essays on the economics of media platforms

ESSAYS ON THE ECONOMICSOF MEDIA PLATFORMS

KATHOLIEKEUNIVERSITEIT

LEUVEN

FACULTEIT ECONOMIEEN BEDRIJFSWETENSCHAPPEN

Proefschrift voorgedragentot het behalen van de graadvan Doctor in de Economische Wetenschappen

door

Dries De Smet

Nummer 367 2011

ii

Daar de proefschriften in de reeks van de Faculteit economie

en bedrijfswetenschappen het persoonlijk werk zijn van hun

auteurs, zijn alleen deze laatsten daarvoor verantwoordelijk.

Since the dissertations in the series published by the Faculty of

Business and Economics are the personal work of their authors,

only the latter bear full responsibility for them.

Doctoral Commission

Chairman: Wim Moesen (KU Leuven)Advisor: Patrick Van Cayseele (KU Leuven)Co-Advisor: Jo Swinnen (KU Leuven)Other Members: Simon Anderson (University of Virginia)

Fabrizio Germano (Universitat Pompeu Fabra)Frank Verboven (KU Leuven)

Preface

Before starting this PhD, I attended an introductory evening on doing aPhD in economics. The big difference between an undergraduate thesisand a PhD, one of the speakers explained, is that an undergraduate thesishas a clear target. You get a bow and arrow and you know what you haveto do to hit the bull’s eye. The question is clear, and so is the answer, oryou know at least in which direction or within which boundaries you haveto search.

For a PhD you’re much better equipped, with graduate courses andthe help of senior academics, but the questions and answers are far lessobvious. A PhD, and by extension any academic research, should be stateof the art, trying to answer new and detailed questions and use newlydiscovered and advanced methodologies. In sum: it is repositioning thefrontiers of knowledge.

If writing a thesis was a journey, an undergraduate thesis would prob-ably resemble a city trip: you know where the highlights are, and the onlything you have to do is reach your destinations and hit the camera button.Writing a PhD is much more like starting an explorative expedition. Itis hard to know what to expect and to predict which road will lead tosuccess.

After the first year in Leuven, PhD students are required to presenta draft proposal of their PhD. Comparing the initial road map with theultimate results would prove that such an initial road map is not veryuseful, at least to find your way in the PhD.

None of the questions answered in this PhD emerged as one of theoriginal questions at the start of my PhD. There are many questions whichlook interesting at first sight, but turn out to be trivial, answered bysomeone else, not relevant anymore or way too difficult to solve adequately.New questions emerge during research and new courses are taken. What

vi Preface

ended up in this PhD is only a fraction of the work done; just as importantare the time and effort put in ideas and drafts that never made it to thefinal version.

The absence of clear questions and answers makes academic researchoften a frustrating task, even for experienced researchers. As a result,more than 50 % of all starters at UGent and KU Leuven quit early andnever finish their PhD (Verlinden et al. 2005).

However, finding answers on well-defined questions, allowed me topresent my work on workshops and conferences, which are inspiring en-vironments to exchange ideas and to catch up new ways to define andanswer interesting questions. With the deadline in sight, I also becamemore focused and more productive and I’m glad to have brought this PhDto a good conclusion.

In this dissertation, I was able to combine two of my biggest passions:media and economics. Since my childhood, I am passionate about media,news and writing. The constant changes in the media sector, due tothe rise of the internet and the loss of advertising revenues following thefinancial crisis, made it extremely interesting to explore and explain thissector from an economic viewpoint.

Whether I will stay in academia and/or in economics remains to beseen. What’s for sure, is that I will remain an economist with an academicbackground. Whatever the problem, I catch myself finding out how agentsinteract, what the incentives are and which trade-offs emerge. Economicsis, contrary to what many people believe, not the science of money buta powerful tool to explain reality. I hope to have explained part of thiscomplex reality in this PhD.

Dries De SmetMannheim, July 2011

Contents

Preface v

Contents vii

1 General Introduction 1

2 Newspapers 72.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.2 Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.1 Monopoly model . . . . . . . . . . . . . . . . . . . . 122.2.1.1 Set-up . . . . . . . . . . . . . . . . . . . . . 122.2.1.2 Results . . . . . . . . . . . . . . . . . . . . 15

2.2.2 Competition Model . . . . . . . . . . . . . . . . . . . 162.2.2.1 Set-Up . . . . . . . . . . . . . . . . . . . . 162.2.2.2 Results . . . . . . . . . . . . . . . . . . . . 18

2.2.3 Welfare Analysis . . . . . . . . . . . . . . . . . . . . 212.2.4 Conclusion of the Model . . . . . . . . . . . . . . . . 23

2.3 Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . 232.3.1 Data Description . . . . . . . . . . . . . . . . . . . . 232.3.2 Empirical Strategy . . . . . . . . . . . . . . . . . . . 262.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.4 Concerns . . . . . . . . . . . . . . . . . . . . . . . . 31

2.4 Survey of Belgian Journalists . . . . . . . . . . . . . . . . . 352.4.1 Related Literature . . . . . . . . . . . . . . . . . . . 352.4.2 Methodology . . . . . . . . . . . . . . . . . . . . . . 372.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.5 Advertising Bias Channels . . . . . . . . . . . . . . . . . . . 452.5.1 Supply and demand channels . . . . . . . . . . . . . 45

viii CONTENTS

2.5.2 Inside channels . . . . . . . . . . . . . . . . . . . . . 462.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 46A Monopoly profits . . . . . . . . . . . . . . . . . . . . . . . . 49B Figures & Tables . . . . . . . . . . . . . . . . . . . . . . . . 50C Overview Surveys on Influence Advertisers . . . . . . . . . . 53D Advertising Bias Channels . . . . . . . . . . . . . . . . . . . 58

D1 Supply and demand channels . . . . . . . . . . . . . 58D2 Inside channels . . . . . . . . . . . . . . . . . . . . . 62

3 Directories 673.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.2 Set-Up of the Theoretical Model . . . . . . . . . . . . . . . 723.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.3.1 The Availability Restriction . . . . . . . . . . . . . . 763.3.2 The Comprehensive Directory Restriction . . . . . . 80

3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 86A Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89B Empirical Investigation of the Yellow Pages Industry . . . . 94

B1 Industry Characteristics . . . . . . . . . . . . . . . . 94B2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 95B3 Identification . . . . . . . . . . . . . . . . . . . . . . 96B4 Results . . . . . . . . . . . . . . . . . . . . . . . . . 99B5 Robustness Check: SURE . . . . . . . . . . . . . . . 101B6 Reconciling Theory with Empirical Findings . . . . . 101

C Extensions to the Model . . . . . . . . . . . . . . . . . . . . 104C1 Negative Utility from Advertisements . . . . . . . . 104C2 Endogenizing r (as a function of s1 and s2) . . . . . 105C3 Price discrimination . . . . . . . . . . . . . . . . . . 107C4 Results for a Ramsey Planner . . . . . . . . . . . . . 108C5 Competition . . . . . . . . . . . . . . . . . . . . . . 110

D Proofs of Extensions . . . . . . . . . . . . . . . . . . . . . . 112

4 Aggregators 1174.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 1184.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . 1204.3 Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . 124

4.3.1 Set-Up of the Model . . . . . . . . . . . . . . . . . . 1244.3.1.1 Framework without metaplatform . . . . . 1254.3.1.2 Framework with metaplatform . . . . . . . 128

CONTENTS ix

4.3.1.3 Timing of the Game . . . . . . . . . . . . . 1304.3.2 Basic Results . . . . . . . . . . . . . . . . . . . . . . 130

4.3.2.1 Stage 2: Seller Quantities . . . . . . . . . . 1304.3.2.2 Stage 1: Allowance Decision . . . . . . . . 132

4.3.3 Extensions . . . . . . . . . . . . . . . . . . . . . . . 1354.3.4 Future research . . . . . . . . . . . . . . . . . . . . . 139

4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 144A Equilibrium quantities . . . . . . . . . . . . . . . . . . . . . 146B A Legal Discussion on Metaplatforms . . . . . . . . . . . . 146

B1 Arguments in favor of the Platform . . . . . . . . . . 146B2 Arguments in favor of the Metaplatform . . . . . . . 148

5 General Conclusion 151

Acknowledgements 157

About the Author 159

Bibliography 160

Dissertations 171

x CONTENTS

List of Figures

2.1 Questions on Relationship Advertisers Newspapers . . . . . 392.2 Suggestion to Write about Advertiser . . . . . . . . . . . . . 412.3 Suggestion to Adjust Article about Advertiser . . . . . . . . 42B1 Total Advertising Expenditure over Time . . . . . . . . . . 50B2 Total Coverage over Time . . . . . . . . . . . . . . . . . . . 50

3.1 Different regimes given availability . . . . . . . . . . . . . . 773.2 Different regimes given comprehensiveness . . . . . . . . . . 82A1 Platform Profit . . . . . . . . . . . . . . . . . . . . . . . . . 90C1 Different Solutions in the r-s-space . . . . . . . . . . . . . . 105C2 Different solutions in the v-s-space . . . . . . . . . . . . . . 106C3 Relative prices (p = p1

p2) . . . . . . . . . . . . . . . . . . . . 107

C4 Optimal welfare solutions in the r-s-space. . . . . . . . . . . 109D1 Different Solutions in the r-s-space with r < 0 . . . . . . . . 113D2 Different solutions in the r-s-space. . . . . . . . . . . . . . . 115D3 Price Ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.1 Set-Up of the Game . . . . . . . . . . . . . . . . . . . . . . 1244.2 Equilibrium Seller Quantities for All Cases . . . . . . . . . 1324.3 Equilibrium Zones for Allowance Decisions . . . . . . . . . . 134

xii LIST OF FIGURES

List of Tables

2.1 Summary Statistics of Advertising and Coverage . . . . . . 252.2 Coverage explained by advertising and dummies . . . . . . 282.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . 312.4 Granger Causality Test . . . . . . . . . . . . . . . . . . . . 332.5 Coverage explained by adv. of last 12 months . . . . . . . . 342.6 Summaries Survey . . . . . . . . . . . . . . . . . . . . . . . 372.7 Survey Results per Group . . . . . . . . . . . . . . . . . . . 44B1 Coverage explained by logged adv. and dummies . . . . . . 51B2 Coverage explained by lagged advertising . . . . . . . . . . 52B3 Advertising explained by lagged coverage . . . . . . . . . . 52C1 Overview Surveys on Influence Advertisers . . . . . . . . . . 53

B1 Variables: summary statistics . . . . . . . . . . . . . . . . . 95B2 Results of the OLS regression . . . . . . . . . . . . . . . . . 100B3 Results of OLS and SURE . . . . . . . . . . . . . . . . . . . 103

xiv LIST OF TABLES

Chapter 1

General Introduction

2 General Introduction

When the global payments technology company Visa came under at-tack from the antitrust authorities in the 1990s, the company hired theeconomists Richard Schmalensee and David Evans (Warsh 2007). In 1999,they wrote Paying with Plastic in which they defended the debit and creditcards industry. As a side product, they also analyzed how the market forplastic cards was functioning. They stated that standard economic modelsfailed to observe the chicken-egg problem that confronted card issuers: ifconsumers don’t possess a certain card, then merchants won’t accept it,and vice versa, if merchants don’t accept it, then consumers don’t wantthis card.

When Schmalensee asked Jean Tirole and Jean-Charles Rochet to havea closer look at the problem, these French economists figured out thatthese chicken-egg problems are not limited to credit cards, but are presentin many markets. Every market where an intermediary is active thatconnects two sides of the market, serves as a candidate to be examinedas a two-sided market. The list of applications seems endless: real estatebrokers, auction houses, internet platforms, dating services, employmentagencies, magazines, radio, television, payment systems such as debit andcredit cards, operating systems for personal computers, video games, mp3players... (Evans & Schmalensee 2008)

The three topics discussed in this PhD thesis, newspapers, telephonedirectories and buyer/seller internet platforms, might seem at first sightunrelated, but a closer look learns that they are all two-sided markets,essentially bringing together readers and advertisers. That immediatelyexplains why this PhD is titled ‘Essays on the Economics of Media Plat-forms’.

To model these markets, I rely on the fast evolving literature on two-sided markets, started by the seminal papers of Rochet & Tirole (2003)and Armstrong (2006)1. Two-sided markets can be loosely defined as mar-kets where one or more platforms serve two distinctly different types ofagents. There is a certain kind of interdependence between both groupsof agents, i.e. they generate an externality on each other. A key featureof two-sided markets is that these externalities (or also: indirect networkeffects) are not internalized by the different groups. As a result, an in-termediary is needed to internalize these externalities. As Rysman (2009)argues, this definition might be overly inclusive. While many markets can

1Other seminal papers are Caillaud & Jullien (2001), Caillaud & Jullien (2003),

Evans (2004), Parker & Van Alstyne (2005) and Rochet & Tirole (2006).

3

be considered as two- or multi-sided markets, these externalities are notalways important for the analysis. Rysman (2009) gives the example ofcars. While the existence of car dealers and gasoline stations was veryimportant for car drivers by the introduction of cars, today it is much lessimportant given the abundance of dealers and gasoline stations. But thesenetworks might become important again when new car brands or new fuelsources are introduced.

Nearly all papers in the two-sided market literature place the platformor intermediary at the foreground. It are in particular the key decisions ofplatforms that are investigated, given the decisions of both agent groupsto join or use the platform. Most of these articles look at price settingplatforms, though some consider quantity setting platforms. A big dividein the literature has been the question whether agents pay a membershipfee (e.g. Armstrong (2006)) or a transaction fee (e.g. Rochet & Tirole(2003)). Weyl (2010) tried to unite both models by introducing two-parttariffs which are combinations of both. He stresses more the heterogeneityof agents: are they heterogeneous in their valuation for interaction (as inRochet & Tirole (2003)), for membership (as in Armstrong (2006)) or forboth?

In the models I have build in the next chapters, I have not chosen inan outspoken way for one type of payment (e.g. membership or transac-tion fee), or for one type of user heterogeneity. I believe that a consistentapproach is important to derive general results, as has been done by Arm-strong (2006), Rochet & Tirole (2003) and Weyl (2010). Though, for thequestions I try to answer in the next chapters, it is more important tomodel the markets under investigation in a logical, sensible way. In sum,I combine price and quantity setting platforms, between membership andtransaction price and between membership and transaction level hetero-geneity.

Structure The main text of this dissertation consists of three chapters.In chapter 2, I investigate the potential favorable treatment of advertis-ers by newspapers. Chapter 3 questions whether the Universal ServiceObligation for telephone directories is efficient. In chapter 4, I discuss theeffect of aggregators on traditional platforms such as real estate websites.

In chapter 2, the main question is: do newspapers write more abouttheir advertisers? I try to answer this question on three levels. In mytheoretical model, I show that the answer is yes, if readers do not mind


bias too much and if bias is relatively effective for advertisers. In a surveyof Belgian journalists, 21 percent answers this question positively, butdirect pressure is almost non-existent. In the empirical part, I find thatadvertisers in Belgian Dutch-language newspapers receive a significantlyhigher coverage.

This chapter is part of a larger and quickly emerging field of the eco-nomic literature, known as forensic economics (Zitzewitz (2009) and Fis-man (2010)). This field is aimed at uncovering the hidden behavior ofdifferent types of agents. These empirical studies use measurement in-consistencies, violations of the efficient market hypothesis, discontinuities,variation in incentives and field experiments to discover all types of corrup-tion. One of the first forensic economists avant la lettre was the Belgianstatistician Adolphe Quetelet. Quetelet (1846) compared the chest size ofFrench conscripts with the expected values, assuming that chest size wasnormally distributed over the population. He observed a puzzling shortageof men with a chest-size slightly above 1.57 meter and an excess numberof men below 1.57 meter. It might not surprise the reader that men below1.57 meter were excused for the Imperial army. Zitzewitz (2009) providesnumerous examples of cheating and corruption on taxes, education, gov-ernment procurement, sports, weather forecasts and yes, also academicperformances. Advertising bias is also a hidden action because newspa-pers will never admit that they write more about advertisers because thereputation loss might be devastating.

In chapter 3, I examine the universal service obligation for telephonedirectories. In most countries, this obligation consists of two parts. First,publishers must offer a comprehensive directory, including all telephonesubscribers. Second, this directory should be available to all users. Build-ing a simple theoretical model, inspired by the literature on two-sidedmarkets, I find that one of these two obligations is redundant. Imposingavailability to all users is sufficient to induce publishers to offer a compre-hensive directory. The reverse however does not hold.

In chapter 4, the effect of metaplatforms on traditional platforms isinvestigated. Aggregators, or metaplatforms as I call them, are both anopportunity and a threat for traditional platforms. The metaplatformredirects customers that would otherwise not have visited the traditionalplatform. But the metaplatform scores off some of the advertiser prof-its generated through customers. In a theoretical model, I show thattraditional platforms might allow the metaplatform, even when it is not

5

profitable, due to a coordination failure.In this chapter, I shed my light on the so-called New Economy that

was created by the rise of the Internet. In chapters 2 and 3, I remainedsilent on the Internet though it goes without saying that it had and has abig impact on newspapers and telephone directories. Especially during thefinancial crisis, when the fast declining advertisement revenues acceleratedthe damage done by the Internet, many newspapers kicked the bucket.Printed telephone directories survived, but is is unlikely that they willsurvive once everyone is connected to the Internet. Note that the focusof chapters 2 and 3 on the traditional media does not degrade my results,since I’m confident that the basic mechanisms in print also hold for onlinenewspapers and telephone directories.

Chapter 2

The Advertiser is

Mentioned Twice.

Media Bias in Belgian

Newspapers

This chapter is based on joint work with Stijn Vanormelingen.

8 Newspapers

2.1 Introduction

Companies are indispensable for newspapers. They are not only an impor-tant source of news, e.g. by the products they release or by the fluctuationsin their stock prices, but are also an important source of revenues throughadvertisements. While this is not unusual in itself, it might become wor-risome when a newspaper mixes up this double role of companies. Whenan advertiser has a higher probability of being covered in the newspa-per, because the newspaper wants to please the advertiser, then we labelthis as advertising bias. The advertiser is mentioned twice: once in theadvertisement and once in advertising bias, which is wrapped as news.

Allegations that editors and journalists are influenced by advertisersare probably as old as advertisements in newspapers1. One of the oldestdocumented cases of advertising bias is found in an editorial in the NewYork Herald. When other pill makers complained about the favorabletreatment of dr. Brandreth cure-all pills, the editor of the New York Her-ald wrote in 1835 in his newspaper: “Send us more advertisement than dr.Brandreth does and we’ll cut dr. Brandreth to dead” (Bagdikian 2004).A more recent example is the newspaper Chicago Sun-Times, who men-tioned in an article on the sales after the death of Versace only depart-ment stores that also advertised in the newspaper. The editor defendedthe approach with “we have to take care of our customers”. The samenewspaper later declared that theaters advertising get preferential edito-rial treatment. Similarly, according to a leading journalist, it was quitenormal in the 1950s that articles on women’s and fashion pages “reflectedthe relative advertising strength”.

The empirical research on advertising bias is relatively new. Reuter& Zitzewitz (2006) find that mutual funds that advertise in the US fi-nancial press receive more as well as more positive coverage. Rinallo &Basuroy (2009) document a strong influence of advertisements by Italianfashion companies on the editorial content of newspapers and magazines.A one percentage increase in the number of pages advertised generates0.4 percent additional coverage (also measured in pages). Similar resultsare found by Bignon & Miscio (2009) for the French financial press re-porting on financial assets in 1907, Gambaro & Puglisi (2009) in theirstudy about Italian newspapers writing about major listed companies and

1For a historical view on advertisements in newspapers, see Baldasty (1992) and

DeLorme & Fedler (2005).

2.1 Introduction 9

Gurun & Butler (2010) investigating coverage of local advertisers by USlocal media. Di Tella & Franceschelli (2009) find that advertisements bythe incumbent government (which can be perceived as subsidies) have astrong negative impact on the coverage of scandals in Argentinia.2

These studies confirm surveys on adverting bias. Soley & Craig (1992)report that 90% of the US editors in their sample have been pressuredby advertisers. More than one third of the respondents answered thatadvertisers succeeded in influencing the news. In a survey of Hays &Reisner (1990), 37% of farm magazine writers report that advertisers’attempts to influence what stories appear harmed the profession. Anequal amount criticized pressure from their editor or publisher to pleaseadvertisers.3 In a study of the American Society of Newspaper Editors,50% of the readers believed that newspapers allow advertisers’ intereststo influence news decisions (ASNE 1999).

In the literature, many different definitions have been used to describemedia bias. There seems to be no accepted definition of bias, but mostauthors follow the slanting definition of Hayakawa (1940): “Slanting is theprocess of selecting details that are favorable or unfavorable to the subjectbeing described”.4

We define advertising bias as the existence of an implicit or explicitbundle of advertisement and favorable coverage. Suppose news storiescan be ranked on newsworthiness. Since a newspaper has only limitedspace, there exists a newsworthiness threshold. An unbiased newspaperwill publish all news above the threshold and ignore all news below thethreshold. Positive bias will occur when a certain news story appears in thenewspaper, despite being under the newsworthiness threshold. Negativebias occurs when a news story is not in the newspaper, despite being abovethe newsworthiness threshold.5

2Evidence of non-existence of advertising bias is scarce, perhaps because of a publi-

cation bias towards statistically significant results. Reuter (2009) does not find any

statistal difference in the ratings of a wine magazine that accepts advertisements ver-

sus a magazine that doesn’t.3For network television correspondents, this figure appears to be much lower. Price

(2003) finds that only 7% reported some advertiser pressure.4For an overview of bias definitions, see McCluskey & Swinnen (2010).5Note that for political bias, the construct of newsworthiness is seldom used, be-

cause it is difficult to operationalise. In the context of political media bias, bias is often

expressed as the relative position towards two extreme viewpoints. A medium that

balances the two sides is seen as unbiased. In empirical studies, these two sides are left

vs. right (Groseclose & Milyo (2005) and Gentzkow & Shapiro (2010)) and government

vs. opposition (Di Tella & Franceschelli (2009) and Durante & Knight (2009)). Theo-

10 Newspapers

In their theoretical models, Ellman & Germano (2009) and Germano& Meier (2010) consider only negative bias, i.e. the suppression of nega-tive news. Bias is considered as something negative for news consumersand something positive for advertisers, creating a clear trade-off. In ourtheoretical model, bias is presented as the bundle of an advertisement withfavorable coverage. This coverage can be favorable both at a general level,i.e. by promoting business or consumption, and at the company level. Atthe general level, the media outlet is inclined to create a positive climatefor all businesses, or to choose content that stimulates consumption. Atthe company level, an outlet pleases advertisers by inserting more posi-tive news and downplaying negative news. This concept is very similarto stealth marketing or covert marketing, where advertisers try to invadeeditorial content with promotional messages. A difference is that this in-vasion mostly happens through a sponsorship, and not as a complementto advertising6.

The empirical exercise exploits the fact that advertisement levels shouldhave no influence on the coverage of companies. We consider a newspaperas unbiased if advertisement expenditures have no influence on the news-paper’s reporting of companies. A statistically significant effect of adver-tisement on coverage, is an indication of advertising bias. This is similarto other empirical studies on advertising bias, such as Reuter & Zitzewitz(2006), Rinallo & Basuroy (2009) and Gambaro & Puglisi (2009).

In this chapter, we attempt to answer the question whether adver-tising bias is present by applying several approaches. First, we build atheoretical model. Second we combine two data sets holding informationabout advertising and coverage in different Belgian newspapers and fi-nally, we conduct a survey among Belgian journalists. The remainder ofthis chapter is structured as follows.

In section 4.3, we build up a theoretical model to explain advertisingbias. In this simple theoretical model, we investigate the incentives of amonopolist newspaper to insert positive advertising bias. We argue thatthe amount of bias – if any – depends on the relative effectiveness ofbias and advertisement and the relative aversion of readers towards biasand advertisement. Whether bias will prevail is ultimately an empiricalquestion.

retical studies often consider a binary state of the world (Baron (2006), Gentzkow &

Shapiro (2006), Sobbrio (2009) and Suen (2004)).6Roy & Chattopadhyay (2010) give an overview of the different definitions of stealth

marketing; Goodman (2006) provides a legal discussion.

2.2 Theoretical Model 11

In section 2.3, we measure advertising bias in the Dutch-language Bel-gian newspapers. For that purpose, we matched a dataset of advertise-ment expenditures in newspapers with the number of articles mentioningthese advertisers. We find that advertisers are mentioned twice as often asnon-advertisers. Controlling for the fact that some newspapers are morebusiness oriented, we document that an expenditure of 100.000 euro gen-erates – depending on the specification – between 0.47 and 1.86 additionalarticles. Alternatively, every four to fourteen full page black and whiteadvertisements generate one additional article in the same month.

The question remains how this bias is implemented. To answer thisquestion we conducted a survey of Belgian Dutch-speaking newspaperjournalists in section 2.4. Based on the qualitative answers of this survey,and the many examples in the literature, we define a number of explicitand implicit channels that might facilitate advertising bias. The channelsare described in section 2.5.

Section 4.4 concludes and discusses policy proposals to reduce adver-tising bias in the future.

2.2 The Newspaper’s Incentives for Adver-

tising Bias

In this section, we sketch a simple theoretical model, in order to under-stand better the incentives and determinants of advertising bias. Thismodel will help us to better understand under which conditions bias islikely, i.e. when we can expect advertiser bias is likely. We model anewspaper as a platform that serves both readers and advertisers7. Thenewspaper has the option to introduce advertising bias. By doing so, thepublisher takes into account two effects. On the one hand, advertisingbias increases the willingness to pay by advertisers. On the other hand,advertising bias scares off readers. The trade-off between both effects de-termines the amount of advertising bias, if any. This mechanism is similarto the one in Ellman & Germano (2009) and Germano & Meier (2010).

To keep the model tractable, we make several non-trivial assumptions.

7For an overview of platforms and the two-sided markets literature, see Rochet &

Tirole (2006) and Rysman (2009). For theoretical applications of two-sided markets

to media markets, see Anderson & Coate (2005), Anderson & Gabszewicz (2006) and

Armstrong (2006). Rysman (2004), Kaiser & Wright (2006) and Argentesi & Filistruc-

chi (2007) provide empirical investigations of media as a two-sided market.

12 Newspapers

First, we only look at one specific type of bias, i.e. positive news – whichis not relevant for readers, in return for advertisement. Other types ofbias (e.g. political bias) are likely to produce other results. Second, weassume that readers don’t pay. Third, we eliminate all marginal and fixedcosts, i.e. producing news (and advertisement) is free. Introducing a costto produce news might, together with the introduction of a reader price,reduce the averseness of readers towards bias and ads, because they mighthelp finance news. Finally, we consider at most 2 media outlets. Addingmore media outlets might provoke different dynamics not captured by thismodel. Most of these assumptions can be relaxed, but this comes at thecost of losing tractability and/or obfuscating the focus of this chapter, i.e.the likeliness of advertising bias in equilibrium.

2.2.1 Monopoly model

In the first subsection, we introduce the model with a monopoly newspaperpublisher to make clear the underlying mechanism behind advertising bias.In section 2.2.2 we introduce competition in the model.

2.2.1.1 Set-up

The monopolist newspaper acts as a platform facing a double demandof readers and advertisers. The newspaper offers them a product thatconsists of three parts: news (n), advertisements (qA) and bias (b). Biasis the mentioning of an advertiser in a news story, despite the fact thatthis mentioning is not newsworthy.8 The newspaper is a quantity setterand chooses the amount of news, bias and ads. The total available spacefor content in the newspaper is fixed and normalized to 1. Therefore, thevariables n, qA and b can be interpreted as percentages and add up to one.Choosing two of these variables automatically determines the third one,therefore we concentrate on b and qA.

Readers Readers make a decision whether to read the newspaper ornot. Readers like news but attribute a disutility to advertisements andbias. To simplify the model, we make abstraction of the reader price.The

8If the advertiser is mentioned in a news story and this mentioning is newsworthy,

then it is counted under news (n). Moreover, we only have two levels of newsworthiness:

either it is newsworthy (news) or it is not (bias). We don’t allow for more detailed levels.


utility of reader k reads

uRk = γ0n− γ1b− γ2qA − wk (2.1)

where γi are the (dis)utilities gained from each type of content. We assumethat there is a unit mass of readers who have an opportunity cost w toread or pick up a copy of the newspaper, uniformly distributed over 0 and1. Note that we (implicitly) assume here that readers know the amountof bias in the newspaper, but cannot distinguish news from bias9. Tosimplify the notation, we use n = 1 − b − qA, and introduce β0 ≡ γ0,β1 ≡ γ0 +γ1 and β2 ≡ γ0 +γ2. All potential readers whose utility is largerthan zero (uRk ≥ 0), read the newspaper, and therefore the number ofreaders is equal to

qR = β0 − β1b− β2qA (2.2)

Advertisers As readers of newspapers act as consumers in the mar-ket, companies want to persuade them to buy their products. There aretwo channels through which a company can reach consumers10. First,by advertisements in the newspaper and second, by bias published bythe newspaper. Companies want to advertise in the newspaper becauseit enables them to persuade consumers to buy their products. Bias andadvertisements differ in their propensity to persuade consumers, labeledα1 respectively α2. We further assume that newspapers spread the totalamount of bias equally over all advertisers, i.e. the amount of bias per

9It might be possible to introduce a more complex model where news and bias can

be verified ex-post, in the spirit of Milgrom & Roberts (1982). Gentzkow & Shapiro

(2010) model ex post verification in the context of political bias; Blasco, Pin & Sobbrio

(2010) do so for advertising bias. We don’t introduce a micro model on verification, but

our modeling might be seen as a long run effect of bias. Suppose that readers detect

bias ex post; and that the choice of bias does not change over time, then readers know

perfectly the amount of bias, but they don’t know which articles are biased. Moreover,

if ex post revelation is not perfect, then our parameter γ1 will capture both the aversion

against bias and the chance that bias is detected.10In the same spirit, one can also replace consumers by voters, donors or investors.

Interestingly, DellaVigna & Gentzkow (2009) show in an overview of empirical literature

on persuasion that these latter categories might be even more prone to persuasive

communication than consumers.

14 Newspapers

advertiser is equal to bqA

11. Therefore, the persuasion per reader is

σ = α1b

qA+ α2 (2.3)

This persuasion is equal to the amount of products or services the averagereader buys from the advertiser. If σ < 1, this can be interpreted as thechance that the average reader will buy the product or service. There is amass 1 of advertisers which obtain a profit of πAk per product or servicesold. The utility function of advertiser k reads

uAk = πAkσqR − pA (2.4)

This parameter πAk is uniformly distributed over 0 and 1. We furtherassume that each company can place only one ad. All potential advertiserswhose utility is larger than zero (uAk ≥ 0) will advertise in the newspaper.Therefore, the implicit price of the quantity setting monopolist is

pA = (1− qA)σqR (2.5)

Newspaper The newspaper is a profit maximizer. It chooses the opti-mal amount of bias b and advertisement qA. The remaining part of thenewspaper consists of news. With a cost of c per copy, the profit functionreads

Π = (pA − c)qA (2.6)

In setting its optimal advertisement quantity, the newspaper takes intoaccount the direct effect of this quantity on profit, but also its indirecteffects through a lower equilibrium price and lower number of readers.Similar for bias, the newspaper will weigh the positive effect of increasingthe willingness to pay of advertisers with the negative effect of chasingaway some readers. For the sake of simplicity, we will assume that c = 0for the remainder of the chapter.

Further assumptions To obtain the results in the next section, news-paper profit is maximized with Kuhn-Tucker constraints. To simplify the

11Note that we implicitly assume here that the newspaper knows the distribution of

πAk, but that the πAk of each company is private information. If this information was

public, then the marginal firms would receive more bias, since they drive the advertiser

demand price (in the absence of price discrimination). In that case, bias would not be

given to infra-marginal advertisers, since they are going to advertise anyway.


analysis, we impose that the parameters β1 and β2 are larger than 1 andβ0, respectively. These assumptions exclude two sets of corner solutions:first, newspapers with only ads and bias, and second full coverage onthe reader side no matter what’s in the newspaper. We maintain theseassumptions throughout the chapter.

2.2.1.2 Results

The monopolist newspaper maximizes his profit Π = pAqA subject to theconstraints n, b, qA, qR ≥ 0. We solve this problem with Kuhn-Tuckerconditions.

Lemma 2.2.1 The optimal quantities are

b =

0 if α1α2

<β1(√

1−β2+β22)

(β2−1)β2

2α1α2β1−α22β2

1+α21(β2−1)β2

3α1β1(α2β1−α1β2)if

β1(√

1−β2+β22)

(β2−1)β2≤ α1

α2≤ 2β1

1+2β21

2β1if α1

α2> 2β1

1+2β2

qA =

1+β2−√

1−β2+β22

3β2if α1

α2<

β1(√

1−β2+β22)

(β2−1)β2

α1−2α2β1+2α1β23(α1β2−α2β1)

ifβ1(√

1−β2+β22)

(β2−1)β2≤ α1

α2≤ 2β1

1+2β2

0 if α1α2

> 2β11+2β2

The second solution is the unconstrained maximum. The first andthird solution are corner solutions where respectively b and qA are zero.Given the assumptions on parameters β0, β1 and β2, the constraintsn, qR ≥ 0 are never binding in this problem.

In the first solution, the newspaper decides not to insert bias, eitherbecause it is not effective enough (α1 too low or α2 too high) or becauseit is too annoying relative to advertising (β1 too high or β2 too low). Forthe newspaper, it is more profitable to earn its revenue directly throughadvertising than offering the bundle of advertisement and bias.

If advertising is relatively uneffective or disliked by consumers, thenewspaper will have only one advertiser (qA → 0). All the biased articleswill be written about the only advertiser. The newspaper can then beseen as a company newspaper. Depending on the annoyance of bias (β1),this amount of bias can increase up to one half of the newspaper content.

In the region where β1(√

1−β2+β22)

(β2−1)β2≤ α1

α2≤ 2β1

1+2β2, the newspaper will

have both bias and advertisements. Each advertiser also gets a share ( bqA

)of the biased news. The higher α1 and β2, the higher the bias and thelower the advertisement quantity. The reverse holds for α2 and β1.

16 Newspapers

We summarize these results in the following proposition.

Proposition 2.2.2 Bias appears if relative effectiveness of bias is highcompared to the relative annoyance. The newspaper will implement adver-tisements if these are relatively effective compared with the relative annoy-ance. Only for intermediate effectiveness of bias (vs. ads), both bias andads will be implemented.

The newspaper profit12 increases the more effective bias and advertise-ment (α1 and α2 high) and the lower the annoyance of bias and adver-tisement (β1 and β2 low). Rather remarkable, for the second solution, thenumber of readers is increasing in β1, but decreasing in β2. This meansthat readership increases when bias becomes more annoying. The reasonfor this increased readership lies in the fact that the increased annoyanceof bias induces the newspaper to lower the bias levels in equilibrium, whichis partly replaced by ads. This reduction in bias overcompensates the neg-ative effect of β1 on readership; the result is that readership increases13.

In sum, we can conclude that the monopolist newspaper will appearbias-free if bias is not effective for advertisers or highly disapproved byreaders. There will be only one advertiser (and a substantial amount ofbias) if an ad is not effective for advertisers or it is highly disapproved byreaders. A newspaper will offer the bundle of advertisement and bias if therelative effectiveness (α1

α2) and annoyance (β1

β2) of bias are intermediate.

2.2.2 Competition Model

2.2.2.1 Set-Up

The set-up of the duopoly model is similar to the monopoly model. Tointroduce competition between newspapers, we assume newspapers are

12The profits corresponding with the three solutions can be found in Appendix A.13Note that these results can also help to explain the impact of ad avoidance tech-

nologies, such as digital video recorders (e.g. TiVo) and whitelisting (e.g Ad Block Plus

for Firefox and Chrome). These technologies make it possible to avoid ads, either by

skipping or blocking them. As a result, advertisements become less annoying (β2 de-

creases), but also less effective (α2 decreases). In the extreme case, where β2 = α2 = 0

and as a result β →∞, α →∞, the newspaper implements the third regime, with bias

and only one advertiser. An advertisement alone has no value for an advertiser and is

therefore complemented with bias. Of course, when the model is altered and bias can

be bought directly (without having to by it as a complement to advertising), it is likely

that bias will be offered to various companies. See Anderson & Gans (2010), Wilbur

(2008a) and Wilbur (2008b) for a theoretical and empirical discussion of ad avoidance

on advertisement levels.


horizontally differentiated in the style of Hotelling competition. This com-petition model is commonly used in media economics models (Armstrong& Weeds (2005), Anderson & Coate (2005)).

Readers In contrast to the monopoly model, where readers decide whetherto read the newspaper or not, readers now decide which of both newspa-pers they’ll read. Readers read one and only one newspaper, so we as-sume that the market is covered. The Hotelling model implies that thenewspapers are horizontally differentiated. One example to explain thisdifferentiation is political differentiation. Suppose that one newspaper ispolitically left-leaning and that the other one is politically right-leaning.In the absence of price and quality differences, readers prefer the newspa-per that is closest to their political opinion. It is important to notice thatthe two newspapers are at the end of the Hotelling line, i.e. newspaperstake the most extreme position, or in other words, there is a maximaldifferentiation between both newspapers. The utility functions reads:

uRik = γ0ni − γ1bi − γ2qAi − twik (2.7)

where wik is the distance of reader k towards the position of the news-paper i; t is the parameter that captures the travel cost, or the disutilityfrom every unit that the newspaper differs in position with the own posi-tion. We assume the Hotelling line has a length of one and that readersare uniformly distributed between 0 and 1. Introducing uRi as the utilitywithout the travel cost (i.e. uRi = γ0ni − γ1bi − γ2qAi), the quantity ofreaders of both newspapers reads:

qR1 =12

+uR1 − uR2

2t(2.8)

qR2 =12

+uR2 − uR1

2t(2.9)

Similar to the monopoly case, we also introduce β0, β1 and β2 tosimplify the computational output.

Advertisers While readers are reading one and only one newspaper,advertisers might advertise in one, two or no newspaper. In the two-sidedmarkets terminology, this means that readers are singlehoming and ad-vertisers are multihoming. As such, the newspapers compete for readers,but act as monopolists with regards to advertisers. Since the readers of

18 Newspapers

both newspapers do not overlap, the decision for companies to advertisein newspaper 1 is independent from the decision to advertise in newspa-per 2. This is similar to the set-up of Armstrong (2006). The utility anddemand functions for advertisers are equal to those in the monopoly caseand therefore omitted.

Newspapers The profits of the newspapers are similar to those in themonopoly case. Newspapers do charge for advertisements but not for bias.Readers also don’t pay. The newspaper is a quantity setter, i.e. it choseshow many place is reserved for news, bias and ads. Since the total placein the newspaper is normalized to one, we only look at the decisions forbias (bi) and ads (qAi). The profit function reads:

Πi = (pAi − c)qAi (2.10)

which is maximized with respect to bi and qAi.

Further assumptions As discussed above, we assume that the marketis covered on the reader-side, i.e. all potential readers read one and onlyone newspaper. It suffices to assume that β0 ≤ 3

2 . In the comparisonbetween monopoly and duopoly, we set β0 equal to 3

2 , but this assump-tion does not affect the competition results, since β0 does not appearin the equilibrium quantities. To reduce the notational burden, we sett = 1, though we relax that assumption below to investigate the effect ofa strengthening or lessening of competition.

2.2.2.2 Results

The competing newspapers maximize their profit Πi with respect to bi

and qAi, subject to the constraints 0 ≤ n, b, qA, qR ≤ 1. We solve thisproblem with Kuhn-Tucker conditions.

Lemma 2.2.3 The optimal quantities are symmetric:

bi =

0 if α1α2

<β1(2+β2−

√4+β2

2)

2β22α2

2β21+α2

1β2−α1α2β1(2+β2)

α1β1(α1β2−α2β1)if

β1(2+β2−√

4+β22)

2β2≤ α1

α2≤ β1

1+β21

β1if α1

α2> β1

1+β2

qAi =

2+β2−√

4+β22

2β2if α1

α2<

β1(2+β2−√

4+β22)

2β2

α1−α2β1+α1β2α1β2−α2β1

ifβ1(2+β2−

√4+β2

2)

2β2≤ α1

α2≤ β1

1+β2

0 if α1α2

> β11+β2


As can be seen, the results are qualitatively similar to the ones obtainedin the monopoly situation. We also obtain three solutions, where thesecond solution is the unconstrained maximum and the first and thirdsolutions are the corner solutions where bi and qAi, respectively, are equalto zero. Similar to the monopoly situation, newspapers will offer thebundle of advertisement and bias if the relative effectiveness (α1

α2) and

annoyance (β1β2

) of bias are intermediate.

Intensity of competition To investigate the effect of the strength ofcompetition on the equilibrium results, we relax the assumption t = 1and consider t as an additional parameter. When this parameter t islow, competing newspapers are close substitutes. Based on a comparativestatics analysis, we can state the following.

Proposition 2.2.4 If newspapers are closer substitutes for the reader, thenewspaper lowers the levels of bias and advertisement. The only exceptionis when the newspaper is offering both bias and advertisement: more com-petition decreases bias but increases the ad level.

These results show that if the competition for the reader becomes moreintense, newspapers react to this by increasing the utility of the readers.Even in the case where the ad level is increased, reader utility increasesdue to the decrease in bias. These results confirm our intuition that moreintense competition will drive out some (but not all) bias, and the amountof news will certainly increase. As can be expected: increased competitiondiminishes newspaper’s profit.

Comparison monopoly and competition It is interesting to com-pare our competition results with those obtained under monopoly. Wesummarize the main findings in the following proposition.

Proposition 2.2.5 Bias is (weakly) higher under competition. The num-ber of ads is higher under competition if the relative effectiveness of bias(vs. ads) is low and/or the relative aversion against bias (vs. ads) is high.The profit of a monopolist is always higher than the profit of a duopolist.

The finding that bias is always higher under competition is rather re-markable. One would have intuitively expected that the bias level wouldbe higher under monopoly. The result might be an artefact of our model-ing strategy where the monopoly is located at one side of the Hoteling line,

20 Newspapers

‘competing’ with an outside good (with utility 0) whereas duopoly news-papers are competing with each other, and unaffected by the height of β0,given that β0 is large enough such that the market is covered. The differ-ence in ad levels are determined by the parameters for effectiveness andaversion of bias and ads. Competitive newspaper have lower levels of adswhen ads are relatively annoying and/or relatively ineffective. The reverseholds when bias is more annoying and more ineffective than ads. Profit isalways lower under competition, which is in line with the expectations.

In sum, we can conclude that bias does not disappear under compe-tition. Our results show that more intense competition for the readerreduces advertising bias, though switching from monopoly to competitionincreases the amount of bias and decreases the amount of (real) news.These results differ from those obtained in the literature.

Ellman & Germano (2009) and Germano & Meier (2010) find that induopoly, at least one newspaper eliminates all bias. In their models, biasis bad news about advertisers that is suppressed. Competitive newspapersare likelier to tell the truth about advertisers, because they do not fullyinternalize the reduction in advertisement spending (in all newspapers)due to the bad news. If a newspaper is the only newspaper revealing thenews, it gains on the reader side.

In models of political bias, it is not uncommon that bias increases ifcompetition is introduced (Mullainathan & Shleifer 2005). The mechanismbehind political bias is quite different from our modeling. Political biasmodels assume that readers have a (positive) preference for bias, andassume that people like to read or hear the opinion that is closest to theirprior belief. Switching from monopoly to duopoly or oligopoly, meansthat newspapers can better cater to the preferences of the reader, byintroducing bias. Note however, that bias and the effect of bias are quitedifferent in this setting. Since these biases are in the opposing direction(e.g. politically left-leaning for one newspaper, right-leaning for the other),the amount of (aggregated) misinformation in society is limited. In thecase of advertising bias, it is unlikely that a newspaper will enter withreverse advertising bias, i.e. blaming its advertisers, since this strategy isunlikely to be successful14 (Reuter & Zitzewitz 2006).

