doelgroep — een

veelvoorkomend scenario bij marktonderzoek — kan een bedrijf online polls

inzetten als enquête-instrument om gegevens te verzamelen die verder verwerkt

zullen worden. Hierbij kunnen statistische methodes zoals weging worden

ingezet om mogelijke vertekening bij te stellen. Derhalve moeten zulke polls

niet alleen onderzoeksvragen bevatten, maar ook vragen naar

achtergrondinformatie. Die informatie is nodig om vertekening te corrigeren

met behulp van weging op basis van steekproeftrekking. Ondanks de potentie

van online polls zijn er nog wel enkele waarschuwingen van toepassing.

Managers moeten juiste internettechnologie toepassen om 'robots' (automatisch

invullen van de poll door software), meervoudige deelname en andere

pogingen om de poll te beïnvloeden te ontwijken.

University of Groningen Online marketing and advertising ... · Processed on: 23-9-2016...

Documents

Transcript of University of Groningen Online marketing and advertising ... · Processed on: 23-9-2016...