14In theory, reverse advertising bias is not impossible. Criticizing advertisers might

be perceived as a strong signal for the trustworthiness of the newspaper. As a conse-

quence, it increases the willingness to pay of readers. Even if readers don’t pay for the

newspaper, a better reputation increases the readership, indirectly increasing revenue.

If this increase from the reader side is large enough, it overcomes the direct revenue


2.2.3 Welfare Analysis

Journalism researchers tend to hold the view that only one level of biasis acceptable for the society: when bias is zero. From an economic pointof view, though, it is not guaranteed that the absence of bias is welfaremaximizing. In this section, we analyze the results of a monopolisticwelfare maximizers (i.e. a Ramsey planner). For clarity, we split theanalysis in reader welfare and advertiser welfare. For simplicity, we onlylook at the case where the Ramsey planner publishes one newspaper.

Reader Welfare Readers gain utility from reading the news in thenewspaper, but value bias, advertisement and a travel cost negatively.Since readers don’t pay to read a newspaper, the consumer surplus of thereaders is equal to

WR =uRqR

2(2.11)

where uR stands for the utility without the travel cost. Written in full,WR = (β0−β1b−β2qA)2

2 . Since the newspaper doesn’t occur costs in ourmodel, reader welfare is feasibly maximized when both the amount of biasand advertisement is zero.

Proposition 2.2.6 Reader welfare is maximized when the newspaper con-tains neither bias nor advertisements.

Note that this resembles a public newspaper where, similar to somepublic broadcasters, no bias and ads appear.

Advertiser Welfare In this two-sided market setting, advertisers area demand side too, and therefore advertiser welfare is also computed as aconsumer welfare. Advertisers gain from every reader, but have to pay aprice pA to appear in the newspaper. The advertiser welfare reads:

WA =uAq2

A

2(2.12)

Maximizing advertiser welfare with respect to bias and advertisement,generates the following results.

decrease from the advertiser side.

22 Newspapers

Proposition 2.2.7 The quantities that maximize welfare are

b =

{0 if α1

α2< 2β1

β2β0(2α2β1−α1β2)3β1(α2β1−α1β2)

if α1α2≥ 2β1

β2

qA =

{2β03β2

if α1α2

< 2β1β2

α1β03(α1β2−α2β1)

if α1α2≥ 2β1

β2

These results learn that an advertiser welfare maximizer will alwaysimplement advertisements, as could have been expected. Bias will appearin the newspaper if the relative effectiveness of bias (vs. ads) is relativelyhigh, or the relative aversion against bias (vs. ads) is relatively low. Notealso the welfare maximizer never brings the advertisement to the maxi-mum level; the highest amount of qA is 2β0

3β2, which is, given our assumption

that β2 > β0, at most 23 . The reason is that advertisers only gain utility

if there are also readers. Implementing a newspaper with only advertise-ments, or only advertisements and bias, would leave the advertisers witha newspaper without readers. Due to this two-sided nature, maximizingadvertiser welfare always ensures that a substantial amount of readers stillreads the newspaper.

The next step in this welfare analysis is a Ramsey planner that maxi-mizes the sum of reader and advertiser welfare. Though equilibrium resultsare not hard to compute, we omit them here because they don’t add muchinsight. As could have been expected, the Ramsey planner weights the in-terests of readers and advertisers. As the advertiser welfare also containspart of the reader welfare, the Ramsey planner has a strong tendency toabolish all bias. Bias might however still occur, e.g. when bias is relativelyeffective (α1 is high) and/or bias is relatively harmless for readers (β1 islow). We conclude that the optimal amount of bias might be positive, atleast when the welfare definition also includes the advertiser’s consumersurplus.

Comparison Welfare under Monopoly and Competition To con-clude this section on welfare, we compare the welfare levels of the profit-maximizing solutions that were discussed in sections 2.2.1 and 2.2.2. Ourcomputations are summarized in the following proposition.

Proposition 2.2.8 The total welfare of readers and advertisers is higherunder competition, if the relative effectiveness of bias over advertisementsis low and/or relative annoyance of bias is high.

2.3 Empirical Evidence 23

While the exact parameter values where welfare is higher under com-petition might differ for reader welfare, advertiser welfare or for both, theresults are qualitatively the same. As we have seen in the comparison be-tween monopoly and competition regime, competitive newspapers alwaysimplement more bias than their monopolistic counterparts; this is likelyto reduce welfare, and therefore welfare is only higher under parametervalues that urge to implement low amounts of bias (because it is ineffectiveor annoyant).

2.2.4 Conclusion of the Model

In sum, we can conclude that a newspaper will appear bias-free if biasis not effective for advertisers or highly disapproved by readers, whetherthe newspaper faces competition or not. There will be only one advertiser(and a substantial amount of bias) if an ad is not effective for advertisersor it is highly disapproved by readers. A newspaper will offer the bundleof advertisement and bias if the relative effectiveness (α1

α2) and annoyance

(β1β2

) of bias are intermediate.A comparison between monopoly and duopoly newspapers learns that

these results are qualitatively the same, but that newspapers under duopolyset more bias and make less profit than their monopolistic counterparts.If duopoly newspapers are closer substitutes for readers, then newspapersset less bias; while the effect on advertisements is ambiguous.

2.3 Empirical Evidence of Advertising Bias

in Belgian Newspapers

The previously presented theoretical model predicts that the combinationof bias and advertisement will only exist for certain parameter values. Itis therefore an empirical question whether advertising bias will prevail.We test the existence of advertising bias in the Belgian Dutch-languagenewspapers for the period 2001-2005.

2.3.1 Data Description

Clearly, in order to measure advertising bias, one needs rich data aboutadvertising spending at the advertiser/newspaper level as well as data onmedia coverage of these advertisers in the different newspapers.

24 Newspapers

Advertisers Data on advertising expenditures are obtained from AegisMedia (see also Van Cayseele & Vanormelingen (2009)). The data set con-tains monthly expenditures by companies in each newspaper separately,from January 2001 till June 2005. From this set, we selected the largecompanies that concentrated their ads over newspapers. First, we focusedon firms with a total advertising spending of over 300000 euros over thewhole sample period. Moreover, we computed for each advertiser the shareof each newspaper in total advertising spending of the advertiser. We usedthese shares to compute for each advertiser a Hirschman-Herfindahl indexof advertising spending and selected all companies with an index above3000. We deleted two categories of companies from this list. First, weeliminated company names which are also common Dutch words or com-mon family names, in order to avoid too many false positives. Second,we deleted the companies closely linked to a newspaper, such as the ownadvertising agency15. Finally, 57 advertisers are withheld in the datasetand we observe for each of them monthly ad spending in each differentnewspaper.

Coverage The second part of our dataset contains data about news-paper coverage of advertisers and we obtain a measure of how often an(potential) advertiser is mentioned in the newspaper. To be precise, wecount for each newspaper edition, published on a given day, the number ofarticles in which the advertiser is mentioned. To this end, we ran keywordsearches on the company name or the major company product on the Me-diargus database, which is an online archive of all Belgian Dutch-languagenewspaper articles. Note that we counted several mentions in one arti-cle only as one observation and did not distinguish between positive andnegative news16.

15The reason for focusing on the companies that typically concentrate their adver-

tising spending in a limited number of newspapers is to optimize the time consuming

collection of data about coverage. In our empirical strategy we are going to exploit the

variation of advertising spending by a specific advertiser across different newspapers in

a given time period and link this with advertiser coverage in that particular newspaper.

High advertising spending by an advertiser in all newspapers simultaneously will be

picked up by control variables and would not help in identifying the advertising bias

parameter. Therefore we opt to focus on those advertisers concentrating their spending

in a limited number of newspapers.16Obviously, it would be interesting to also include a content analysis. However,

given the size of the dataset on coverage, this would be very time consuming. Moreo-

ever, it is not easy to define some objective measures of positive versus negative news


Table 2.1: Summary Statistics of Advertising and Coverage

Variable Sample Mean S.D. max obs

Advertising All 2249 20039 842108 24624

(euros) Non-Zero 28558 65200 842108 1978

Coverage All 1.58 3.25 96 24624

(nr. of articles Non-Zero 3.43 4.06 96 11389

per month) Advertisers 3.39 5.66 94 1978

Non-Advertisers 1.43 2.89 96 22624

Our final dataset contains information about coverage and advertisingspending in all newspapers active in the Belgian-Dutch language newspa-per market17. The market consisted of 3 elite newspapers and 5 popularnewspapers18 during the period of study. Coverage data is aggregated atthe newspaper-advertiser-month level as we observe advertising spendingat the monthly level instead of the daily level. We present some summarystatistics in table 2.1. In total there are 24624 observations19.

From this table, we can infer that only 1978 of the 24626 observationscontain positive advertisement expenditures, i.e. the average advertiserin our sample advertises 34 newspaper months. The average monthly ex-penditure is 28558 euro for non-zero expenditures, which equals roughlytwo black and white full pages ads. The last two rows indicate an adver-tiser is mentioned twice as much as a non-advertiser. This is a very roughindication that advertising bias might exist.

against which the article could be screened. Therefore, we refrain from doing so and

leave this issue to further research.17The dataset excludes the free newspaper ”Metro” as we do not have any content

information about this newspaper.18These newspapers are: De Morgen, De Standaard, De Tijd (formerly Financieel-

Economische Tijd), Gazet van Antwerpen, Het Belang van Limburg, Het Laatste

Nieuws, Het Nieuwsblad and Het Volk. The first three newspapers are elite newspapers

that focus more on political and international news compared to the popular newspa-

pers who typically spend more articles on showbusiness and criminal news. These are

all the Dutch-language national newspapers, except Metro, which is a free newspaper

that is not available for subscription.19 We have data on 57 firms advertising in 8 newspapers over a period of 54 months

and 57*8*54 = 24624.

26 Newspapers

2.3.2 Empirical Strategy

In order to test for advertising bias, we investigate the effect of adver-tisement expenditure on news coverage. The summary statistics alreadyshowed how coverage of firms that advertised in the newspaper in a givenmonth was already higher compared to the coverage of a firm that didnot advertise. However, there could be various mechanisms going onhere besides advertising bias to explain this positive correlation. First,larger firms would tend to spend more on advertising and would also bemore likely to be covered in newspapers. Second, some characteristics ofthe newspaper will make it more likely to publish articles about compa-nies as well as receiving more advertisements. For example, the averagemention in the only business newspaper in our sample, De Tijd, is with2.27 articles higher than the average of the other newspapers, althoughthe difference is not statistically significant. Third, there could be someunderlying common trends in both advertising spending and newspapercoverage. Moreover, these trends could be newspaper specific. Fourth,due to some unobserved characteristics of the readership of a particularnewspaper, firms could be more interested in that newspaper compared toother newspapers and that particular readership could be more interestedin the products of the advertiser as well. Given that newspapers tend tochoose articles based on what their readers interests, a positive correlationbetween coverage and advertising would arise. Fifth, firms could be morelikely to advertise when there has happened something newsworhty at thefirm (e.g. a new product launch) and again a positive correlation betweenadvertising spending and coverage would be found.

In our empirical specification, we will control for these factors. If thenewspaper is unbiased, one would expect the positive relation betweencoverage and advertising spending to fade away when controlling for theabove mentioned confounding effects. More precisely, we will estimate thefollowing equation:

Cijt = αAijt + γi + γj + γt + γij + γjt + γit + εijt (2.13)

where Cijt represents coverage of advertiser i by newspaper j in month t.Coverage is measured by the number of articles in which the advertiser orits main product is mentioned. Aijt is advertising spending of advertiseri in newspaper j in month t, and advertising expenditure is measuredin thousands of euros. As discussed above, there could be some adver-tiser specific effects γi leading to higher coverage. Newspaper time and


newspaper-time fixed effects are represented by γj , γt and γjt respectivelyand γij are advertiser-newspaper idiosyncrasies in coverage. Finally, γit

are advertiser-time specific fixed effects on coverage.By controlling for all these factors, we solve for most of the concerns

in identifying a causal impact of advertising spending on coverage men-tioned in the literature. Moreover, we provide some Granger causalitytests as well as include lagged advertising spending to address additionalconcerns about the interpretation of the correlation between advertisingand coverage.

2.3.3 Results

The results of estimating equation 2.13 are displayed in table 2.2.

28 Newspapers

Tab

le2.

2:C

over

age

expl

aine

dby

adve

rtis

ing

and

dum

mie

s

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Adver

tisi

ng

0.0

186***

0.0

155***

0.0

147***

0.0

0946***

0.0

0891***

0.0

0657***

0.0

0472***

[0.0

0352]

[0.0

0384]

[0.0

0379]

[0.0

0263]

[0.0

0244]

[0.0

0242]

[0.0

0156]

Tota

lA

ds

0.0

0185***

[0.0

00702]

Dum

mie

sA

dv

Adv,N

ewsp

Adv

XN

ewsp

Adv

XN

ewsp

Adv

XN

ewsp

Adv

XN

ewsp

Month

Month

New

spX

Month

New

spX

Month

New

spX

Month

Adv

XM

onth

N24624

24624

24624

24624

24624

24624

24624

R2

0.0

13

0.1

75

0.1

95

0.3

83

0.4

01

0.4

02

0.6

28

Sta

ndard

Err

ors

inbra

cket

s.*

p<

0.1

0,**

p<

0.0

5,***

p<

0.0

1

Num

ber

ofart

icle

sas

dep

enden

tvari

able

Tota

lA

ds

isth

eadver

tisi

ng

spen

din

gofth

esa

me

adver

tise

rin

oth

ernew

spaper

s


As a first indication, we regress coverage on advertising. The coefficientof advertising is statistically different from zero. The magnitude is surpris-ing. An additional 54000 euro (=1000 × 1

0.0186 ) advertising expendituregenerates one additional article in the same month. Note that only a verysmall part of the variation is explained (R2=0.013). Again, this is just acorrelation between coverage and advertising spending and all above con-founding effects are also at play. To account for these effects, we includein each specification more and more controls. For example moving fromthe first to the second column, we include advertising dummies to takeinto account the above decribed advertiser specific effects. As expectedthe coefficient on advertising spending goes down, but remains positiveand significant. From columns (4) and (5), it can be seen that controllingfor advertiser-newspaper effects, has a substantial impact on our estimatefor the impact of advertising on coverage. Finally, we also want to controlfor the advertiser specific changes in coverage over time. First, we includeadvertising spending in other newspapers by the particular advertiser asa control variable and second we include interactions between time andadvertiser dummies in the regression. The second strategy is the most ro-bust one, but has the disadvantage of slicing away much potentially usefulvariation in both coverage and advertising. The first strategy is computa-tionally less intensive and preserves more variation in the dependent andindependent variable but rests on the assumption that changes in news-worthiness of advertisers over time can be represented by a linear functionin advertising spending in other newspapers. Even with these controls, theeffect of advertising on coverage remains statistically significant and pos-itive as displayed in columns (6) and (7). The coefficients imply thatrespectively an extra 152000 or 212000 euro of advertising spending willgenerate an additional article in the newspaper.

To put these results in perspective, remember that the average (non-zero) advertiser spends 28858 euro in one newspaper in one month. Thiscorresponds with less than one seventh of 212000 euro, the additionalspending necessary to generate one additional article. As such, the averageadvertiser has a 14% chance of being mentioned. The average full pageblack and white advertisement cost 14450 euro (in the sample period).Stated differently, an advertiser has to buy over 14 full page black andwhite advertisements to receive one additional article. Whether this is agood deal for newspapers or advertisers, depends on the reputation lossfor newspapers (think of the reader averseness β1 in our theoretical model)

30 Newspapers

and the effectiveness of bias (α1 in our theoretical model).Based on these results, we find that if an advertisers stops advertising,

his coverage drops on average from 1.58 to between 1.05 and 1.45 articlesper month per newspaper20.

These results also allow to determine the total amount of newspaperarticles that are due to advertiser bias. The average amount of advertisingmoney spend in newspaper between 2001 and 2005 is 240 million per year.This means that, on average, 1132 up to 4464 articles per year are relatedto advertiser bias. There are many caveats with this simple extrapolationfrom our sample to the total population of advertisers, though these resultsgive at least an indication of the order of magnitude.

The finding of advertising bias in general newspapers is remarkable.Reuter & Zitzewitz (2006) document that the coverage of financial mag-azines is affected by mutual funds advertisement, but general newspapers(New York Times and Wall Street Journal) remain unaffected. Gam-baro & Puglisi (2009) find a positive effect of advertising expenditure onthe daily coverage in general newspapers, but the effect disappears whennewspapers times company fixed effects are included.

It is not so straightforward to compare the magnitude of our coefficientswith previous studies, since advertisement prices also depend on the read-ership; some studies are also in logs. The results of Reuter & Zitzewitz(2006) show that spending 1 million dollar on advertising increases theprobability of a mention with 0.1 to 0.2 percent. Stated differently, of allmutual funds mentions roughly 8 up to 23 percent are due to advertiserbias. Rinallo & Basuroy (2009) measure bias in pages and find that ad-vertising one percentage more pages generates 0.4 percent more articlesin fashion magazines.

Robustness Checks We carry out a number of robustness checks intable 2.3. We exclude the first eight months, because the aggregate datashow a peak in coverage in these months (see Appendix B, graph B2).Next, we exclude the business newspaper De Tijd because this newspaperreports a lot more on companies. We also exclude the high coverage adver-tisers and finally include advertising squared. Our results are unaltered.The quadratic function seems to better fit the data, though the coefficientfor the quadratic term is rather small. Furthermore, we also put adver-

20The difference is the average amount spent, divided by 1000 and multiplied by the

coefficient of advertising, e.g. 285581000

× 0.00472 = 0.135 for specification (7).


Table 2.3: Robustness Checks(1) (2) (3) (4)

Advertising 000532*** 0.00547*** 0.00286*** 0.0157***[0.00196] [0.00203] [00.000862 [0.00464]

Total Ads 0.00129** 0.00189** 0.000569 0.00175**[0.000529] [0.000770] [0.000393] [0.000750]

Advertising2 -0.0000219*[0.00000642]

Dummies Adv X Newsp Adv X Newsp Adv X Newsp Adv X NewspNewsp X Month Newsp X Month Newsp X Month Newsp X Month

N 20976 21546 24094 21546R2 0.406 0.368 0.458 0.369

Standard Errors in brackets. *p<0.10, ** p<0.05, *** p < 0.01Number of articles as dependent variableTotal Ads is the advertising spending of the same advertiser in other newspapersModel (1) First eight months excludedModel (2) De Tijd excludedModel (3) Observations for which coverage > 10 excludedModel (4) Advertising squared included

tising in logs. Since most of our observations included zero advertisementexpenditure, more than 90 percent of our observations is dropped. Westill obtain a positive effect but significance decreases to a 95% confidenceinterval or disappears in the specification with newspaper times monthinteraction dummies. These results are represented in Appendix B, tableB1.

2.3.4 Concerns

Killing negative news In our empirical identification, we did not dis-tinguish between good and bad news. We adopted the assumption thatany news is good news. In section 2.5, we provide some anecdotal evidenceon newspapers and magazines suppressing negative news on advertisers.This might affect our coverage variable. We refrained from distinguishingpositive and negative news because it is hard to find an objective mea-sure. Some authors have attempted to separate positive from negativenews; though only in specific contexts. Reuter & Zitzewitz (2006) investi-gate mutual funds and classify recommendations as positive, dissuasionsas negative. Gambaro & Puglisi (2009) analyse changes in stock prices –news following an increase is considered as positive, and vice versa. Gurun& Butler (2010) count the number of negative words used in articles aboutcompanies. Lott Jr. & Hassett (2004) look at the coverage of economicnews. Based on the economic variable discussed in the title of a newsarticle, and the adjacent verb (e.g. increase or decrease) or adjective (e.g.

32 Newspapers

strong or weak), they coded the news. A similar approach is followed byDoms & Morin (2004). These two latter papers are more sophisticatedversions of the R-word index used by the Economist21. Hamilton (2004)uses the DICTION software to analyse hard vs. soft news. This softwareanalyses word use and word length. For general news on companies, it ismuch harder to reach agreement how to classify articles. Content analysismight prove helpful (see Fico, Lacy & Riffe (2008) for an overview), butit is very hard to set up an automated procedure. Wood, Nelson, Cho& Yaros (2004) manually coded the appearance of brands or companiesin television news or news shows as negative, neutral and positive. Theyclaim a high percentage match amongst independent coders. Upshaw,Chernov & Koranda (2007) conduct a similar analysis. Williams (2009)uses a combination of automated evaluation and coding by humans in acase study evaluating the tone of the news on a financial institution duringthe financial crisis.

If suppressing negative news is an important force, then our test isinconclusive about advertising bias. A statistically positive coefficient canthen be explained by more a higher positive coverage (i.e., more advertis-ing bias) or by less killing of stories (i.e., less advertising bias). However,previous empirical studies that distinguish between positive and negativenews, find that advertising has an influence on positive news but not onnegative news (Reuter & Zitzewitz (2006) and Gambaro & Puglisi (2009)).

Timing In this article, we suggest that advertising generates additionalcoverage. One might argue that the causality can also be the other wayaround. It might be the case that a newspaper acts over-friendly towardssome companies in order to attract them as potential advertisers. Totest for this, we perform a Granger causality test. We regress coverage onlagged ads and lagged coverage; and we also regress ads on lagged coverageand lagged ads. If the coefficient of the lagged ads are jointly significantin the first specification while the coefficients of lagged coverage aren’t inthe second, then we can conclude that advertising Granger causes coveragebut not the other way around. The results in table 2.4 indeed confirmsour hypothesis.

We also ran regressions with lags, showing a similar picture (see tables

21The R-word index is a simple count of the number of times the word ”recession”

appears in newspapers. It was proposed in 1992 as an alternative indicator of economic

activity. (Rrrrrrrecession?, The Economist, 16-07-1998.


Table 2.4: Granger Causality Test

Coveraget Advertisingt

Coveraget−1 0.359*** -0.0476

[0.00663] [0.0300]

Coveraget−2 0.138*** 0.0110

[0.00691] [0.0313]

Coveraget−3 0.132*** 0.0105

[0.00682] [0.0308]

Coveraget−4 0.102*** 0.0397

[0.00628] [0.0284]

Advertisingt−1 0.00188 0.564***

[0.00148] [0.00668]

Advertisingt−2 0.00381** 0.218***

[0.00172] [0.00777]

Advertisingt−3 -0.00176 0.0972***

[0.00173] [0.00781]

Advertisingt−4 0.00145 0.103***

[0.00156] [0.00704]

N 22800 22800

R2 0.413 0.746

Joint significance (F) 10.58 16429.05

Lagged Ads (Prob > F) 0.000 0.000

Joint significance (F) 3859.97 1.14

Lagged Coverage (Prob > F) 0.000 0.3373

Standard errors in brackets

* p < 0.10, ** p < 0.05, *** p < 0.01

34 Newspapers

Table 2.5: Coverage explained by advertising of last 12 months

(1) (2) (3) (4)

Ads (lag year) 0.0237*** 0.00925*** 0.00795*** 0.0202***[0.00412] [0.00158] [0.00167] [0.00476]

Total Ads (lag year) 0.000867[0.000686]

Dummies Adv Adv X Newsp Adv X MonthNewsp X Month

N 19152 19152 19152 19152R2 0.019 0.404 0.404 0.462Standard errors in bracketsNumber of articles as dependent variable* p < 0.10, ** p < 0.05, *** p < 0.01

B2 and B3 in Appendix B). In the specification with lagged advertising oncoverage, coefficients are still significantly (at least jointly) different fromzero. In the reverse regression, the significance disappears. This can alsobe interpreted as weak evidence that advertising influences coverage, butnot the other way around. Reuter & Zitzewitz (2006) show that the reverseeffect (an increase in coverage followed by an increase in advertisement)is much stronger for those companies that already advertise. This alsosuggests a causal advertisement-coverage relation.

Targeted Advertising It might be that our results do not pick up onlybias, but also targeted advertising. Suppose that a certain firm releases anew product. If the readers of a particular newspapers are very interestedin this type of product, it is not illogical that the firm will advertise inthis newspaper (and not in others); and that the newspaper will reporton the product (while others don’t). Our specifications, that exploit thedifferences between newspapers and differences over time, will pick up thiseffect as bias, while it is not. However, note that in our most demand-ing specification, we include not only interaction between advertiser andnewspaper dummies but also between advertiser and month dummies. Sowe control for the average targeted advertising in a certain newspaper aswell as the different advertising intensities by an advertiser over time.

Although we control to a large extent for the targeted advertising prob-lem, there could still be an effect present, i.e. advertisement and coverageincrease due to a product launch (see above). To circumvent this problem,we also run a regression with all the lagged advertisement expenditure in

2.4 Survey of Belgian Journalists 35

the specific newspaper over the last year. The coefficients, shown in table2.5, remain significant throughout these specifications.

2.4 Survey of Belgian Journalists

Conditional on the fact that our sample is representative for other Belgianadvertisers, we find that advertiser bias counts for 1000 up to 5000 addi-tional articles per year. An important question is how these articles endup in a newspaper and how the practice of advertiser bias is perceived byjournalists. In this section, we present a survey conducted among journal-ists working for Dutch-language Belgian newspapers. First we provide anoverview of other surveys on journalists, editors and related parties. Thenwe describe the methodology and the most important results.

2.4.1 Related Literature

Given the potential impact of mass media on society, the people working inthis sector have been repeatedly surveyed. Some of these surveys questionthe perceived influence of advertisers on coverage. Not surprisingly, theperceived advertising bias strongly depends on what you ask and whomyou ask. Outside parties, such as marketeers and the audience, are moreconcerned about advertiser pressure than inside parties, such as journal-ists and editors. In this section, we present an overview of the existingsurvey results on advertiser influence and news. Results are difficult tocompare, because the methodology differs in nearly every study. Besides,media actors interviewed work in different countries and in different mediabranches. A summary of survey results can be found in Appendix C.

Journalists Journalists are at the heart of the news publishing processand are often surveyed on their views of their own profession. All studiespresented here, except for Price (2003), question the advertising bias issueas a part of a longer questionnaire. One remarkable conclusion is that thepercentage of journalists that perceives advertiser influence varies betweena mere 7% and a worrisome 85%. Note however that this last resultmight be biased upward because the survey questioned whether there wasinfluence from advertisers, owners and marketeers. If we look at results ofquestions asking for advertiser influence only, then the percentage variesbetween 7% and 38%. This means that a majority of journalists is not

36 Newspapers

pressured or influenced by advertisers. However, this does not rule outadvertising bias since on average 25% feels the heat from the advertisers.

Editors A similar variance in answers is found when we look at studiesthat question editors. Pew (2008) finds that only 15% of newspaper andtelevision editors perceive that advertiser concerns influence news cover-age decisions, while in Soley & Craig (1992) 90% of the newspaper editorsreport that advertisers try to influence the news. Interestingly, when askedwhether their news organization has engaged in a new revenue experimentthat has raised concerns about editorial independence, 36% of journalismexecutives answers ‘yes’ (Pew 2010). Detailed answers included ‘ad spon-sorships of specific content’ and ‘blurring the lines between ads and news’.These large differences might be explained by the way the questions areasked. Soley & Craig (1992) ask whether they know of one case wherean advertiser tried to influence while the Pew studies (Pew (2000), Pew(2004) and Pew (2008)) and ASNE (1999) ask whether advertiser influenceexists in general.

Other Parties There are only a limited number of studies on otherparties involved in advertising bias, such as marketing and sales inside andoutside the newspaper, readers and owners. Managers and owners considerthe amount of advertiser influence as very limited (10 to 16 %) while halfof the readers in a study by ASNE (1999) think that newspapers allowadvertiser’s interests to influence the news. An & Bergen (2007) questionadvertising directors at US newspapers. One third finds it acceptable thata salesperson asks to favor advertisers as news source, while more than90% find it unacceptable to suppress a negative story because an advertiserasked it. Roughly 60% of marketing experts and PR people believe thatPolish newspapers insert advertisement bias (Tsetsura 2005).

The overview provided above showed that differences on advertisingbias perception are large, but one should be cautious to compare thesestudies, given the differences in methodology, professions, countries andmedia sectors. In general, some respondents perceive influence of ad-vertisers on editorial decisions but this concerns only a minority of therespondents.


Table 2.6: Summaries SurveyNumber Percentage

Age 91 100%20-30 21 23.1%31-40 19 20.8%41-50 23 25.2%50+ 28 30.8%

Experience 90 100%0-5 years 26 28.9%6-10 years 14 15.6%+10 years 50 55.6%

Media Group 63 100%Persgroep 26 41.3%Corelio 21 33.3%Concentra 16 25.4%

Editorial Office 89 100%Politics/Domestic/Foreign 23 25.8%Business 11 12.4%Editor (in chief / graphical) 25 28.1%Other 30 33.7%

2.4.2 Methodology

In order to measure advertisers’ pressure as perceived by journalists, weconducted an anonymous survey among journalists working for Dutch lan-guage newspapers in Belgium. To our knowledge, there is only one studythat questions Belgian journalists on advertiser pressure. Ppress (2010)reports that 85% of the journalists think that “Objectivity of journalismis under increasing influence from advertisers, marketeers and owners”.Their study is not comparable to ours, since this is the only questionon advertiser pressure and since the respondents are magazine writers,compared to newspaper journalists in ours.

The survey was created on a secured online web page. The invitation

38 Newspapers

to participate in the survey with the link to the website was sent by e-mail to the population of around 750 newspaper journalists working forDutch language newspapers in Belgium. We retrieved the e-mail addressesfrom the Association of Belgian Journalists (VVJ), which also encouragedto its members to participate in the survey. Note that only contractualjournalists were contacted, so freelancers were excluded. In total, 131journalists filled out the survey, which was reduced to 101 after droppingthe respondents that spent less than 2 minutes on the survey. This impliesa response rate of around 13.5%. In the introduction to the survey itwas clearly mentioned that the survey was about the relation betweenadvertisers and newspapers.

Table 2.6 displays some summary characteristics of the respondents.The journalists included in the survey are relatively equally spread outover the different age groups, although there are slighltly more journalistsolder than 50 years. Concerning the experience at their current employer,the majority has been working for over 10 years at the same newspaperand 30% of the respondents has less than 5 years experience. In accor-dance with their market shares, most journalists are working for Pers-groep, followed by Corelio and Concentra, although a substantial numberof journalists opted not to respond to this question. Finally, we groupedthe journalists according the editorial office they are working for. Aroundone fourth of the respondents writes about politics or brings more generaldomestic or foreign news and 12% works for the business (economics) edi-torial office. Around 30% works as editor (in chief) or as graphical editor.The rest (33%) writes regional, lifestyle, sports or other news.

2.4.3 Results

The first part of the questionnaire asked the respondents after their gen-eral opinion about the influence of advertisers on newspaper content. Therespondents were presented several statements which they had to judge ona 5-point Likert scale, namely 1 (Strongly disagree) 2 (Disagree) 3 (Nei-ther Agree nor Disagree) 4 (Agree) 5 (Strongly Agree). The responsesto a selection of these questions are reported in figure 2.1, where for ex-positional purposes we have pooled Strongly (dis)Agree and (dis)Agreetogether. First we asked whether advertisers received as much attentionin the newspaper as non-advertisers. Over 60% of the journalists thinksthat there exists no advertising bias as advertisers are mentioned as muchas non-advertisers, while 20.8% (strongly) disagrees with the statement.


Figure 2.1: Questions on Relationship Advertisers Newspapers

Almost one fourth of journalists agree with the statement that advertisersare trying to exercise some influence on the content of the newspapers,although only 12% of them believes they succeed in doing so. Accordingto results, not reported here for brevity’s sake, over 90% of the journalistsstrongly dislikes advertising bias as they state it would be inappropriate ordetrimental for newspaper quality when advertisers manage to influencethe editorial content.

Next, the correspondents were asked directly how often they are ap-proached by somebody, either the editor in chief, direct editor, marketingdepartment or advertiser, to write an article about an advertiser. Again,the responses were measured on a five-point scale, namely 1 (Very Often)2 (Often) 3 (Sometimes) 4 (Seldom) 5 (Never)22. The results are reportedin figure 2.2. Pressure to write an article about an advertiser comes mostlyfrom the marketing department. Over 40% of the journalist are at leastseldom contacted by the marketing department to write an article about

22In an explanatory note, we indicated that ”Very Often” means approximately on

a daily basis, ”Often” means approximately once a week, ”Sometimes” means approx-

imately once a month and ”Seldom” means approximately once a year.

40 Newspapers

an advertiser and for over 20% this happens more or less frequently (atleast once a month). Around 35% journalists approached by the marketingdepartment say they never follow the advice to write about an advertiser(which leaves of course 65% of the journalists following at least seldomthis advice)23. Also advertisers themselves try to put some pressure onjournalists as 20% of the journalists report that they are at least some-times contacted by firms to write an article about them. However, 60% ofthe journalists say they never follow this suggestion. Suggestions to writean article about an advertiser by the editors occur much less frequently,but if they occur, they are, not surprisingly, more often followed.

When we combine the different channels through which the advertiserscan influence the editorial content, we find that 13% of the journalists getsat least often approached by at least one of the four different agents (theeditor in chief, direct editor, marketing department or advertiser) to writean article. 36 % gets at least sometimes approached and almost 60% getsat least seldom approached. These numbers seem to indicate that there issome advertisers’ pressure, although limited to a subset of journalists andonly a part of them gets approached regularly.

We perform this exercise for different groups seperately and report theresults in the first three columns of Table 2.7. First we look at whetherthere exist differences in advertiser pressure across different experiencecategories. Although it is difficult to do draw firm conclusions due to thelow number of observations, it appears that, in line with our priors, lessexperienced workers get approached more often compared to more expe-rienced workers. Splitting the sample by the editorial office the journalistis working for, displays no clear picture, Looking at the percentage ofjournalists that are approached at least sometimes, it appears that mostlyjournalists working at “Other” editorial offices are more prone to experi-ence some advertisers’ pressure. The reason that we do not find a cleareffect could be the relatively low number of observations which obligesus to divide the sample into coarsely defined subgroups. Finally, we in-vestigate whether there exists differences across the different newspapergroups in Belgium. Again, the results should be interpreted with cau-tion, but advertiser pressure seems somewhat larger for the Concentranewspaper group.

23These results come from a second question where we asked whether the journalists

followed the advice. The full results are not reported in the paper, but are available

on request.


Figure 2.2: Suggestion to Write about Advertiser

Advertising bias can not only appear through the simple writing ofan article about the firm in question, but also through the adjustmentof articles about the advertiser, bringing it more in line with advertisers’expectations and wishes. We asked our respondents about the occurrenceof this kind of pressure and results are reported in Figure 2.3. Clearly,the prevalence of pressure to adjust an article about an advertiser is lowerthan the pressure to write an article about an advertiser. Only 10% ofthe journalists report that they receive sometimes or often the suggestionfrom the marketing department to rewrite an article. Similar figures areachieved for advertisers, editors in chief and direct editors. This low figurecan also be due to the fact that the marketing department, and certainlyadvertisers, normally do not preview the articles before they are published.Computing the percentage of journalists that is approached by at least oneof the four different agents, shows 5.2% is at least often approached, 17.5%is at least sometimes approached and 34% is at least seldom approachedto adjust the content of an article about an advertiser. When we split upthe sample by different categories of journalists, we obtain similar resultsas for the part where we asked whether they were approached or not.

Besides these quantitative questions, we also asked qualitative ques-

42 Newspapers

Figure 2.3: Suggestion to Adjust Article about Advertiser

tions where journalists could comment or give examples. These commentsand examples are incorporated in the next section.

To conclude one could say that apparently there is some pressure ofadvertisers, either directly or indirectly, to steer the content of newspapers.However, only a limited number of journalists is affected. Unfortunately,our sample of respondents is too small to split up the sample by thedifferent characteristics of the journalists and perform regressions.

The finding that only a small minority is often approached, does notconflict with the results found in section 2.3. We found that advertiser biasaccounts for roughly 1000 up to 5000 additional articles per year. Giventhat Flanders counts 755 newspaper journalists (freelancers not included)and that the average journalist in our sample writes 10 articles per week,it is not illogical that journalists are approached only seldom, and thatmany journalists are not approached at all. Moreover, advertiser biasmight be implemented without the journalist’s consent and therefore notbe perceived as bias. In the next chapter, we shed light on how this biasmight be implemented.

Moreover, note that the survey measures the perception of journalistsabout advertising bias, which is not necessarily equal to actual pressure of


advertisers to influence newspaper content. As is typically the case withsurveys, respondents might not have the right incentives to reveal thetruth. In this survey, one might argue that journalists have no incentiveto reveal advertisers pressure or bias, since this would hurt the reputationof their profession. Moreover, the - legally not binding - Belgian Code ofthe Council of Journalism (Raad voor de Journalistiek 2010) stipulatesin article 11 that “the journalist does not engage in advertising or propa-ganda, and does not give in to pressure from advertisers or stakeholders ofinformation”. However, we hope to have mitigated these effects by makingour survey anonymous.

44 Newspapers

Tab

le2.

7:Su

rvey

Res

ults

per

Gro

upA

ppro

ach

edA

sked

toch

ange

Oft

enSom

etim

esSel

dom

Oft

enSom

etim

esSel

dom

Tota

l(N

=101)

12.9

%35.6

%59.4

%5.2

%17.5

%33.7

%

Exper

ience

(N=

90)

0-5

yea

rs(N

=26)

23%

54%

65%

4%

19%

42%

6-1

0yea

rs(N

=14)

14%

36%

64%

7%

21%

21%

+10

yea

rs(N

=50)

10%

24%

54%

6%

16%

36%

Edit

ori

al(N

=89)

Politi

cs/D

om

./For.

(N=

23)

17%

35%

56%

0%

17%

30%

Busi

nes

s(N

=11)

9%

18%

55%

9%

18%

36%

Edit

or

(chie

f/

gra

ph.)

(N=

25)

12%

28%

64%

4%

13%

28%

Oth

er(N

=30)

17%

47%

57%

10%

25%

40%

Med

iaG

roup

(N=

63)

Per

sgro

ep(N

=26)

19%

27%

54%

8%

15%

31%

Core

lio

(N=

21)

0%

33%

52%

0%

10%

14%

Conce

ntr

a(N

=16)

13%

50%

75%

13%

27%

56%

The

num

ber

sin

this

table

are

the

per

centa

ge

ofjo

urn

alist

sth

at

are

appro

ach

edat

least

oft

en,so

met

imes

or

seld

om

by

adver

tise

rs,m

ark

etin

gdep

art

men

t,dir

ect

or

indir

ect

chie

f.

2.5 Advertising Bias Channels 45

2.5 Advertising Bias Channels

In the previous sections, we showed that it is likely that the bundle ofadvertisements and coverage exists. It remains an open question however,how this bundle might be implemented. Based on our survey and anecdo-tal evidence worldwide, we try to uncover these channels. We determinefour channels. We label the first two as supply and demand channels,because they uncover how the bundle might be offered (supply) or asked(demand). We complement them with inside channels which explain howthe journalistic ethics might be circumvented inside the news firm. Wediscuss these channels only briefly here, a more elaborate version withexamples can be found in Appendix D.

2.5.1 Supply and demand channels

In this subsection, we discuss how bundles of advertisements and coveragemight realize, by investigating the supply (i.e. the newspaper) and thedemand (i.e. the advertisers).

Supply channel The clearest way how the advertising bias is imple-mented, is when the media outlet offers news for sale. This can be verydirect, i.e. by selling news space directly; or more indirectly by offeringa bundle of coverage and advertisement. This bundled coverage mightcontain additional news about the advertiser, or more in general positivenews that encourages consumption.

Demand channel It might also be the case that the bundle is not of-fered by the media outlet, but exacted by the advertiser. Advertisersmight explicitly negotiate such a bundle; or withhold future advertise-ments if it is not supplemented by favorable coverage. Some advertisershave successfully used withdrawal threats, especially to suppress negativenews. Other apply a withdrawal rule, i.e. they automatically withdrawtheir advertisements if certain content is published. It goes without sayingthat media outlets are not bound by such rules, though they are financiallypunished if they publish on sensitive topics.

46 Newspapers

2.5.2 Inside channels

While media outlets might offer bundles, or advertisers might commandthem, in many media firms, there is a strict wall of separation between thebusiness side and the editorial staff. If the editorial staff can freely choosethe topics they write about, then the scope for advertising bias is limited.That is exactly why such a wall of separation was constructed. There area number of channels though, that allow the business side to circumventthis wall. We summarize them as ‘crossing the wall’ and ‘editorial staffcomposition’.

Crossing the wall If journalists don’t want to implement this bundle,then the business side might do it themselves. In many newspapers, thebusiness side engages in journalism, especially for the lighter sections suchas lifestyle, travel, real estate, cars and food. But the business side mightalso invite journalists to cross the wall, by having them perform in adver-tisements, or organizing meetings between advertisers and journalists.

Editorial staff composition If some journalists don’t want to engagein advertising bias, then media outlets might hire simply other journal-ists. The most visible composition changes are lay-offs. A more subtlemechanism is dedicating more journalists to the fields that are interestingto advertisers.

2.6 Conclusion

Many scholars have argued that there should be a firm wall between theadvertisement department and the editorial staff of a newspaper. Theadvertising department should not exert pressure on the editorial staff toshape articles. And if they do the professional ethics of journalists shouldprevent any direct or indirect influence of advertisers on the selectionof news. The analogy with a church-state separation of journalism andcommerce is never far away.

Newspaper owners and editors might nevertheless have strong incen-tives to tear down the wall between the advertisement department and theeditorial staff. Our theoretical model shows that it is profitable for bothmonopoly and duopoly newspapers to offer a bundle of bias and adver-tisement if the effectiveness of bias is relatively high for advertisers and if

2.6 Conclusion 47

the annoyance of bias is relatively low for readers. If duopoly newspapersare closer substitutes for readers, then newspapers set less bias; while theeffect on advertisements is ambiguous.

But, as Davies (2008) argues, “to the outsider’s eye, [advertising bias]is very tempting. Advertisers have money, the media outlets need money,so they must be vulnerable to some kind of pressure from the advertisers.It’s a fine theory, but its truth is very limited.” Whether bias will prevailis indeed ultimately an empirical question.

We have tried to answer this empirical question for the Belgian Dutch-language newspapers. We investigated to which extent advertisementsaffect the coverage of companies in these newspapers in the period 2001-2005. Depending on the specification, we find that spending 54000 to212000 euro in a newspaper in one month generates on average one ad-ditional article in this newspaper in the same month. Our results appearrobust to alternative specifications.

Our results are unsurprising and surprising at the same time. On theone hand, newspapers are profit maximizing businesses. And as our theo-retical model predicts, pleasing advertisers might generate higher profits.Therefore, it is not illogical that newspapers insert bias. But on the otherhand, newspapers are no normal business. Correctness of informationgenerates a lot of externalities on society. Many decisions, be it voting,consumption or investment, are based on information retrieved from news-papers. Therefore, newspaper owners and editors should not allow thatadvertisers or the advertisement department attempt to influence editorialcontent.

Policy proposals If corporate and society interests on bias do not align,there might be some solutions to root out advertising bias. A first anddrastic solution is to eliminate all advertising. Without advertising, ad-vertising bias becomes useless and will not be implemented. Though itis doubtful whether a complete ban on advertising is desirable. First, asargued by Baker (1994), less advertising also means less media outlets,which hurts differentiation. Second, the absence of advertising revenuesopens the door for influence by other revenue sources, such as politicalinfluence. Petrova (2009) shows for American newspapers in the 19thcentury that newspapers were more likely to be independent from politi-cal parties if advertising revenues were high. Third, the experience of the

48 Newspapers

women’s magazine Ms. learns that a ban on advertisements24 also indi-rectly resulted in a lower audience. Two of the main reasons were thatMs. had to give up full color and reduce the frequency to a bi-monthlypublication (Cunningham & Haley 2000)25.

Baker (1994) offers two interesting proposals to reduce the reliance onadvertisers. A first is his TA-SR proposal, which stands for ‘Tax Advertis-ing / Subsidize Readers’. The aim is to redistribute advertising revenuesaccording to the audience size. This tilts the trade-off of inserting ad-vertising bias more in the direction of the readers26. Another proposal isto randomize advertising over programs or content, such that advertiserscannot know to which content their advertisement is adjacent. This pro-posal is most relevant for television. This randomization already happensto some extent in newspapers, magazines and internet websites.

Another solution is to install a government body that watches overthe general interest. However, determining whether an article crosses theline between bias and news is not always easy to do. Note that we haveonly tried to estimate bias in general and not in specific articles. It isvery difficult to determine which articles about advertisers are genuinelynewsworthy, and which constitute bias. Moreover, a government mighthave its own incentives to influence news.

A last solution is to make the reader aware of the existence of ad-vertising bias. Many books (e.g. Seldes (1943), Bagdikian (2004), Baker(1994)), citizen activists (e.g. Fairness and Accuracy in Reporting27) andscientific articles (e.g. Reuter & Zitzewitz (2006), Rinallo & Basuroy(2009) and other references in this article) revealed advertising bias. Thisincreased awareness might reduce the effectiveness of bias and render itunprofitable. This article is also a step in this direction.

24For the rationale of this ban, see the editorial by Gloria Steinem in Ms. Magazine,

July/August 1990.25Interestingly, Kaiser & Song (2009) show that German magazine readers do not

dislike ads. Most readers like ads, especially in magazines where ads are informative,

such as TV and women magazines. This runs contra the common finding that news

consumers are adverse.26Some governments have already implicitly implemented such a scheme. The Bel-

gian government subsidizes the reader side by giving a tax exemption for newspapers

(VAT 0%), and applying a very advantageous postal tariff (Cochez 2010). This is no

pure TA-SR scheme as Baker (1994) proposed since this subsidization is not paid by a

tax on advertising.27See www.fair.org

A Monopoly profits 49

Appendix

A Monopoly profits

The monopoly profits corresponding with the three regimes in proposition4.3.5, are:

Π =

α2(z−3−2β2)(6−2β2+z)(4β2−3+z)

216β22

if α1α2

<β1(

√1−β2+β2

2)

(β2−1)β2

(α1(2β2−3)−2α2β1)3

216α1β1(α1β2−α2β1)if

β1(√

1−β2+β22)

(β2−1)β2≤ α1

α2≤ 2β1

1+2β29α116β1

if α1α2

> 2β11+2β2

with z =√

9− 6β2 + 4β22 .

50 Newspapers

B Figures & Tables

Figure B1: Total Advertising Expenditure over Time

Figure B2: Total Coverage over Time

B Figures & Tables 51

Tab

leB

1:C

over

age

expl

aine

dby

logg

edad

vert

isin

gan

ddu

mm

ies

(1)

(2)

(3)

(4)

(5)

(6)

(7)

ln(A

dver

tisi

ng)

0.4

66**

0.3

63**

0.3

28**

0.3

62**

0.3

19*

0.1

77

0.0

392

[0.2

04]

[0.1

44]

[0.1

60]

[0.1

53]

[0.1

81]

[0.2

07]

[0.2

37]

ln(T

ota

lA

ds)

0.1

15**

[0.0

539]

Dum

mie

sA

dv

Adv,N

ewsp

Adv,N

ewsp

Adv

XN

ewsp

Adv

XN

ewsp

Adv

XN

ewsp

Month

Month

New

spX

Month

New

spX

Month

N1978

1978

1978

1978

1978

1978

1693

R2

0.0

10

0.3

53

0.3

73

0.4

10

0.5

02

0.6

10

0.6

35

Sta

ndard

erro

rsin

bra

cket

s

*p

<0.1

0,**

p<

0.0

5,***

p<

0.0

1

52 Newspapers

Table B2: Coverage explained by lagged advertising

(1) (2) (3) (4)

Advertisingt−1 0.00501* 0.00421 0.00249 0.00314[0.00294] [0.00327] [0.00341] [0.00196]

Advertisingt−2 0.00615** 0.00549** 0.00357 0.00666***[0.00254] [0.00248] [0.00279] [0.00202]

Advertisingt−3 0.00250 0.00153 -0.000735 0.00184[0.00208] [0.00198] [0.00181] [0.00214]

Advertisingt−4 0.00812*** 0.00663** 0.00391* 0.00678***[0.00284] [0.00256] [0.00229] [0.00234]


N 22800 22800 22800 22800R2 0.016 0.192 0.402 0.432

Joint significance (F) 15.99 10.26 11.45 11.30Lagged Ads (Prob > F) 0.000 0.000 0.000 0.000

Standard errors in bracketsNumber of articles as dependent variable* p < 0.10, ** p < 0.05, *** p < 0.01

Table B3: Advertising explained by lagged coverage

(1) (2) (3) (4)

Coveraget−1 0.387* 0.344** 0.169 0.353*[0.214] [0.169] [0.112] [0.198]

Coveraget−2 0.261 0.225* 0.116 0.258*[0.163] [0.126] [0.0970] [0.147]

Coveraget−3 0.134 0.137 0.0654 0.165[0.109] [0.100] [0.0792] [0.108]

Coveraget−4 0.205* 0.218** 0.141* 0.216**[0.108] [0.0978] [0.0727] [0.106]


N 22800 22800 22800 22800R2 0.015 0.140 0.434 0.303

Joint significance (F) 1.63 2.15 1.48 1.66Lagged Coverage (Prob > F) 0.1666 0.0738 0.2083 0.1585

Standard errors in bracketsAdvertising expenditure as dependent variable* p < 0.10, ** p < 0.05, *** p < 0.01

C Overview Surveys on Influence Advertisers 53

CO

verv

iew

Surv

eys

on

Influence

Advert

isers

Table

C1:

Overv

iew

Surv

eys

on

Influence

Advert

isers

Proposit

ion

Pro

Neutral

Contra

Research

Year

Country

N

Edit

ors

Allow

advert

iser’

sin

tere

sts

toin

fluence

na

na

75%

ASN

E(1

999)

1997

US

50

the

new

s

Inyour

opin

ion,to

what

exte

nt

do

22%

0%

78%

Pew

(2000)

1999

US

162

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

Have

there

been

inst

ances

inw

hic

h25%

6%

71%

Pew

(2004)

2004

US

92

your

new

sroom

was

encoura

ged

todo

a

story

because

itre

late

dto

an

ow

ner,

advert

iser,

or

sponso

r?

Inyour

opin

ion,to

what

exte

nt

do

15%

1%

84%

Pew

(2008)

2007

US

156

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

Our

new

spaper

seld

om

runs

stori

es

15%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

whic

hour

advert

isers

would

find

cri

tical

or

harm

ful.

Has

any

advert

iser

succeeded

in37%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

influencin

gth

enew

sor

featu

res

inyour

new

spaper?

Has

there

been

pre

ssure

from

wit

hin

55%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

your

paper

tow

rite

or

tailor

new

s

stori

es

tople

ase

advert

isers

?

Has

any

advert

iser

trie

dto

kill

71%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

ast

ory

at

your

new

spaper?

Has

any

advert

iser

thre

ate

ned

to83%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

wit

hdra

wadvert

isng

from

your

paper

54 Newspapers

because

ofth

econte

nt

ofth

est

ori

es?

Are

you

aw

are

ofan

att

em

pt

by

any

88%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

advert

iser

toin

fluence

what

new

sor

featu

res

apeare

din

your

new

spaper?

Has

any

advert

iser

ever

wit

hdra

wn

89%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

advert

isin

gin

resp

onse

toth

econte

nt

ofyour

new

spaper?

Has

any

advert

iser

trie

dto

influence

90%

na

na

Sole

y&

Cra

ig(1

992)

1991

US

147

the

conte

nt

ofa

new

sor

featu

rest

ory

?

Edit

ors

&Journalists

Have

you

ever

been

pre

ssure

dby

a7%

na

na

William

s(1

992)

1990

US

41

senio

redit

or

or

their

publish

er

todela

y

or

killa

story

are

alest

ate

advert

iser

mig

ht

find

offensi

ve?

Do

you

know

ofone

case

where

32%

na

na

William

s(1

992)

1990

US

41

advert

isers

actu

ally

pulled

ads

because

ofa

story

?

Our

realest

ate

secti

ons

do

not

inclu

de

44%

na

na

William

s(1

992)

1990

US

41

covera

ge

that

would

offend

are

alest

ate

advert

iser

as

am

att

er

ofpolicy.

Did

realest

ate

advert

isers

ever

thre

ate

n83%

na

na

William

s(1

992)

1990

US

41

topull

ads

from

the

papers

because

of

the

covera

ge?

Journalists

Allow

advert

iser’

sin

tere

sts

toin

fluence

10%

25%

65%

ASN

E(1

999)

1997

US

1714

the

new

s

New

sroom

shave

been

pre

ssure

dto

do

55%

11%

33%

Hollin

gs

et

al(2

007)

2007

New

Zeala

nd

391

ast

ory

because

itre

late

dto

an

advert

iser,

ow

ner

or

sponso

r.

Inyour

opin

ion,to

what

exte

nt

do

28%

2%

70%

Pew

(2000)

1999

US

228

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?


(Tra

dit

ionalM

edia

)

Inyour

opin

ion,to

what

exte

nt

do

36%

0%

64%

Pew

(2000)

1999

US

124

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

(Inte

rnet)

Have

there

been

inst

ances

inw

hic

hyour

26%

7%

67%

Pew

(2004)

2004

US

96

new

sroom

was

encoura

ged

todo

ast

ory

because

itre

late

dto

an

ow

ner,

advert

iser,

or

sponso

r?(T

radit

ional

Media

)

Have

there

been

inst

ances

inw

hic

hyour

35%

8%

57%

Pew

(2004)

2004

US

45

new

sroom

was

encoura

ged

todo

ast

ory

because

itre

late

dto

an

ow

ner,

advert

iser,

or

sponso

r?(I

nte

rnet)

Inyour

opin

ion,to

what

exte

nt

do

34%

2%

65%

Pew

(2008)

2007

US

232

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

Obje

cti

vity

ofjo

urn

alism

isunder

85%

6%

10%

Ppre

ss(2

010)

2009

Belg

ium

140

incre

asi

ng

influence

from

advert

isers

,

mark

ete

ers

and

ow

ners

Have

you

felt

pre

ssure

dto

report

a7%

na

93%

Pri

ce

(2003)

1999

US

131

story

because

ofadvert

isers

Have

you

felt

pre

ssure

dnot

tore

port

a7%

na

93%

Pri

ce

(2003)

1999

US

131

story

because

ofadvert

isers

Nati

onalm

edia

publish

mate

rials

about

38%

49%

13%

Tse

tsura

(2005)

2005

Pola

nd

99

acom

pany

or

apro

duct

inexch

ange

for

apaid

advert

isem

ent

appeari

ng

els

ew

here

inth

esa

me

mediu

m.

Marketin

g&

Sale

s(in

tern)

Itis

accepta

ble

that

anegati

ve

story

on

3%

6%

91%

An

&B

erg

en

(2007)

2010

US

211

an

advert

iser

issu

ppre

ssed

on

the

request

ofth

isadvert

iser.

Itis

accepta

ble

that

afe

atu

rest

ory

on

14%

24%

62%

An

&B

erg

en

(2007)

2009

US

211

are

staura

nt

isw

ritt

en

by

the

56 Newspapers

rest

aura

nt,

whic

his

als

oan

advert

iser

inth

at

secti

on.

Itis

accepta

ble

that

the

advert

isin

g23%

21%

56%

An

&B

erg

en

(2007)

2008

US

210

depart

ment

encoura

ges

aphoto

edit

or

touse

photo

wit

hth

elo

go

ofa

shir

t

sponso

r(a

nd

advert

iser)

pro

min

entl

y

dis

pla

yed.

Itis

accepta

ble

that

asa

lesp

ers

on

ask

s33%

12%

55%

An

&B

erg

en

(2007)

2007

US

210

tofa

vor

advert

isers

as

new

sso

urc

eand

avoid

non-a

dvert

isers

.

Marketin

g&

Sale

s(extern)

Nati

onalm

edia

publish

mate

rials

about

58%

33%

9%

Tse

tsura

(2005)

2005

Pola

nd

98

acom

pany

or

apro

duct

inexch

ange

for

apaid

advert

isem

ent

appeari

ng

els

ew

here

inth

esa

me

mediu

m.

(PR

)

Nati

onalm

edia

publish

mate

rials

about

59%

32%

9%

Tse

tsura

(2005)

2005

Pola

nd

90

acom

pany

or

apro

duct

inexch

ange

for

apaid

advert

isem

ent

appeari

ng

els

ew

here

inth

esa

me

mediu

m.

(Mark

eti

ng)

Readers

Allow

advert

iser’

sin

tere

sts

toin

fluence

50%

32%

18%

ASN

E(1

999)

1997

US

3000

the

new

s

Ow

ners

&M

anagers

Inyour

opin

ion,to

what

exte

nt

do

14%

1%

85%

Pew

(2000)

1999

US

101

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

Have

there

been

inst

ances

inw

hic

h16%

7%

77%

Pew

(2004)

2004

US

59

your

new

sroom

was

encoura

ged

todo

a

story

because

itre

late

dto

an

ow

ner,

advert

iser,

or

sponso

r?


Inyour

opin

ion,to

what

exte

nt

do

10%

0%

90%

Pew

(2008)

2007

US

79

advert

isin

gconcern

sin

fluence

new

s

org

aniz

ati

on’s

decis

ions

about

whic

h

stori

es

tocover

or

em

phasi

ze?

Inth

ela

sttw

oyears

,has

your

new

s36%

10%

53%

Pew

(2010)

2010

US

353

org

aniz

ati

on

trie

dor

dis

cuss

ed

anew

revenue

experi

ment

that

has

rais

ed

concern

sabout

edit

ori

alin

depence

or

eth

ics?

58 Newspapers

D Advertising Bias Channels

In many theoretical models, the implementation of bias is fairly simple:the business side of the newspaper decides it and therefore it is imple-mented. In reality though, explicit money for bias schemes almost neveroccur, or at least not open, because this would reduce the reputation to-wards news consumers. Moreover, the business side does not make thenews items themselves, therefore it is worthwhile to stress the role of thejournalist. In this section, we expand the advertising bias channels thatare discussed in section 2.5. We try to illustrate the channels with some ex-amples, though advertising bias typically suffers a tip-of-the-iceberg prob-lem, where only the most visible cases come to the surface. As Fleetwood(1999) puts it, “for every printed story that draws advertiser ire, thereare thousands that never see the light of the day”. Many cases, where themedia outlet inserts self-censorship, are never known, perhaps not evenby the journalists who implement it.

D1 Supply and demand channels

In this section, we discuss how the advertising bundle might look like inpractice. Basically, this bundle of advertising and favorable coverage isan (implicit) agreement between the business side of the media outlet andthe advertiser. Both parties can initiate this bundle.

Supply channel The purest form of bias is where the media outlet sellsnews space to the advertiser. This practice was commonly accepted forUS newspapers during the late 19th and beginning 20th century (Baker(1994), Baldasty (1992)). Companies wrote or commanded news articlesor editorials, and these were published without any disclaimer. For read-ers, it was impossible to distinguish these so-called reader notices fromother news, apart from its overoptimistic content. The practice steadilydeclined with the Post Office Appropriation Act of 1912, prohibiting pay-ment for news. Though milder forms of these reader notices still appear inthe media. For example, in some news segments, Seattle television stationKiro TV featured only experts who had paid a fee of 1000 dollars to thestation28. Rush Limbaugh, a conservative American radio talk show host,allows companies to buy words that he then weaves into his monologues29.

28Seatle Times, 07-06-2002, documented in Jackson, Hart & Coen (2003)29New York Times, 07-06-2008, documented in Hart (2009)

D Advertising Bias Channels 59

Similarly, AOL had an agreement with Infinity’s Radio to be mentionedat least six times a day in the programming30.

While these recent examples illustrate that selling news space directlystill occurs, we focus on the bundle of selling an advertisement and favor-able coverage. In a few cases, the news outlet makes this bundle explicit.Press-Enterprise announced in 2001 “advertise your restaurant and get afree feature story”31. Richard Shortway, publisher of Vogue, said “he cold,hard facts of magazine publishingmean that those who advertise get editorial coverage”32. This schemewas also applied in the Chicago Sun-Times, where “theaters advertisingget preferential treatment on the news pages”33. When other pill mak-ers complained about the favorable treatise of Doctor Brandreth cure-allpills, James Bennett, editor of the New York Herald wrote in 1835 in hisnewspaper: “Send us more advertisement than dr. Brandreth does [...][and] we’ll cut dr. Brandreth to dead”34.

More often, and perhaps more importantly, advertising bias is an im-plicit bundle where the media outlet attracts or thanks advertisers byinserting favorable coverage, or by avoiding negative news. This self-censorship by the media is proven by a quote of Helen Gurley Brown,editor of Cosmopolitan, who empathizes with advertisers: “Who needssomebody you’re paying thousands of dollars a year to come back andbite you on the ankle?”35. And further, on a question why she didn’tcriticize tobacco products: “We just don’t say rotten things about ouradvertisers. There is very little to warn our readers about, because adver-tisers’ products are so good”36. Seldes (1943), for example, investigatednewspapers in 1940 and found that many newspapers did not report theFederal Trade Commission’s reports on mistakes in advertisements, if theyalso ran these ads. According to Nan Robertson, a New York Timesjournalist and author, it was quite normal in the 1950’s that articles onwoman’s and fashion pages “reflected the relative advertising strength”(Fleetwood 1999). Boston Herald ordered its consumer columnist to stopwriting about a local bank merger. The reporter attributed this decisionto the fact that one of the banks was also a big advertiser and lender

30Metro Times, 11-06-2003, documented in Hart & Hollar (2004)31Columbia Journalism Review, 09/10-2001, documented in Jackson & Hart (2002)32Advertising Age, 17-04-1972, quoted in Bagdikian (2004)33New York Times, 03-06-2002, documented in Jackson, Hart & Coen (2003)34McClung Lee (1937), documented in Bagdikian (2004)35Quoted in Bogart (2000).36Quoted in Bogart (2000).

60 Newspapers

to the newspaper37. In the earliest edition of The Wall Street JournalOctober 3 edition in 2002, the newspaper put an advertisement of BearStearns next to an article about an error in a stock market order. Insteadof moving the advertisement, the paper decided to eliminate the news ar-ticle on Bear Stearns and place it as a subsection of another story38. Ina story about increased sales after the murder on Versace in 1997, theChicago Sun Times mentioned only department store advertisers. “Wehave to take care of our customers”, said editor Larry Green39. Moresubtle is the creation of additional reporting to attract advertisers. CBSadmitted to reduce the number of prime-time war specials to please adver-tisers40; US News & World created in 2003 a war-free zone, called SecondFront41. Similarly, some respondents of our survey mentioned that thereis disproportionate attention to soft news that sells advertisements, suchas a ‘winter special’ for the travel section and ‘interior and design’ forthe lifestyle section. Ironically, more advertiser-friendly soft news sectionsmight indicate that the hard news section is perhaps less affected by adver-tiser influence. Abraham Rosenthal, New York Times executive director(1977-1988) said on soft news: “If they keep my A section pure, then Iwill let them have all those other sections.” (Fleetwood 1999)42.

Demand channel Though the media outlet is the central player in ad-vertising bias, advertisers play their role too. In the most basic form, someadvertisers at least expect that such a bundle exists. The head of the Gro-cery Manufacturers Association, Paul Willis, wrote in 1962 to televisionstations that “their editorial department and business department mightbetter understand their interdependent relationships”, and later claimedthe publication of favorable articles in print magazines as a success ofhis action43. One respondent of our survey mentioned that organizers ofcultural events often expect an article if they buy an advertisement.

Apart from creating expectations, bigger advertisers apply a more ef-

37Washington Post, 01-05-2000, documented in Jackson & Hart (2001).38thestreet.com, 11-10-2002, documented in Jackson, Hart & Coen (2003)39Documented in Fleetwood (1999)40See Baker (1994).41MediaWeek, 24-02-2002, documented in Hart & Hollar (2004)42In the same spirit, Claussen (2002) relates the story of an older man asking a

journalist “What is the first responsibility of the press?” The young journalist answers

enthousiastically: “To print the truth”. He is corrected by the older man. “No, the

first responsibility is to make profit so that it can afford to print the truth.43Advertising Age, 17-04-1972, documented in Bagdikian (2004)


fective weapon to obtain favorable coverage: they withdraw their adver-tisements if coverage is unfavorable or not favorable enough. While thiswithdrawal does not undo the negative coverage, it might serve as a rep-utation device to impede negative coverage in the future. Baker (1994)and Bagdikian (2004) give many examples of withdrawals. Piano adver-tisers withdrew from Esquire in 1940, after it declared that the guitar isa better accompaniment to singing than the piano. Pharmaceutical com-panies threatened to withdraw all advertisements from Modern Medicine,after parent company New York Times ran a series of articles on medicalmalpractices in 1976. WDHD lost 4.5 million of advertising from the au-tomobile industry in 1991; and 1 million from Proctor & Gamble in 1990after it ran an article that criticized Proctor & Gamble. Cosmetics firmRevlon withdraw all its advertisements from Ms. because women on thecover were not wearing make-up, 60 Minutes lost advertising money be-cause it alarmed for health issues on apples. But often the threat of with-drawing already suffices to change the content. Time magazine was saidto retreat Osama Bin Laden as person of the year in 2001 after pressurefrom Walmart44. Georgia Governor Sonny Perdue threatened to withhold500000 dollar; as a result an incident during the campaign was not airedon the CBS television station WGCL (Atlanta)45. Nordstrom pulled itsadvertising from the Seattle Times and the Post-Intelligencer after theypublished stories on unfair labor practices (Bogart 2000).

While the cases above might seem incidental, some bigger companieshave an explicit withdrawal rule. They automatically withdraw adver-tisements if certain content is published. During the Federal Commu-nications Commission’s hearing in the 1960s, many of these rules wererevealed. Proctor & Gamble refused to advertise in issues depicting busi-ness as cold or ruthless (besides other stipulations); Brown & Williamsontobacco directed that tobacco products should not be used in a derogatoryor harmful way46. General Motors announced in 2004 that they refused toadvertise on TV programs about the atrocities in Iraq47. BP and MorganStanley issued directives demanding their ads be pulled from any editionthat included “objectionable content”48. Google Adsense installed sensi-

44New York Post, 2002, documented in Jackson, Hart & Coen (2003)45Creative Loafing, 13-12-2006, documented in Jackson (2007)46See Bagdikian (2004) for a discussion of these withdrawal rules and for additional

examples.47USA Today, 18-05-2004, documented in Hart & Hollar (2005)48AdAge.com, 24-05-2005, documented in Hollar, Jackson & Goldstein (2006)

62 Newspapers

tivity filters in 2003, to avoid that advertisements run next to damagingcontent. Many websites that ran stories on war (e.g. Iraq), sex (e.g. ex-plicit content, rape) or business bashing (e.g. boycotting Sony) lost alltheir Google ads49. Note that, even with these withdrawal rules, mediacompanies can still write about sensitive topics, though they are punishedby obtaining lower advertisement revenues. This business concern mightinfluence the news selecting process.

To sum up, we argued that advertisement bias might be initiated bymedia outlets, advertisers or it might be a mutual agreement. Media out-lets might offer an explicit bundle of advertisement and favorable coverage,but in many cases the offer is implicit, by self-censoring the news to avoidconflicts. Advertisers can threaten to withdraw advertisement money inorder to obtain favorable coverage, or they can constitute a withdrawalrule.

D2 Inside channels

The examples and mechanisms above only considered the business side ofthe newspaper, and the advertisers. This overlooked the role of the jour-nalist. Journalists perceive themselves as independent from the businessside. The ethics of the journalists determine that journalists should careabout the public interest, by describing the news as truthfully as possibleand by collecting information independently50. The New York Times codeexplicitly stresses the wall of separation between the sales department andthe editorial staff51.

Our company and our local units treat advertisers as fairlyand openly as they treat our audiences and news sources. Therelationship between the company and advertisers rests on theunderstanding that news and advertising are separate – thatthose who deal with either one have distinct obligations andinterests, and each group respects the other’s professional re-sponsibilities.

49East Bay Express, 02-08-2006, documented in Jackson (2007)50 The Belgian Code of the Council of Journalism (Raad voor de Journalistiek

2010). For an extensive list of Codes of Ethics for journalists, see

http://www.rjionline.org/mas/codes-of-ethics.php.51The New York Times Company Policy on Ethics in Journalism, article 80

(http://www.nytco.com/press/ethics.html)


Inserting advertising bias does not fulfill these codes and therefore journal-ists might be reluctant to do so. Some respondents of our survey testifiedthat they were asked to insert bias, but refused. We do not question theethics of journalists here. On the contrary, the paragraph below describeshow advertising bias can be implemented, despite the existence of ethicaljournalists. Pressure doesn’t make journalists happier, write Hamilton &Krimsky (1996): “Newsroom moral plummets when a publisher strollsinto the editor’s office and asks for special handling of a story relating toan advertiser.” Direct pressure might even work contra-productive, theyargue. “Reporters balk at puffing advertisers; and when they uncoversomething negative about one of them, they expect that the bad newswill be published”. Therefore, media outlets use more indirect forms ofpressure.

In general, we observe two ways in which media outlets can circumventthe independence of journalists. The first is to tear down the wall betweenthe business side and the editorial staff. The second way is to changethe staff composition such that independence from the business goals islimited.

Crossing the wall When the wall between the sales and editorial staffis torn down, then it is easier to cross the line between both, in both direc-tions. This becomes very clear when the sales department is also involvedin the activities of the editorial staff. In many newspapers, ‘lighter’ news,such as real estate, travel and food is written by the ad department. Aslabeled by the editor of the Houston Chronicle, this is “nothing controver-sial” 52. In 2003, the marketing department of the Pioneer Press wrote anarticle about a restaurant to compensate for another negative article onthe same restaurant53. To extend this habit further, the Fairbanks DailyNews-Miner hired an advertorial writer54.

A more subtle form of the journalistic ambitions of the sales or market-ing department are suggestions to the journalists. In fact, such suggestionsare not per se wrong, as long as the journalists have the freedom to act ornot. Though, as Bagdikian (2004) writes, these suggestions were knownas business office musts – it was hard not to follow these suggestions, asrecently as the 1980s. Some editors also provide a list of the biggest ad-

52Editor & Publisher, 31-03-1979, documented in Bagdikian (2004)53Chicago Reader, 05-09-2003, documented in Hart & Hollar (2004)54Columbia Journalism Review, 9/10-2005, documented in Hollar, Jackson & Gold-

stein (2006)

64 Newspapers

vertisers. One of our respondents mentions that the directors named thethree biggest advertisers, directly followed by the statement that it wasnot the goal to adjust articles toward these advertisers. The statementwas just to inform the staff. After some negative coverage, CNN informedits staff that Nasdaq is a major advertiser55. A step further, televisionstation KTVO urged its journalists always to go to station advertisersfirst for expert opinion and industry comment56.

The business side can also facilitate direct contact between journalistsand advertisers. Again, this is not wrong per se, but it might be consid-ered as a further attempt to break down the wall between business andjournalism. In 2009, Washington Post organized off-the-record salons withadvertisers and journalists57; Des Moines Register asked advertisers to seewhat they can do together for the coverage of high profile events58. Or,as summarized by Mackenzie Warren of News-Press: “Keeping reportersaway from the business side of the paper is old-school snobbery”59.

But it is not always the business side that visits the journalistic side.Journalists also turn to the business side, or the business side appliesjournalistic practices in its advertisements. By doing so, the journalis-tic reputation is used to increase the effectiveness of the advertisement.Though this is not directly advertising bias (since there is no bundle andthere is no biased coverage), the mechanism exploited here is the same. Inboth cases, the outlet uses the fact that reader can not fully distinguishnews from bias; or here: news from advertisements with journalists. Eitherthe ad looks like a news item (NBC ad in the LA Times60, insurer ad ontelevision channel Griffin Communications61), or journalists are featuredin ads or purchased items (KVVU, KLAS and KTNV used reporters toconduct interviews for purchased segments62). Some journalists are paidby companies, but they fail to disclose their interdependence. Similarly,some media rely heavily on business-paid experts. The technology editorof Child magazine, for example, was paid by Kodak to promote products,which he did on a local television channel and NBC’s Today show63. Tele-

55Daily News, 06-09-2001, documented in Jackson & Hart (2002)56Columbia Journalism Review, 9/10-2004, documented in Hart & Hollar (2005)57Politico, 03-07-2009, documented in Hart (2010)58Forbes, 17-04-2008, documented in Hart (2009)59Washington Post, 04-12-2006, documented in Jackson (2007)60New York Times, 10-04-2009, documented in Hart (2010)61Tulsa World, 05-04-2009, documented in Hart (2010)62Las Vegas Review-Journal, 09-10-2009, documented in Hart (2010)63Wall Street Journal, 19-04-2005, documented in Hollar, Jackson & Goldstein


com analyst Jeff Kagan was cited in various media, such as Kansas CityStar, seemingly as an independent expert, but in reality paid by severaltelecom companies64.

Editorial staff composition A second mechanism to reduce the in-dependence of the staff is to change the composition of this staff. Themost visible composition change is firing unwilling journalists and replac-ing them by easier journalists – or less ethical ones. These dismissalsmight serve as a signal to other journalists. Similarly, but less visible, is amedia organisation’s decision not to advance unwilling journalists. Thereare many examples of journalists fired due to pressure of advertisers. In2009, Summit Daily News fired a reporter due to a conflict on ski re-sorts65; Hartford Courant did the same with a columnist after a negativeconsumer column on mattress company Sleepy66. In 2006, Evening Sunfired its long-serving columnist after calling Walmart products ‘Made inChina thingy-ma-bob’67. CityBusiness (New Orleans) fired its editor aftershe objected to the introduction of an advertiser-sponsored news page68.

A more subtle mechanism is dedicating more journalists to the fieldsthat are interesting to advertisers. As such, bias slips in without theconsent of the journalists involved. By hiring relatively more freelancejournalists, news outlets are also able to shift power away from journalists.Although freelancers are in principle more free to write about what theylike, the market logic predicts that they are going to write articles thatare bought by outlets.

(2006).64Pitch,28-07-2008, documented in Hart (2009).65CBS4Denver.com, 09-12-2009, documented in Hart (2010)66New York Times, 18-08-2009, documented in Hart (2010)67New York Times, 04-12-2006, documented in Jackson (2007)68Columbia Journalism Review, 09/10-02, documented in Jackson, Hart & Coen

(2003)

66 Newspapers

Chapter 3

The Universal Service

Obligation for Telephone

Directories: Regulating

the Redundant

This chapter is joint work with Patrick Van Cayseele and is adapted fromDe Smet & Van Cayseele (2010) and De Smet & Van Cayseele (2011).

68 Directories

3.1 Introduction

After the liberalization of the telecommunications sector in the 1990’s, themember states of the European Union decided to implement a universalservice obligation1, to ensure that in every country the necessary servicesare available at “determined quality and an affordable price, even if themarket would not provide it”2. In many other countries, a similar frame-work was implemented3. One of these obligations concerns the publishingof telephone directories. Article 5, paragraph 1(a), of the European Uni-versal Service Directive specifies that “member states shall ensure that atleast one comprehensive directory is available to end-users”. Article 25adds that “subscribers to publicly available telephone services have theright to have an entry in the publicly available directory”.

We can disentangle this obligation in two parts. First, the telephonedirectory should be comprehensive, i.e. it should list all the persons andcompanies that have a phone4. Second, the directory should be availableto all users.

In this chapter, we develop a simple model to investigate whether theinclusion of both conditions is sensible. Is the imposition of both con-ditions necessary, or is it sufficient to impose one condition, to obtainalso the other? More specific, we check first whether imposing completeavailability implies that the directory will be comprehensive. Second, weinvestigate whether the reverse holds, i.e. whether imposing a comprehen-sive directory results in a free directory. A priori, it is not clear why onecondition would imply the other. By providing a comprehensive list, thepublisher gives away a valuable listing for which some persons or compa-nies may want to pay. Similarly, certain users perceive the directory as avaluable search instrument for which they may be willing to pay. There-fore, it might be reasonable to charge not only (some of) the listed personsor companies, but also the users.

1Universal Service Directive, Directive 2002/22/EC2See http://ec.europa.eu/information_society/policy/ecomm/toda

ys_framework/universal_service/index_en.htm, emphasis added3See for example the Telecommunications Act of 1996 for the United

States (P.L. No. 104-104, 110 Stat. 56 (1996)) or Telecommunica-

tions (Consumer Protection and Service Standards) Act 1999 for Australia

(http://www.austlii.edu.au/au/legis/cth/consol_act/tpassa1999620/).4Note that telephone directories will not be comprehensive in practice, since every-

one has the right to opt-out. Furthermore, mobile numbers are, as a rule, not included

in the directory, though subscribers have the right to opt in.

3.1 Introduction 69

Our model is particularly relevant in the policy discussion about theuniversal service obligations. Since the introduction of a universal serviceframework for telecommunication, postal services and electricity in the1990s, the literature on universal service obligations has boomed. Schol-ars have been investigating the rationale of these obligations (Cremer,Gasmi, Grimaud & Laffont (1998), Crew & Kleindorfer (2002), Graham,Cornford & Marvin (1996)), the costs and funding (Garbacz & Thompson(2005), Jaag, Koller & Trinkner (2008), Mirabel & Poudou (2004), Ro-driguez & Storer (2000), Weller (1999)) and the effects on industrial or-ganization topics like competition, entry and pricing (Armstrong (2008),De Donder (2006), Riordan (2001), Valletti, Hoernig & Barros (2002)).Due to technological changes, such as the Internet and mobile phones, alot of research is devoted to the question whether the universal serviceshould be adapted, extended or even abolished (Cremer (2000), Downes& Greenstein (2007), Xavier (2003))5.

Theoretical contribution This article expands the understanding ofthe strategic decisions of telephone directories publishers in the context oftwo-sided markets. Inspired by this literature, started by Rochet & Tirole(2003), Parker & Van Alstyne (2005) and Armstrong (2006), the directoryis modeled as a platform that connects two distinct sides of the market:receivers, also labeled as users, buyers or readers, and senders, also referredto as sellers or advertisers. While the roles of receivers and senders canoverlap in reality, i.e. sellers can also be buyers and vice versa, we keepboth sides of the market strictly separated. This clarifies the analysisin section 4.3, and makes the interpretation of telephone directories as atwo-sided markets easier.

Note that the framework of two-sided markets is important for the

5At the end of this chapter, we come back to the policy debate on telephone di-

rectories. Note that this policy discussion is focused on White Pages. There are three

generally accepted types of telephone directories: White Pages, Yellow Pages and Grey

Pages. The first type contains an alphabetical list of persons, with address and tele-

phone number. The directory is divided in regions. Yellow Pages is synonymous with

a business directory. It classifies firms by their business type or goods or services pro-

vided. Grey Pages are less known. These are so-called reverse telephone directories

where one can browse the numbers and find the associated customer details. The latter

was mostly used by emergency services, phone companies, law enforcement, and public

libraries. Our model applies to both White Pages and Yellow Pages. In what follows,

we use the more general term telephone directory which can both refer to White and

Yellow Pages.

70 Directories

research question dealt with in this article. The universal service obliga-tion for telephone directories simultaneously imposes a restriction on bothsides of the market. The platform has to provide access to the receiverside (availability), and similarly, it has to provide access to the sender side(comprehensiveness). The focus of this article is on whether it is sufficientto have a restriction on only one side to obtain a market outcome thatsatisfies the restrictions on both sides.

Although dealing with a precise policy problem, we add to the theoryof two-sided markets by including product differentiation on the platform.The sender side types can self-select by choosing their listing format; theother side of the market can see the different formats on one single plat-form. In the seminal contributions, platforms can be differentiated (e.g.through the Hotelling model), but this differentiation only takes place be-tween platforms. If the other side wants to see differentiated content, ithas to connect with multiple platforms (i.e. multihoming).

As a result of this differentiation, platforms have the possibility tocharge different prices. This is quite unusual for the models used in thetwo-sided markets literature. As Weyl (2010) summarizes the literature,he emphasizes that “platforms are price setters on both sides and typi-cally set uniform prices” as one of the key features of the current models.Weyl (2010) also adds that the possibility for platforms to offer multipleproducts is “an important question for future research”.

We try to fill this gap in the literature, and by doing so we complementthe small subsegment of two-sided markets articles that allow for multipleproducts on a platform. In Damiano & Li (2007), a monopoly matchmakerdifferentiates both sides, men and women, in many types by setting aschedule of prices. The matchmaker uses price discrimination to allowmen and women to select their optimal type, which generates an efficientmatching process. The argumentation is that differentiation is better thanthe uniform pricing used in online dating markets. A price that is used asa signal mitigates the misrepresentation in markets with uniform pricing.Our model is similar to their model to some extent since we also modelthe platform as a monopolist6 and we also allow the differentiated side

6Interestingly, in their accompanying paper Damiano & Li (2008), they compare

the results of monopoly with competition. They find that while monopolistic match

makers can use prices to sort high types from low types, their duopolistic counterparts

are involved too much in price competition and therefore they are less efficient. In our

article, we choose for a monopoly model, for computational simplicity, but also because

this resembles the reality of telephone directory publishing in most European countries.

3.1 Introduction 71

to self-select. Though, we do not model the potential inference of qualityby a price mechanism, i.e. we do not allow to use the listing formats asa screening device for users and hence do not address potential qualitydistortion as in Mussa & Rosen (1978)7.

Viecens (2006) presents a duopoly model inspired by shopping malls.Similar to our model, for the buyer side, not only the number but alsothe type of shops present in the mall matters. The relative importanceof quality vs. quantity determines whether the platforms will both engagein attracting low or high quality shops. This relates to our approachwhere we distinguish between the type and the volume of the sender side.Contrary to our model, shop keepers in Viecens (2006) cannot self-selecttheir presence form in the mall.

Other papers that implement non-uniform pricing are Gomes (2010)and Hagiu & Lee (2008). In Gomes (2010), the platform auctions itsaudience to advertisers. A platform charging different prices related tothe action space of the demand side is introduced by Hagiu & Lee (2008).

Final remark. The universal service obligation for telephone directoriesis at the heart of this article. This might mislead the reader to concludethat our contribution is old-fashioned and therefore irrelevant for two rea-sons. First, the universal service obligation is subject to debate, as wediscuss in section 4.4 Second, it is beyond doubt that the paper telephonedirectories will be replaced by the internet. Though, we have at leastthree reason to firmly believe that these two trends do not devaluate therelevancy of this chapter.

First, there is no qualitative difference between paper and online di-rectories. While paper directories are likely to vanish, online directoriesbelong to most visited websites worldwide. Second, with minor modifi-cations, our model also applies to online markets such as search engines.Similar to directories, search engines apply differentiation on the platform,by offering different formats: regular links which are free and sponsoredlinks which are paid. The ranking of these results also introduces differ-entiation. Therefore, our results might be interpreted also in the lightof search engines. Third, though search results are not regulated, it isnot unthinkable that some regulation will emerge in the future (New YorkTimes (2010) and Mayer (2010)). The European Commission and the Fed-

7An empirical investigation of price discrimination in the Yellow Pages industry is

found in Busse & Rysman (2005).

72 Directories

eral Trade Commission (US) announced to scrutinize Google’s search re-sults for cross-promotion, i.e. favoring its own related services over rival’sin both regular and paid results (European Commission (2010), Singhal(2011)). Even if the universal service obligation for telephone directoriesis abolished, our results remain relevant because they provide guidancetoo for potential regulation on search engines.

The remainder of the chapter is organized as follows. In section 4.3, weintroduce the model that subsequently is used to investigate the relevantpolicy issues: the consequences of imposing the two main restrictions of theuniversal service obligation for directories. In section 3.3, we analyze foreach restriction the equilibrium outcome, conduct a comparative staticsanalysis and question whether it is necessary to supplement the restrictionfocussed on with the other restriction. Section 4.4 concludes and adds thepolicy debate on universal service. In the appendix of this chapter, wesuggest an empirical method to calibrate the relevant parameters, based onthe empirical work of Rysman (2004). We also work out some extensionsof the model and suggest how our model can be applied to other industries,such as dating events, shopping streets, online search and online media.

3.2 Set-Up of the Theoretical Model

In this section, we set up a theoretical model that is tailored to the tele-phone directories industry. The goal of this model is twofold. First, thismodel is one of the first attempts to model the important dynamics of thedirectories business. Second, we try to shed light on the two main featuresof the universal service obligation for directories: comprehensiveness andavailability.

The telephone directory industry is characterized by three players: re-ceivers, senders and publishers. Receivers and senders can be both naturalpersons or companies. The reasons why receivers and senders want to con-nect with each other can be various. To clarify the analysis, we assumethat the phone is used only for business reasons. Senders are retailers orcompanies that advertise in the directory to connect with users in orderto sell their products or services. Receivers are consumers that use the di-rectory to connect with senders in order to buy their products or services.This narrowing of the scope of telephone use allows to clearly distinct bothsides of the market. An alternative interpretation is that the decisions ofreading the directory as a receiver on the one hand and being listed in the

3.2 Set-Up of the Theoretical Model 73

directory as a sender on the other hand are made independently.Our analysis is focused on the platform. The platform is the publisher

who connects receivers and senders. We assume that the platform is aquantity setter8 on the sender side, and a price setter on the receiver side.We further assume that there are two types of listings in the directory:small listings (indexed 1), and large listings (indexed 2). Small listingscontain only limited information; large listings contain more information,but it is not guaranteed that receivers value this additional information.The platform sets the optimal quantities q1 and q2 and determines theoptimal price for receivers pR. We investigate a monopolistic market sit-uation9.

Receivers Receivers value both small and large listings. The valuationfor small listings is given by r1, the valuation for large listings by r2.Small and large listings are valued differently but it is not necessarily thecase that readers appreciate large listings more than small listings. Largelistings might contain more information, such as a website, e-mail addressor map. Or it might contain pictures or just being printed in a largerfont size. Some receivers might perceive this as a nuisance, hampering thebrowsability of the telephone directory. Utility is negatively affected bythe price pR and the opportunity cost k. This opportunity cost can berelated to the availability of outside options, such as going to a companyone already knows, asking a friend for advice, or driving to a city andsearching randomly for retailers. The net utility of using the directory isequal to:

UR = max(r1q1 + r2q2 − pR − k, 0) (3.1)

8While publishers have a tradition of announcing list prices each year, the practice

learns that they adjust these prices with discounts to accommodate the number of

senders, i.e. list prices can be seen as maximum prices but can differ substantially from

the real prices.9In Europe, 70% of the countries has only one player in printed directories. Some

countries have two publishers. These countries include the bigger countries (UK,

Spain), Scandinavian countries (Sweden, Denmark, Finland) and alpine countries (Aus-

tria, Switzerland). For most countries, the market structure is not monopolistic if one

considers the market for internet directories. We can expect further consolidation as

was recently the case in the Netherlands (merger approved in 2008, NMa Decision

6246, case European Directories – Truvo Nederland). (Data collected from the EADP

website (EADP.org). EADP is the European Association of Directory and Database

Publishers and coordinates most European directory publishers. We counted for each

country the number of publishers in the category Telecommunication Directories that

publish print directories.)

74 Directories

Receivers are homogenous in the valuations r1 and r2 but are heteroge-neous in their opportunity cost. We assume there is a mass one of potentialreceivers and the opportunity cost k is uniformly distributed between 0and 1, so the number of receivers qR is given as:

qR = r1q1 + r2q2 − pR (3.2)

Since qR is the number of potential receivers that reads the directory, wealso refer to it as the readership. It is clear from expression 3.2 that adirectory without listings will have no receivers.

Senders Senders value looks at their listing since they generate profitπ from each look. We assume that this profit is related to the sales ofservices and goods. The number of looks for each listing depends on thesize10 of the listing si and the number of receivers:

L1 = s1qR

L2 = s2qR (3.3)

We assume that large listings attract more receivers than small listings(s2 > s1)11. Further we assume that each sender buys at most one ad.Senders pay a price pi to use the platform. The net benefit of a senderreads:

πLi − pi (3.4)

with i = 1, 2 being the type of listings. Senders are of mass one and areheterogeneous in their profit per look π which is uniformly distributedbetween 0 and 1. Define π ≡ p2−p1

L2−L1to be the profit level where senders

are indifferent between both ad types, and π1 ≡ p1L1

is the lowest profit forwhich the sender’s net benefit is positive. Then

q1 = π − π1

q2 = 1− π (3.5)

10Both parameters s1 and s2 stand for the size of the listing format. Note that it

can also be interpreted as salience, as in the framework of Haan & Moraga-Gonzalez

(2009). Salience is the chance that a certain firm is remembered.11This assumption can be made without loss of generality. If it would be the case

that small listings attract more attention than large listings, then small listings can be

relabeled q2 and large listings q1.

3.2 Set-Up of the Theoretical Model 75

Since we model the platform as a quantity setting monopolist on the senderside, we transform the demand functions to inverse demand functions:

p1 = (1− q1 − q2)L1

p2 = (1− q2)L2 − q1L1 (3.6)

Publisher The platform maximizes its profit:

Π = p1q1 + p2q2 + pRqR (3.7)

While deciding on the quantities, a platform has to take into accountthe direct and indirect effects of changing these quantities. An increasein quantity directly affects profits by a decrease in prices, which is com-plicated by product differentiation. Another, indirect effect is the effectof quantity on usage. An increase in the number of listings increasesthe number of receivers and therefore increases the willingness to pay ofsenders. This effect, the feedback loop effect, dampens the negative ef-fect of a quantity increase on sender price. A similar mechanism existsalong the receiver side: an increase in receiver price reduces the numberof receivers, which further decreases the willingness to pay of senders.

In the next sections, we adapt this set-up to shed light on the researchquestions. In order to present these optimal decisions more clearly, wedefine three additional parameters: r = r1

r2, R = r1 + r2 and s = s1

s2. Thus

r is the relative appreciation of q1 versus q2, whereas s is the relativeattention-attraction of q1 versus q2. We replace r1, r2 and s1 to obtainexpressions that contain only r, R, s and s2

12. To simplify the analysis,we assume that the platform has no costs. Including a fixed cost wouldnot change our results. Printing industries typically have large fixed costscompared to marginal costs13. Therefore, it seems reasonable to focus onfixed costs only. As a consequence, there is no cost difference betweensmall and large listings.

12We replace r1 by rR1+r

, r2 by R1+r

and s1 by ss2.13For digital media, one can argue that there is only a first copy cost, since marginal

costs are often nearly zero.

76 Directories

3.3 Investigation of the Universal Service Obli-

gations for Telephone Directories

In this section, we try to answer the following questions. Will the platformprovide a comprehensive telephone book, given that is available to allend-users? And will the book be available to all end-users given that theplatform provides a comprehensive directory? Subsections 3.3.1 and 3.3.2are structured as follows. First we explain the restriction imposed (i.e.availability in 3.3.1 and comprehensiveness in 3.3.2). Then we compute theoptimal outcomes for the monopoly platform and perform a comparativestatics analysis. Third, we extend the choice set of the platform with therelative size of the listing (s) We end with a discussion on the necessity ofthe other restriction, given the outcomes obtained.

3.3.1 The Availability Restriction

In this section, we examine whether the monopolist platform will providea comprehensive directory, including all potential senders, given that thedirectory is available to all receivers. In order to publish such a directory,the platform must offer the small listings for free. Otherwise, some senders(i.e. those with a low willingness to pay) will opt out and prefer not toappear in the directory. With the restriction that the directory is avail-able to all receivers, we assume that receivers don’t have to pay for thedirectory, i.e. that pR = 0. Note that even with pR = 0, not all potentialreceivers use the directory.

Therefore, the maximization problem of the platform reads:

maxq1,q2

Π = p1q1 + p2q2 (3.1)

s.t. q1, q2 ≥ 0

s.t. q1 + q2 ≤ 1

Solving this problem allows to state the following results:

Proposition 3.3.1 The optimal outcome for the monopoly platform issuch that:

Regime 1 A “large listings only” platform (i.e. q∗1 = 0, q∗2 > 0) prevails if r ≤ s

Regime 2 A “small listings only” platform (i.e. q∗1 > 0, q∗2 = 0) prevails if 34

< s < 1 and

r ≥ s4s−3

Regime 3 A “comprehensive” platform (i.e. q1 + q2 = 1: everyone is listed) prevails if

s < 14

and r ≥ s−4s2+2√

s2(1−5s+4s2)

1−4s

3.3 Results 77

Figure 3.1: Different regimes in the r-s-space, given availability

Regime 4 A differentiated platform (i.e. q∗1 > 0, q∗2 > 0: small and large listings are

offered) prevails otherwise

Note that all proofs are relegated to Appendix A.When s is large vis-a-vis r, as in regime 1, this means that the at-

tractiveness of a small listing is quite high, though its contribution toreadership is quite low. A high s makes a large listing more attractive forsenders, and will increase the willingness to pay for large listings. Thishigh s will also induce self-selection of senders, who trade in their largelisting for a small listing. The platform incurs a double loss because theprice paid by a sender will be lower (p1 ≤ p2), but also because the direc-tory will attract less receivers (small r), reducing the general price level.Therefore, a monopolist will set q1 as low as possible – even negative if itwere possible. If r ≤ s, the monopolist will offer only large listings.

The reverse occurs in regime 2 when both s and r are sufficiently large.Then the small listings q1 contribute (much) more to the readership thanlarge listings q2. Since both listings appear to be quite similar in termsof attractiveness – when s approaches 1, they are equal – the difference

78 Directories

in willingness to pay for both types becomes smaller. Given this higherreadership (and therefore the general price level), it becomes optimal tosteer senders away from q2 to q1. If s and r are sufficiently large, then themonopoly platform will offer only small listings.

A third regime occurs when s is sufficiently small, and r is sufficientlylarge. With these parameter values, q1 has a relative high contributionto readership. But even in the case where this contribution is smallerthan large listings (r < 1), only a limited amount of the more profitablesenders self-select into q1 (against q2) because the difference in size isquite large. In this case, the platform offers more quantity space thanthere are potential senders, leading to a negative price for type 1. Thisis compensated by the profit made through q2. Since negative prices areimpossible in our model, we cap q1 + q2 to 1, i.e. the amount of potentialsenders. This results in an implicit price of zero for the lowest category.Figure 3.1 graphically summarizes these regimes in the r-s-space14.

When both s and r are intermediate, both listing formats figure nextto each other in the directory. Regime 4 is the unconstrained optimumfor this maximization problem. As shown above, this regime will notalways be feasible, because for some values of s and r, this will violate thenon-negativity conditions on q1 and q2, or the upper limit q1 + q2 ≤ 1.In this range, s is small enough to allow self-selection, and large enoughto charge the small listings. Simultaneously, r is large enough such thatreaders value both listing formats.

Performing comparative statics analysis allows us to formulate threeinterdependent propositions on the impact of R and s2, r and s on theequilibrium quantities and profit.

Proposition 3.3.2 • R and s2 positively affect profit, but do not af-fect the optimal quantities.

• The proportion of r1 to r2 (= r) does affect the optimal quantities,but only in the comprehensive and differentiated optimum. The big-ger r1 relative to r2, the higher the amount of small listings and thelower the amount of large listings.

• The proportion of s1 to s2 (= s) does affect the optimal quantities,but only in regime 4. If r1 is substantially larger than r2, an increase

14As can be seen, the boundaries between regimes 3 and 4, r =s−4s2+2

√s2(1−5s+4s2)

1−4s, and between regimes 4 and 2, r = s

4s−3, go asymptotically

to 14

and 34.

3.3 Results 79

in s increases q1 and decreases q2. If r1 is smaller than r2, anincrease in s decreases q1 and increases q2.

The effect of r on the optimal quantities is straightforward. The morea listing type is appreciated by receivers, the more prominent this typewill be on the directory. The effect of s is slightly more complicated. Ahigher s increases the willingness to pay for q1, but it might also lure awaysenders from choosing a large ad. Therefore, the ultimate decision of theplatform depends on the contribution of both types to readership. If q1

listings contribute sufficiently more to readership, then an increase in s

results in an increase of q1 and a decrease of q2. The reverse holds whenq1 contributes less than q2.

An important remark on these comparative static findings is that theeffect of r and s is not limited to play a role only within these regimes, butalso may affect which regime is optimal. A change in r or s can induce aregime shift.

Note that regime 3 is the only regime where the platform offers acomprehensive directory. In this regime, the constraint q1 + q2 ≤ 1 isbinding, i.e. q1 + q2 = 1. This also means that q1 is offered for free tothe potential senders. The platform only makes profit from the categoryof large listings, q2.

Endogenous Size Decision The results of the constrained problemwith pR = 0, indicate that the platform will offer a comprehensive di-rectory if s < 1

4 and r ≥ s−4s2+2√

s2(1−5s+4s2)1−4s . In all the other cases, the

platform will either offer only one type (regime 1 and 2) or two types at apositive price. All the latter cases result in a non-comprehensive directory.

This clearly shows how the choice to offer a comprehensive directorydepends on the parameters r and s. In our analysis, we treat size asexogenous, which gave us a good insight in the interplay between r ands. In reality, however, platforms can choose the size of the listings. Sincewe assumed no costs to size in the model, platforms will set the levelparameter s2 of the listings as high as possible. It is more interesting toinvestigate the choice of s (for a given s2).

Proposition 3.3.3 If a profit maximizing monopolist can choose the sizeof the listings, it makes the large listing as large as possible and the smalllisting as small as possible. It offers receivers a directory with all potentialsenders.

80 Directories

Except for the special cases r = 0 or r = ∞, the profit maximizingmonopolist sets s = 0, i.e. it reduces the attractiveness of the small listingsas much as possible. The profit maximizing monopolist implements regime3, i.e. it offers receivers a directory with all potential senders (q1+q2 = 1).It offers two possibilities to senders: a large listing that attracts manyreceivers, or a small listing that attracts virtually no receivers. The latteris offered for free15. Casual observation of many European directoriesshows the prevalence of this outcome.

It goes without saying that this proposition has an important corollaryin the debate whether a government should impose a universal serviceconstraint on comprehensiveness.

Corollary 3.3.4 If receivers don’t pay, the inclusion of a comprehensive-ness constraint in the universal service obligation is not necessary, sincea profit maximizing publisher will always choose to open the directory forall potential senders.

If the receiver price is zero, and the platform can choose the relativesize s, then it will offer a comprehensive directory. Note that the inclusionof costs, marginal or fixed, might potentially change these results. Whens is fixed (i.e. by receiver preferences), our regimes show that offering acomprehensive directory is not always the profit maximizing regime for aplatform.

If we add more listing categories to our model – a closer fit with tele-phone directories in reality – the likely result will be that the largest listingis as large as possible, the smallest as small as possible. This is similarto the quality degradation models of Mussa & Rosen (1978), Maskin &Riley (1984) and Besanko, Donnenfeld & White (1988). In these models,a monopolist deteriorates the quality offered to the groups with the low-est willingness to pay for quality. If a regulator rules that a mere listingis a product with too low quality, it can still implement regulatory cor-rections, such as minimum quality standards though this is not alwayswelfare improving (Besanko, Donnenfeld & White 1988).

3.3.2 The Comprehensive Directory Restriction

In the previous section, we investigated whether the directory is compre-hensive if the price for receivers was zero. In this section, we look at

15Note that in regime 3, q1 is always offered for free, even if s > 0.

3.3 Results 81

the reverse problem: will the platform offer the directory for free if it isimposed that the directory should be comprehensive.

The maximization problem of the platform reads:

maxq1,q2,pR

Π = p1q1 + p2q2 + pRqR (3.2)

s.t. q1, q2, pR ≥ 0

s.t. q1 + q2 = 1

Solving this problem allows to state the following results:

Proposition 3.3.5 There are four regimes for a monopoly platform:Regime 1 A “large listings only” platform (i.e. q∗1 = 0, q∗2 = 1, p∗R > 0) prevails if

s2−Rs2

≤ s < 1 and r ≤ r1 ≡ R−(1−s)s2R+(1−s)s2

Regime 2 A “small listings only” platform (i.e. q∗1 = 1, q∗2 = 0, p∗R > 0) prevails ifs2−R

s2≤ s < 1 and r ≥ r2 ≡ R+(1−s)s2

R−(1−s)s2

Regime 3 A differentiated platform not charging users (i.e. q∗1 > 0, q∗2 > 0, p∗R = 0)

prevails if s ≤ s2−3Rs2

; or s2−3Rs2

≤ s ≤ s2−2Rs2

and

r3 ≡ 3R2−2R(1−s)s2+(1−s)2s22−2

√3(R2(1−s)s2((1−s)s2−2R))

3R2+2R(1−s)s2−(1−s)2s22

≤ r

≤ r4 ≡ 3R2−2R(1−s)s2+(1−s)2s22+2

√3(R2(1−s)s2((1−s)s2−2R))

3R2+2R(1−s)s2−(1−s)2s22

Regime 4 A differentiated platform charging users (i.e. q∗1 > 0, q∗2 > 0, p∗R > 0) prevails

otherwise

Similar to the problem of section 3.3.1, regime 4 is the unconstrainedregime (given the constraint q1 + q2 = 1). This regime is not alwaysfeasible. The other regimes are corner solutions where q1, q2 and pR areequal to zero, respectively.

When s and r are both sufficiently large (regime 2), then the same self-selection problem pops up as in the specification where pR = 0. Since bothlistings appear very similar, the platform gains more by steering sendersto the listing type that maximizes readership (and therefore profit). Thesame holds when r is sufficiently small (regime 1), i.e. then it is optimalto offer only large listings. Note that it depends on both R and s2 todetermine whether s is sufficiently large. If R is relatively large withrespect to s2, then even with a small s, the platform will choose to offeronly one type if r is sufficiently large or small.

If s is small and R is sufficiently small with respect to s2, then thedirectory is offered for free to users (regime 3). If R is small, then thewillingness to pay of receivers is low and therefore, it might be optimalto give away the directory for free. It might even be profit maximizing

82 Directories

Figure 3.2: Different regimes in the r-s-space, given comprehensiveness

to subsidize the use of the directory, i.e. a negative price (pR < 0). Inprinciple, it is not strictly necessary to impose that pR ≥ 0. A platformcould subsidize its receivers, although in practice it is not directly clearhow. Note that a small R is not sufficient, but should be accompanied witha small s as well. Since the restriction holds that the directory should becomprehensive, the platform will only make profits from large listings andreceivers. The more interesting it is to extract money from large senders,the more the platform is inclined to charge the receivers no price. Thishappens when s is sufficiently small.

Figure 3.2 shows these regimes graphically in the r-s-space. In thisgraph, R and s2 are equal to 0.25 and 1. These parameters are crucial indetermining the boundaries of the regime. If 2R > s2, then the platformwill never set a receiver price pR equal to zero, as is the case in the thirdregime. Contrary to the problem in 3.3.1, the boundary conditions arealso determined by the parameters (R and s2).

Within these regimes, it is interesting to see what happens if the pa-rameters change. The comparative statics for the regimes “large listings

3.3 Results 83

only” and “small listings only” are the easiest to interpret. These regimesare not affected by the size parameters s and s2. For both regimes, anincrease in R affects price, readership and profit positively. An increasein r also has a positive effect on price, readership and profit in the “smalllistings only” outcome, but a negative effect on all three in the “largelistings only” outcome.

For the outcome where pR = 0, only r affects the equilibrium quantities(dq1

dr > 0, dq2dr < 0). Profits are affected positively if r increases, but only if

r > 1. The same holds for the effect of r on readership. An increase in s

is negative for profits which anticipates already proposition 3.3.7. R ands2 affect profit positively.

Comparative statics for the “differentiated paid platform” are slightlymore complicated. The relative contribution to readership r affects q1

positively and q2 negatively. An increase in r has a positive effect onprice, readership and profit, if r > 1. The reverse holds if r < 1. Both s

and R affect q1 positively and q2 negatively if r > 1. An increase in s2

affects q1 negatively if r > 1. The increased profit from senders allows theplatform to lower receiver prices. As before, profit is affected positivelyby R and s2, but negatively by s.

These insights can be summarized as follows.

Proposition 3.3.6 • R affects profit positively. For regimes 3 and 4,s2 affect profit positively, an increase in s has a negative effect onprofit.

• R, s2 and s only matter for the equilibrium quantities in regime 4(the “differentiated paid platform”). Their effect is crucially deter-mined by whether r is bigger or smaller than 1.

• r increases q1, but decreases q2. It has a positive effect on profit if r

is already quite high (above 1). The reverse holds if r < 1. Profit ishigher if receivers have extreme preferences, i.e. highly favoring onead type over the other.

These comparative statics are very similar to those presented in propo-sition 3.3.2. Apart from some differences, the general insights on the effectsof changes in the corresponding parameters holds. We conclude this sec-tion with the observation that a platform prefers receivers with extremepreferences, i.e. preferences where the difference between both types islarge. The closer r to 1 (i.e. where tastes are equal), the lower the profitwill be.

84 Directories

Endogenous Size Decision The results of the constrained problemwith q1 + q2 = 1, indicate that the platform will offer the directory forfree to receivers if s ≤ s2−3R

s2; or s2−3R

s2≤ s ≤ s2−2R

s2and r3 ≤ r ≤ r4.

In all the other cases, the platform will charge the receivers a positiveprice. Similar to the question on comprehensiveness, we allow now thatthe platform also determines the relative size s.

Proposition 3.3.7 If a profit maximizing monopolist can choose the sizeof the listings, it makes the large listing as large as possible and the smalllisting as small as possible. Which regime will prevail, depends on themagnitude of R and s2. The smaller R, the more likely that receiversdon’t pay.

If R < s23 , then the receiver price will be zero. In the range s2

3 ≤ R ≤s22 , the receiver price is zero if r3|s = 0 ≤ r ≤ r4|s = 0. If R > s2

2 , thenreceivers will always be charged to access the platform. We can summarizethis in the following corollary.

Corollary 3.3.8 The comprehensiveness restriction is not sufficient toguarantee that receivers don’t have to pay.

In sum, whereas our model sustains the claim that the directory willbe comprehensive if the price is zero, the reverse is less likely. Receiverswill only be offered a free directory if their total valuation for listings R isrelatively small16, or the gain from a large listing s2 is relatively high. Inthis case, the senders cross-subsidize the receivers. If R is relatively high,or s2 is relatively low, the reverse holds: receivers cross-subsidize senders.

Before we conclude, we discuss two further issues, readability and uni-form pricing.

16In the past, it was often assumed that receivers were very price elastic. Given

the two-sided markets models a la Armstrong, a receiver price of zero is then a likely

outcome, at least if senders elasticity is (much) lower. Recent research shows that

receivers are not that averse to pay a price. van Caspel, Moerman, Vermeer & Hermans

(2002) show that a price increase of 1 euro causes a modest drop in demand of 1.75

tot 2 percent. A price around 5 euro might be optimal, although no calculations on

profit maximization are given. A likely effect of a positive price is that people might

keep their directory for a longer period. In most countries, publishers offer an annual

update, because that is mandatory for White Pages in the European universal service

directive. Of course, if directories are provided online, updates can occur permanently.

3.3 Results 85

Readability Our conclusion that availability implies comprehensive-ness, but not the other way around, is based on the premise that it isfeasible for the platform to implement a relative size s = 0. This impliesthat the size of a small listing is infinitesimally small. As a result, the in-formation might become unreadable. A drawback of our model is that thisextremely small size gets accepted by the reader, because the valuationfor small listings r1 is not related to size. It is reasonable to assume that alisting should have a minimum size to be readable. In that case, for a fixeds2, s also has a lower limit s. This implies that our claim on redundancyof the comprehensiveness condition given availability only holds if s < 1

4

and r ≥ s−4s2+2√

s2(1−5s+4s2)

1−4s . The reverse claim, whether availabilityimplies comprehensiveness, only partially relies on the minimum size s. IfR > s2

2 , then it will never be the case that receivers don’t pay, no matterhow small s is17.

In sum, the existence of a lower limit for s softens the basic claim in thisarticle. Our results are only guaranteed if this lower limit is sufficientlysmall with regard to r. If not, then both conditions are necessary toimplement and it does not suffice to impose only the availability condition.

Uniform Pricing In models on two-sided markets, nearly all contribu-tions assume that platforms set uniform prices on both sides (Weyl 2010).Our model diverts from these models by allowing multiple products andhence different prices. Interestingly, our results show that in two of thefour regimes (for both restrictions), the platform chooses not to use thepossibility to offer multiple listing formats. As a result, the platform offersonly small, or only large, listings. As such, to the casual observer, marketoutcomes for these regimes seem similar to the earlier models that allowonly for one product and one price.

The traditional explanation for not offering the full range of products,is that it is neither feasible nor desirable in the presence of returns to scale(Spence 1976). Further, the self-selection of consumers prevents that aseller can fully exploit discriminatory prices. We complement these expla-nations with a two-sided market logic: it also depends on the other side of

17Another solution is to make the valuations r1 and r2 dependent on the size. We

are reluctant to do so, because the advantage of keeping the size and the informational

value separate is that it does not determine a specific relationship between both. It

might be that this relationship is non-monotic: an increase in the size when size is

small is probably positively valued; an increase when size is already large is negatively

valued.

86 Directories

the market (captured by parameter r) whether offering multiple productsis optimal. We showed that platforms also offer a less profitable format ifreceivers value this type sufficiently (and the size difference between bothlisting formats is large enough).

3.4 Conclusion

In this article, we examined the main conditions of the universal ser-vice obligation for telephone directories. These conditions are compre-hensiveness and availability. Telephone directories should contain all tele-phone subscribers and should be available to all end-users. We investigatewhether one of these conditions is redundant, i.e. if imposing one con-dition also implies that one automatically gets the other. We find thatimposing a reader price equal to zero (and hence availability to all end-users) results in the publishing of a comprehensive directory. Conversely,it is not guaranteed that imposing a comprehensive directory results inavailability to all end-users. An important caveat to our model is thatcosts are ignored and that an exogenously imposed minimum size couldsoften our redundancy result.

These results might provide guidance in the discussion on regulatingsearch engines. These engines share the same micro-economic mechanismwith directories: paid listings (on the top of the ranking) feature next toalgorithmic listings (with unknown ranking). The debate on the regula-tion of search engines is centered around a fair ranking; the arguments aresimilar to the rationale for a universal service obligation for directories.Search engines structure vital information for a society and the marketmight not provide valuable search results. Both conditions of the obliga-tion, availability and comprehensiveness, are at the heart to achieve aninformed society. Our results suggest that the availability condition mightbe favored over the comprehensiveness condition.

We end with two remarks related to the current policy debate. First,is a universal service obligation for directories desirable? And second, isall data on subscribers readily available?

Discussion on desirability Our model investigated the positive ques-tion whether one of the conditions of the universal service obligation fordirectories is redundant. The normative question, whether such an obliga-tion is desirable, is beyond the scope of this article and can probably not

3.4 Conclusion 87

be solved by a simple theoretical model. As shown by Cremer, Gasmi, Gri-maud & Laffont (1998), many reasons might justify the implementationof a universal service regulatory framework. A universal service obligationmight be the solution for the non-internalization of network externalitiesor the under-provision of a public good. Besides, it might serve a redis-tributive or regional purpose. It is currently under discussion whether atelephone directory fits these policy goals.

In 2005, the European Commission asked “For how long will there bea need to keep directories and directory enquiry services within the scopeof universal service?”18 According to the same document, the commonview of the industry was that a universal service obligation in the retailmarket is unnecessary and harmful to competition. This was followedby a Commission staff working document in 2006 that proposed to “re-move the provision of directories and directory inquiry services from thescope of universal service and leave the market to meet demand for theseservices”19.

This proposal was not implemented, and as a response to the Pub-lic Consultation on Universal Service Principles in E-Communications ofMarch 2010, many telecom operators reiterated that this obligation wasan “anachronistic obligation” (Telecom Italia20) and “increasingly unjus-tified” (Telefonica21).

It seems likely that directories and directory assistance will be left outfrom the universal service directive in a future revision. Moreover, 9 ofthe 27 countries of the European Union have not implemented the part of

18COM(2005) 203, On the Review of the Scope of Univer-

sal Service in accordance with Article 15 of Directive 2002/22/EC,

http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2005:02

03:FIN:EN:PDF.19SEC(2006) 816, Annex to COM(2006) 334 final,

http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=CELEX:52006

SC0816:EN:NOT20Telecom Italia (2010), Telecom Italia response to the Pub-

lic Consultation on Universal Service Principles in E-Communications,

http://ec.europa.eu/information_society/policy/ecomm/doc/library/

public_consult/universal_service2010/comments/telecom_italia.pdf.21Telefonica (2010), Telefonica Response On The Public Con-

sultation On Universal Service Principles In E-Communications,


public_consult/universal_service2010/comments/telefonica.pdf.Other tele-

com operators, such as British Telecom, Deutsche Telekom, Orange, Fonecta and

Vodafone, joined these arguments.

88 Directories

the universal service directive on directory services22. In 2003, it was alsonoted at an OECD round table that there is no evidence that universaldirectory services would not be provided in absence of a universal require-ment23. Besides, a participant noted that directory service is mostly usedby business users and not general consumers and should therefore not beregarded as a universal service. In the United States, the official viewpointremains that directory service is one of the core services included in theuniversal service funds. At the same OECD round table, the Federal Com-munications Commission noted that “access to directory service is usedby a substantial majority of residential customers, is widely available, isessential for education, public health, and safety, and is consistent withthe public interest, convenience, and necessity”.

Discussion on availability of data Another point that has been putforward in the policy debate on directories, is the availability of infor-mation. In our model we assumed that all information on companies isreadily available to the platform. In reality, this might not be the case.Therefore, many parties have pointed out that there is a difference be-tween the universal service obligation for directories at the wholesale andthe retail level. The retail level concerns the comprehensiveness and avail-ability of the directory to the end user. The wholesale level includes theobligation on network operators to make directory data available to thirdparties. The non-availability of contact data might constitute a seriousbarrier to entry. It is reasonable to impose the obligation on all marketplayers to share contact information, at least for the purpose of composinga directory.

22Vodafone (2010), Vodafone comments on European Commission questionnaire

for the public consultation on universal service principles in e-communications,


public_consult/universal_service2010/comments/vodafone_uk.pdf23DAFFE/COMP(2004)19, OECD Policy Round Table, Non-Commercial Service

Obligations 2003, http://www.oecd.org/dataoecd/43/35/33691140.pdf

A Proofs 89

Appendix

A Proofs

Proof of Proposition 3.3.1 If we solve equation 4.5 with respect tothe constraints q1, q2 ≥ 0 and q1 + q2 ≤ 1, we find four optimal regimes:

q∗1 q∗2regime 1 0 2

3regime 2 2

3 0

regime 32−r−

√1−r+r2

3(1−r)1−2r+

√1−r+r2

3(1−r)

regime 4 r(s−1)s+s2−s3+√

z

6(1−s)s(r2+s−2rs)3r3(s−1)−8r2(s−1)s+2s2−2s3−√z+r(5s3−3s2−2s+

√z)

6(s−r)(1−s)s(r2+s−2rs)

where z = (r−s)2(s−1)s(−3r2+6rs+s2−4s). Regime 4, the differentiatedoptimum, is a local maximum, at least when r > s. The global maximumis reached when q1 and q2 approach −∞, because in that case prices arealso negative and profit approaches +∞. Off course, this regime will notbe feasible. Regime 4 will also be infeasible for some parameter values.

Since we have solved this model with Karush-Kuhn-Tucker conditions,we should check whether and when these conditions hold.

The only non-zero Lagrange operator related to the constraint q1+q2 ≤1, prevails in regime 3. Therefore, if the Lagrange operator, which is equaltoRs2(r

3(1−4s)−3(1+√

1−r+r2)s+r2(−4+√

1−r+r2+(7−4√

1−r+r2)s))9(−1+r)2(1+r)

+ Rs2(r(1+√

1−r+r2+(2+5√

1−r+r2)s))9(−1+r)2(1+r) , is positive, then regime 3 should be

implemented and q1 + q2 will be equal to 1. The corresponding price p1

is equal to 0, i.e. small listings are not paid.The condition q1 ≥ 0 is violated if r ≤ s. Moreover, the second order

derivatives show that regime 4 is a maximum only when r > s. In thiscase regime 1 will be implemented where q1 = 0. The other non-negativitycondition q2 ≥ 0 is violated if r > s

4s−3 , in this case regime 2 will beimplemented where q2 = 0.

90 Directories

Figure A1: Platform profit. Left panel: r=0.5, right panel: r=1000

Proof of Proposition 3.3.2 To trace back the effects of R, s2, r ands, we perform a comparative statics analysis on the equilibrium quantitiesof the different regimes and on the profits under the different regimes.

First, we find that the derivatives under all regimes of quantity withrespect to R and s2 are zero: dq1

dR = dq2dR = dq1

ds2= dq2

ds2= 0. If we derive the

equilibrium profits Πi (with index i for the regime) towards R and s2, weobtain:

dΠ1

dR=

dΠ1

ds2=

4α

27(1 + r)

dΠ2

dR=

dΠ2

ds2=

4rαs

27(1 + r)

dΠ3

dR=

dΠ3

ds2=

(1 − 2r +

√1 − r + r2

) (−1 + 2r2 −

√1 − r + r2 + 2r

(−1 +

√1 − r + r2

))α(−1 + s)

27(−1 + r)2(1 + r)

dΠ4

dR=

dΠ4

ds2= −

α(−9r3(−1 + s)s + 3r2(−9s2 + 9s3 +√

(r − s)2(−1 + s)s(−3r2 + 6rs + (−4 + s)s))

108(1 + r)(r − s)(−1 + s)s(r2 + s − 2rs)

−rs(−8s − 11s2 + 19s3 + 6

√(r − s)2(−1 + s)s(−3r2 + 6rs + (−4 + s)s))

108(1 + r)(r − s)(−1 + s)s(r2 + s − 2rs)

+s(−8s2 + 7s3 + s4 + 4

√(r − s)2(−1 + s)s(−3r2 + 6rs + (−4 + s)s))

108(1 + r)(r − s)(−1 + s)s(r2 + s − 2rs)

−s2

√(r − s)2(−1 + s)s(−3r2 + 6rs + (−4 + s)s)))

108(1 + r)(r − s)(−1 + s)s(r2 + s − 2rs)

with α = s2 in the case of dΠdR and α = R in the case of dΠ

ds2. If r ≥ 0,

R > 0, s2 > 0 and 0 ≤ s < 1, then these derivatives are unambiguouslypositive. An increase in R or s2 always increases profit.

A Proofs 91

Second, if we derive the optimal quantities to r, we obtain:

dq11

dr= 0

dq21

dr= 0

dq31

dr=

−1 − r + 2√

1 − r + r2

6(−1 + r)2√

1 − r + r2

dq41

dr=

(r − s)(

r(−1 + s)s + s2 − s3 +√

(r − s)2(−1 + s)s(−3r2 + 6rs + (−4 + s)s

))

3(−1 + s)s(

r2 + s − 2rs)2

−−6r3 + 18r2s + 4s2 + 2s3 − 2rs(2 + 7s) +

√(r − s)2(−1 + s)s

(−3r2 + 6rs + (−4 + s)s

)

6(

r2 + s − 2rs) √

−(r − s)2(−1 + s)s(3r2 − 6rs − (−4 + s)s

)

Results for dq2dr are similar, and, as can be expected in the comprehensive

case, dq32

dr is exactly the same as dq31

dr , but with the opposite sign. With thesame restrictions as above, we can show that dq1

dr > 0 and dq2dr < 0 in the

differentiated and comprehensive case.Third, we can derive the optimal quantities towards s. It is easy to

show that dq11

ds = dq21

ds = dq31

ds = 0. The derivative dq41

ds is somewhat morecomplicated; and crosses the horizontal axes once in the r-space, i.e. itcan be both negative and positive, depending on the value of r.

92 Directories

We can summarize all the comparative statics as follows:

Comparative statics of the Free Directory Restriction

regime 1 dq1 dq2 dπ

dr 0 0 < 0

ds 0 0 0

dR 0 0 > 0

ds2 0 0 > 0


dr 0 0 > 0

ds 0 0 > 0

dR 0 0 > 0

ds2 0 0 > 0


dr > 0 < 0 < 0 if r < 1

> 0 if r > 1

ds 0 0 <0

dR 0 0 > 0

ds2 0 0 > 0


dr >0 <0 < 0 if r < ρ2

> 0 if r > ρ2

ds < 0 if r < ρ1 < 0 if r > 1 < 0 if r < ρ3

> 0 if r > ρ1 > 0 if r < 1 > 0 if r > ρ3

dR 0 0 > 0

ds2 0 0 > 0

Interpretation: each cell is the derivative of column with respect to row.

where ρ1, ρ2 and ρ3 are parameter values above 1 where there is aswitch in the comparative statics, i.e. ρ1 is the value of r for which dq2

1ds = 0

and mutatis mutandis the same holds for ρ2 and ρ3.

Proof of Proposition 3.3.3 To find the solution of the private mo-nopolist who decides on quantities and size (s), we maximize equation4.5 with respect these variables. We find two regimes for this problem,which are equivalent to regimes 2 and 3 of the problem in section 3.3.1,supplemented with a size choice of s = 1 in regime 2 and s = 0 in regime3. Further we find that regime 3 dominates for all values of r. The onlyexception is r →∞; then the platform is indifferent between both regimes.In sum, one can say that s = 0 is optimal for all values of r. Moreover, itis the only solution if r2 > 0.

Proof of Proposition 3.3.5 Solving equation 3.3, we obtain four op-timal regimes.

A Proofs 93

q∗1 q∗2 p∗Rregime 1 0 1 R

2(1+r)regime 2 1 0 rR

2(1+r)

regime 32−r−

√1−r+r2

3(1−r)1−2r+

√1−r+r2

3(1−r) 0

regime 4 12 +

R(r−1)2(1+r)(1−s)s2

12 +

R(1−r)2(1+r)(1−s)s2

3(r−1)2R2+2(1+r)2R(1−s)s28(1+r)2(1−s)s2

-(1+r)2(1−s)2s228(1+r)2(1−s)s2

Similar to the proof of proposition 3.3.1, we solved this problem underKarush-Kuhn-Tucker conditions. Checking for non-negativity of the keyvariables and computing the Hessian matrices for these regimes, we con-clude that that these are the four regimes that might prevail. The bound-aries where these regimes apply are in the main part of this article.

Proof of Proposition 3.3.6 In the table below, we summarize allsigns of the first order derivatives. The results in proposition 3.3.6 fol-low straightforward from these tables.

Comparative statics of the Comprehensive Directory Restrictionregime 1 dq1 dq2 dpR dπ dqRdr 0 0 < 0 < 0 < 0ds 0 0 0 0 0dR 0 0 > 0 > 0 > 0ds2 0 0 0 0 0

regime 2 dq1 dq2 dpR dπ dqRdr 0 0 > 0 > 0 > 0ds 0 0 0 0 0dR 0 0 > 0 > 0 > 0ds2 0 0 0 0 0

regime 3 dq1 dq2 dpR dπ dqRdr > 0 < 0 0 > 0 if r > 1 > 0 if r > 1

< 0 if r < 1 < 0 if r < 1ds 0 0 0 < 0 0dR 0 0 0 > 0 > 0ds2 0 0 0 > 0 0

regime 4 dq1 dq2 dpR dπ dqRdr > 0 < 0 > 0 if r > 1 > 0 if r > 1 > 0 if r > 1

< 0 if r < 1 < 0 if r < 1 < 0 if r < 1ds > 0 if r > 1 > 0 if r < 1 > 0 < 0 < 0

< 0 if r < 1 < 0 if r > 1dR > 0 if r > 1 > 0 if r < 1 > 0 > 0 > 0

< 0 if r < 1 < 0 if r > 1ds2 > 0 if r < 1 > 0 if r > 1 < 0 > 0 > 0

< 0 if r > 1 < 0 if r < 1

Interpretation: each cell is the derivative of column with respect to row.

Proof of Proposition 3.3.7 The proof of proposition is similar to theproof of proposition 3.3.4 and therefore omitted. It can also be derivedfrom the comparative statics results for dπ

ds .

94 Directories

B Empirical Investigation of the Yellow Pages

Industry

As we stressed in the introduction of this dissertation, the interactionbetween theory and empirics is often lacking. In this section, we try tobridge the gap between theory and empirics.

In propositions 3.3.1 and 3.3.5, we make the claim that it is possible topredict the different regimes based on the parameters r and s. Off course,these predictions are conditional on the restrictive assumptions made, butat least, it gives an empirically testable hypothesis. An empirical exercisecan exploit variation in both parameters to test whether these parameterscan explain the equilibrium quantities of listings. A major concern of suchan application is endogeneity. Since r and s are already determined bythe equilibrium quantities, we don’t obtain valuable results. Therefore,in this section, we propose only a first move, i.e. by suggesting how r

and s might be estimated. For this application, we collected data on theEuropean Yellow Pages. We estimate the major parameters and reconcileour results with our theoretical model.

B1 Industry Characteristics

The total Yellow Pages revenues worldwide are estimated at 30.9 billionUS dollar in 2008 (Kelsey 2009). Yellow Pages attract 5.2 % of the globaladvertising market and employ 82300 people, nearly half of them are salesrepresentatives.

Print media are in turmoil everywhere in the world, and also the printYellow Pages do not escape the crisis. Yellow Pages companies hold outrelatively well in 2008, with revenues declining only 2.3% vis-a-vis 2007,but share prices nosedived at the end of the first decennium of the thirdmillennium. If we compare the shares of the beginning of 2010 with thebeginning of 2005, Seat (Italy) lost 99.7% of its price. Yell (UK) lost 91%and Pages Jaunes (France) 56%.

The reason for this decline, besides worldwide financial problems, isclear cut: the Internet. In 2008, Yellow Pages companies got 85% of theirrevenues from their print division and 15% from online outlets. It is likelythat the print edition will vanish, though the decline is slowed by twofacts24. First, Yellow Pages are still seen as more reliable and extensive

24The Economist, Dial I for internet, May 22nd 2008.

B Empirical Investigation of the Yellow Pages Industry 95

Table B1: Variables: summary statistics

Variable Mean Std.

Quantity Persuasive Ad 20937 10955

Quantity Informational Ad 96805 67842

Quantity Free Ad 657706 694509

Price Persuasive Ad 1622 1092

Price Informational Ad 310 199

Usage Penetration 0.40 0.12

Internet Penetration 0.29 0.17

Circulation 5807407 2423109

Inflation (Index=100 in 2000) 101 8

Income per Capita 26038 5511

than online search engines. Second, firms are loyal to Yellow Pages, helpedby an extensive sales force.

Below, we describe the data, discuss the relevance of the advertisementtypes and present the results of our empirical study on Yellow Pages.

B2 Data

To test the mutual effect of usage and advertisements in Yellow Pages, thefirst thing we need are data on usage and ads. Data on usage and adver-tisements are rare, even for those who advertise. Therefore, we contactedseveral companies to collect data.

We received book level data on advertising, i.e. quantities, list pricesand real prices. The latter is important because it is common practice togive a discount off the price. This creates a gap between list prices andreal prices. Since advertisers decide on real prices and not on list prices,we use real prices in the investigation.

Considering usage, we have book level data on distribution, but surveyson book usage are only available at the country level. To approximate thenumber of users, we multiply the percentage of people frequently using thebook with the circulation in a certain area. Because the lack of detailedusage data, we carry out our analysis on country level and not on booklevel.

We also have data on country characteristics which are used as controlvariables. These data are obtained from international sources, such asIMF, OECD and Eurostat.

The sample includes 5 European countries. Those countries are rela-

96 Directories

tively homogenous in population and market structure. It is dominatedby countries with a single market player. In two countries, there is somecompetition, e.g. from local city directories, but we treat those countriesas monopolies too. The earliest data point is 1995, the latest 2006. Sincethe data is compiled from different sources, we often lack data for the com-plete time span; i.e. the data set is an unbalanced panel. The statisticsof the variables are summarized in table B1.

B3 Identification

Reader Side Most users pick up the Yellow Pages to find a particularcompany or a particular good or service. If the directory is used to finda company which is already known to the user, then this is labeled asknown search. If the directory is used to find new suppliers, then this isunknown search. The distinction between both types of look-ups can havea substantial impact on usage.

At one extreme, one might argue that the users only value “raw” in-formation: the name and phone number of each supplier, be it classifiedby the category of the good or service that is supplied. In such a setting,it even might be the case that the users would pay for receiving a wellclassified directory that alphabetically lists all the suppliers in a certaincategory. This would be the case when each consumer is in a satisfac-tory or even optimal relationship with a certain supplier. The directorythen is an ideal instrument for retrieving the coordinates of the particularsupplier a consumer wants to patronize. But it merely serves the “admin-istrative” purpose of an organized inventory of one’s business contacts.Large colored ads floating around then could disturb the user that is onlyinterested in finding the coordinates of his trusted supplier. Hence theselarge ads might reduce the attractiveness of the directory.

At the other extreme, the large flashy ads in the directory serve topersuade consumers that have no relationship with a supplier yet, or lookfor change. These ads aim at the starting of a relationship with the par-ticular supplier that uses the directory for this purpose. In this setting,larger and flashy ads may convey information that this type of user is keenfor. It could separate the good suppliers from the bad in a signaling envi-ronment: the most efficient suppliers who have substantial turnover canpay for the larger ads, while the less efficient or inexperienced supplierscan not. The “nuisance” then comes from the small entries that merelyprovide for contact data. Since they signal no quality, they are not looked


at and hence redundant. They could be dismissed entirely by the phonedirectory provider.

In the case of the first extreme, people only use the directory as anindex: only the company name and number suffices. In the other extreme,people use the directory to find out the reputation of a company: onlypersuasive ads count. But perhaps there is also a category in between:informative ads. Users do not search only for the telephone number, butalso for more information about a company, i.e. fax number, e-mail or webaddress, opening hours or mobile number. Besides, it can also be interest-ing to have additional information about the activities of the company. Ifone is are looking for a replacement of one’s central heating, when look-ing at the category central heating, from a simple list of companies andtelephone numbers, one can not find out whether a supplier is specializedin gas heaters or oil-fired central heating. An informative ad can providemore information.

We can model Yellow Pages as a usage generating market. By choosingthe right amounts of quantities, a platform manages to create look-ups forthe advertisers. One potential production function is a Cobb-Douglasfunction.

U = BQβ00 Qβ1

1 Qβ22

∏

i

controlβi

i (B1)

If we take logs, then we obtain:

ln(U) = β + β0 ln Q0 + β1 ln Q1 + β2 ln Q2 +∑

βi ln controli (B2)

We will estimate this equation in the results section.Usage is captured by the percentage of people answering yes on the

survey question “Have you used the printed Yellow Pages last month?”.These are considered regular users of the directory. To count the num-ber of persuasive ads, we add up the larger advertisements with logo orgraphic. The informational ads are smaller ads that still fit into one col-umn. These ads are in the within category alphabetical list of companies.Large persuasive ads are always accompanied by a regular entry in the al-phabetical list, with a reference to the ad (e.g. see also advertisement onprevious page). We don’t have data on the free listings, i.e. what compa-nies get if they don’t advertise, but we can approach them by subtractingthe total number of ads of the total number of companies in a country.

98 Directories

Advertiser Side Rysman (2004) treats ads as a homogeneous serviceand explains the amount or volume of advertising chosen by an individualbusiness, but not the type. Equilibrium is reached when two offsettingnetwork effects keep each other in balance. On the one hand, more adver-tising leads to increased usage of the directory, i.e. users like ads in YellowPages unlike what seems to be the case in other media. And since adver-tisers like eyeballs, they buy more advertising as usage increases. Thisis the positive feedback loop that links the two sides of the market. Acountervailing force exists because of the negative network effect of con-gestion that takes place when too much advertising crowds a category:an overwhelming number of large ads for plumbers running over severalpages of a directory is not likely the medium that still another plumberwill choose to list his services in. The two effects taken together lead todirectories in which not every business decides to buy ad space.

As noticed above however, the type of ad that is requested by adver-tisers can be quite different and hence this can result in a different impactregarding the network effects just mentioned. Large ads may attract morepotential customers, but are perhaps annoying to users. At the same timethey trigger the possibility that another business’ ad becomes unnoticedor that trade is diverted rather than created.

The small ads on the other hand probably convince fewer potentialclients. But because they contain additional information in a condensedand surveyable way, they probably contribute more to the usage of thedirectory. By this, they might inflict a positive externality on the otheradvertisers of the directory. At the same time they might steal less busi-ness from the larger ads. In this respect, they might be complementary tolarger ads, and the directory that has one large and two small ads mightbe a better product than the directory of the same size that is composedof two large ads.

It is not hard to understand that the providers of directories will takethese differences into account. The result will be inter alia a differentpricing strategy for each type of ad. These pricing strategies will take intoaccount that the contribution to usage of each type of ad is different andthe effect of the ad on reaching potential customers. The congestion effectmight be circumvented to some extent by including different types of adsthat do not compete for the attention of the user. This leads to the claimthat the different types of ads might be heterogeneous services.

Following Rysman (2004), we model the advertising site as a Cobb-


Douglas function. Similar to the usage production function, it containsthe major ad type and control variables. Besides, the price also dependson the number of users U .

pj = AQα00 Qα1

1 Qα22 Uα3

∏

i

controlαii (B3)

If we take logs, then we obtain:

ln pj = α+α0 ln Q0+α1 ln Q1+α2 ln Q2+α3 ln U+∑

αi ln controli (B4)

We will estimate this equation in the results section.In this price equation, the quantities are the same as in the usage equa-

tion. As explained above we prefer real prices above rate card prices. Forusage, i.e. the number of consumers, we have no direct data. Therefore,we approach this number by multiplying the percentage of people regularlyusing the directory with the circulation. This may lead to an underestima-tion of the number of users, because directories are distributed to familieswhich consist usually of more than one person. Robustness checks showthough that multiplying this usage figure with average family size doesnot change the results below.

B4 Results

Reader side Do users value a directory mostly for its listings, small adsor large ads? In this section we test this question and present the results.

Because we work with an unbalanced panel, consisting of 5 differentcountries, we use a least squares estimator with fixed effects. This esti-mation technique allows to control possible characteristics of particularcountries - even without measuring them, as long as those characteristicsdo not change over time.

The number of observations is quite limited, though our results appearto be quite robust to other specifications of the model (see table B3).

Rysman (2004) proposed instrumental variables in his study on theYellow Pages in the United States. In the usage equation, the advertise-ment level is instrumented by the number of people covered by a directory.In the advertisement equation, he instruments usage by the people thatrecently moved. For advertisements he uses the earnings level in a county,because this approximates the hourly wage and can be seen as a costshifter. We do not apply these instruments because we have only a lim-ited number of observations and we do not have the necessary data to

100 Directories

Table B2: Results of the OLS regression

Usage Price Small Ads Price Large Ads

Quantity Listings 0.371 -0.023 0.137

(2.98)*** (0.45) (3.02)***

Quantity Small Ads 1.697 -0.851 1.744

(1.82)* (2.26)** (5.29)***

Quantity Large Ads 0.207 -0.155 -1.352

(0.25) (0.57) (5.71)***

Internet -0.109 0.373 0

(0.22) (2.35)** 0

Income per Capita 0.18 0.346 -0.423

(0.23) (1.07) (1.5)

Usage 0.253 0.17

(3.07)*** (2.35)**

CPI -0.004 0.027

(0.64) (4.92)***

Constant -13.52 9.905 -2.52

(1.21) (2.63)** (0.76)

Observations 24 24 24

R-squared 0.73 0.96 0.94

Absolute value of t statistics in parentheses

* significant at 10%; ** significant at 5%; *** significant at 1%

instrument. In this section, we present the results for the OLS regression.The robustness checks with seemlingly unrelated regression estimation canbe found in table B3.

As can be seen from table B2, the coefficients of the quantities of thesmall advertisements and the free listings are significantly different fromzero in the usage equation. The coefficient of small ads is quite large:increasing the number of small ads with 1 percent increases the numberof users with 1.7 percent. Large ads seem to have no effect on usage.

These results indicate that readers value most the small ads in thebook, but listings (=only name and telephone number) are also appre-ciated. Large advertisements do not affect usage. While large adver-tisements can be interesting for advertisers because they attract a lot ofeyeballs, users are not interested in the additional (persuasive) elements.

Advertiser side In table B2, the results for the price equation are givenin the third and fourth column. Since price, quantity and usage are givenin logs, we can easily interpret the coefficients. As expected, both adcategories have a negative and significant own elasticity.

Higher usage increases the willingness to pay in both the small and


large ad category.In the large ads price equation, the quantity of small ads has a positive

and significant effect on the price. This means that more small ads increasethe willingness to pay for large ads. The amount of large ads has no effecton the price of small ads. The quantity of free ads affects the price of largeads positively.

The fact that internet penetration has a significantly positive effect onthe price of small ads, might be related to the fact that the internet is nota substitute, but a complement for additional information, e.g. it wouldbe of no use to add a mail and website address if no users had internetaccess. As explained above, internet might rather be a substitute for largeads, because it serves the same needs: in depth search for a product orservice.

B5 Robustness Check: SURE

To check for the robustness of our OLS results, we apply seemingly unre-lated regression (SUR, Zellner (1962)). SURE allows for correlated errorsbetween equations that are seemingly unrelated but based on the samedata. As can be seen in table B3, the results of the seemingly unrelatedregression are similar to the OLS regression.

B6 Reconciling the Theoretical Model with Empirical

Findings

With this empirical results in mind, it is interesting to look back to thetheoretical model. The most interesting application of these results comesfrom the usage equation. If we add a third advertisement category tothe theoretical model, say q0 (which is associated with s0 < s1), thenwe get results for estimations of r0, r1 and r2. Point estimates for theseparameters are 0.37, 1.70 and 0.21 respectively25. If we ignore r0 for amoment, the estimates of r1 and r2 can be merged to a r estimate. Sincer = r1

r2, r = 8.2. One can wonder, if ratio is really that large, why is q2

still in the directory? The answer can be seen in figure 3.1. With a highr, the result depends on the value of s. If s is smaller than 0.25, then q1

is offered for free; if s is larger than 0.75, then only q1 is offered. Between

25See table B2 in this appendix.

102 Directories

0.25 and 0.75 both quantities are offered and a positive price occurs in themarket.

Even if no one likes the large advertisements, publishers will still offerthem as long as the difference in views is large enough. If our modelsextends to three categories and the listing is small enough, then it will beoffered for free and the whole market is covered. If the large advertise-ment is substantially larger than the small advertisement, then all threeadvertisements will be offered. In most directories, the free listing is onlyone line, which fits with our prediction that, if platforms can choose thesize difference, they will make it as large as possible.

We don’t have size information for all observations, though rough esti-mations show that the size of the average small advertisement is between0.04 and 0.11 of size of the average large advertisement. This is relativelysmall but still a lot larger than the listing. The average listing is between0.002 and 0.006 of the size of a large advertisement. With the assumptionsof our model, this size difference would lead to a solution where all adver-tisers are included and the smallest advertisement is given away for free.Our model does not incorporate the possibility of a third advertisementtype, which explains why small advertisements are paid in reality.

It is harder to reconcile the estimation of the advertisement equationwith the theoretical model. The reason is that we have not modeled thepotential congestion effect in the directory. We have assumed that thenumber of rival advertisements have no influence on the visibility of theadvertisement, or ds1

dq1= ds2

dq1= ds1

dq2= ds2

dq2= 0. What we do model is

the self-selection effect. Therefore, an increase in the amount of smalladvertisement should not only decrease the price the small ads, but alsoof the large ads. We do find the negative effect of an increased quantity onthe own price, but we do not find an effect on the cross price. A shift inlarge ads has no effect on small ads, while small ads have a positive effecton large ads. The theoretical model falls short in explaining this sign.


Tab

leB

3:R

esul

tsof

OLS

and

SUR

E

OLS

SU

RE

Usage

Pric

eSm

all

Ads

Pric

eLarge

Ads

Usage

Pric

eSm

all

Ads

Pric

eLarge

Ads

Quanti

tyLis

ting

0.3

71

-0.0

23

0.1

37

0.3

71

-0.0

23

0.1

37

(2.9

8)*

**

(0.4

5)

(3.0

2)*

**

(3.4

4)*

**

(0.5

5)

(3.7

0)*

**

Quanti

tySm

all

Ads

1.6

97

-0.8

51

1.7

44

1.6

97

-0.8

51

1.7

44

(1.8

2)*

(2.2

6)*

*(5

.29)*

**

(2.1

0)*

*(2

.77)*

**

(6.4

8)*

**

Quanti

tyLarg

eA

ds

0.2

07

-0.1

55

-1.3

52

0.2

07

-0.1

55

-1.3

52

(0.2

5)

(0.5

7)

(5.7

1)*

**

(0.2

9)

(0.7

)(6

.99)*

**

Inte

rnet

-0.1

09

0.3

73

0-0

.109

0.3

73

0

(0.2

2)

(2.3

5)*

*0

(0.2

6)

(2.8

8)*

**

0

Incom

eper

Capit

a0.1

80.3

46

-0.4

23

0.1

80.3

46

-0.4

23

(0.2

3)

(1.0

7)

(1.5

)(0

.27)

(1.3

2)

(1.8

4)*

Usa

ge

0.2

53

0.1

70.2

53

0.1

7

(3.0

7)*

**

(2.3

5)*

*(3

.76)*

**

(2.8

8)*

**

CPI

-0.0

04

0.0

27

-0.0

04

0.0

27

(0.6

4)

(4.9

2)*

**

(0.7

8)

(6.0

3)*

**

Const

ant

-13.5

29.9

05

-2.5

2-1

3.5

29.9

05

-2.5

2

(1.2

1)

(2.6

3)*

*(0

.76)

(1.4

)(3

.22)*

**

(0.9

3)

Obse

rvati

ons

24

24

24

24

24

24

R-s

quare

d0.7

30.9

60.9

4

Abso

lute

valu

eofz

stati

stic

sin

pare

nth

ese

s

*si

gnifi

cant

at

10%

;**

signifi

cant

at

5%

;***

signifi

cant

at

1%

104 Directories

C Extensions to the Model

In this appendix, we enrich our initial model with some extensions. Theseextensions do not shed light on the main question of this article, i.e.whether a universal service obligation is necessary and desirable. Though,these extensions provide additional insight on the functioning of the YellowPages industry. All extensions are based on the model where pR = 0.

C1 Negative Utility from Advertisements

In section 4.3, we implicitly assumed that r1 and r2 are positive thoughit is likely that in some markets this assumption is violated. Especiallyin media markets, advertisements are often seen as a nuisance (Anderson& Coate (2005) and Peitz & Valletti (2008)). In telephone directories,this is less likely since advertisements contain relevant information forusers, though it might be that some advertisement types do not contributeto usage, on the contrary, they decrease usage (see also the empiricalapplication).

In our model, it should hold that at least one of the listing formatscontributes to usage, i.e. either r1 or r2 is positive. Otherwise there wouldbe no usage on the platform. But the model can perfectly cope with onenegative parameter (either r1 < 0 or r2 < 0).

Proposition C.1 • If r1 < 0 and r2 > 0, then the platform onlyoffers large listings.

• If r2 < 0 and r1 > 0, then the platform always offers small listings.The platform will also offer large advertisements if the size differenceis large enough.

This proposition is visualized in figure C1. In the left panel, below thex-axis, we plotted the case where r1 < 0; in the right panel the case wherer2 < 0. As be seen from the graphs, the border r < s extends to negativevalues of r as well in the case where r1 < 0. Since small listings are lessprofitable and are disliked by users, they do not appear anymore in thedirectory. In the case r2 < 0, things are somewhat more complicated. Theplatform will keep out the large listings only if s is large; more precisely ifr > s

4s−3 (and for all r-values if r < 0 and 34 < s < 1). Small listings will

be offered for free if |r| ≥ s−4s2±2√

s2(1−5s+4s2)

1−4s .

C Extensions to the Model 105

Figure C1: Different Solutions in the r-s-space. Left panel: if r < 0, thenr1 < 0, right panel: if r < 0, then r2 < 0

C2 Endogenizing r (as a function of s1 and s2)

The parameters r and s were treated independently in section 4.3, thoughone can argue that they are related. We argued that users might moreeasily choose a large listing when they compare both types. Though whendeciding on using the directory or not, users might prefer smaller list-ings because these listings use less space for the information provided.Therefore, it makes the directory easier to handle and provides a betteroverview.

We assumed that the number of looks is directly related to size. Onecan argue that the valuations r1 and r2 are also related to the size of alisting. After all, it is exactly the size of the listings that often blurs theinformational value of a listing. Therefore, we can change usage equation3.2 in order to include this nuisance effect of size.

qR = v(q1 + q2)− n(s1q1 + s2q2) (C1)

The first term, with parameter v, captures the taste for variety, i.e. thepure network externality of adding another advertiser to the platform. Thesecond term, with parameter n, captures the nuisance of the thickness ofthe directory. To simplify the analysis, we normalize n to 1. Note that wecan rewrite equation C1 as qR = (v − s1)q1 + (v − s2)q2, i.e. r1 = v − s1

and r2 = v − s2. The ratio r is equal to v−ss2v−s2

. Since s1 < s2, r1 > r2

always holds. Therefore, we can use the right panel of figure C1 to analysethe evolution of the equilibrium r-ratio.

106 Directories

Figure C2: Different solutions in the v-s-space

Proposition C.2 • If v ≤ s1, there will be no platform since thereare no users.

• In all other cases, the platform will offer small listings. If s1 < v ≤4s13 , then the platform offers only small listings.

If v >4s2s2−2

√s2(1−5s+4s2)s22

−1+5s , then the platform implements the

comprehensive solution. If 4s13 < v ≤ 4s2s2−2

√s2(1−5s+4s2)s22

−1+5s , thenthe platform is differentiated.

In figure C2, the outcomes are shown in function of v and s. Thehorizontal line is the size of the large listing s2. Three out of the foursolutions of the model in section 4.3 persist in this adapted model. Inaddition, if the taste for variety is not sufficiently large, then there will beno platform on the market. The claim that the absolute size of the listingsdoes not matter in the choice of the quantities does not hold anymore.

Proposition C.3 If size acts as a nuisance factor to the informationalvalue of the platform, then an increase in s2 will lead to an increase insmall listings and a decrease in large listings.

This proposition holds for solutions 3 and 4 (comprehensive and dif-ferentiated platform). In the case where only one listing format is offered,


Figure C3: Relative prices (p = p1p2

). Left panel: r=0.5, right panel:r=1000

none of the parameters affect the solution. Note that an increase in s2 canalso induce a shift between the solutions and makes it likelier that therewill be no platform in the market.

If size is not seen as a nuisance to the composition, but as a valueadded, the results change drastically. We can write the usage function asqR = v(q1 + q2)+n(s1q1 + s2q2). With n normalized to 1, the r of section4.3 now reads as r = v+s1

v+s2or r = v+ss2

v+s2.

Proposition C.4 If size generates a positive effect on usage, then theplatform always offers both types. If the size difference is large enough,then small listings are offered for free. (If v = 0, then only large listingswill be offered.)

Since r = v+ss2v+s2

and 0 ≤ s < 1, for every combination between v ≥ 0and s2 ≥ 0, it holds that s ≤ r < 1. If we have a look at figure 3.1 again,then we see that the only possible solutions are the comprehensive andthe differentiated solution.

C3 Price discrimination

If a listing of size one costs one euro, then we would expect that a listing oftwice the size would cost less than two euro. If this does not hold, everysender can buy two listings and obtain the same effect. This explainswhy the bundle costs less than the sum of the individual parts. Busse &Rysman (2005) find the existence of this second degree price discriminationin Yellow Pages and link it to competition. If there are more platforms

108 Directories

in the market, then the price discrimination is larger, i.e. large ads arerelatively cheaper vis-a-vis small ads.

Using the model, we can have a look at the price ratio in function ofthe size ratio. We note that the absolute size (s2) has no influence onprices. Prices are only determined by the relative size (s) and the relativevaluation (r).

Proposition C.5 If both types are offered, then large listings always paymore per view than small listings. There is a cross-subsidization fromlarge senders to small senders.

The reason behind this cross-subsidization is, in the case two formatsare offered, that the relative contribution of small listings to usage is largerthan their visibility (r > s).

Figure C3 shows the curvature, under r < 1 (left panel) and r > 1(right panel). On the horizontal axes is the ratio of sizes (s). On thevertical axes, you find p = p1

p2, the ratio of prices. If there is second degree

price discrimination, we would expect a price ratio above the 45 degreeline. In both situations, we observe the reverse: if the size difference ismaximal (s=small), then p1 = 0 and therefore p = 0. The ratio graduallyincreases towards 1 when both sizes are the same.

One of the reasons why we don’t find price discrimination is that weimpose that senders buy at most one listing. Hence it is not possible tosubstitute a large ad by several small ads. Multiple listings are seldomobserved in Yellow Pages, though this does not tell anything about thepossibility. In sum, our model abstracts from the second degree pricediscrimination observed in Yellow Pages. On the contrary, the effects weinvestigate predict relative higher prices for the large ads.

C4 Results for a Ramsey Planner

We contrast the optimal quantities of the private monopoly with a publicmonopoly platform that maximizes welfare under a break-even constraint.Therefore, it takes into account the welfare of receivers and senders. Thewelfare of the receivers is equal to the sum of the net utilities (see equation3.1). The welfare of the senders is also the sum of the net utilities andconsists of two parts: senders buying a small listing and senders buyinga large listing (see equation 3.4). Parameters α and β present the impor-tance of sender welfare versus receiver welfare. The objective function of


Figure C4: Optimal welfare solutions in the r-s-space. Between brackets,the optimal values for q1 and q2 are shown.

the Ramsey planner reads:

W = αWA + βWR (C2)

We impose the same restrictions as in the private monopoly case: q1, q2 ≥0 and q1 + q2 ≤ 1.

Proposition C.6 A welfare maximizing monopolist will always offer itslistings for free and fully cover the market, though it offers only one type.The platform will offer only small listings if r > ρ ≡ s2α+Rβ

ss2α+Rβ ; only largelistings if r ≤ ρ.

Note that if the Ramsey planner cares only about receivers (α = 0),then ρ = 1. If he cares only about senders (β = 0), then ρ = 1

s . If r < 1 orr > 1

s , then there is no conflicting interest between receivers and senders:the optimal solution is the same. But if 1 < r < 1

s , then the Ramseyplanner has to disappoint either the receivers or the senders. This can beseen in figure C4.

If a Ramsey planner can choose the size endogenously, then he choosess = 0 if r > ρ ≡ s2α+Rβ

ss2α+Rβ and s = 1 otherwise. In other words, s = 0 is

110 Directories

implemented if q2 = 1, s = 1 if q1 = 1.

C5 Competition

We assumed the market structure is a monopoly, i.e. there is a singleplatform. We can examine what happens if two or more platforms enterthe market within the context of the framework introduced.

Suppose that receivers singlehome, i.e. they use at most one platformto perform their searches. Competition will maximize the utility of thereceivers. Therefore, platforms will offer only one listing format and giveit away for free. If there exist an infinitesimally small entry cost, then theincumbent monopoly stays a monopoly but it makes no longer profits26.

Proposition C.7 If users singlehome and an infinitesimally small entrycost exist, then the incumbent monopolist becomes a contested monopolist.He will only offer the listing format that receivers like most and will provideit for free.

This proposition is partly driven by the fact that receivers are homoge-nous (except for their opportunity cost). Therefore, they all choose theplatform with the largest gross utility. This result can change if heteroge-neous receivers are introduced.

Note that competition causes the contested monopolist to implementsolutions that are similar to the welfare solutions.

Corollary C.8 If the monopoly is contested, then it implements the op-timal quantities that prevail under welfare optimizations. This does nothold if 1 < r < s2α+Rβ

ss2α+Rβ . In this case, it offers the optimal welfare solutionfor readers only, but not for total welfare.

This corollary can easily be explained by observing figure C4. Sincewe assume that readers singlehome, competition is focused on this side ofthe market. Since readers pay no price, the only option is to maximizewelfare for them, leading the monopolist to implement always the readerwelfare solution. The only conflict zone of reader welfare and total welfareis when 1 < r < s2α+Rβ

ss2α+Rβ . Again, this result can be altered too, e.g. whensinglehoming on the advertisement side is introduced.

26If there is no entry cost, results depend on how receivers decide using a certain

platform or not when utility is exactly the same. If the market is split evenly among

the platforms, there will be several platforms that all offer exactly the same directory.

The argument is similar to the classical analysis of Bertrand competition.


Other Applications An interesting avenue for further research is toapply the framework to other markets. These markets should have thefollowing in common with our telephone directories model. There shouldbe two distinctly different demand sides that value each other’s presence.One side can access the platform under different forms and it is importantthat they can self-select their type. The other side values one type overanother in the direct comparison, but potentially values them differentlyon their contribution to the platform.

One example is a shopping street. Consider a shopping street as aplatform that connects buyers with shopkeepers. Assume now that shopsare differentiated by their display windows. Shops with large windows aremore interesting to shopkeepers because they have a larger chance to bevisited. It is a priori unclear whether buyers prefer small or large windows.Buyers can have a double preference for the size of the windows. On theone hand, a shop with a large window is more attractive than a shop witha small window. On the other hand, if a buyer considers the entiretyof shops, i.e. the shopping street, the large windows might be relativelyannoying because they reduce the overall visibility and make the shoppingstreet considerably longer. Similar to the telephone directories example,the equilibrium number of shops with large and small windows can bedescribed by parameters r and s, at least if there is one platform thatcontrols all the shops in a certain area27.

Another example might be a dating event. Suppose that men andwomen are looking for a partner. Women are not charged. Men arecharged but they can choose out of two types: they can appear at theevent in suit or casual wear. If they compare them directly, women prefermen in suits over casually dressed men. But when they consider to go tothe dating event or not, they might prefer an event with more casual men,because it means that they don’t have to dress formally themselves. Again,the potential difference between direct comparison and contribution tothe entirety of the platform might drive the decision of the platform todifferentiate or not.

With some adaptations, our framework can also be applied to searchengines. Many search engines, such as Google and Bing, allow companiesto buy the highest ranking in the search results. As such, companies have

27While this is not so likely in an older city, it might be true in a shopping mall or

newly developed area. For the implications of monopoly versus dispersed ownership of

shops, see Nocke, Peitz & Stahl (2007).

112 Directories

the choice to appear as a regular entry or a sponsored entry. For thesurfers, the number of regular and sponsored search results determine theuse of a search engine. When they use the engine, it is likely that a surferclicks more on sponsored links (at least on average). Therefore, our resultsmight also hold for search engines.

Note that the model can also be applied to online media. One of thebig questions is whether online media can successfully introduce differentprices, i.e. by charging at least one group of readers for preferential access.This so-called freemium model - a combination of free and premium - hasbeen proposed as one of the solutions for online media. To analyze thismodel within our theoretical model, replace advertisers by readers andreaders by (voluntary, non-paid) bloggers. QR becomes then the numberof blog posts, s1 and s2 can then be interpreted as the amount of blogarticles that can be read by both reader types respectively. Bloggers valuethe amount of exposure. Whether it will be optimal to charge differentprices depends on the parameters of the model.

Further adaptations can be made to ’upgrade’ the model to a realnews platform with paid journalists. In that case, journalists will enter asa cost factor or they will become part of the platform management. Inboth cases, the profit function of the platform will change and thereforethe result will be different from those presented in this article.

D Proofs of Extensions

Proof of Proposition C.1 The rationale behind this proposition issimilar to proposition 3.3.1. It can be seen as an extension or rectificationof this proposition. As in proposition 3.3.1, solution 4 is the unconstrainedoptimum of the problem. The same constraints apply: q1 + q2 ≤ 1 andq1, q2 ≥ 0.

In the case where r2 > 0 and r1 < 0, it suffices to investigate q1 ≤ 0since the platform will never set q2 ≤ 0 or q1 + q2 > 1, even if that wouldbe possible. The constraint q1 ≤ 0 is violated if r ≤ s, also for valuesr < 0. Therefore, the analysis is not altered.

Things are somewhat more complicated in the case where r1 > 0 andr2 < 0. The constraint q1 ≥ 0 will never be violated since this wouldinduce a no-receivers platform. To check whether q1 + q2 ≤ 1 is notviolated in solution 4, we have to investigate in which range the Lagrangeoperator is positive. We find two r-regions where the assumptions are

D Proofs of Extensions 113

Figure D1: Different Solutions in the r − s-space. Left panel: if r < 0,then r1 < 0, r2 > 0, right panel: if r < 0, then r1 > 0, r2 < 0.

violated: r ≥ s−4s2−2√

s2(1−5s+4s2)

1−4s if r > 0 and r ≤ s−4s2+2√

s2(1−5s+4s2)

1−4s

if r < 0. We also have to check whether q2 = 3r3(s−1)−8r2(s−1)s+2s2

6(s−r)(1−s)s(r2+s−2rs)

− 2s3+√

z−r(5s3−3s2−2s+√

z)6(s−r)(1−s)s(r2+s−2rs) ≥ 0 (with z = (r − s)2(s − 1)s(−3r2 + 6rs +

s2− 4s)). This is violated in two r-regions: r > s4s−3 in both cases (r < 0

and r > 0) and for all r-values if r < 0 and 34 < s < 1.

Proof of Proposition C.2 We solve the same model as in section 4.3,though we alter the receiver demand. Since the new receiver equationcan be written in a similar way as the old usage equation, we obtainqualitatively the same results. Though, solution 1, where (q1, q2) = (0, 2

3 )is not an optimal solution anymore. This solution can be replaced bysolution 5, (q1, q2) = (0, 0), which prevents that qR would drop belowzero.

Rewriting the old parameters r1 and r2 as v − s1 and v − s2, allowsinterpreting the solutions in the r − s-space. r is equal to v−ss2

v−s2, i.e. r

is determined by s (in the graph) and v and s2 (which are fixed in thegraph). There are two cases that should be investigated. First, if v > s2,then both quantities are valued positively in the usage equation. Second,if v < s2, then at least one quantity is valued negatively. If s1 < v < s2,then small listings generate a positive effect on readership, large listingsgenerate a negative effect. If v < s1, then both listings generate a negativeeffect and it is optimal for the platform to offer no listings at all.

In figure D1, the optimal solution choice is given for v > s2 (left panel)and v < s2 (right panel). The r(s) corresponding with the s-values is given

114 Directories

by the straight green line. In both cases, the relative value of v versus s2

determines the slope. If v > s2, the larger v, the flatter this line. If v < s2,the larger v, the steeper this line. As can be seen in the left panel, ther determination crosses solution 3 and 4 (for v set twice as large as s2),but if v was smaller, then it would have crossed solution 2 as well. If weknow the value s, then we can immediately derive from the graph whichsolution will prevail. The same holds for the case v < s2. Note that ifr(s) > 0, this indicates that both r1 and r2 are smaller than zero andthat no listings are offered in optimum. If r(s) < 0, then the line alwayscrosses the solution 2, 3 and 4. On the right panel, r(s) is drawn for v

half as large as s2.

Proof of Proposition C.3 In solutions 3 and 4, the optimal quantitiesread:

Solution 3:

q∗1(−2+s)s2+v−

√(1−s+s2)s2

2−(1+s)s2v+v2

3(−1+s)s2

q∗2(−1+2s)s2−v+

√(1−s+s2)s2

2−(1+s)s2v+v2

3(−1+s)s2

Solution 4:

q∗1s(s2−v)v+

√sv2(3v2+s(4s2

2−8s2v+v2))6s(ss2(s2−2v)+v2)

q∗2ss2(2s2−5v)v+3v3−s2

√sv2(3v2+s(4s2

2−8s2v+v2))6v(ss2(s2−2v)+v2)

If we derive these quantities to s2, then dq1ds2

> 0 and dq2ds2

< 0, at leastin the regions where these solutions prevail.

Proof of Proposition C.4 We can draw the same graph as for proposi-tion C.2. The r(s) corresponding with the s-values is given by the straightgreen line. This line always ends in point (1,1). The starting point (0, v

v+s2)

depends on the values of v and s2 and should lie between zero and one.If v = 0, then r = s and we obtain solution 1, i.e. only large listings areoffered (see the proof of proposition 3.3.1. In graph D2, the r(s)-line isshown for v = 2 and s2 = 1.

Proof of Proposition C.5 To check whether the price per view ishigher for large listings, one simply has to check whether p2

s2> p1

s1, for

every solution in the relevant parameter space. The ratios p1s1

and p2s2

D Proofs of Extensions 115

Figure D2: Different solutions in the r-s-space.

are plotted in figure D3. On the left panel, r = 0.5, on the right panelr = 1000. The s-values on the left hand side of the dotted line are therelevant ones, because on the right hand side of this line only one type isoffered in the market.

Proof of Proposition C.6 We solve equation C2 with the same Karush-Kuhn-Tucker conditions as under the private monopolist, i.e. q1, q2 ≥ 0and q1 + q2 ≤ 1. We obtain three groups of solutions: an unconstraint op-timum, two solutions where either q1 or q2 is zero and two solutions whereq1 + q2 = 1. We find further that only the Lagrange operator related tothis last solution is always binding in the region 0 < s < 1. Therefore,(1,0) and (0,1) are the only relevant solutions to our problem.

We obtain the critical r = ρ ≡ s2α+Rβss2α+Rβ by comparing the two welfare

levels, under the first and second solution. Equivalently we can say that aRamsey planner implements solution 1 if s < s2α+Rβ−rRβ

rs2α and solution 2otherwise. The critical values for the monopolist maximizing the welfareof one side of the market only, can be found easily by setting α or β equalto zero.

Proof of Proposition C.7 Suppose that a platform would offer a quan-tity combination that does not maximize receiver welfare. Then another

116 Directories

Figure D3: Price ratios. Left panel: r=0.5, right panel: r=1000

platform can offer a combination that slightly improves receiver welfareand the first platform will have no users and hence no senders. Since thereare no costs in our model, it is always feasible to offer all listings for free.If we introduce costs to our model, then the platform can never offer thesolution where there is only one type and that type is free of charge.

Chapter 4

Metaplatforms.

The Effect of Aggregators

on Content and Profits

118 Aggregators

4.1 Introduction

In this article, we investigate the role of aggregators or metaplatforms.Platforms, in the context of two-sided markets, bring together two dis-tinctly different demands, e.g. buyers and sellers, readers and advertisersor card holders and merchants. Metaplatforms, as we define them, arevehicles that aggregate one side of the market over all platforms, to servethe other side of the market.

To clarify the meaning of metaplatforms, consider the market for on-line real estate classified ads. A traditional real estate platform bringspotential buyers and sellers together, by setting up a website where sell-ers pay to announce their property for sale. Potential buyers might becharged to see these announcements. Metaplatforms collect these sellerannouncements from the different platforms and present them together ontheir own website. Potential buyers can access all announcements and areredirected to the original platform.

A metaplatform creates both opportunities and threats for the tra-ditional platform. In our theoretical model, we show the two adversaryeffects of a metaplatform. On the one hand, a metaplatform increases thenumber of buyers in the web ecosystem, but on the other hand some ofthe revenues related to the buyers are absorbed by the metaplatform, atthe expense of the traditional platforms. Platforms can decide whetherto allow or disallow the metaplatform. Our results show that despite thischoice, platforms do not always take the most profitable decision. Thoughdisallowance of the platform would often be more profitable, both plat-forms will allow the metaplatform, because they don’t want to end up asthe only platform not allowing (while the other does).

Our results are particularly relevant for the current debates on aggre-gators, e.g. on news aggregators such as Google News and real estate ag-gregators such as Funda in the Netherlands and Google Maps in Australia.In these debates, believers and non-believers are diametrically opposed toeach other. Believers are proponents of the so-called link economy, andtherefore also of aggregators. As Jarvis (2008) argues: “Links are the newcurrency of media. Links can be exploited and monetized; get links andyou can grab audience and show ads and make money.” But non-believersfirmly think that aggregators steal more attention than they redistribute.Mishkin (2009) provides evidence that news aggregators retain the ma-jority of the readers. Readers scan the headlines, but rarely click further.

4.1 Introduction 119

The same observation is made by research firm Outsell, as documentedin Wauters (2010). It is clear that proponents of aggregators stress thefirst effect we described, i.e. the market expansion, while the opponentsstress the second effect, i.e. demand shifting in favor of the aggregator.News Corporation, owned by Rupert Murdoch, and the press agenciessuch as Associated Press, are well known as strong opponents of aggrega-tors. Belgian newspapers also tried to restrict the use of links, withoutmuch success (Souffreau, 2010a, 2010b). Whether traditional mediagain or lose from aggregators, remains an open question. It is for sure,though, that aggregators play an important role in the content distribu-tion and therefore also indirectly in the content production (see also Pew(2011)).

Similar discussions are also held in court. In 2006, the Dutch societyof real estate agencies (NVM) took two aggregators to court to stop theaggregating on the base of copyright law. In a preliminary statement, thecourt argued that copying the address, price and 1 to 1.5 rule descriptionare not considered as a copyright infringement, but as a quote. In thisarticle, we focus on the economic effects of metaplatforms. In AppendixB, we provide a discussion on the legal arguments1.

Metaplatforms and aggregators are by no means a brand new concept.For newspapers, aggregators such as Dialog and Lexis Nexis were erectedin 1972 and 1973. The main difference, though, is that they were noreal competitors for newspapers, but served more as an archive than asa direct news medium. But also other real world examples, such as bookstores can be interpreted as metaplatforms. Book stores aggregate booksfrom different publishers, who are platforms themselves since they bringtogether authors and readers.

We like to stress here the importance of IT for the concept of meta-platforms. At least two features give metaplatforms a lot more relevance

1Note that the discussion on the economics of copyright infringements by aggre-

gators is related to the discussion on digital music and movies a couple of years ago.

Similar to the spidered websites, it has been shown that a potential copyright in-

fringement for digital music should not necessary harm music distributors, due to the

sampling effect (Peitz & Waelbroeck 2006). A similar discussion is whether the copying

of pages from books by the Google Books project can be considered as fair use. Travis

(2006) argues that “like the samples on iTunes and similar digital music services, the

primary utility of Google Book Search will be to enable internet users to preview works

about which they lack the information to make a purchasing decision. [...] Google is

salvaging entire libraries full of dusty, crumbling books while creating a highly efficient

marketing platform for authors.”

120 Aggregators

in the internet age: digitalization and hyperlinks. Digitalization makes itmuch more efficient to store and reproduce content. While the metaplat-form described for online real estate classified ads would also be feasiblefor offline classified ads, it would cost a lot more effort to retrieve andreproduce the information. Due to a far-reaching digitalization, the use ofso-called bots, crawlers, harvesters, scrapers or spiders makes it easy to filldatabases with information. A second important, and possibly determin-ing feature of the internet is the existence of hyperlinks. These hyperlinksmake it possible to switch from one website to another; or from platformto platform. The possibility of websites to redirect consumers easily toother platforms might be perceived as a threat, because it is easier to loseconsumers. But linking also increases the value of a website significantly,without having to invest in own content or storage space.

The remainder of this chapter is organized as follows. The next sectionprovides an overview of the literature on aggregators and the economicconcepts that are related to metaplatforms. Section 4.3 describes themodel and the equilibrium results on the effects of a metaplatform. Wealso discuss some extensions to our model. Section 4.4 concludes and shedslight on the insight of our model in the context of the current debates onaggregators.

4.2 Related Literature

In this section, we discuss the two streams of literature that are relatedto our model: network economics and two-sided markets.

As was discussed in the introduction, one of the defining features ofthe internet is the use of hyperlinks. Through these links, a networkis formed. If links are costly to form and maintain, then individuals orfirms will trade off the cost against the expected benefits. Bala & Goyal(2000) investigate how equilibria are formed in social networks. Jackson& Wolinsky (1996) study the stability and efficiency of these networks,also in a non-cooperative game where individuals can form or sever links.Jackson (2003, 2008) provides a survey of the literature.

The article closest to ours is Dellarocas, Katona & Rand (2010). Theyapply these link formation theories to websites, and question whether siteswill link to each other. They show that links can be beneficial by reduc-ing competition and duplication effort. Links do not always form, eventhough they would lead to more content and higher aggregate profits. Ag-

4.2 Related Literature 121

gregators prove to be profitable if they offset the loss in ad revenues byattracting more traffic. Their findings also show that aggregators increasecompetition. When sites cannot link to each other, they react by overin-vesting in content. One way out of this overinvesting is linking to anotherwebsite. As the aggregator shrinks profits, it becomes more interesting tolink to another website (and stop investing in own content). In an exten-sion, they also make the decision to allow the aggregator endogenously.In a market where there is no outside option, sites are always better off ifthey coordinate to refuse links.

From Dellarocas, Katona & Rand (2010), we borrow the search pro-cess of buyers. The set-up of their model deviates in three importantpoints. First, they apply a two-sided market model with readers andsellers (whereas we apply a three-sided model with buyers, sellers and ad-vertisers). The authors treat content as a cost factor (e.g. newspapers),while we perceive content as a revenue side (e.g. real estate platforms).Second, aggregators can at most copy one website (or link to it in theirterminology) whereas in our model the aggregator can copy information ofany website. Strictly speaking, their type of aggregator does not aggregateinformation, but acts as a quality filter for the reader. Third, platformscan also copy (or link) each others information. We abstract from thelatter possibility.2

Despite the difference in assumptions and subsequently the modeling,their results are qualitatively similar to ours. We also find that sellerquantity is increased if the metaplatform is allowed. We also show thatthere is a coordination failure, which is not limited to the case where thereis no market expansion. This shows that the results of Dellarocas, Katona& Rand (2010) also extend to a framework with a real aggregator and amulti-sided market with externalities.

Further economic literature on aggregators is scarce. George & Hogen-dorn (2011) also stress the market expansion effect of an aggregator,through a reduction of the transaction costs of viewers. Chiou & Tucker(2011) exploit a natural experiment to test the impact of Google News

2Another article that allows platforms to link to each other, is Mayzlin & Yoga-

narasimhan (2010). They investigate blogs which can establish links between each

other. Linking increases the value for readers, but might also lead to a loss in read-

ership, on the short term directly because the reader follows the link and on the long

term because credibility and promotion is given to a rival. Whether links are formed,

depends on the magnitude of both effects. Unfortunately, they don’t discuss the pos-

sibility of an aggregator.

122 Aggregators

on news websites. For a short period of time, AP was no longer avail-able on Google News. Chiou & Tucker (2011) find that Google Newsusers were less likely to visit the news websites (compared to Yahoo Newswhere AP was still allowed). Their results suggest that aggregators mightprove useful to attract additional visitors. This is in line with our the-oretical findings, though we should add that increased (indirect) visitorsdo not automatically lead to higher revenues for the platforms, due to thepotential loss in ad revenue.

The two-sided market literature, started with Caillaud & Jullien (2003),Rochet & Tirole (2003) and Armstrong (2006), focuses on how a plat-form gets both sides on board. Many two-sided markets were inspiredby internet platforms, since the internet made different business mod-els plausible, such as pay per transaction (e.g. pay per click advertise-ments) (Caillaud & Jullien 2001). Internet industries that received sub-stantial attention are B2B, B2C and C2C platforms (Yoo, Choudhary &Mukhopadhyay 2007), online publishing (McCabe & Snyder 2010), onlineadvertising (Evans 2008), (open source) software (Economides & Kat-samakas (2006), Gallaugher & Wang (2002)) and net neutrality (Econo-mides & Tag (2008), Hahn & Wallsten (2006)). Not so much has been doneyet on real estate platforms, the example extensively used in this chap-ter. Hendel, Nevo & Ortalo-Magne (2009) apply the matching model ofColes & Muthoo (1998) to real estate marketing platforms. Starting froma stock-and-flow model, they test platform choice empirically by investi-gating a broker’s platform and a ‘for sale by owner’ platform in Madison,Winconsin, (United States). They find that a listing on the more expen-sive platform does not lead to higher prices, but on average houses aresold quicker. Unfortunately, there is no aggregator active in this market.For an overview of the two-sided markets literature and the strategies ap-plied, see Rochet & Tirole (2006), Rysman (2009) and Eisenmann, Parker& Van Alstyne (2006).

Our contribution to the literature is that we complement the litera-ture of two-sided markets with the introduction of a metaplatform, whichremained undiscussed in the recent literature. In the link economy lit-erature, we are one of the first to analyse the effect of aggregators onplatforms and its market sides.

Besides, our model is related to other economic concepts. Our meta-platform can be interpreted as an adapter in the spirit of Katz & Shapiro(1985). Metaplatforms make platforms compatible, since a buyer can

4.2 Related Literature 123

access the platform content through the metaplatform. Pollock (2007)provides an interesting analysis on porting, the conversion of software orservices developed for one platform to run on another. In a sense, thisis also what a metaplatform does. Pollock (2007) gives the platforms thepossibility to make porting more costly for other platforms. A monopolistplatform decides on both prices and porting cost. The author shows that,under certain conditions, this decision on porting has far more impact onconsumer welfare than the pricing decision. To some extent, this is alsowhat we find: for buyers, the decision whether to allow the metaplatformmatters much more than the difference in seller quantity set due to thisdecision. Katz & Shapiro (1985) investigate whether the construction ofan adapter is welfare enhancing. Their results show that unequally sizedfirms don’t have enough incentives to create an adapter while unequallysized firms might have socially excessive incentives. The focus of thisstrand of literature is more on the welfare aspects, while we focus on theimpact for the platforms (and less on the demand sides).

Another related concept is vertical separation. This means that themanufacturer does not sell its products directly to the consumers, butsells to a retailer who sells in turn to consumers3. In two-sided markets, aplatform normally connects one side (e.g. buyers) with the other side (e.g.sellers). Vertical separation can then be seen as two connected platforms,where one platform focuses on one side, whereas the other focuses on theother side. Consider again the market for real estate. A traditional plat-form aims at connecting both buyers and sellers. If a metaplatform comesin the market, it can be seen as vertical separation, since the metaplat-form particularly aims at reaching buyers, while the traditional platformis now focused more on the seller side.

Most scientific articles on aggregators are written either in the field ofinformation technology or law. The former mostly stresses the technicaldetails to build an aggregator, or the impact of spiders on website per-formance (e.g. Boswell (2003)). Legal analysis focuses on the argumentsused by courts (e.g. Jennings & Yates (2009), Fischer (2001)). Furtherdetails on this legal analysis can be found in Appendix B.

3See (Bonanno & Vickers 1988) for the concept of vertical separation. Their analy-

sis, though, is focused on the price effect of vertical separation and therefore the results

are less related to ours.

124 Aggregators

Figure 4.1: Set-Up of the Game

4.3 Theoretical Model

4.3.1 Set-Up of the Model

Our metaplatform model consists essentially of five groups of players: theplatforms, the respective demand sides of these platforms and metaplat-forms. In the context of real estate classified ads, these demand sides arepotential buyers and sellers of houses and advertisers who try to reach thebuyers (e.g. banks who are selling loans). The platform is a website wheresellers can announce their property for sale. The website attracts buyersby collecting and representing this information on sellers. The metaplat-form is also a website that is read by buyers; sellers cannot announcedirectly on the metaplatform, but their content appears on it if the meta-platform is allowed to copy the platform. Advertisers can place their adsboth on platforms and metaplatforms. Our set-up is summarized in figure4.1 and discussed in detail below.

We maintain the real estate classified ads example throughout thechapter since it clarifies our model. It might be clear that our model isnot limited to this particular example. It might be applied to other two-sided markets, such as auction platforms, matching platforms for datingor employment or media platforms in general.


4.3.1.1 Framework without metaplatform

Sellers Sellers are house owners who try to sell their house. Sellers valuethe amount of buyers positively and the membership price negatively.Furthermore, we assume that sellers are multihoming (i.e. joining oneor more platforms); buyers are singlehoming (i.e. joining at most oneplatform). This means that buyers visit only one platform; sellers thatwant to reach these potential buyers can only reach them by announcing onthe platform chosen by these buyers. As such, the platform is a bottleneckfor sellers: and platforms can act as monopolists on the potential buyers.This is a common assumption in two-sided markets (see Armstrong (2006))and in line with the evidence presented by Evans (2008) on search engines.

The utility a seller j derives from platform i is equal to:

USij = αS

ijQBi − pS

i (4.1)

with QBi the number of potential buyers on platform i, pS

i the membershipsellers pay on platform i and αS

ij the value seller j derives from being seenby a potential buyer on platform i. This αS

ij is seller and platform specific.Sellers differ regarding to the chances they have to sell a house, or by theprofits they can make to sell their house. The valuation of a buyer mightalso be different across platforms. Implicitly, we assume here that thetype of information presented on both platforms differs. One platformmight put more emphasis on the location or construction type, while theother platform presents more information on the look (e.g. with pictures).Some sellers value one type of information more than the other, becauseit affects their chances to sell the house. We assume that the valuation forbuyers αS

ij is uniformly distributed over 0 and 1. Therefore, the demandon platform i reads:

QSi = 1− pS

i

QBi

(4.2)

Buyers Buyers visit only one platform and don’t pay. To model how thebuyers choose which platform to visit, we rely on the anchor site trafficspecification of Dellarocas, Katona & Rand (2010). They assume that thedirect viewers are proportional to the utility UB

i a website generates.

QBi =

UBi∑n

k=1 UBk + µ

(4.3)

where 0 ≤ µ ≤ 1 represents the outside option, e.g. searching for houses onthe street, instead of using the platform. As argued by Dellarocas, Katona

126 Aggregators

& Rand (2010), this consumer demand can be the result of a dynamicsearch model. Suppose that new buyers are unaware of the quality of eachplatform and therefore choose randomly. Over time, these buyers canadopt various strategies to find out the value of the platforms, such as amulti-armed bandit exploration-exploitation process. The static formulaabove can be considered as the steady state property of the dynamic game,where a number of new consumers are added each period. One can thinkof new generations that buy a property.

This resembles a logit structure, a discrete choice model where theprobability of choosing product A depends on the relative utility of A

vis-a-vis the sum of the utility of all the products4.We assume that the utility of buyers is equal to:

UBi = αBQS

i (4.4)

αB is not buyer or platform specific and is normalized to 15.

Advertisers Advertisers are, similar to sellers, also looking for the at-tention of the buyers, but for other reasons. We assume that advertiserssell other products than houses. This might in principle be any product(think of advertisers in general newspapers). Logical advertisers in thiscontext are companies that sell goods complementary to houses, such asbanks offering mortgage loans, moving services, notaries or do-it-yourselfshops. We don’t model these advertisers explicitly; though it would notincrease the complexity of the model, it would not add much insight. Onecan see the optimal price pA as the result of a maximization process onthe advertiser market. An increase in pA might be the result of a positiveexternal shock in the size of the advertiser market, or an increase in thewillingness to pay of advertisers. We assume further that buyers and sell-ers are indifferent about the amount of advertisers on the platform. Assuch, this third demand side does not generate externalities on the othertwo demand sides.

Note that we chose for a different modeling approach for sellers, buyersand advertisers. Most two- or multi-sided markets articles treat all sidessymmetric. While a symmetric approach is certainly most apt for general

4One might even extend this framework to a nested logit, with platforms and

metaplatforms in different nests. We abstract from this possibility.5This normalization has no qualitative impact on the results, but simplifies the

equilibrium analysis


models, we prefer a more fine-tuned approach for an applied model. Sellershave complete information about the game, and are allowed to multihome.Buyers, on the contrary, have incomplete information and can only sin-glehome. This structure, imposing that one side singlehomes whereas theother multihomes, is not unusual in two-sided markets. Indeed, if bothwould multihome, it would be possible for one group to stop multihomingwithout losing much (since they are still connected to all multihomers atthe other side). By assuming that buyers do not all choose the largestplatform, we avoid an outcome where all buyers would be attracted to themetaplatform, leaving the platform without direct buyers. Modeling theadvertisers explicitly would not change our results, and would not add toa better understanding; therefore, we assume that pA is the result of amaximization process on the advertiser side.

Platforms The platforms are websites that connect buyers and sellers.Platforms generate revenues on both sides. We assume that platformsare quantity setters on the seller side. Therefore, equation 4.2 should beinterpreted implicitly: pS(QS

i ) = QBi (1 − QS

i ). On the buyer side, theplatform does not gain direct revenues from the buyers since they are notcharged. Platform can also gain revenues by charging advertisers. Theseadvertisers pay for attention time, which is a function of the total timebuyers spend on the website. We assume that the time a buyer spendson a website is proportional to the number of sellers. Therefore, the totalamount of buyer attention tA is equal to the number of buyers that visitthe platform times the time they spend on the website: tAi = QB

i × tQSi ,

where t is the average time a buyers looks at a seller announcement. Forsimplicity, we assume t is equal to one. pA is the price advertisers pay forevery unit of attention time (tA).

As a result, the profit function of the platform reads:

Πi = pSi QS

i + pAQBi QS

i (4.5)

Note that we have normalized costs to zero. The only decision variablein this profit equation is QS

i . To simplify the analysis, we assume thatthere are two platforms active in the market, indexed 1 and 2. We relaxthis assumption in section 4.3.3.

128 Aggregators

4.3.1.2 Framework with metaplatform

The metaplatform is a website that automatically collects the informationof other platforms. Therefore, it might also be labeled as an aggregator orcrawler or spider. It copies the seller information to its own database andcan represent this information on its website. This might provoke a legaldiscussion, because the copying of information might be considered as acopyright infringement. Other arguments against the automatic copyingof information are the hot news doctrine and trespass to chattels (prop-erty infringement). We discuss these arguments in Appendix B. In themodel, we assume that automatic copying is not allowed, and the plat-form should give its approval to the metaplatform to copy the information.This variable ci takes the value 1 if copying is allowed by platform i, and0 otherwise.

The metaplatform competes with normal platforms in the buyer mar-ket and advertiser market, but not in the seller market. Since some buyerswill visit the metaplatform (in stead of the platform), the metaplatformcan also attract advertisers and generate revenues. These potential buyersvisiting the metaplatform are still interesting for platforms. Although theydon’t generate advertiser attention time, they watch the announcementsof sellers, increasing the willingness to pay of the sellers (and therefore theprice). The metaplatform has no own sellers; it relies only on the contentprovided by other platforms – therefore it doesn’t make money on theseller side.

It is important to stress the difference between direct and indirectbuyers. Direct buyers (QB,d

i ) visit the platform directly, while indirectbuyers (QB,i

i ) are redirected by the metaplatform. The number of indirectbuyers depends on the number of buyers that visit the metaplatform, andthe ratio of these buyers who watch the seller announcements of platformi. We assume that this ratio is proportionate to the ratio of sellers of bothplatforms on the metaplatform. Therefore, this ratio (ri) is equal to

ri =

{0 if c1, c2 = 0

ciQSi

c1QS1 +c2QS

2otherwise

An important question is how buyers value sellers that multihome.Sellers that join the two platforms will appear twice on the metaplatform.One can argue that buyers value these sellers only once, since the under-lying property is the same. In the basic model, we assume that buyers


also value the second mentioning of the same seller. Since we implicitlyassumed that both platforms are differentiated in the type of informationthey collect on sellers, the second seller announcement will contain differ-ent information, and will therefore also be valuable for buyers. We discussthis issue further in section 4.3.4.

The number of direct and indirect buyers are equal to:

QB,di =

QSi

(1 + c1)QS1 + (1 + c2)QS

2 + µ(4.6)

QB,ii =

ri(c1QS1 + c2Q

S2 )

(1 + c1)QS1 + (1 + c2)QS

2 + µ(4.7)

The total amount of buyers QBi is

QBi = QB,d

i + QB,ii (4.8)

This formula replaces the definition of QBi in equation 4.3. If c1 and c2

are equal to zero, both definitions coincide. Sellers value both direct andindirect buyers and therefore equation 4.2 remains unchanged. Since theplatform only gains advertisement revenue from direct buyers, we replacethe profit function of the platform (equation 4.5) by6

Πi = pSi QS

i + pAQB,di QS

i (4.9)

The metaplatform (indexed with m) only gains profit from advertisers,and therefore their profit function is equal to

Πm = pAQBmQS

m (4.10)

where QBm = c1QS

1 +c2QS2

(1+c1)QS1 +(1+c2)QS

2 +µand QS

m = c1QS1 + c2Q

S2 . As can be

seen, the metaplatform cannot influence its own profits (since it doesn’tdecide on the amount of sellers on its platform) and therefore it is nottreated as a strategic player. An extension of the model might give themetaplatform tools to behave strategically, such as attracting own sellers,or paying side payments7. In the basic model, we assume there is onlyone metaplatform. This assumption is relaxed in 4.3.3.

6After simplification, the full form of the profit function reads Πi =(pA+(1+ci)(1−QS

i ))(QSi )2

(1+c1)QS1 +(1+c2)QS

2 +µ7We discuss these extensions in sections 4.3.3 and 4.3.4.

130 Aggregators

4.3.1.3 Timing of the Game

Platforms maximize their profits with respect to the quantity of sellers QSi

and the decision to allow copying by the metaplatform ci. Platforms firstdecide simultaneously whether to allow the copying and second decide onthe quantity of sellers. The timing of the game is:

1. Allowance metaplatform (c1, c2)

2. Quantity sellers (QS1 , QS

2 )

In their extension where sites can veto links, Dellarocas, Katona & Rand(2010) apply a reverse timing. Sites first decide how much they will in-vest in content and then decide whether to allow the link source our not.Both approaches are defendable, but it seems more natural to regard theallowance as a slow variable, since this probably implies the implemen-tation of protection software by the website, whereas seller quantity (orcontent in their case) is easily adjustable.

4.3.2 Basic Results

In this section, we examine the optimal seller quantities for a marketsituation with two platforms and one metaplatform. We solve this gameby backward induction, i.e. we first solve for the equilibrium quantities(QS

1 , QS2 ) for given decisions on the allowance (c1, c2).

4.3.2.1 Stage 2: Optimal Seller Quantities

There are four possibilities concerning allowance: both platforms disallowthe metaplatform (c1, c2 = 0), both platforms allow the metaplatform(c1, c2 = 1) or only one of the platforms allows (c1 = 1, c2 = 0 or c1 =0, c2 = 1). We compute the equilibrium quantities in each case.

Both platforms disallow the metaplatform The optimal seller quan-tities when both platforms disallow the metaplatform read:

Lemma 4.3.1

QSi =

110 (3(1 + pA − µ) +

√9((1 + pA)2 + µ2) + 22µ(1 + pA))

if pA ≤ 2+µ3+2µ

1if pA > 2+µ

3+2µ


These solutions are symmetric and unique; both platforms set a positiveamount of seller quantity. If the price for advertisement attention is rela-tively high, then platforms attract all sellers. This means that the impliedseller price goes to zero, i.e. the platform only generates revenues fromadvertisements and not from sellers anymore.

Both platforms allow the metaplatform The optimal seller quanti-ties when both platforms allow the metaplatform read:

Lemma 4.3.2

QSi =

120 (3(2 + pA − µ) +

√9((2 + pA)2 + µ2) + 22µ(2 + pA))

if pA ≤ 4+µ3+µ

1if pA > 4+µ

3+µ

These solutions are quite similar in structure as when both disallow themetaplatform.

One player allows, the other disallows the metaplatform Need-less to say, case 3 and 4 are symmetric and therefore treated as one case.We relegate the equilibrium quantities to Appendix A because they aretoo lengthy to discuss here. The graphical presentation of the solutions infigure 4.2 shows that the equilibrium quantity of platform i, who asym-metrically allows the metaplatform (ci = 1, cj = 0), is close but belowthe equilibrium quantity when both allow the metaplatform. The quan-tity of the platform i, who asymmetrically disallows the metaplatform(ci = 0, cj = 1) is close but above the quantity when both disallow. Theintuition behind this result is that less can be gained from advertisementrevenues if the metaplatform is allowed, and therefore the quantity of adsis lowered to increase the implicit price of sellers. This increases the rev-enues for sellers, since platforms are over-providing seller quantity (com-pared to the case where a platform would only make profits from sellers)because more sellers also means more advertisement revenues (throughthe advertisement attention time). In the asymmetric case, the platformdisallowing the metaplatform will set a seller quantity even higher thanunder the symmetric disallowance case, because it faces a stronger com-petition on the market for buyers and attempts to attract more buyers byincreasing the number of sellers.

132 Aggregators

Figure 4.2: Equilibrium Seller Quantities for All Cases, with pA = 0.25

These results also allow for a comparative statics analysis of quantitiesand profits.

Proposition 4.3.3 A higher µ and lower pA reduce the profits of theplatforms. Increases in µ and pA lead to an increase in the optimal amountof sellers.

This proposition holds for all four cases. A higher µ reduces the marketfor buyers and therefore puts pressure on profits. Moreover, less buyersare attracted for the same amount of sellers, and therefore the numberof sellers is decreased. An increased advertiser demand increases profitsand induces platforms to increase the number of sellers – this lowers theimplicit seller price but is offset by the higher advertiser price.

4.3.2.2 Stage 1: Optimal Allowance Decision

The strategic decision on allowance of copying of seller information canbe represented by a two by two game-theoretic table (two players, twoactions):


Player 2

c2 = 0 c2 = 1Player c1 = 0 Π1|(c1, c2 = 0), Π2|(c1, c2 = 0) Π1|(c1 = 0, c2 = 1), Π2|(c1 = 0, c2 = 1)

1 c1 = 1 Π1|(c1 = 1, c2 = 0), Π2|(c1 = 1, c2 = 0) Π1|(c1, c2 = 1), Π2|(c1, c2 = 1)

We search for equilibria in pure strategies in this game. We summarizethe equilibrium conditions in the following lemma:

Lemma 4.3.4

c1, c2 = 0 is an equilibrium if Πi|(c1, c2 = 0) > Πi|(c1 = 1, c2 = 0)

c1, c2 = 1 is an equilibrium if Πi|(c1, c2 = 1) > Πi|(c1 = 0, c2 = 1)

ci = 1, cj = 0 is an equilibrium if Πi|(ci = 1, cj = 0) > Πi|(ci, cj = 0) and

Πj |(ci = 1, cj = 0) > Πj |(ci, cj = 1)

Based on the equilibria in lemmas 4.3.1 and 4.3.2, we can state thefollowing proposition.

Proposition 4.3.5 For all values of pA and µ, there is always at leastone symmetric equilibrium in pure strategies. If pA is relatively low, bothplatforms allow the metaplatform; if pA is relatively high, both platformsdisallow the metaplatform. For moderate values of pA, both equilibria aresupported. A higher µ increases the likelihood that the metaplatform willbe accepted.

The intuition behind this proposition is straightforward. A high µ

means that a substantial fraction of the buyers is not in the ecosystem, i.e.there is a large potential for the metaplatform to attract new buyers. Thismakes the allowance of a metaplatform more profitable. A high pA meansthat a platform has a lot to lose by allowing the metaplatform. The higherpA, the less likely that the loss in advertisement revenue is compensatedby an increase in seller revenue (through the increased amount of buyers).

Computing the sum of the profits of both player learns that they donot always take the profit maximizing allowance decision.

Proposition 4.3.6 For some values of µ and pA, there is a coordinationproblem. Both platforms allow the metaplatform although a joint disal-lowance would be more profitable.

The proposition stated above is quite similar to the classical prisoner’sdilemma. If both platforms would be able to coordinate their actions,they would disallow the metaplatform whenever (disallow,disallow) this is

134 Aggregators

Figure 4.3: Equilibrium Zones for Allowance Decisions, withEQ=equilibrium PM=profit maximizing

more profitable. Though, even if (disallow, disallow) is the most profitablefor both, it might be more profitable for one player to deviate to (allow,disallow). The reason why they both choose to allow the metaplatform isthat they don’t want to end up as the only platform disallowing.

In figure 4.3, the four equilibrium zones are shown. In the upper andlower part of the graph, the platforms’s equilibrium actions are in line withthe outcome which maximizes profit. In these cases, there is no coordina-tion failure. In the lower intermediate zone, both players choose to allowthe metaplatform, while joint disallowance would be more profitable. Thisis the case described in proposition 4.3.6. In the upper intermediate zone,disallowance is the most profitable, but both equilibria are supported.There might be a coordination problem, but it is also possible that theequilibrium actions are optimal.

The results in lemmata 4.3.1 and 4.3.2 also allow to do a welfare analy-sis of the equilibrium outcome. Therefore, we compute the buyer and sellersurplus, and the profits of platforms and metaplatforms. Total welfare is


the sum of these demand surpluses and the (meta)platforms profit.

Lemma 4.3.7 Buyers are always better off if the metaplatform is allowed.Sellers favor the metaplatform if µ is relatively high and pA is relativelylow. Total welfare is always higher if the metaplatform is allowed.

The allowance of the metaplatform provokes the platforms to set alower seller quantity, though for buyers, this is always compensated by thecreation of an additional outlet to look for properties (i.e. the metaplat-form). The result on the seller surplus is similar to the profit of platforms(which is also higher under the metaplatform if µ is high and pA low).In terms of figure 4.3, the seller surplus equivalence line (where surpluswith or without the metaplatform is equal), is always below the platform’sprofit equivalence line. Based on lemma 4.3.7, we can review the impactof the coordination failure on all market players.

Proposition 4.3.8 The coordination failure to disallow the metaplatformprovokes a welfare transfer from sellers to buyers and from platforms tometaplatforms.

Indeed, since the seller surplus equivalence line is always below theplatform’s profit equivalence line, sellers would even disallow when allow-ing is more profitable and an equilibrium for the platforms. This guar-antees that sellers always lose if the metaplatform is allowed due to thecoordination failure.

4.3.3 Extensions

In this section, we alter the basic framework to investigate interestingextensions to the model. First, we investigate what happens if both plat-forms are jointly owned. Second, we extend our framework to a marketsituation with n (symmetric) platforms and m metaplatforms. We con-clude the section with an analysis of side payments.

Jointly Owned Platforms In the previous section, we analyzed theseller quantity decision of two competing platforms. Suppose now thatthese actions are taken by a joint entity, then we can state the following:

Under joint ownership, the optimal quantities for both platforms read:

136 Aggregators

Lemma 4.3.9

QSi =

18 (2 + 2pA − 3µ +

√(2 + 2pA + 3µ)2 + 8µ(1 + a))

if c1, c2 = 0 and pA ≤ 2+µ2+2µ

1if c1, c2 = 0 and pA > 2+µ

2+2µ116 (4 + 2pA − 3µ +

√(4 + 2pA + 3µ)2 + 8µ(2 + a))

if c1, c2 = 1 and pA ≤ 4+µ2+µ

1if c1, c2 = 1 and pA > 4+µ

2+µ

As can be seen, these results are similar in structure as the duopolycase; the comparative statics for the duopoly case hold here as well. Com-paring the quantities and profits with those from separately owned plat-forms, we can state:

Proposition 4.3.10 If both platforms are jointly owned by one monop-olist, the monopolist lowers the amount of sellers on the platform. Theprofit of the jointly owned platforms is higher than the total profit of theseparately owned platforms.

These results are in line with the standard monopoly models, wherequantity is reduced, but profits increased. Note that the largest gain fromthe joint decision is made when there is a coordination failure. Indeed,the result of a merged decision entity for both platforms is double: first,there is a monopoly profit gain (by setting the monopoly quantity) andsecond, there is a gain from implementing the profit maximizing solution(in terms of the allowance decision)8.

Generalization to n platforms We assumed that there are only twoplatforms active in the market, though reality learns that in many indus-tries multiple platforms are operating. Besides, there might be more thanone metaplatform available in the market. We generalize our model to m

metaplatforms and n platforms. To keep the model tractable, we assumethat firms are symmetric and that platforms decide to either allow allmetaplatforms, or allow none. In this setting, the implicit price for sell-ers and profit function of the platforms remain the same, but the buyersquantity changes to

8The profit equivalence line (where Π|(c1, c2 = 0) = Π|(c1, c2 = 1)) is slightly lower

than in the case where both platforms decide independently


QB,dn,i =

QSi∑n

k (1 + mck)QSk + µ

(4.11)

QB,in,i =

rn,imciQSi∑n

k (1 + mck)QSk + µ

(4.12)

with

rn,i =

{0 if ∀ci : ci = 0

cimQSi∑n

k ckmQSk

otherwise

Assuming that all platforms are symmetric, the equilibrium quantitiesfor all firms are equal to:

Lemma 4.3.11

QSi =

12(3n−1) ((2n − 1)(1 + pA) − 3µ +

√4(2 + 2pA)(3n − 1)µ + (1 + pA − 2n − 2npA + 3µ)2)

if (1)

1 if (2)

(2n−1)(1+m+pA)−3µ+√

8µ(3n−1)(1+pA+m)+(1+pA−2n−2pAn+3µ)2(1+m)(3n−1)

if (3)

1 if (4)

with (1): c1, c2 = 0 and pA ≤ n+µ2n+2µ−1 , (2): c1, c2 = 0 and pA > n+µ

2n+2µ−1 ,

(3): c1, c2 = 1 and pA ≤ (1+m)(n+mn+µ)2(n+mn+µ)−(1+m) and (4): c1, c2 = 1 and pA >

(1+m)(n+mn+µ)2(n+mn+µ)−(1+m) . Note that these general formulas boil down to thequantities of lemmas 4.3.1 and 4.3.2 for m = 1 and n = 2. Computingthe comparative statics of these quantities and profits, we can state thefollowing proposition.

Proposition 4.3.12 The more platforms in the market, the lower theprofits of these platforms. The effect of m on profits is ambiguous. IfpA is low and/or µ is high, more metaplatforms increase the platforms’sprofits. An increase in m decreases the amount of seller quantity, whereasan increase in n increases the seller quantity.

It is not surprising that increased competition (n ↑) decreases profit.n increases competition on the buyer side. As a result, platforms increasethe amount of sellers and profits go down. The effect of m is more nuanced.A higher m increases the impact of the allowance of metaplatforms, andtherefore reduces profits of advertisements, but increases profits from sell-ers (through an increased number of buyers in the ecosystem). Therefore,this revenue shift is the most profitable when the potential of metaplat-forms to increase the amount of buyers in the ecosystem is high (i.e. µ

138 Aggregators

is high) and when revenues from advertisements are relatively low. Notsurprising, platforms decrease the number of sellers as the relative overin-vesting in seller quantity does pay off less because this is less compensatedby advertisement revenues.

There is no equilibrium in the deviating case (i.e. one platform disal-lows when all other platforms allow, or vice versa), therefore it is unclearwhat the optimal ci’s are. It is clear, however, that an increase in m shiftsthe profit equivalence line upwards, whereas n decreases this line.

Side Payments and Bargaining Until now, we assumed that therewas no possibility for the metaplatforms to buy off the copying of theirinformation on the metaplatform. From the previous analysis, we knowthat the existence of metaplatforms can be unprofitable, therefore plat-forms might be tempted to obstruct the metaplatform. The possibilityof side payments might convince these platforms to allow the metaplat-form. Whether side payments can compensate the losses of the platforms,depends on the profit of the metaplatform (see equation 4.10).

Proposition 4.3.13 The total profits in the ecosystem are always higherif a metaplatform can operate. Therefore, if side payments are feasible,then the metaplatform will always be allowed.

Since total profits are higher under a metaplatform, it is always pos-sible to compensate the potential losses of platforms occurred due to theexistence of this metaplatform. This is in line with the analysis of Katz& Shapiro (1985). They also first analyse compatibility as an exogenouscharacteristic of the market. Afterwards, they investigate the private (andsocial) incentives for compatibility. Since total profit is strictly higher ifµ > 0, this additional profit might be divided by a bargaining processbetween the platforms and the metaplatform.

It is interesting to see that Google Fast Flip uses such a compensationscheme. This website shows snap shots from news websites such as NewYork Times, Washington Post and Newsweek. Google Fast Flip showsadvertisements on its website and shares the revenues with the news web-sites (Bharat 2009). Platforms can also ask indirectly for side payments.Legal actions might be inspired by the aim for a settlement on side pay-ments. In the Belgian case Copiepresse vs Google9, Sullivan (2007) arguedthat it was perfectly possible to stop Google accessing the news pages of

9A discussion of this case can be found in Van Asbroeck & Cock (2007).


newspapers by simply putting a robots.txt file on their servers. It is likelythat Google would have complied voluntarily with this request. Sullivan(2007) poses:

This case was never about getting content out. It was abouttrying to blackmail Google into including content. Now thenewspapers may get a large fine coming their way, but whetherGoogle will feel it still wants to cut a deal with them remainsunseen.

Note also that it should not always be the metaplatform who compen-sates the platform for potential losses. It can also be the other way around.If a metaplatform is beneficial to the platform, then the platform can paythe aggregator to be in the market. This is not the case in our model, butit might be if a fixed cost is introduced. If the advertisement market issmall, then the platform profits will be higher if a metaplatform exists, butthe metaplatform will only make moderate profits. But again, to makecompensation possible, total profit with the existence of a metaplatformshould be higher, otherwise there is not enough to redistribute.

Another way to settle side payments is a vertical integration betweenthe metaplatform and one or more platforms. The additional profits aredirectly internalized, but side payments might still be needed for thoseplatforms that do not share the same owner as the metaplatform. Someplatforms have set up their own aggregator, such as Dutch real estateplatform Funda, in reaction to other aggregators. Content providers ofnews, such as NewsCorp, were also examining the launch of an aggregatorbut the project was shelved in October 2010 (Levy 2010).

4.3.4 Future research

Even with our simple model, we were able to gain interesting insights onthe effect of metaplatforms or aggregators. Here we present some exten-sions that might be interesting as well, though we only provide the in-tuition. These extensions concern multihoming, strategic metaplatforms,congestion, pricing and bias.

Multihoming and doublecounting The specification in the previoussections did count a house that was multiple listed twice. This is reason-able, if both platforms contain other information about the properties, e.g.different photos, or present the information in a different way. Though,

140 Aggregators

one can also argue that a house that is multiple listed should be countedonly once. A reasonable specification assumes that α1j and α2j are in-dependently drawn from a uniform distribution between 0 and 1. Thisleads to a multihoming (or double-counting or overlap) of c1c2Q

S1 QS

2 . Ofcourse, when at least one platform disallows the metaplatform, the overlapis zero. The problem though, is that adding this factor to the denominatorof equation 4.7 makes the model untractable. Another set-up is to assumethat the overlap is a constant fraction χ of the total amount of quantityon the metaplatform. This solution is unsatisfactory, but provides insightin the direction of the multihoming correction.

Consider a market with n platforms and 1 metaplatform. Now weintroduce χ(n), the overlap factor, which is equal to

χ(n) =

{0 if ∀ci : ci = 0∑n

k ck−1∑nk ck

χ otherwise

The quantity of indirect buyers is corrected with this overlap factor:

QB,in,i =

rn,ici(1− χ(n))QSi∑n

k (1 + ck(1− χ(n)))QSk + µ

(4.13)

For n = 2, the equilibrium quantities in the case where both allow themetaplatform read

Lemma 4.3.14

QSi =

120(4−χ) (12(2 + pA − µ)

+3(µ + 2)χ√

9(4pA + (µ− 2)(χ− 4))2 + 80µ(χ− 4)(χ− 4− 2pA))

if pA ≤ (4−χ)(4+µ)4(3+µ)

1

if pA >(4−χ)(4+µ)

4(3+µ)

The equilibrium quantities for the other cases remain unchanged; in allother cases χ(n) = 0. The comparative statics show that dQS

i |(c1,c2=1)χ > 0,

i.e. the platform increases its quantity if there is more overlap. As ex-pected, more overlap reduces the profits (dΠi|(c1,c2=1)

χ < 0). The intro-duction of χ also moves down the profit equivalence line. This is logical,since χ decreases the profits when both allow, but doesn’t affect the profitswhen both disallow the metaplatform.

Metaplatforms as strategic players In our model, the metaplatformhad no instruments to influence its own profit. This might not correspond


to reality. Real estate aggregators, such as zoekallehuizen.nl and jaap.nlin the Netherlands, also offer the possibility to add a house to their data-base, without having to buy an announcement on a traditional platform.Moreover, metaplatforms might also have the possibility to refuse to copycontent, or copy only a certain amount of the content of traditional plat-forms. Therefore, future research might consider a specification where themetaplatform can also decide to place its own seller quantity (QS

m) andchoose whether to act as an aggregator: cm is 1 if it copies content fromother platforms, cm is 0 otherwise10. One possibility is to replace thenon-strategic profit function of the metaplatform (equation 4.10) by

Πm = pSmQS

m + pAQB,am QS

m (4.14)

with

pSm = QB,d

m (1−QSm)

QB,am =

c1cmQS1 + c2cmQS

2 + QSm

(1 + c1cm)QS1 + (1 + c2cm)QS

2 + QSm + µ

In the demand function of the buyers, ci is then replaced by ci × cm

and QSm is added to the nominator.

The timing becomes particularly relevant here. A logical timing wouldbe that the metaplatform decides first whether it is willing to copy con-tent (cm). Second, the platforms decide whether to allow the copying bythe metaplatform and finally platforms and metaplatform set the content.Unlike in the case without metaplatform, the game would change substan-tially if the seller quantities are determined before the allowance decisionsare made, for two reasons. First, the metaplatform might strategically setits quantity to influence the allowance decision, and second, the platformsmight strategically set quantity to influence the willingness to copy of themetaplatform.

This blur in the line between metaplatforms and platforms might alsobe particulary relevant to expand our model to other media markets, suchas news websites. An interesting trend is that the most visited newsaggregator, Yahoo!News, has expanded its staff with journalists in 2010.While the number of own stories remained low in 2010 (4 %), it is a relativehuge increase with regards to 2009 where 1% of the stories came from theYahoo staff (Pew 2011).

10This copying might also be continuous, e.g. cm = 0.5 means that the metaplatform

copies 50% of the seller announcements of traditional platforms.

142 Aggregators

Congestion In our model, the utility of buyers (sellers) does not dependon the number of buyers (sellers). This means that there is no congestionor push away effect from other advertisers. It is unlikely that this holdsin reality. We would expect that the chance of a match or sale is reducedby an increase in seller announcements. This congestion effect is widelydocumented in the advertisment literature. As Comanor & Wilson (1974)put it: “To the extent that the advertising of others creates noise in themarket, one must shout louder to be heard, so that the effectiveness ofeach advertising message declines as the aggregate volume of industryadvertising increases.”

Most two-sided markets articles, such as the seminal contribution ofRochet & Tirole (2003) ignore congestion or own-side effects. They as-sume that both sides of the market always interact once, irrespective thenumber of agents on each side. This means that a potential congestioneffect is absent. Other authors put congestion at the heart of their expo-sition. Belleflamme & Toulemonde (2009) investigate how negative intra-group externalities influence the potential of a for-profit platform to entersuccessfully in a market where the two sides are already trading on anon-profit platform. The for-profit platform enters by subsidizing oneside and charging the other11. They show that entry is only possible ifthis externality is intermediate. Church & Gandal (1992) argue that soft-ware providers weigh the network effect (cross-group externality) with thecompetition effect of more agents on their side of the market, generatinga lower chance of being seen or chosen. The same question pops up in theempirical article of Tucker & Zhang (2008): should platforms announcethe number of sellers or not? On the one hand, a high number of sellerssignals that there are also a high number of buyers (cross-group external-ity). On the other hand, it signals harsher competition for these buyers(congestion or own-side effect).

Theoretically such a congestion effect can easily be operationalized inour model by adding a negative own side quantity term to our utilityfunction, e.g. US

ij = αSijQ

Bi −βQS

i − pSi where β measures the disutility of

congestion. The problem though, is that the inclusion makes our modelintractable.

11While most articles do not allow for negative prices, subsidizing one side and

charging the other side is a strategy very well suited for two-sided platforms. Caillaud

& Jullien (2003) discuss these divide & conquer strategies and the efficiency of the

resulting equilibria.


Effects on Pricing and Competition within the Demand SidesRelated to the previous extension, it might be interesting to investigatehow a metaplatform is changing the competition on the demand sides andthe subsequent prices and profits on this side. Rosenfeld (2002) provides afirst attempt to describe these effects. He argues that the likely effect of anaggregator is that prices drop, as was also suggested by the empirical workof Brown & Goolsbee (2002). Nevertheless, in his article Rosenfeld (2002)mainly warns for the potential negative effects of aggregators or spiderson competition. His argument is threefold. First, collusion might preventthe expected price drops. Second, spiders might trigger price wars. Third,other factors might still keep prices above marginal cost. We think thathis claim that “reduced search costs of the internet create as many dangersas they offer remedies” is exaggerated12. Further work needs to be done todisentangle the effects of a metaplatform on prices and competition withinthe demand sides.

Search Results Bias In our analysis we implicitly assumed that everyseller had equal chances of being reached by a buyer. In reality, though,the position in search results matters a lot (Battelle 2007). A platform,but also a metaplatform, might use his power to charge a menu of prices, oreven auction its top positions, as is done by search engines (Varian 2007).Similarly to advertisement bias, buyers are probably worse off becauseof the reduced matching quality, but (meta)platforms can extract morerevenues13. In practice, this type of search results bias often occurs in

12The arguments provided by Rosenfeld (2002) are not fully satisfactory. First, it

is true that lower prices increase the gains from collusion, though it is not clear why

spiders would make collusion likelier. It is more likely that other players enter with

lower prices, since the spider eliminates many of the advertisement-related fixed costs.

Therefore, collusion will be harder to sustain. For the second argument, on price wars,

they cite the article of Greenwald & Kephart (1999). In this article, Greenwald &

Kephart (1999) argue that price wars might result in a framework where sellers have

incomplete information or follow naive strategies. Though, when sellers are rational

and behave game-theoretically, this does not lead to price wars; the arrival of the

spider results in price drops. Third, Rosenfeld (2002) argues that the elimination of

search costs is not sufficient to reduce prices to marginal cost. Product heterogeneity,

consumer heterogeneity, branding and loyalty programs might keep prices high. This

is true, though the question is whether an aggregator makes these factors worse. There

seems to be no reason to assume this is the case.13Note that many matching models would argue that introducing menu pricing

increases the matching quality, e.g. see Damiano & Li (2007). Different prices allow

players to signal a high type, whereas uniform pricing doesn’t. If seller’s profit π is

144 Aggregators

real estate markets. The Dutch platform Funda.nl always puts the housesof its own brokers on top. The content they spider from other brokersis at the end of the search results. The Belgian metaplatform Zimmo.beapplies a similar hierarchical system. On top appear the properties frombrokers which website is developed by MaxImmo, a company affiliatedwith Zimmo. Then are the brokers listed that are affiliated, followed bythe unaffiliated brokers. At the bottom of the list appear the propertiesthat are privately sold (without a broker).

4.4 Conclusion

With the risk of stating the obvious, the internet has changed and trans-formed many businesses; especially those where information plays a bigrole. In this article, we shed light on what we named metaplatforms:platforms that aggregate the content of other platforms.

We argued that the introduction of such an aggregator has two impor-tant opposite effects on a normal platform’s profit. First, a metaplatformattracts and redirects customers that would not have visited the plat-form otherwise. Second, the metaplatform scores off some of the profitsgenerated by these customers, e.g. by absorbing the advertising revenues.

We show that a metaplatform is profitable for normal platforms if thefirst effect dominates the second. This will be the case if two conditionsare fulfilled. First, the outside option for customers should be relativelyimportant, i.e. the entrance of a metaplatform can attract many newbuyers in the ecosystem. Second, the importance of the advertising marketis relatively low, i.e. the platform doesn’t lose a lot when he has to tradein direct customers for redirected customers.

Even if platforms can decide to allow or disallow the metaplatform,they do not always choose the most profitable option. For intermediateadvertisement prices, it would be optimal for both platforms to disallowthe metaplatform, but they decide both to allow the metaplatform. Thisis, similar to the prisoner’s dilemma, the result of a coordination failure.Both have an incentive to deviate from the disallow-disallow strategy,and no platform wants to end up as the only platform not allowing themetaplatform.

If platforms and the metaplatform can bargain over the initiation andallowance of links, then the metaplatform will never be obstructed by the

uncorrelated with house quality, then this signalling is worthless for buyers.

4.4 Conclusion 145

platforms since total profits of all players is higher under the existence ofa metaplatform. In other words, the existence of side payments overcomesthe negative effect that a metaplatform might generate. An increase inthe number of platforms and metaplatforms reduces the profit for eachplatform. An increase in platforms increases the amount of seller quantityset by the platform, the reverse occurs with an increase in metaplatforms.Adding more (meta)platforms does not change the basic insights on themarket expansion and demand shifting effect of a metaplatform.

With this article on metaplatforms, we hope we opened a new line ofresearch on the functioning and effects of metaplatforms or aggregators.Further research might add more sophistication to our model. An inter-esting avenue might be to allow that the metaplatform can also attractsellers, and therefore behave strategically. Second, adding congestion tothe model might make the matching role of the platform more realistically.The chance of a match, e.g. in the housing market, does not only dependon the number of actors on the other side of the market, but also on theown side of the market. Finally, an interesting application of this researchmight be the empirical testing of the hypotheses formed in this article.It would particularly interesting to see whether auctions, matching andmedia platforms fit in the theoretical framework proposed here.

146 Aggregators

Appendix

A Equilibrium quantities

The equilibrium quantities for the asymmetric case can be found onhttp://www.driesdesmet.be/appendix_a.pdf.

B A Legal Discussion on Metaplatforms

Our simple theoretical model allowed to gain insight on the threats andopportunities of metaplatforms. The legal discussion on metaplatforms isoften focused on the question who should be granted the right for links: theplatform or the metaplatform. Our framework provides a partial answer tothis question, because it sheds light on who can gain and who is harmed indifferent circumstances. But apart from these calculations, some generalarguments in favor of platforms and metaplatforms have been developped.

On the one hand, one can argue that the platform should decidewhether to allow the metaplatform copying its content or not. The contentowner can assert its intellectual property rights or argue that the consump-tion of its bandwidth capacity is a trespass to chattels. On the other hand,one can bring forward that the metaplatform should be able to use the in-formation of the platform. From a welfare point, the metaplatform mightenhance competition by reducing search costs. Denying access can be in-terpreted as foreclosure. Besides, one can argue that copying information,with deeplinking, is fair use of information.

We provide these legal arguments below, because they are key to under-stand the past, current and future legal battles on metaplatforms. Theycomplement the insights we derived from the model in the main part ofthis article.

B1 Arguments in favor of the Platform

Intellectual Property Rights Websites contain a lot of elements pro-tected by copyright, such as texts and coding (literary works), photographsand animations (artistic works), music and audio content (musical works)and lay-out (Jennings & Yates 2009). As such, the content of a real estateplatform is protected by these rights.

B A Legal Discussion on Metaplatforms 147

The economic rationale for intellectual property rights is to allow thatthose who invest in products or services can take advantage of it. Sincemost intellectual products can be reproduced at a small marginal cost,companies would have no incentives to invest in future innovations with-out intellectual property rights. This would reduce growth, i.e. reducingwelfare for society. If real estate platforms do not invest anymore in brandfamiliarity, website quality and the collection of additional information,then the likely result is a less informed market, leading to an increase insearch costs. This leads to a dead weight loss in society (Stahl 1989).

Similar to the argument of copyright, some countries also have a data-base regulation in order to protect the content of database holders. TheBritish regulation for example stipulates that “a person infringes the da-tabase right if, without the consent of the owner of the right, he extractsor re-utilizes all, or a substantial part, of the contents of the database”(Jennings & Yates 2009). This argument was used in the German caseStepstone vs Ofir and the French case Cadremploi.fr vs Keljob.

Hot News Doctrine But even if content is not protected by copyright,it can still be prevented from distribution by the so-called Hot News Doc-trine. News gets a specific treatment in US courts since the SupremeCourt’s decision on International News Service vs. Associated Press in191814. The court argued that news or facts are not copyrightable, be-cause these facts are not “the creation of a writer but is a report of mattersthat ordinarily are publici juris; it is the history of the day.” But the courtargued in favor of Associated Press, whose news was copied and rewrittenby INS, saying that this missappropriation of information resulted in un-fair competition. This protection is limited in time. The competitor mightnot engage in distribution of reproduction till the original distributor canvalorize the news facts he has done research on.

The digitalization of news makes it much easier to free-ride on time-sensitive news. As argued by Freedman & Pozza (2010), the Hot NewsDoctrine has taken on new found significance. Recent court decisions aremixed, e.g. in NBA vs. Motorola15 the court rejected the claim of NBAwhile in the case Barclays Capital vs. Theflyonthewall.com16 they ruledin favor of the content provider. The debates between proponents (such

14International News Service vs. Associated Press, 248 US 215 (1918)15National Basketball Assoc. v. Motorola, Inc., 105 F.3d 84116Barclays Capital Inc. v. Theflyonthewall.com, Inc., No. 06 Civ. 4908(DLC), 2010

WL 1005160

148 Aggregators

as AP, New York Times and Time) and opponents (such as Google andTwitter) of the Doctrine are ongoing.

Trespass to Chattels Spidering can be considered a trespass to chat-tels, i.e. the unauthorized use, dispossession or interference with the tan-gible property of another17. While this early common law required aphysical touching of another’s chattel, modern interpretation stipulatesthat indirect touching is sufficient. In recent history, trespass to chattelshas been used in many technological cases, such as telephone lines, radio,television broadcastings and e-mail spam (Fischer 2001).

This claim was used in the case eBay vs Bidder’s Edge and OysterSoftware vs Forms Processing18. Bidder’s Edge collected data on auctionsof eBay and other platforms. While eBay first granted the right to spiderto Bidder’s Edge, after a while eBay asked to cease collecting information.Bidder’s Edge complied, but resumed later. Technical protection measuresof eBay failed, therefore the case ended up in court (see Fischer (2001) fora discussion of the case).

The focus of the case was based on the illegal access of Bidder’s Edgeto eBay’s computer system. The court argued that “eBay’s servers areprivate property, conditional access to which eBay grants the public”.Therefore, Bidder’s Edge has to comply with the terms of use of eBay.

While other interests might have played in the eBay vs Bidder’s Edgecase19, it is a legitimate question of platforms to ask protection of theirbandwidth. If spiders overuse the platform’s servers, then this reduces thewebsite speed, reducing the surfing experience of potential buyers.

B2 Arguments in favor of the Metaplatform

Fair use A counterargument to the intellectual property right argu-ment, is the fair use doctrine. This doctrine, only applicable in the U.S.,says that “the fair use of a copyrighted work for purposes such as crit-icism, comment, news reporting, teaching (including multiple copies forclassroom use), scholarship, or research, is not an infringement of copy-right”20. This is similar to the French ‘droit de citation’ and British ‘right

17Restatement (Second) of Torts § 217.18Oyster Software vs Forms Processing, 2001 WL 1736382 (N.D.Cal. 2001), United

States.19See Fischer (2001) and Cisneros (2007).20US Copyright Act 1976 (USC), 17, section 107

B A Legal Discussion on Metaplatforms 149

of parody’ (Depoorter & Parisi 2002).Gordon (1982) argues that use of a copyrighted work should be allowed

without remuneration, if there is a market failure. This means that thehassle cost of selling or transfering money would be so hard that usagewould be discouraged. In a broader argumentation, one can see fair useas a correction to the copyright law. As the court argued in the caseBill Graham Archives vs Dorling Kindersley, “the ultimate test of fair useshould be whether the copyright law’s goal of promoting the Progress ofScience and useful Arts would be better served by allowing the use thanby preventing it” (quoted in Travis (2006)).

The interpretation of real estate content as Science or Art is a bit farstretched, but the citation right was used in the case NVM vs Zoekalle-huizen.nl21. The Dutch court argued in a preliminary statement thatcopying the address, price and 1 to 1.5 rule description are not consideredas a copyright infringement, but as a quote. The same holds for pictures,given that their size is sufficiently reduced22.

Foreclosure Foreclosure means that a dominant firm denies access to anessential good it produces, with the intent of extending monopoly powerfrom that segment to an adjacent segment (Rey & Tirole 2007). Translatedto our theoretical model, one can argue that the platform forecloses thedirect seller market, in order to extend its monopoly to the broker market.If the metaplatform cannot survive without the content of the platform,then the metaplatform exits (or doesn’t enter) the market. As such, theplatform doesn’t face competition, neither in the brokers market nor inthe buyers market.

Such a monopoly power might reduce welfare and therefore a competi-tion authority will act against foreclosure. Note that metaplatforms haveto prove that the content of the traditional platform is essential, i.e. itcannot easily or cheaply be reproduced. In the metaplatform case, one canargue that it is not so hard for the metaplatform to set up a traditionalplatform as well to attract direct sellers. But in the case of the Dutch caseof Funda.nl (the platform of NVM brokers), it can be argued that it ishard to reproduce the supply of NVM brokers if Funda has an exclusivitycontract with NVM brokers.

21Arrest nr AV5236, Court Arnhem, 16-03-2006. Available on www.rechtspraak.nl.22In the Kelly vs. Ariba Soft Corporation, the court also decided that using com-

pressed pictures by the image search engine was considered as fair use and therefore

allowed (California, US, Case SA CV 99-560).

150 Aggregators

Chapter 5

General Conclusion

152 General Conclusion

In this dissertation, I investigated three related two-sided markets:newspapers, telephone directories and internet platforms. Each of thesemarkets shared the same characteristics. One or more platforms serve twodistinctly different groups, readers and advertisers. Since these demandsides have different utility and profit functions, their interests do not al-ways align. As a result, the platform will trade-off to what extent it willserve (and therefore attract) both sides. This trade-off is made by opti-mizing its own profits. When the newspaper implements advertiser bias,it serves advertisers but it chases away readers. But the existence of theother side of the market is not necessarily bad for the demand side. In thecase of the real estate platforms in chapter 4, the price is lower for sellersbecause of the feedback effect they generate on buyers. More sellers meansmore buyers and therefore the platform lowers the price compared to amarket without these feedback effects. While the framework used sharesmany similarities, in each chapter I tried to solve different questions. Theanswers provided in each chapter are summarized below.

Summary Chapter 2 In chapter 2, I investigated the existence of awall between the advertisement department and the editorial staff of anewspaper. Most scholars agree that the advertising department shouldnot exert pressure on the editorial staff to shape articles. And if they dothe professional ethics of journalists should prevent any direct or indirectinfluence of advertisers on the selection of news. The analogy between achurch-state separation of journalism and commerce is never far away.

Newspaper owners and editors might nevertheless have strong incen-tives to tear down the wall between the advertisement department and theeditorial staff. My theoretical model shows that it is profitable for bothmonopoly and duopoly newspapers to offer a bundle of bias and adver-tisement if the effectiveness of bias is relatively high for advertisers and ifthe annoyance of bias is relatively low for readers. If duopoly newspapersare closer substitutes for readers, then newspapers set less bias; while theeffect on advertisements is ambiguous.

But, as Davies (2008) argues, “to the outsider’s eye, [advertising bias]is very tempting. Advertisers have money, the media outlets need money,so they must be vulnerable to some kind of pressure from the advertisers.It’s a fine theory, but its truth is very limited.” Whether bias will prevailis indeed ultimately an empirical question.

I have tried to answer this empirical question for the Belgian Dutch-

153

language newspapers. I investigated to which extent advertisements affectthe coverage of companies in these newspapers in the period 2001-2005.Depending on the specification, I find that spending 54000 to 212000 euroin a newspaper in one month generates on average one additional article inthis newspaper in the same month. My results appear robust to alternativespecifications.

My results are unsurprising and surprising at the same time. On theone hand, newspapers are profit maximizing business. And as my theo-retical model predicts, pleasing advertisers might generate higher profits.Therefore, it is not unlogical that newspapers insert bias. But on theother hand, newspapers are no normal business. Correctness of infor-mation generates a lot of externalities on society. Many decisions, be itvoting, consumption or investment, are based on information retrievedfrom newspapers. Therefore, newspaper owners and editors should not al-low that advertisers or the advertisement department attempt to influenceeditorial content.

Summary Chapter 3 In chapter 3, I examined the main conditionsof the universal service obligation for telephone directories. These condi-tions are comprehensiveness and availability. Telephone directories shouldcontain all telephone subscribers and should be available to all end-users.I investigate whether one of these conditions is redundant, i.e. if impos-ing one condition also implies that one automatically gets the other. Ifind that imposing a reader price equal to zero (and hence availabilityto all end-users) results in the publishing of a comprehensive directory.Conversely, it is not guaranteed that imposing a comprehensive directoryresults in availability to all end-users. An important caveat to my modelis that costs are ignored and that an exogenously imposed minimum sizecould soften my redundancy result.

Beyond this specific result on universal service regulation, I also showedthat my model can be extended to cover other interesting topics that mightbe playing in the market for telephone directories. In Appendix C, I pro-posed extensions to the basic model to gain additional insights on thefunctioning of the telephone directory market. I relaxed some assump-tions (e.g. on the sign of the valuation of advertisements), endogenizedparameters, and looked at issues such as price discrimination and welfareimplications.


Summary Chapter 4 With the risk of stating the obvious, the internethas changed and transformed many businesses; especially those whereinformation plays a big role. In this article, I shed light on what I namedmetaplatforms: platforms that aggregate the content of other platforms.

I argued that the introduction of such an aggregator has two impor-tant opposite effects on a normal platform’s profit. First, a metaplatformattracts and redirects customers that would not have visited the plat-form otherwise. Second, the metaplatform scores off some of the profitsgenerated by these customers, e.g. by absorbing the advertising revenues.

I show that a metaplatform is profitable for normal platforms if thefirst effect dominates the second. This will be the case if two conditionsare fulfilled. First, the outside option for customers should be relativelyimportant, i.e. the entrance of a metaplatform can attract many newbuyers in the ecosystem. Second, the importance of the advertising marketis relatively low, i.e. the platform doesn’t lose a lot when he has to tradein direct customers for redirected customers.

Even if platforms can decide to allow or disallow the metaplatform,they do not always choose the most profitable option. For intermediateadvertisement prices, it would be optimal for both platforms to disallowthe metaplatform, but they decide both to allow the metaplatform. Thisis, similar to the prisoner’s dilemma, the result of a coordination failure.Both have an incentive to deviate from the disallow-disallow strategy,and no platform wants to end up as the only platform not allowing themetaplatform.

If platforms and the metaplatform can bargain over the initiation andallowance of links, then the metaplatform will never be obstructed by theplatforms since total profits of all players is higher under the existence ofa metaplatform. In other words, the existence of side payments overcomesthe negative effect that a metaplatform might generate. An increase inthe number of platforms and metaplatforms reduces the profit for eachplatform. An increase in platforms increases the amount of seller quantityset by the platform, the reverse occurs with an increase in metaplatforms.Adding more (meta)platforms does not change the basic insights on themarket expansion and demand shifting effect of a metaplatform.

With this article on metaplatforms, I hope I opened a new line ofresearch on the functioning and effects of metaplatforms or aggregators.Further research might add more sophistication to my model. An inter-esting avenue might be to allow that the metaplatform can also attract

155

sellers, and therefore behave strategically. Second, adding congestion tothe model might make the matching role of the platform more realistically.The chance of a match, e.g. in the housing market, does not only dependon the number of actors on the other side of the market, but also on theown side of the market. Finally, an interesting application of this researchmight be the empirical testing of the hypothesis formed in this article.It would particularly interesting to see whether auctions, matching andmedia platforms fit in the theoretical framework proposed here.

Acknowledgements

I would like to thank the many people who supported me through the fiveyears I worked on my PhD.

In the first place, I thank my advisor, Patrick Van Cayseele, for theexcellent guidance in the last five years. I especially liked the inspirationaltalks and the abundance of good research ideas. It is clear that withoutPatrick’s insights and expertise, I wouldn’t have been able to write thisPhD.

Much appreciation goes to the members of committee, Simon Ander-son, Fabrizo Germano, Jo Swinnen and Frank Verboven, for the adviceduring my PhD and on the pre-defence. As one of the members noticedcorrectly, the pre-defence often resembled a brainstorming session andprovoked many valuable suggestions to improve my work.

I am also indebted to Stijn Vanormelingen, my co-author of the secondchapter. Despite the travel distance between Barcelona and Mannheim, wemanaged to work together in an efficient way. For this chapter, I also liketo thank the Vereniging voor Vlaamse Journalisten (VVJ) and Mediargus.Laurens De Koster provided much appreciated research assistance, whichwas partly published as De Koster (2010).

I would like to show my gratitude to the Center of Economic Studies(KU Leuven), Licos (KU Leuven) and ZEW (Mannheim) who providedfinancial support to start and finish this dissertation.

I also take the opportunity to thank the colleagues in Leuven andMannheim of which many became good friends.

With the lunch-partners in crime of the first hour – Thomas, Kristof,Peter and Eline – and those who joined sooner or later – Laura, Annelore,Ellen, Gerd, Thomas, Joris and many others – I held amusing, but never-ending discussions often followed by mails in the afternoon which provedthat Peter was right, again. With my room mates in Leuven – Thomas,

158 Acknowledgements

Alex, Laura and Maryam – and Mannheim – Mario, Matthias and Uli – Idid not only share the office, but also many thoughts, questions and ideas.My fellow MASE students who also joined the PhD programm in Leuven– Thijs and Jo – remained much appreciated sounding boards throughoutthe PhD.

Besides, I would also thank all my other colleagues in Leuven andMannheim. Thanks to An, Andre, Anna, Anna, Annie, Aysegul, Bert,Bram, Catherine, Catherine, Christophe, Damiaan, Dirk, Dirk, Eline, Els,Francine, Frank, Giulia, Heidi, Isabelle, Jacqueline, Jan, Javier, Jeroen,Jeroen, Julien, Karen, Karla, Kobe, Koen, Kris, Kristine, Kristof, Laura,Leonardo, Liesbeth, Lotte, Lydia, Maarten, Marijke, Mark, Mathias, Nico-las, Pablo, Pieter, Priscilla, Rien, Romain, Sandra, Saskia, Sebastian, Si-mon, Sophie, Stef, Tom, Toon, Wim, Wouter, Xavier and Yuemei; and toGeorg, Isabel, Kai, Martin, Nina, Tobias, Ulrich and Vigen.

I am also indebted to those who brought the necessary diversion through-out my PhD: the leaders of the youth movements KSA Ter Straeten andPirlewiet, my squash partner Geert and all the members of the Begijnhofcommittee. I also thank numerous friends for the support, even those whokept on asking “How is your PhD going?” and “When do you get a realjob?”. I also express my gratitude for those who have helped me to learnhow business and journalism work in practice. Many thanks to Ari, Hilde,Paul and Peter of Antwerp by Bike and Bram, Georges, Jan, Tim andTom of Apache.

I am most grateful to my parents, grandparents, brothers, sister, sisters-in-law and parents-in-law who supported me in any respect during thecompletion of this project. Lastly, and most importantly, I wish to thankmy wife Nele. Many thanks for the never-ending support and criticalreflection during my PhD, and thanks to be an inexhaustible source ofwarmth and joy in the past nine years. I dedicate this PhD to her.

About the Author

Dries De Smet studied Economic Sciences at the KU Leuven. Afterwards,he completed a one-year master in Journalism at the Erasmushogeschoolin Brussels and wrote for a short time for the Belgian newspaper De Stan-daard. In 2006, he enrolled for the PhD program and entered the re-search program Master in Science of Economics. His PhD, titled ‘Essayson the economics of media platforms’ investigates media such as newspa-pers, telephone directories and websites as two-sided platforms attractingboth advertisers and readers. Dries works since September 2010 at ZEW(Mannheim, Germany) and is also involved in the online news magazineApache. His main research interests are media economics and competitionpolicy.

160 About the Author

Bibliography

An, S. & Bergen, L. (2007), ‘Advertiser pressure on daily newspapers. a survey of

advertising sales executives’, Journal of Advertising 36(2), 111–121.

Anderson, S. & Gabszewicz, J. (2006), Handbook of the Economics of Art and Culture,

Elsevier Science, Amsterdam, chapter The media and advertising: a tale of two-

sided markets.

Anderson, S. P. & Coate, S. (2005), ‘Market provision of broadcasting: A welfare

analysis’, Review of Economic Studies 72(4), 947–972.

Anderson, S. P. & Gans, J. (2010), Platform siphoning: Ad-avoidance and media con-

tent, CEPR Discussion Papers 7729, C.E.P.R. Discussion Papers.

Argentesi, E. & Filistrucchi, L. (2007), ‘Estimating market power in a two-sided market:

The case of newspapers’, Journal of Applied Econometrics 22(7), 1247–1266.

Armstrong, M. (2006), ‘Competition in two-sided markets’, RAND Journal of Eco-

nomics 37(3), 668–91.

Armstrong, M. (2008), ‘Access pricing, bypass and universal service in post’, Review

of Network Economics 7(2), 172–187.

Armstrong, M. & Weeds, H. (2005), Public service broadcasting in the digital world,

Industrial Organization 0507010, EconWPA.

ASNE (1999), Examining our credibility. Perspectives of the public and the press.,

Technical report, American Society of Newspaper Editors.

Bagdikian, B. H. (2004), The New Media Monopoly, Beacon Press.

Baker, C. E. (1994), Advertising and a Democratic Press, Princeton University Press.

Bala, V. & Goyal, S. (2000), ‘A noncooperative model of network formation’, Econo-

metrica 68(5), 1181–1230.

Baldasty, G. J. (1992), The commercialization of news in the nineteenth century, The

University of Wisconsin Press.

Baron, D. P. (2006), ‘Persistent media bias’, Journal of Public Economics 90(1-2), 1–

36.

Battelle, J. (2007), The Search: how Google and It’s Rivals Rewrote the Rules of

Business and Transformed Our Culture, Portfolio.

Belleflamme, P. & Toulemonde, E. (2009), ‘Negative intra-group externalities in two-

sided markets’, International Economic Review 50(1), 245–272.

162 BIBLIOGRAPHY

Besanko, D., Donnenfeld, S. & White, L. J. (1988), ‘The multiproduct firm, quality

choice, and regulation’, Journal of Industrial Economics 36(4), 411–29.

Bharat, K. (2009), ‘Read news fast with Google fast flip’, The Official Google

Blog (14-09-2009). http://googleblog.blogspot.com/2009/09/read-news-

fast-with-google-fast-flip.html.

Bignon, V. & Miscio, A. (2009), Media bias in financial newspapers: Evidence from

early 20th century france, Technical report.

Blasco, A., Pin, P. & Sobbrio, F. (2010), Paying positive to go negative: Advertisers’

competition and media reports, Working paper, Universita di Bologna.

Bogart, L. (2000), Commercial culture: the media system and the public interest,

Transaction Publishers.

Bonanno, G. & Vickers, J. (1988), ‘Vertical separation’, Journal of Industrial Eco-

nomics 36(3), 257–65.

Boswell, D. (2003), ‘Distributed high-performance web crawlers: A survey of the state

of the art’.

Brown, J. R. & Goolsbee, A. (2002), ‘Does the internet make markets more compet-

itive? Evidence from the life insurance industry’, Journal of Political Economy

110, 481–507.

Busse, M. & Rysman, M. (2005), ‘Competition and price discrimination in yellow pages

advertising’, RAND Journal of Economics 36(2), 378–390.

Caillaud, B. & Jullien, B. (2001), ‘Competing cybermediaries’, European Economic

Review 45(4-6), 797–808.

Caillaud, B. & Jullien, B. (2003), ‘Chicken and egg: Competition among intermediation

service providers’, RAND Journal of Economics 34(2), 309–28.

Chiou, L. & Tucker, C. (2011), ‘News and Online Aggregators’, SSRN eLibrary .

Church, J. & Gandal, N. (1992), ‘Network effects, software provision, and standardiza-

tion’, Journal of Industrial Economics 40(1), 85–103.

Cisneros, O. (2007), ‘Ebay fights spiders on the web’, Wired Magazine (31-07-2000).

http://www.wired.com/politics/law/news/2000/07/37 643.

Claussen, D. S. (2002), American journalism: History, principles, practices, McFar-

land, chapter Economics, business, and financial motivations, pp. 106–115.

Cochez, T. (2010), ‘De post laat niet in zijn kaarten kijken over perssteun’, Apache

(10-08-2010). http://www.apache.be/2010/08/de-post.

Coles, M. G. & Muthoo, A. (1998), ‘Strategic bargaining and competitive bidding in a

dynamic market equilibrium’, Review of Economic Studies 65(2), 235–60.

Comanor, W. S. & Wilson, T. A. (1974), Advertising and Market Power, Harvard

University Press.

Cremer, H., Gasmi, F., Grimaud, A. & Laffont, J.-J. (1998), ‘Universal services: An

economic perspective’, Annals of Public and Cooperative Economics 72(1), 5–43.

Crew, M. & Kleindorfer, P. (2002), Postal and Delivery Services: Pricing, Productivity,

Regulation and Strategy, Kluwer.

BIBLIOGRAPHY 163

Cremer, J. (2000), ‘Network externalities and universal service obligation in the inter-

net’, European Economic Review 44(4-6), 1021–1031.

Cunningham, A. & Haley, E. (2000), ‘A look inside the world of advertising-free pub-

lishing: A case study of ms. magazine’, Journal of Current Issues & Research in

Advertising 22(2), 17–30.

Damiano, E. & Li, H. (2007), ‘Price discrimination and efficient matching’, Economic

Theory 30(2), 243–263.

Damiano, E. & Li, H. (2008), ‘Competing matchmaking’, Journal of the European

Economic Association 6(4), 789–818.

Davies, N. (2008), Flat Earth News. An Award-winning Reporter Exposes Falsehood,

Distortion and Propaganda in the Global Media, Chatto & Windus.

De Donder, P. (2006), ‘Access pricing in the postal sector: theory and simulations’,

Review of industrial organization 28(3), 307–326.

De Koster, L. (2010), An empirical study of commercial media bias in Belgian news-

papers, Master’s thesis, K.U.Leuven.

De Smet, D. & Van Cayseele, P. (2010), Product differentiation on a platform: the in-

formative and persuasive role of advertising, Technical Report 10-03, CES KULeu-

ven.

De Smet, D. & Van Cayseele, P. (2011), The universal service obligation for tele-

phone directories: Regulating the redundant, Technical Report 2011-01, Amster-

dam Center for Law & Economics.

Dellarocas, C., Katona, Z. & Rand, W. (2010), Media, aggregators and the link econ-

omy: Strategic hyperlink formation in content networks, Working Papers 10-13,

NET Institute.

DellaVigna, S. & Gentzkow, M. (2009), Persuasion: Empirical evidence, NBER Work-

ing Papers 15298, National Bureau of Economic Research, Inc.

DeLorme, D. & Fedler, F. (2005), ‘An historical analysis of journalists’ attitudes toward

advertisers and advertising’s influence’, American Journalism 22(2), 7–40.

Depoorter, B. & Parisi, F. (2002), ‘Fair use and copyright protection: a price theory

explanation’, International Review of Law and Economics 21(4), 453–473.

Di Tella, R. & Franceschelli, I. (2009), Government advertising and media coverage of

corruption scandals, NBER Working Papers 15402, National Bureau of Economic

Research, Inc.

Doms, M. & Morin, N. (2004), Consumer sentiment, the economy, and the news media,

Technical report.

Downes, T. & Greenstein, S. (2007), ‘Understanding why universal service obligations

may be unnecessary: The private development of local internet access markets’,

Journal of Urban Economics 62(1), 2–26.

Durante, R. & Knight, B. (2009), Partisan control, media bias, and viewer responses:

Evidence from Berlusconi’s Italy, NBER Working Papers 14762, National Bureau

of Economic Research, Inc.

164 BIBLIOGRAPHY

Economides, N. & Katsamakas, E. (2006), ‘Two-sided competition of proprietary vs.

open source technology platforms and the implications for the software industry’,

Management Science 52(7), 1057–1071.

Economides, N. & Tag, J. (2008), Net neutrality on the internet: A two-sided market

analysis, Working Paper Series 727, Research Institute of Industrial Economics.

Eisenmann, T., Parker, G. & Van Alstyne, M. W. (2006), ‘Strategies for two-sided

markets’, Harvard Business Review p. 12.

Ellman, M. & Germano, F. (2009), ‘What do the papers sell? a model of advertising

and media bias’, Economic Journal 119(537), 680–704.

European Commission (2010), ‘Antitrust: Commission probes allegations of antitrust

violations by Google’, http://europa.eu/rapid/pressReleasesAction.do?ref

erence=IP/10/1624.

Evans, D. S. (2004), ‘The antitrust economics of two-sided markets’, SSRN eLibrary .

Evans, D. S. (2008), ‘The economics of the online advertising industry’, Review of

Network Economics 7(3).

Evans, D. & Schmalensee, R. (2008), ‘Markets with two-sided platforms’, Issues in

Competition Law and Policy 1(28), 667–693.

Fico, F., Lacy, S. & Riffe, D. (2008), ‘A content analysis guide for media economics

scholars’, Journal of Media Economics 21, 114–130.

Fischer, S. (2001), ‘When animals attack: Spiders and internet trespass’, Minnesota

Intellectual Property Review 2(2), 139–182.

Fisman, R. (2010), ‘Industrial espionage. how the CIA got the world to buy American

during the Cold War.’, Slate (24-05-2010).

Fleetwood, B. (1999), ‘The broken wall: How newspapers are selling their credibility

to advertisers.’, Washington Monthly 1(1).

Freedman, J. O. & Pozza, D. C. (2010), ‘Renewed interest in “hot news” misappro-

priation claims against online aggregators of news and information’, Intellectual

Property & Technology Law Journal 22(10), 1–6.

Gallaugher, J. & Wang, Y.-M. (2002), ‘Understanding network effects in software mar-

kets: Evidence from web server pricing’, MIS Quarterly 26(4), 303–327.

Gambaro, M. & Puglisi, R. (2009), What do ads buy? Daily coverage of listed compa-

nies on the italian press, Departemental Working Papers 2009-36, Department of

Economics University of Milan Italy.

Garbacz, C. & Thompson, H. G. (2005), ‘Universal telecommunication service: A world

perspective’, Information Economics and Policy 17(4), 495–512.

Gentzkow, M. & Shapiro, J. M. (2006), ‘Media bias and reputation’, Journal of Political

Economy 114(2), 280–316.

Gentzkow, M. & Shapiro, J. M. (2010), ‘What drives media slant? Evidence from U.S.

daily newspapers’, Econometrica 78(1), 35–71.

George, L. & Hogendorn, C. (2011), Aggregators, search and the economics of new

media institutions, in ‘9th ZEW Conference on the Economics of Information and

Communication’, ZEW.

BIBLIOGRAPHY 165

Germano, F. & Meier, M. (2010), Concentration and self-censorship in commercial

media, Technical Report 1256, Universitat Pompeu Fabra.

Gomes, R. (2010), Mechanism design in two-sided markets: Auctioning users. mimeo.

Goodman, E. (2006), ‘Stealth marketing and editorial integrity’, Texas Law Review

85(1), 83–152.

Gordon, W. (1982), ‘Fair use as market failure: A structural and economic analysis of

the betamax case and its predecessors’, Columbia Law Review 82, 1600.

Graham, S., Cornford, J. & Marvin, S. (1996), ‘The socio-economic benefits of a uni-

versal telephone network: A demand-side view of universal service’, Telecommu-

nications Policy 20(1), 3–10.

Greenwald, A. & Kephart, J. O. (1999), ‘Shopbots and pricebots’.

Groseclose, T. & Milyo, J. (2005), ‘A measure of media bias’, The Quarterly Journal

of Economics 120(4), 1191–1237.

Gurun, U. G. & Butler, A. W. (2010), Don’t believe the hype: Local media slant, local

advertising, and firm value, Technical report, AFA 2010 Atlanta Meetings.

Haan, M. A. & Moraga-Gonzalez, J. L. (2009), Advertising for attention in a consumer

search model, IESE Research Papers D/794, IESE Business School.

Hagiu, A. & Lee, R. S. (2008), ‘Exclusivity and control’, SSRN eLibrary .

Hahn, R. W. & Wallsten, S. (2006), ‘The economics of net neutrality’, The Economists’

Voice 3(6).

Hamilton, J. M. & Krimsky, G. A. (1996), Hold the Press: The Inside Story on News-

papers, Lousiana State University Press.

Hamilton, J. T. (2004), All The News That’s Fit To Sell, Princeton, New Jersey.

Hart, P. (2009), ‘Fear & favor 2008: Financial woes accelerate corporate pressure in

the newsroom’, Extra! .

Hart, P. (2010), ‘Fear & favor 10th annual report: Hidden interference in the news-

room’, Extra! .

Hart, P. & Hollar, J. (2004), ‘Fear & favor 2003 – the fourth annual report: More

examples of media’s vulnerability to power’, Extra! .

Hart, P. & Hollar, J. (2005), ‘Fear & favor 2004 – the fifth annual report: How power

shapes the news’, Extra! .

Hayakawa, S. (1940), Language in thought and action, Harcourt, Brace & Company.

Hays, R. G. & Reisner, A. E. (1990), ‘Feeling the heat from advertisers: Farm magazine

writers and ethical pressures’, Journalism Quarterly 67(4), 936–942.

Hendel, I., Nevo, A. & Ortalo-Magne, F. (2009), ‘The relative performance of real

estate marketing platforms: Mls versus fsbomadison.com’, American Economic

Review 99(5), 1878–98.

Hollar, J., Jackson, J. & Goldstein, H. (2006), ‘Fear & favor 2005 – the sixth annual

report: Outside (and inside) influence on the news’, Extra! .

Hollings, J., Lealand, G., Samson, A. & Tilley, E. (2007), ‘The big NZ journalism

survey: underpaid, under-trained, under-resourced, unsure about the future – but

still idealistic’, Pacific Journalism Review 13(2), 175–197.

166 BIBLIOGRAPHY

Jaag, C., Koller, M. & Trinkner, U. (2008), Calculating the cost of the universal ser-

vice obligation: The need for a global approach, Working Papers 0010, Swiss

Economics.

Jackson, J. (2007), ‘Fear & favor 2006 – the seventh annual report: Encroachment

without apology’, Extra! .

Jackson, J. & Hart, P. (2001), ‘Fear & favor 2000 – the first annual report: How power

shapes the news’, Extra! .

Jackson, J. & Hart, P. (2002), ‘Fear & favor 2001 – the second annual report: How

power shapes the news’, Extra! .

Jackson, J., Hart, P. & Coen, R. (2003), ‘Fear & favor 2002 – the third annual report:

How power shapes the news’, Extra! .

Jackson, M. (2008), Social and Economic Networks, Princeton University Press.

Jackson, M. O. (2003), A survey of models of network formation: Stability and effi-

ciency, Game Theory and Information 0303011, EconWPA.

Jackson, M. O. & Wolinsky, A. (1996), ‘A strategic model of social and economic

networks’, Journal of Economic Theory 71(1), 44–74.

Jarvis, J. (2008), ‘The link economy v. the content economy’, BuzzMachine (18-06-

2008). http://www.buzzmachine.com/2008/06/18/the-link-economy-v-the-cont

ent-economy/.

Jennings, F. & Yates, J. (2009), ‘Scrapping over data: are the data scrapers’ days

numbered?’, Journal of Intellectual Property Law & Practice 4, 120–129.

Kaiser, U. & Song, M. (2009), ‘Do media consumers really dislike advertising? An em-

pirical assessment of the role of advertising in print media markets’, International

Journal of Industrial Organization 27(2), 292–301.

Kaiser, U. & Wright, J. (2006), ‘Price structure in two-sided markets: Evidence from

the magazine industry’, International Journal of Industrial Organization 24(1), 1

– 28.

Katz, M. L. & Shapiro, C. (1985), ‘Network externalities, competition, and compati-

bility’, American Economic Review 75(3), 424–40.

Kelsey (2009), Global yellow pages 2009-2010, Technical report, Kelsey Group.

Levy, K. (2010), ‘News corp dumps news aggregator plan’, Campaign 42, 1–1.

Lott Jr., J. R. & Hassett, K. A. (2004), ‘Is newspaper coverage of economic events

politically biased?’, http://www.aei.org/docLib/20040913_588453[1].pdf.

Maskin, E. & Riley, J. (1984), ‘Monopoly with incomplete information’, RAND Journal

of Economics 15(2), 171–196.

Mayer, M. (2010), ‘Do not neutralise the web’s endless search’, Financial Times (14-

07-2010).

Mayzlin, D. & Yoganarasimhan, H. (2010), Link to success: How blogs build an audi-

ence by promoting rivals, Technical report, Yale School of Management.

McCabe, M. & Snyder, C. (2010), The economics of open-access journals, Working

Papers 914525, SSRN.

BIBLIOGRAPHY 167

McClung Lee, A. (1937), The Daily Newspaper in America, Macmillan, New York.

McCluskey, J. J. & Swinnen, J. F. M. (2010), The Oxford Handbook of Business

and Government, Oxford Handbooks, chapter Media Economics and the Polit-

ical Economy of Information.

Milgrom, P. & Roberts, J. (1982), ‘Limit pricing and entry under incomplete informa-

tion: An equilibrium analysis’, Econometrica 50(2), 443–459.

Mirabel, F. & Poudou, J.-C. (2004), ‘Mechanisms of funding for universal service obli-

gations: the electricity case’, Energy Economics 26(5), 801–823.

Mishkin, A. (2009), ‘The fallacy of the link economy’, paidContent.org (13-08-2009).

http://paidcontent.org/article/419-the-fallacy-of -the-link-economy/.

Mullainathan, S. & Shleifer, A. (2005), ‘The market for news’, American Economic

Review 95(4), 1031–1053.

Mussa, M. & Rosen, S. (1978), ‘Monopoly and product quality’, Journal of Economic

Theory 18, 301–17.

New York Times (2010), ‘The Google algorithm’, The New York Times (14-07-2010).

http://www.nytimes.com/2010/07/15/opinion/15thu3.html.

Nocke, V., Peitz, M. & Stahl, K. (2007), ‘Platform ownership’, Journal of the European

Economic Association 5(6), 1130–1160.

Parker, G. & Van Alstyne, M. (2005), ‘Two-sided network effects: A theory of infor-

mation product design’, Management Science 51(10).

Peitz, M. & Valletti, T. M. (2008), ‘Content and advertising in the media: Pay-tv versus

free-to-air’, International Journal of Industrial Organization 26(4), 949–965.

Peitz, M. & Waelbroeck, P. (2006), ‘Why the music industry may gain from free down-

loading – the role of sampling’, International Journal of Industrial Organization

24(5), 907–913.

Petrova, M. (2009), Newspapers and parties: How advertising revenues created an

independent press, Working Papers w0131, Center for Economic and Financial

Research (CEFIR).

Pew (2000), ‘Self censorship: How often and why’,

http://people-press.org/report/39/.

Pew (2004), ‘Bottom-line pressures now hurting coverage, say journalists’,


Pew (2008), ‘Financial woes now overshadow all other concerns for journalists’,


Pew (2010), ‘State of the media 2010: News leaders and the

future’, http://www.stateofthemedia.org/2010/specialreport

s_survey_executives.php.

Pew (2011), ‘State of the news media 2011: An annual report on American journalism’,

http://stateofthemedia.org/2011/.

Pollock, R. (2007), The control of porting in two-sided markets, Cambridge Working

Papers in Economics 0754, Faculty of Economics, University of Cambridge.

168 BIBLIOGRAPHY

Ppress (2010), ‘Onderzoek naar de opleidingsbehoeften van magazine- en vrije pers

journalisten’, mimeo.

Price, C. J. (2003), ‘Interfering owners or meddling advertisers: How network televi-

sion news correspondents feel about ownership and advertisers influence on news

stories’, The Journal of Media Economics 16(3), 175–188.

Quetelet, A. (1846), Letters addressed to H.R.H. the Grand Duke of Saxe Coburg and

Gotha, on the Theory of Probabilities, Charles & Edwin Layton.

Raad voor de Journalistiek (2010), ‘Code van de Raad voor de Journalistiek’,

http://www.rvdj.be/sites/default/files/pdf/code20 10.pdf.

Reuter, J. (2009), ‘Does advertising bias product reviews? An analysis of wine ratings’,

Journal of Wine Economics 4(3), 125–151.

Reuter, J. & Zitzewitz, E. (2006), ‘Do ads influence editors? Advertising and bias in

the financial media’, The Quarterly Journal of Economics 121(1), 197–227.

Rey, P. & Tirole, J. (2007), A Primer on Foreclosure, Vol. 3 of Handbook of Industrial

Organization, Elsevier, chapter 33, pp. 2145–2220.

Rinallo, D. & Basuroy, S. (2009), ‘Does advertising spending influence media coverage

of the advertiser?’, Journal of Marketing 73(6), 33–46.

Riordan, M. (2001), Handbook of Telecommunications Economics, Elsevier Science,

chapter Universal Residential Telephone Service.

Rochet, J.-C. & Tirole, J. (2003), ‘Platform competition in two-sided markets’, Journal

of the European Economic Association 1(4), 990–1029.

Rochet, J.-C. & Tirole, J. (2006), ‘Two-sided markets: A progress report’, RAND

Journal of Economics 37(3), 645–667.

Rodriguez, F. & Storer, D. (2000), ‘Alternative approaches to estimating the cost of

the uso in posts’, Information Economics and Policy 12(3), 285 – 299.

Rosenfeld, J. (2002), ‘Spiders and crawlers and bots, oh my: The economic efficiency

and public policy of online contracts that restrict data collection’, Stanford Tech-

nology Law Review 3.

Roy, A. & Chattopadhyay, S. (2010), ‘Stealth marketing as a strategy’, Business Hori-

zons 53(1), 69–79.

Rysman, M. (2004), ‘Competition between networks: A study of the market for yellow

pages’, Review of Economic Studies 71(2), 483–512.

Rysman, M. (2009), ‘The economics of two-sided markets’, Journal of Economic Per-

spectives 23(3), 125–43.

Seldes, G. (1943), The Facts Are... A Guide to Falshood and Propaganda in the Press

and Radio, 5 edn, In Fact.

Singhal, A. (2011), ‘Supporting choice, ensuring economic opportu-

nity’, http://googleblog.blogspot.com/2011/06/supporting-choice-

ensuring-economic.html.

Sobbrio, F. (2009), A citizens-editors model of news media, MPRA Paper 18213, Uni-

versity Library of Munich, Germany.

BIBLIOGRAPHY 169

Soley, L. C. & Craig, R. L. (1992), ‘Advertising pressures on newspapers: A survey’,

Journal of Advertising 21(4), 1–10.

Souffreau, B. (2010a), ‘Krantenuitgevers snappen het internet niet’, Apache.be (12-07-

2010). http://www.apache.be/2010/07/krantenuitgevers-snappen-het-interne

t-niet/.

Souffreau, B. (2010b), ‘Reprocopy past richtlijn dieplinken aan’, Apache.be (23-09-

2010). http://www.apache.be/2010/09/reprocopy-past-richtlijn-dieplinken

-aan/.

Spence, M. (1976), ‘Product differentiation and welfare’, American Economic Review

66(2), 407–14.

Stahl, D. (1989), ‘Oligopolistic pricing with sequential consumer search’, American

Economic Review 79(4), 700–712.

Suen, W. (2004), ‘The self-perpetuation of biased beliefs’, Economic Journal

114(495), 377–396.

Sullivan, D. (2007), ‘Google loses in Belgium newspaper case’, SearchEngineLand.com

(13-02-2007). http://searchengineland.com/google-loses-in-belgi

um-newspaper-case-10500.

Travis, H. (2006), ‘Google book search and fair use: iTunes for authors, or Napster for

books?’, University of Miami Law Review 61, 601–681.

Tsetsura, K. (2005), Bribery for news coverage: Research in Poland., Technical report,

Institute for Public Relations Online: International Research.

Tucker, C. & Zhang, J. (2008), Decomposing the congestion effect and the cross-

platform effect in two-sided networks: A field experiment, Technical Report 8-12,

NET Institute.

Upshaw, J., Chernov, G. & Koranda, D. (2007), ‘Telling more than news: Commercial

influence in local television stations’, Electronic News 1(2), 67–87.

Valletti, T. M., Hoernig, S. & Barros, P. P. (2002), ‘Universal service and entry: The

role of uniform pricing and coverage constraints’, Journal of Regulatory Economics

21(2), 169–90.

Van Asbroeck, B. & Cock, M. (2007), ‘Belgian newspapers v Google news: 2-0’, Journal

of Intellectual Property Law & Practice 2(7), 463–466.

van Caspel, M., Moerman, B., Vermeer, L. & Hermans, E. (2002), Eindrapport. uni-

versele telecomdiensten. marktconsultatie vraagzijde en aanbodzijde, Technical

report, Ministerie van Economische Zaken.

Van Cayseele, P. & Vanormelingen, S. (2009), Prices and network effects in two-sided

markets: the Belgian newspaper industry, Technical Report 1404392, SSRN.

Varian, H. R. (2007), ‘Position auctions’, International Journal of Industrial Organi-

zation 25(6), 1163–1178.

Verlinden, A., Billiet, J., Smedts, D., Pyck, H., Page, H. & Van de

Velde, M. C. (2005), ‘Doctoreren in vlaanderen. verslag van de sur-

vey aan de universiteit gent en de katholieke universiteit leuven’,

http://www.ugent.be/nl/onderzoek/doctoreren/visie/rapportdoctoreren/sur

vey2006.pdf.

170 BIBLIOGRAPHY

Viecens, M. F. (2006), Two-sided platforms with endogenous quality differentiation,

Economics Working Papers we061204, Universidad Carlos III, Departamento de

Economıa.

Warsh, D. (2007), ‘Everything you wanted to know (but were afraid to ask) about

two-sided markets’, EconomicPrincipals.com.

Wauters, R. (2010), ‘Report: 44 percent of Google news visitors scan headlines, don’t

click through’, TechCrunch (19-01-2010).

Weller, D. (1999), ‘Auctions for universal service obligations’, Telecommunications Pol-

icy 23(9), 645–674.

Weyl, E. G. (2010), ‘A price theory of multi-sided platforms’, American Economic

Review 100(4), 1642–72.

Wilbur, K. C. (2008a), ‘How the digital video recorder changes traditional television

advertising’, Journal of Advertising 37(1), 143–149.

Wilbur, K. C. (2008b), ‘A two-sided, empirical model of television advertising and

viewing markets’, Marketing Science 27(3), 356–378.

Williams, S. (2009), Measuring “company A”. A case study and critique of a news

media content analysis program, Technical report, Institute for Public Relations.

Williams, W. S. (1992), ‘For sale! real estate advertising & editorial decisions about

real estate news’, Newspaper Research Journal 13, 160–169.

Wood, M. L. M., Nelson, M. R., Cho, J. & Yaros, R. A. (2004), ‘Tonight’s top

story: Commercial content in television news’, Journalism & Mass Communi-

cation Quarterly 81(4), 807–823.

Xavier, P. (2003), ‘Should broadband be part of universal service obligations?’, info

5(1), 8–25.

Yoo, B., Choudhary, V. & Mukhopadhyay, T. (2007), ‘Electronic b2b marketplaces

with different ownership structures’, Management Science 53(6), 952–961.

Zellner, A. (1962), ‘An efficient method of estimating seemingly unrelated regression

equations and tests for aggregation bias’, Journal of the American Statistical

Association 57, 348–368.

Zitzewitz, E. (2009), ‘Forensic economics’, http://www.dartmouth.edu/~ericz/forensi

c.pdf.

Doctoral dissertations

from the Faculty of

Business and Economics

(from August 1, 1971 till July 4, 2011)For a more recent list, see http://www.kuleuven.be/doctoraatsverdediging/.

1. GEPTS, Stefaan Stability and efficiency of resource allocation processes in discrete commodity spaces,1971.2. PEETERS, Theo Determinanten van de internationale handel in fabrikaten, 1971.3. VAN LOOY, Wim Personeelsopleiding: een onderzoek naar investeringsaspekten van opleiding, 1971.4. THARAKAN, Mathew Indian exports to the European community: problems and prospects, 1972.5. HERROELEN, Willy Heuristische programmatie: methodologische benadering en praktische toepassingop complexe combinatorische problemen, 1972.6. VANDENBULCKE, Jacques De studie en de evaluatie van data-organisatiemethodes en data-zoekmethodes,1973.7. PENNYCUICK, Roy A. The economics of the ecological syndrome, 1973.8. KAWATA, T. Bualum Formation du capital d’origine belge, dette publique et strategie du developpementau Zaire, 1973.9. DONCKELS, Rik Doelmatige orientering van de sectorale subsidiepolitiek in Belgie: een theoretischonderzoek met empirische toetsing, 1974.10. VERHELST, Maurice Contribution to the analysis of organizational information systems and theirfinancial benefits, 1974.11. CLEMEUR, Hugo Enkele verzekeringstechnische vraagstukken in het licht van de nutstheorie, 1974.12. HEYVAERT, Edward De ontwikkeling van de moderne bank- en krediettechniek tijdens de zestiende enzeventiende eeuw in Europa en te Amsterdam in het bijzonder, 1975.13. VERTONGHEN, Robert Investeringscriteria voor publieke investeringen: het uitwerken van een opera-tionele theorie met een toepassing op de verkeersinfrastructuur, 1975.14. Niet toegekend.15. VANOVERBEKE, Lieven Microeconomisch onderzoek van de sectoriele arbeidsmobiliteit, 1975.16. DAEMS, Herman The holding company: essays on financial intermediation, concentration and capitalmarket imperfections in the Belgian economy, 1975.17. VAN ROMPUY, Eric Groot-Brittannie en de Europese monetaire integratie: een onderzoek naar degevolgen van de Britse toetreding op de geplande Europese monetaire unie. Leuven, Acco, 1975. XIII, 222pp.18. MOESEN, Wim Het beheer van de staatsschuld en de termijnstructuur van de intrestvoeten met eentoepassing voor Belgie, 1975.19. LAMBRECHT, Marc Capacity constrained multi-facility dynamic lot-size problem, 1976.20. RAYMAECKERS, Erik De mens in de onderneming en de theorie van het producenten-gedrag: eenbijdrage tot transdisciplinaire analyse, 1976.21. TEJANO, Albert Econometric and input-output models in development planning: the case of thePhilippines, 1976.22. MARTENS, Bernard Prijsbeleid en inflatie met een toepassing op Belgie, 1977.23. VERHEIRSTRAETEN, Albert Geld, krediet en intrest in de Belgische financiele sector, 1977.24. GHEYSSENS, Lieven International diversification through the government bond market: a risk-returnanalysis, 1977.25. LEFEBVRE, Chris Boekhoudkundige verwerking en financiele verslaggeving van huurkooptransactiesen verkopen op afbetaling bij ondernemingen die aan consumenten verkopen, 1977.26. KESENNE, Stefan Tijdsallocatie en vrijetijdsbesteding: een econometrisch onderzoek, 1978.

172 Dissertations

27. VAN HERCK, Gustaaf Aspecten van optimaal bedrijfsbeleid volgens het marktwaardecriterium: eenrisico-rendementsanalyse, 1978.28. VAN POECK, Andre World price trends and price and wage development in Belgium: an investigationinto the relevance of the Scandinavian model of inflation for Belgium, 1979.29. VOS, Herman De industriele technologieverwerving in Brazilie: een analyse, 1978.30. DOMBRECHT, Michel Financial markets, employment and prices in open economies, 1979.31. DE PRIL, Nelson Bijdrage tot de actuariele studie van het bonus-malussysteem, 1979.32. CARRIN, Guy Economic aspects of social security: a public economics approach, 1979.33. REGIDOR, Baldomero An empirical investigation of the distribution of stock-market prices and weak-form efficiency of the Brussels stock exchange, 1979.34. DE GROOT, Roger Ongelijkheden voor stop loss premies gebaseerd op E.T. systemen in het kader vande veralgemeende convexe analyse, 1979.35. CEYSSENS, Martin On the peak load problem in the presence of rationizing by waiting, 1979.36. ABDUL RAZK ABDUL Mixed enterprise in Malaysia: the case study of joint venture between Malysianpublic corporations and foreign enterprises, 1979.37. DE BRUYNE, Guido Coordination of economic policy: a game-theoretic approach, 1980.38. KELLES, Gerard Demand, supply, price change and trading volume on financial markets of thematching-order type. = Vraag, aanbod, koersontwikkeling en omzet op financiele markten van het Eu-ropese type, 1980.39. VAN EECKHOUDT, Marc De invloed van de looptijd, de coupon en de verwachte inflatie op hetopbrengstverloop van vastrentende financiele activa, 1980.40. SERCU, Piet Mean-variance asset pricing with deviations from purchasing power parity, 1981.41. DEQUAE, Marie-Gemma Inflatie, belastingsysteem en waarde van de onderneming, 1981.42. BRENNAN, John An empirical investigation of Belgian price regulation by prior notification: 1975 -1979 - 1982, 1982.43. COLLA, Annie Een econometrische analyse van ziekenhuiszorgen, 1982.44. Niet toegekend.45. SCHOKKAERT, Eric Modelling consumer preference formation, 1982.46. DEGADT, Jan Specificatie van een econometrisch model voor vervuilingsproblemen met proeven vantoepassing op de waterverontreiniging in Belgie, 1982.47. LANJONG, Mohammad Nasir A study of market efficiency and risk-return relationships in the Malaysiancapital market, 1983.48. PROOST, Stef De allocatie van lokale publieke goederen in een economie met een centrale overheid enlokale overheden, 1983.49. VAN HULLE, Cynthia ( /08/83) Shareholders’ unanimity and optimal corporate decision making inimperfect capital markets, 1983.50. VAN WOUWE, Martine (2/12/83) Ordening van risico’s met toepassing op de berekening van ultiemeruınekansen, 1983.51. D’ALCANTARA, Gonzague (15/12/83) SERENA: a macroeconomic sectoral regional and national ac-count econometric model for the Belgian economy, 1983.52. D’HAVE, Piet (24/02/84) De vraag naar geld in Belgie, 1984.53. MAES, Ivo (16/03/84) The contribution of J.R. Hicks to macro-economic and monetary theory, 1984.54. SUBIANTO, Bambang (13/09/84) A study of the effects of specific taxes and subsidies on a firms’ R& D investment plan, 1984.55. SLEUWAEGEN, Leo (26/10/84) Location and investment decisions by multinational enterprises inBelgium and Europe, 1984.56. GEYSKENS, Erik (27/03/85) Produktietheorie en dualiteit, 1985.57. COLE, Frank (26/06/85) Some algorithms for geometric programming, 1985.58. STANDAERT, Stan (26/09/86) A study in the economics of repressed consumption, 1986.59. DELBEKE, Jos (03/11/86) Trendperioden in de geldhoeveelheid van Belgie 1877-1983: een theoretischeen empirische analyse van de ”Banking school” hypothese, 1986.60. VANTHIENEN, Jan (08/12/86) Automatiseringsaspecten van de specificatie, constructie en manipulatievan beslissingstabellen, 1986.61. LUYTEN, Robert (30/04/87) A systems-based approach for multi-echelon production/inventory sys-tems, 1987.62. MERCKEN, Roger (27/04/87) De invloed van de data base benadering op de interne controle, 1987.63. VAN CAYSEELE, Patrick (20/05/87) Regulation and international innovative activities in the phar-maceutical industry, 1987.64. FRANCOIS, Pierre (21/09/87) De empirische relevantie van de independence from irrelevant alterna-tives. Assumptie indiscrete keuzemodellen, 1987.65. DECOSTER, Andre (23/09/88) Family size, welfare and public policy, 1988.66. HEIJNEN, Bart (09/09/88) Risicowijziging onder invloed van vrijstellingen en herverzekeringen: eentheoretische analyse van optimaliteit en premiebepaling, 1988.67. GEEROMS, Hans (14/10/88) Belastingvermijding. Theoretische analyse van de determinanten van debelastingontduiking en de belastingontwijking met empirische verificaties, 1988.68. PUT, Ferdi (19/12/88) Introducing dynamic and temporal aspects in a conceptual (database) schema,1988.69. VAN ROMPUY, Guido (13/01/89) A supply-side approach to tax reform programs. Theory and empir-ical evidence for Belgium, 1989.70. PEETERS, Ludo (19/06/89) Een ruimtelijk evenwichtsmodel van de graanmarkten in de E.G.: em-pirische specificatie en beleidstoepassingen, 1989.71. PACOLET, Jozef (10/11/89) Marktstructuur en operationele efficientie in de Belgische financiele sector,1989.72. VANDEBROEK, Martina (13/12/89) Optimalisatie van verzekeringscontracten en premieberekenings-principes, 1989.

173

73. WILLEKENS, Francois () Determinance of government growth in industrialized countries with appli-cations to Belgium, 1990.74. VEUGELERS, Reinhilde (02/04/90) Scope decisions of multinational enterprises, 1990.75. KESTELOOT, Katrien (18/06/90) Essays on performance diagnosis and tacit cooperation in interna-tional oligopolies, 1990.76. WU, Changqi (23/10/90) Strategic aspects of oligopolistic vertical integration, 1990.77. ZHANG, Zhaoyong (08/07/91) A disequilibrium model of China’s foreign trade behaviour, 1991.78. DHAENE, Jan (25/11/91) Verdelingsfuncties, benaderingen en foutengrenzen van stochastische groothe-den geassocieerd aan verzekeringspolissen en -portefeuilles, 1991.79. BAUWELINCKX, Thierry (07/01/92) Hierarchical credibility techniques, 1992.80. DEMEULEMEESTER, Erik (23/3/92) Optimal algorithms for various classes of multiple resource-constrained project scheduling problems, 1992.81. STEENACKERS, Anna (1/10/92) Risk analysis with the classical actuarial risk model: theoreticalextensions and applications to Reinsurance, 1992.82. COCKX, Bart (24/09/92) The minimum income guarantee. Some views from a dynamic perspective,1992.83. MEYERMANS, Eric (06/11/92) Econometric allocation systems for the foreign exchange market: Spec-ification, estimation and testing of transmission mechanisms under currency substitution, 1992.84. CHEN, Guoqing (04/12/92) Design of fuzzy relational databases based on fuzzy functional dependency,1992.85. CLAEYS, Christel (18/02/93) Vertical and horizontal category structures in consumer decision making:The nature of product hierarchies and the effect of brand typicality, 1993.86. CHEN, Shaoxiang (25/03/93) The optimal monitoring policies for some stochastic and dynamic pro-duction processes, 1993.87. OVERWEG, Dirk (23/04/93) Approximate parametric analysis and study of cost capacity managementof computer configurations, 1993.88. DEWACHTER, Hans (22/06/93) Nonlinearities in speculative prices: The existence and persistence ofnonlinearity in foreign exchange rates, 1993.89. LIN, Liangqi (05/07/93) Economic determinants of voluntary accounting choices for R & D expendituresin Belgium, 1993.90. DHAENE, Geert (09/07/93) Encompassing: formulation, properties and testing, 1993.91. LAGAE, Wim (20/09/93) Marktconforme verlichting van soevereine buitenlandse schuld door privatecrediteuren: een neo-institutionele analyse, 1993.92. VAN DE GAER, Dirk (27/09/93) Equality of opportunity and investment in human capital, 1993.93. SCHROYEN, Alfred (28/02/94) Essays on redistributive taxation when monitoring is costly, 1994.94. STEURS, Geert (15/07/94) Spillovers and cooperation in research and development, 1994.95. BARAS, Johan (15/09/94) The small sample distribution of the Wald, Lagrange multiplier and likeli-hood ratio tests for homogeneity and symmetry in demand analysis: a Monte Carlo study, 1994.96. GAEREMYNCK, Ann (08/09/94) The use of depreciation in accounting as a signalling device, 1994.97. BETTENDORF, Leon (22/09/94) A dynamic applied general equilibrium model for a small open econ-omy, 1994.98. TEUNEN, Marleen (10/11/94) Evaluation of interest randomness in actuarial quantities, 1994.99. VAN OOTEGEM, Luc (17/01/95) An economic theory of private donations, 1995.100. DE SCHEPPER, Ann (20/03/95) Stochastic interest rates and the probabilistic behaviour of actuarialfunctions, 1995.101. LAUWERS, Luc (13/06/95) Social choice with infinite populations, 1995.102. WU, Guang (27/06/95) A systematic approach to object-oriented business modeling, 1995.103. WU, Xueping (21/08/95) Term structures in the Belgian market: model estimation and pricing erroranalysis, 1995.104. PEPERMANS, Guido (30/08/95) Four essays on retirement from the labor force, 1995.105. ALGOED, Koen (11/09/95) Essays on insurance: a view from a dynamic perspective, 1995.106. DEGRYSE, Hans (10/10/95) Essays on financial intermediation, product differentiation, and marketstructure, 1995.107. MEIR, Jos (05/12/95) Het strategisch groepsconcept toegepast op de Belgische financiele sector, 1995.108. WIJAYA, Miryam Lilian (08/01/96) Voluntary reciprocity as an informal social insurance mechanism:a game theoretic approach, 1996.109. VANDAELE, Nico (12/02/96) The impact of lot sizing on queueing delays: multi product, multimachine models, 1996.110. GIELENS, Geert (27/02/96) Some essays on discrete time target zones and their tails, 1996.111. GUILLAUME, Dominique (20/03/96) Chaos, randomness and order in the foreign exchange markets.Essays on the modelling of the markets, 1996.112. DEWIT, Gerda (03/06/96) Essays on export insurance subsidization, 1996.113. VAN DEN ACKER, Carine (08/07/96) Belief-function theory and its application to the modeling ofuncertainty in financial statement auditing, 1996.114. IMAM, Mahmood Osman (31/07/96) Choice of IPO Flotation Methods in Belgium in an AsymmetricInformation Framework and Pricing of IPO’s in the Long-Run, 1996.115. NICAISE, Ides (06/09/96) Poverty and Human Capital, 1996.116. EYCKMANS, Johan (18/09/97) On the Incentives of Nations to Join International EnvironmentalAgreements, 1997.117. CRISOLOGO-MENDOZA, Lorelei (16/10/97) Essays on Decision Making in Rural Households: astudy of three villages in the Cordillera Region of the Philippines, 1997.118. DE REYCK, Bert (26/01/98) Scheduling Projects with Generalized Precedence Relations: Exact andHeuristic Procedures, 1998.119. VANDEMAELE Sigrid (30/04/98) Determinants of Issue Procedure Choice within the Context of theFrench IPO Market: Analysis within an Asymmetric Information Framework, 1998.

174 Dissertations

120. VERGAUWEN Filip (30/04/98) Firm Efficiency and Compensation Schemes for the Management ofInnovative Activities and Knowledge Transfers, 1998.121. LEEMANS Herlinde (29/05/98) The Two-Class Two-Server Queueing Model with NonpreemptiveHeterogeneous Priority Structures, 1998.122. GEYSKENS Inge (4/09/98) Trust, Satisfaction, and Equity in Marketing Channel Relationships, 1998.123. SWEENEY John (19/10/98) Why Hold a Job ? The Labour Market Choice of the Low-Skilled, 1998.124. GOEDHUYS Micheline (17/03/99) Industrial Organisation in Developing Countries, Evidence fromCote d’Ivoire, 1999.125. POELS Geert (16/04/99) On the Formal Aspects of the Measurement of Object-Oriented SoftwareSpecifications, 1999.126. MAYERES Inge (25/05/99) The Control of Transport Externalities: A General Equilibrium Analysis,1999.127. LEMAHIEU Wilfried (5/07/99) Improved Navigation and Maintenance through an Object-OrientedApproach to Hypermedia Modelling, 1999.128. VAN PUYENBROECK Tom (8/07/99) Informational Aspects of Fiscal Federalism, 1999.129. VAN DEN POEL Dirk (5/08/99) Response Modeling for Database Marketing Using Binary Classifica-tion, 1999.130. GIELENS Katrijn (27/08/99) International Entry Decisions in the Retailing Industry: Antecedentsand Performance Consequences, 1999.131. PEETERS Anneleen (16/12/99) Labour Turnover Costs, Employment and Temporary Work, 1999.132. VANHOENACKER Jurgen (17/12/99) Formalizing a Knowledge Management Architecture Meta-Model for Integrated Business Process Management, 1999.133. NUNES Paulo (20/03/2000) Contingent Valuation of the Benefits of Natural Areas and its WarmglowComponent, 2000.134. VAN DEN CRUYCE Bart (7/04/2000) Statistische discriminatie van allochtonen op jobmarkten metrigide lonen, 2000.135. REPKINE Alexandre (15/03/2000) Industrial restructuring in countries of Central and Eastern Eu-rope: Combining branch-, firm- and product-level data for a better understanding of Enterprises’ behaviourduring transition towards market economy, 2000.136. AKSOY, Yunus (21/06/2000) Essays on international price rigidities and exchange rates, 2000.137. RIYANTO, Yohanes Eko (22/06/2000) Essays on the internal and external delegation of authority infirms, 2000.138. HUYGHEBAERT, Nancy (20/12/2000) The Capital Structure of Business Start-ups, 2000.139. FRANCKX Laurent (22/01/2001) Ambient Inspections and Commitment in Environmental Enforce-ment, 2001140. VANDILLE Guy (16/02/2001) Essays on the Impact of Income Redistribution on Trade, 2001141. MARQUERING Wessel (27/04/2001) Modeling and Forecasting Stock Market Returns and Volatility,2001.142. FAGGIO Giulia (07/05/2001) Labor Market Adjustment and Enterprise Behavior in Transition, 2001.143. GOOS Peter (30/05/2001) The Optimal Design of Blocked and Split-plot experiments, 2001.144. LABRO Eva (01/06/2001) Total Cost of Ownership Supplier Selection based on Activity Based Costingand Mathematical Programming, 2001.145. VANHOUCKE Mario (07/06/2001) Exact Algorithms for various Types of Project Scheduling Prob-lems. Nonregular Objectives and time/cost Trade-offs, 2001.146. BILSEN Valentijn (28/08/2001) Entrepreneurship and Private Sector Development in Central Euro-pean Transition Countries, 2001.147. NIJS Vincent (10/08/2001) Essays on the dynamic Category-level Impact of Price promotions, 2001.148. CHERCHYE Laurens (24/09/2001) Topics in Non-parametric Production and Efficiency Analysis,2001.149. VAN DENDER Kurt (15/10/2001) Aspects of Congestion Pricing for Urban Transport, 2001.150. CAPEAU Bart (26/10/2001) In defence of the excess demand approach to poor peasants’ economicbehaviour. Theory and Empirics of non-recursive agricultural household modelling, 2001.151. CALTHROP Edward (09/11/2001) Essays in urban transport economics, 2001.152. VANDER BAUWHEDE Heidi (03/12/2001) Earnings management in an Non-Anglo-Saxon environ-ment, 2001.153. DE BACKER Koenraad (22/01/2002) Multinational firms and industry dynamics in host countries :the case of Belgium, 2002.154. BOUWEN Jan (08/02/2002) Transactive memory in operational workgroups. Concept elaboration andcase study, 2002.155. VAN DEN BRANDE Inge (13/03/2002) The psychological contract between employer and employee :a survey among Flemish employees, 2002.156. VEESTRAETEN Dirk (19/04/2002) Asset Price Dynamics under Announced Policy Switching, 2002.157. PEETERS Marc (16/05/2002) One Dimensional Cutting and Packing : New Problems and Algorithms,2002.158. SKUDELNY Frauke (21/05/2002) Essays on The Economic Consequences of the European MonetaryUnion, 2002.159. DE WEERDT Joachim (07/06/2002) Social Networks, Transfers and Insurance in Developing countries,2002.160. TACK Lieven (25/06/2002) Optimal Run Orders in Design of Experiments, 2002.161. POELMANS Stephan (10/07/2002) Making Workflow Systems work. An investigation into the Im-portance of Task-appropriation fit, End-user Support and other Technological Characteristics, 2002. 237pp.162. JANS Raf (26/09/2002) Capacitated Lot Sizing Problems : New Applications, Formulations andAlgorithms, 2002.163. VIAENE Stijn (25/10/2002) Learning to Detect Fraud from enriched Insurance Claims Data (Context,Theory and Applications), 2002.

175

164. AYALEW Tekabe (08/11/2002) Inequality and Capital Investment in a Subsistence Economy, 2002.165. MUES Christophe (12/11/2002) On the Use of Decision Tables and Diagrams in Knowledge Modelingand Verification, 2002.166. BROCK Ellen (13/03/2003) The Impact of International Trade on European Labour Markets, 2002.167. VERMEULEN Frederic (29/11/2002) Essays on the collective Approach to Household Labour Supply,2002.168. CLUDTS Stephan (11/12/2002) Combining participation in decision-making with financial participa-tion : theoretical and empirical perspectives, 2002.169. WARZYNSKI Frederic (09/01/2003) The dynamic effect of competition on price cost margins andinnovation, 2003.170. VERWIMP Philip (14/01/2003) Development and genocide in Rwanda ; a political economy analysisof peasants and power under the Habyarimana regime, 2003.171. BIGANO Andrea (25/02/2003) Environmental regulation of the electricity sector in a European MarketFramework, 2003.172. MAES Konstantijn (24/03/2003) Modeling the Term Structure of Interest Rates Across Countries,2003. 173. VINAIMONT Tom (26/02/2003) The performance of One- versus Two-Factor Models of theTerm Structure of Interest Rates, 2003.174. OOGHE Erwin (15/04/2003) Essays in multi-dimensional social choice, 2003.175. FORRIER Anneleen (25/04/2003) Temporary employment, employability and training, 2003.176. CARDINAELS Eddy (28/04/2003) The role of cost system accuracy in managerial decision making,2003.177. DE GOEIJ Peter (02/07/2003) Modeling Time-Varying Volatility and Interest Rates, 2003.178. LEUS Roel (19/09/2003) The generation of stable project plans. Complexity and exact algorithms,2003.179.MARINHEIRO Carlos (23/09/2003) EMU and fiscal stabilisation policy : the case of small countries,2003.180. BAESENS Bart (24/09/2003) Developing intelligent systems for credit scoring using machine learningtechniques, 2003.181. KOCZY Laszlo (18/09/2003) Solution concepts and outsider behaviour in coalition formation games,2003.182. ALTOMONTE Carlo (25/09/2003) Essays on Foreign Direct Investment in transition countries : learn-ing from the evidence, 2003.183. DRIES Liesbeth (10/11/2003) Transition, Globalisation and Sectoral Restructuring: Theory and Evi-dence from the Polish Agri-Food Sector, 2003.184. DEVOOGHT Kurt (18/11/2003) Essays On Responsibility-Sensitive Egalitarianism and the Measure-ment of Income Inequality, 2003.185. DELEERSNYDER Barbara (28/11/2003) Marketing in Turbulent Times, 2003.186. ALI Daniel (19/12/2003) Essays on Household Consumption and Production Decisions under Uncer-tainty in Rural Ethiopia, 2003.187. WILLEMS Bert (14/01/2004) Electricity networks and generation market power, 2004.188. JANSSENS Gust (30/01/2004) Advanced Modelling of Conditional Volatility and Correlation in Fi-nancial Markets, 2004.189. THOEN Vincent (19/01/2004) ”On the valuation and disclosure practices implemented by venturecapital providers”, 2004.190. MARTENS Jurgen (16/02/2004) ”A fuzzy set and stochastic system theoretic technique to validatesimulation models”, 2004.191. ALTAVILLA Carlo (21/05/2004) ”Monetary policy implementation and transmission mechanisms inthe Euro area.”, 2004.192. DE BRUYNE Karolien (07/06/2004) ”Essays in the location of economic activity”, 2004.193. ADEM Jan (25/06/2004) ”Mathematical programming approaches for the supervised classificationproblem.”, 2004.194. LEROUGE Davy (08/07/2004) ”Predicting Product Preferences : the effect of internal and externalcues.”, 2004.195. VANDENBROECK Katleen (16/07/2004) ”Essays on output growth, social learning and land alloca-tion in agriculture : micro-evidence from Ethiopia and Tanzania”, 2004.196. GRIMALDI Maria (03/09/004) ”The exchange rate, heterogeneity of agents and bounded rationality”., 2004.197. SMEDTS Kristien (26/10/2004) ”Financial integration in EMU in the framework of the no-arbitragetheory”, 2004.198. KOEVOETS Wim (12/11/2004) ”Essays on Unions, Wages and Employment”, 2004.199. CALLENS Marc (22/11/2004) ”Essays on multilevel logistic Regression”, 2004.200. RUGGOO Arvind (13/12/2004) ”Two stage designs robust to model uncertainty, 2004.201. HOORELBEKE Dirk (28/01/2005) ”Bootstrap and Pivoting Techniques for Testing Multiple Hypothe-ses”, 2005.202. ROUSSEAU Sandra (17/02/2005) ”Selecting Environmental Policy Instruments in the Presence ofIncomplete Compiance”, 2005.203. VAN DER MEULEN Sofie (17/02/2005) ”Quality of Financial Statements : Impact of the externalauditor and applied accounting standards”, 2005.204. DIMOVA Ralitza (21/02/2005) ”Winners and Losers during Structural Reform and Crisis : the Bul-garian Labour Market Perspective”, 2005.205. DARKIEWICZ Grzegorz (28/02/2005) ”Value-at-risk in Insurance and Finance : the ComonotonicityApproach”, 2005.206. DE MOOR Lieven (20/05/2005) ”The Structure of International Stock Returns : Size, Country andSector Effects in Capital Asset Pricing”, 2005.207. EVERAERT Greetje (27/06/2005) ”Soft Budget Constraints and Trade Policies : The Role of Insti-tutional and External Constraints”, 2005.

176 Dissertations

208. SIMON Steven (06/07/2005) ”The Modeling and Valuation of complex Derivatives : the Impact of theChoice of the term structure model”, 2005.209. MOONEN Linda (23/09/2005) ”Algorithms for some graph-theoretical optimization problems”, 2005.210. COUCKE Kristien (21/09/2005) ”Firm and industry adjustment under de-industrialisation and glob-alization of the Belgian economy”, 2005.211. DECAMPS MARC (21/10/2005) ”Some actuarial and financial applications of generalized diffusionprocesses”, 2005.212. KIM HELENA (29/11/2005) ”Escalation games: an instrument to analyze conflicts. The strategicapproach to the bargaining problem”, 2005.213. GERMENJI ETLEVA (06/01/2006) ”Essays on the economics of emigration from Albania”, 2006.214. BELIEN JEROEN (18/01/2006) ”Exact and heuristic methodologies for scheduling in hospitals: prob-lems, formulations and algorithms”, 2006.215. JOOSSENS KRISTEL (10/02/2006) ”Robust discriminant analysis”, 2006.216. VRANKEN LIESBET (13/02/2006) ”Land markets and production efficiency in transition economies”,2006.217. VANSTEENKISTE ISABEL (22/02/2006) ”Essays on non-linear modelling in international macroeco-nomics”, 2006.218. WUYTS Gunther (31/03/2006) ”Essays on the liquidity of financial markets”, 2006.219. DE BLANDER Rembert (28/04/2006) ”Essays on endogeneity and parameter heterogeneity in cross-section and panel data”, 2006.220. DE LOECKER Jan (12/05/2006) ”Industry dynamics and productivity”, 2006.221. LEMMENS Aurelie (12/05/2006) ”Advanced classification and time-series methods in marketing”,2006.222. VERPOORTEN Marijke (22/05/2006) ”Conflict and survival: an analysis of shocks, coping strategiesand economic mobility in Rwanda, 1990-2002”, 2006.223. BOSMANS Kristof (26/05/2006) ”Measuring economic inequality and inequality aversion”, 2006.224. BRENKERS Randy (29/05/2006) ”Policy reform in a market with differentiated products: applicationsfrom the car market”, 2006.225. BRUYNEEL Sabrina (02/06/2006) ”Self-econtrol depletion: Mechanisms and its effects on consumerbehavior”, 2006.226. FAEMS Dries (09/06/2006) ”Collaboration for innovation: Processes of governance and learning”,2006.227. BRIERS Barbara (28/06/2006) ”Countering the scrooge in each of us: on the marketing of cooperativebehavior”, 2006.228. ZANONI Patrizia (04/07/2006) ”Beyond demography: Essays on diversity in organizations”, 2006.229. VAN DEN ABBEELE Alexandra (11/09/2006) ”Management control of interfirm relations: the roleof information”, 2006.230. DEWAELHEYNS Nico (18/09/2006) ”Essays on internal capital markets, bankruptcy and bankruptcyreform”, 2006.231. RINALDI Laura (19/09/2006) ”Essays on card payments and household debt”, 2006.232. DUTORDOIR Marie (22/09/2006) ”Determinants and stock price effects of Western European con-vertible debt offerings: an empirical analysis”, 2006.233. LYKOGIANNI Elissavet (20/09/2006) ”Essays on strategic decisions of multinational enterprises: R& D decentralization, technology transfers and modes of foreign entry”, 2006.234. ZOU Jianglei (03/10/2006) ”Inter-firm ties, plant networks, and multinational firms: essays on FDIand trade by Japanse firms.”, 2006.235. GEYSKENS Kelly (12/10/2006) ”The ironic effects of food temptations on self-control performance”,2006.236. BRUYNSEELS Liesbeth (17/10/2006) ”Client strategic actions, going-concern audit opinions andaudit reporting errors”, 2006.237. KESSELS Roselinde (23/10/2006) ”Optimal designs for the measurement of consumer preferences”,2006.238. HUTCHINSON John (25/10/2006) ”The size distribution and growth of firms in transition countries”,2006.239. RENDERS Annelies (26/10/2006) ”Corporate governance in Europe: The relation with accountingstandards choice, performance and benefits of control”, 2006.240. DE WINNE Sophie (30/10/2006) ”Exploring terra incognita: human resource management and firmperformance in small and medium-sized businesses”, 2006.241. KADITI Eleni (10/11/2006) ”Foreign direct investments in transition economies”, 2006.242. ANDRIES Petra (17/11/2006) ”Technology-based ventures in emerging industries: the quest for aviable business model”, 2006.243. BOUTE Robert (04/12/2006) ”The impact of replenishment rules with endogenous lead times onsupply chain performance”, 2006.244. MAES Johan (20/12/2006) ”Corporate entrepreneurship: an integrative analysis of a resource-basedmodel. Evidence from Flemish enterprises”, 2006.245. GOOSSENS Dries (20/12/2006) ”Exact methods for combinatorial auctions”, 2006.246. GOETHALS Frank (22/12/2006) ”Classifying and assessing extended enterprise integration approa-ches”, 2006.247. VAN DE VONDER Stijn (22/12/2006) ”Proactive-reactive procedures for robust project scheduling”,2006.248. SAVEYN Bert (27/02/2007) ”Environmental policy in a federal state”, 2007.249. CLEEREN Kathleen (13/03/2007) ”Essays on competitive structure and product-harm crises”, 2007.250. THUYSBAERT Bram (27/04/2007) ”Econometric essays on the measurement of poverty”, 2007.251. DE BACKER Manu (07/05/2007) ”The use of Petri net theory for business process verification”, 2007.252. MILLET Kobe (15/05/2007) ”Prenatal testosterone, personality, and economic behavior”, 2007.

177

253. HUYSMANS Johan (13/06/2007) ”Comprehensible predictive models: New methods and insights”,2007.254. FRANCKEN Nathalie (26/06/2007) ”Mass Media, Government Policies and Economic Development:Evidence from Madagascar”, 2007.255. SCHOUBBEN Frederiek (02/07/2007) ”The impact of a stock listing on the determinants of firmperformance and investment policy”, 2007.256. DELHAYE Eef (04/07/2007) ”Economic analysis of traffic safety”, 2007.257. VAN ACHTER Mark (06/07/2007) ”Essays on the market microstructure of financial markets”, 2007.258. GOUKENS Caroline (20/08/2007) ”Desire for variety: understanding consumers’ preferences for vari-ety seeking”, 2007.259. KELCHTERMANS Stijn (12/09/2007) ”In pursuit of excellence: essays on the organization of highereducation and research”, 2007.260. HUSSINGER Katrin (14/09/2007) ”Essays on internationalization, innovation and firm performance”,2007.261. CUMPS Bjorn (04/10/2007) ”Business-ICT alignment and determinants”, 2007.262. LYRIO Marco (02/11/2007) ”Modeling the yield curve with macro factors”, 2007.263. VANPEE Rosanne (16/11/2007) ”Home bias and the implicit costs of investing abroad”, 2007.264. LAMBRECHTS Olivier (27/11/2007) ”Robust project scheduling subject to resource breakdowns”,2007.265. DE ROCK Bram (03/12/2007) ”Collective choice behaviour: non parametric characterization”, 2007.266. MARTENS David (08/01/2008) ”Building acceptable classification models for financial engineeringapplications”, 2008.267. VAN KERCKHOVEN Johan (17/01/2008) ”Predictive modelling: variable selection and classificationefficiencies”, 2008.268. CIAIAN Pavel (12/02/2008) ”Land, EU accession and market imperfections”, 2008.269. TRUYTS Tom (27/02/2008) ”Diamonds are a girl’s best friend: five essays on the economics of socialstatus”, 2008.270. LEWIS Vivien (17/03/2008) ”Applications in dynamic general equilibrium macroeconomics”, 2008.271. CAPPELLEN Tineke (04/04/2008) ”Worldwide coordination in a transnational environment: Aninquiry into the work and careers of global managers”, 2008.272. RODRIGUEZ Victor (18/04/2008) ”Material transfer agreements: research agenda choice, co-publicationactivity and visibility in biotechnology”, 2008.273. QUAN Qi (14/04/2008) ”Privatization in China: Examining the endogeneity of the process and itsimplications for the performance of newly privatized firms”, 2008.274. DELMOTTE Jeroen (30/04/2008) ”Evaluating the HR function: Empirical studies on HRM architec-ture and HRM system strength”, 2008.275. ORSINI Kristian (05/05/2008) ”Making work pay: Insights from microsimulation and random utilitymodels”, 2008.276. HOUSSA Romain (13/05/2008) ”Macroeconomic fluctuations in developing countries”, 2008.277. SCHAUMANS Catherine (20/05/2008) ”Entry, regulation and economic efficiency: essays on healthprofessionals”, 2008.278. CRABBE Karen (21/05/2008) ”Essays on corporate tax competition in Europe”, 2008279. GELPER Sarah (30/05/2008) ”Economic time series analysis: Granger causality and robustness”,2008280. VAN HOVE Jan (20/06/2008) ”The impact of technological innovation and spillovers on the patternand direction of international trade”, 2008281. DE VILLE DE GOYET Cedric (04/07/2008) ”Hedging with futures in agricultural commodity mar-kets”, 2008282. FRANCK Tom (15/07/2008) ”Capital structure and product market interactions: evidence from busi-ness start-ups and private firms”, 2008283. ILBAS Pelin (15/09/2008) ”Optimal monetary policy design in dynamic macroeconomics”, 2008284. GOEDERTIER Stijn (16/09/2008) ”Declarative techniques for modeling and mining business pro-cesses”, 2008285. LAMEY Lien (22/09/2008) ”The private-label nightmare: can national brands ever wake up?”, 2008286. VANDEKERCKHOVE Jan (23/09/2008) ”Essays on research and development with spillovers”, 2008287. PERNOT Eli (25/09/2008) ”Management control system design for supplier relationships in manufac-turing: Case study evidence from the automotive industry”, 2008288. AERTS Kris (16/10/2008) ”Essays on the economics of evaluation: public policy and corporate strate-gies in innovation”, 2008289. BOUDT Kris (27/11/2008) ”Estimation of financial risk under non-normal distributions”, 2008290. MARKIEWICZ Agnieszka (01.12.2008) ”Essays in exchange rate economics”, 2008291. GLADY Nicolas (08.12.2008) ”Customer profitability modeling”, 2008292. MIERZEJEWSKI Fernando (11.12.2008) ”Essays on liquidity-preference in markets with borrowingrestrictions”, 2008293. VAN BEVEREN Ilke (15.12.2008) ”Globalization and firm dynamics”, 2008294. VERBRUGGEN Marijke (16.12.2008) ”The role of career counseling in the new career era”, 2008295. CORNIL Anneleen (15.12.2008) ”Essays on financial statement quality: The influence and the effectsof accounting restatements”, 2008296. SOENEN Helke (16.12.2008) ”Between vulnerability and creativity: an anthropological economicanalysis of undocumented migrants in Brussels”, 2008297. VERGOTE Olivier (17.12.2008) ”Financial transaction data and volatility”, 2008298. COLANTONE Italo (18.12.2008) ”Essays on entrepreneurship and industrial dynamics in an openeconomic context”, 2008299. LETEN Bart (19.12.2008) ”Technology scope decisions in large firms”, 2008300. HERBOTS Jade (19.12.2008) ”Dynamic project portfolio management: Selection and capacity-allocationdecisions”, 2008

178 Dissertations

301. JURZYK Emilia (22.12.2008) ”Essays on the effects of foreign bank entry into Central and EasternEuropean countries”, 2008302. CARDOEN Brecht (23.01.2009) ”Operating room planning and scheduling: solving a surgical casesequencing problem”, 2009303. VERMEULEN Bart (27.01.2009) ”Design issues in conjoint analysis for market and non-market valu-ation”, 2009304. JIE YU (09.02.2009) ”Optimal design methodology for choice experiments in the presence of modeluncertainty and consumer heterogeneity”, 2009305. DE RORE Lotte (03.03.2009) ”Measuring productivity and improving efficiency in software develop-ment environments”, 2009306. LIU Fang (20.03.2009) ”The prediction power of interest rates with respect to exchange rates andconsumption”, 2009307. SLOOTMAEKERS Veerle (20.03.2009) ”Institutions and market structure in transition countries”,2009308. DE WITTE Kristof (31.03.2009) ”On analyzing drinking water monopolies by robust non-parametricefficiency estimations”, 2009309. CONSENTINO Fabrizio (27.04.2009) ”Modelling the missingness: estimation, testing and model se-lection”, 2009310. COENE Sofie (26.05.2009) ”Routing problems with profits and periodicity”, 2009311. CREEMERS Stefan (15.06.2009) ”Appointment-driven queueing systems”, 2009312. CHENG Xiaoqiang (19.06.2009) ”Legal system, financial development and economic performance:evidence from China”, 2009313. VERRIEST Arnt (22.06.2009) ”Institutional factors, firm profitability and financial reporting quality”,2009314. DECANCQ Koen (15.06.2009) ”Essays on the measurement of multidimensional inequality”, 2009315. VAN DEN BERGH Bram (02.07.2009) ”Basis instinct: the fire of desire in economic decisions”, 2009316. GIJSENBERG Maarten (06.07.2009) ”Timing is money. In the search of the role of timing in marketingdecisions and effectiveness”, 2009317. CALLAERT Julie (16.09.2009) ”Enterprising academics: (when) does the combination of scientificand patenting activities make sense?”, 2009318. JOCHMANS Koen (18.09.2009) ”Essays on inference in econometric models with heterogeneity andendogeneity”, 2009319. FERRARI Stijn (23.09.2009) ”Empirical models of restricted entry: efficiency in ATM deploymentand retail magazine distribution, 2009320. HAESEN Raf (21.10.2009) ”Designing information system services in information-intensive organiza-tions”, 2009321. VAN CAMPENHOUT Bjorn (29.10.2009) ”Sample splitting and threshold estimation techniques. Ap-plications in development studies”, 2009322. DUTILLIEUX Wouter (04.11.2009) ”Audit research opportunities in European institutional settings”,2009323. LUYPAERT Mathieu (17.12.2009) ”Mergers and acquisitions: when do they really create value?”, 2009324. CHEN Xinliang (17.12.2009) ”Super-replicating exotic options, allocating capital and approximatingaggregate distributions using the comonotonic approach”, 2009325. GRINE Slimane (18.01.2010) ”Multi-layer model of correlated instruments. Application to energymarkets”, 2010326. DUNKERLEY Fay (25.02.2010) ”Essays in transport economics”, 2010327. VERGHOTE Patrick (24.02.2010) ”Essays on international account management”, 2010328. BUI THI NGOC Tuan (25.02.2010) ”Spot and forward prices in the Brussels stock exchange: Differ-ential costs, informativeness and price discovery”, 2010329. ZARNIC Ziga (19.03.2010) ”Trade, institutional measures and competitiveness”, 2010330. VANORMELINGEN Stijn (31.03.2010) ”Essays on empirical industrial organization”, 2010331. STOUTHUYSEN Kristof (17.05.2010) ”Formal control in interfirm service exchanges”, 2010332. BINZ Hanna (25.05.2010) ”Who does (not) want to pay for innovation - the financing gap storyrevisited”, 2010333. PERSYN Damiaan (28.05.2010) ”Essays on unionised labour markets in a globalising world”, 2010334. MASSCHELEIN Stijn (31.05.2010) ”Information and control mechanisms in collaborations: threeexperimental studies”, 2010335. NGUYEN THI TUONG Van (31.08.2010) ”Essays on commodity prices”, 2010336. ZULUAGA DIAZ Blanca (03.09.2010) ”Essays on schooling investment decisions and poverty”, 2010337. FATICA Serena (03.09.2010) ”Three essays on firms’ location choices and public policy”, 2010338. XIANG Tao (10.09.2010) ”Food standards and development: partial and general equilibrium analysiswith applications to China”, 2010339. DEBLAERE Filip (17.09.2010) ”Resource-constrained project scheduling under uncertainty”, 2010340. TALLA NOBIBON Fabrice (24.09.2010) ”Algorithms for selection and graph-coloring problems withapplications in marketing and micro-economics”, 2010341. VAN DE VOORDE CARINE (24.09.2010) ”Essays on risk adjustment in health insurance”, 2010342. MAHIEU Koen (08.10.2010) ”Robut smoothing and forecasting of multivariate time series”, 2010343. ZHAONING Shang (08.10.2010) ”A recursive approach to general diffusion processes: with applicationsto financial and insurance pricing”, 2010344. CARBONEZ Katelijne (11.10.2010) ”On the behavior of commodity prices: statistics and economics”,2010345. KUPPER Gerd (27.10.2010) ”Contributions to European energy policy”, 2010346. VAN DER LOO Saskia (29.10.2010) ”Transport infrastructure pricing and investment issues”, 2010347. LECOCQ Cathy (26.10.2010) ”Technological performance of regions (and firms). The case of biotech-nology”, 2010348. VAN LANDEGHEM Bert (22.11.2010) ”Essays on subjective well-being”, 2010

179

349. POELMANS Jonas (13.12.2010) ”Essays on using formal concept analysis in information engineering”,2010350. SABBE Jeroen (14.12.2010) ”Revealed preference analysis of consumption and production behavior:applications of the collective model”, 2010351. MONSIEUR Geert (17.12.2010) ”Pattern-based coordination in process-based service compositions”,2010352. VISNJIC Ivanka (22.12.2010) ”Conceiving and implementing service business models within manufac-turing firms”, 2010353. CLAERHOUT Diederik (17.01.2011) ”The supply chain operations planning: the lead time settingproblem”, 2011354. VAN LAERE Elisabeth (26.01.2011) ”Capital regulation of financial institutions, the role of ratingsand the tension field between regulation and economic reality”, 2011355. JI YUEMEI (01.03.2011) ”Three essays on the economics of education: a behavioral approach”, 2011356. NEYENS Inge (03.03.2011) ”Composing and managing technological alliance portfolios”, 2011357. ROVIRA KALTWASSER Pablo (02.03.2011) ”Heterogeneous expectations in the foreign exchangemarket”, 2011358. THORWARTH Susanne (18.03.2011) ”Productivity effects of research, development and design activ-ities”, 2011359. DE JAEGER Simon (13.05.2011) ”Assessing incentive-based environmental policies for reducing mu-nicipal solid waste disposal”, 2011360. SLAETS Leen (26.05.2011) ”Analyzing phase and amplitude variation of functional data”, 2011361. VANDEMOORTELE Thijs (27.05.2011) ”Political and economic theory of standards”, 2011362. VAN WEERT Koen (10.06.2011) ”Optimal portfolio selection: the comonotonic approach”, 2011363. ROGGE Nicky (14.06.2011) ”Evaluating teaching and research performances: a non-parametric ’benefit-of-the-doubt’ approach”, 2011364. VANDEPLAS Anneleen (16.06.2011) ”The efficiency and equity effects of emerging agricultural valuechains: four essays on institutions and development”, 2011365. WANG Lihong (28.06.2011) ”Value creation through financing and governance: a comparison of listedSOEs and private companies in China”, 2011366. THARMARATNAM KUKATHARMINI (04.07.2011) ”Robust estimation and model selection in semi-parametric regression models”, 2011

essays on the economics of media platforms

Documents