Social networks and their economics influencing consumer choicedr soc

CONTENTS vii5 How social networks influence consumer choice of mobile phone 5.2 Case study: Using homophily for social network marketing 92 5.4.2 Description of the statistical approach

Social

Networks and their Economics Influencing

Consumer Choice

Daniel Birke

Social Networks and their Economics

Social Networks and their Economics

Influencing Consumer Choice

Daniel Birke

Visiting Researcher, Aston Business School, Birmingham, and works in a leading international management consultancy

in Germany.

This edition first published 2013

All rights reserved No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher.

Wiley also publishes its books in a variety of electronic formats Some content that appears in print may not be available in electronic books.

Designations used by companies to distinguish their products are often claimed as trademarks All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners The publisher is not associated with any product or vendor mentioned in this book.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of

merchantability or fitness for a particular purpose It is sold on the understanding that the publisher is not engaged in rendering professional services and neither the publisher nor the author shall be liable for damages arising herefrom If professional advice or other expert assistance is required, the services of a competent professional should be sought.

Library of Congress Cataloging-in-Publication Data

2.4 Key findings from physics research into complex networks 302.5 Empirical research on social networks and network effects 32

CONTENTS vii

5 How social networks influence consumer choice of mobile phone

5.2 Case study: Using homophily for social network marketing 92

5.4.2 Description of the statistical approach used: Quadratic

5.10 Multi-country case study of operator choice in social networks 122

7.5 The model: Price discrimination between on- and off-net calls 1697.6 Estimation results: Adapting consumption choice

8.3 Looking ahead: How social network analysis is changing research

Appendix A Success factors for viral marketing campaigns 183

A.3 Design the campaign around a good understanding of the specific

List of figures

2.3 Increase in the use of ‘social networks’ in the title of academic

2.4 A sociogram with undirected (a) and directed relationships (b) 25

2.6 Sociogram of an undirected graph (a) and a direct graph (b) 282.7 Transition from a regular (a) to a random (c) graph via a small-world

2.9 (a) Outside-in and (b) inside-out approach for measuring network

3.2 Applications of social network analysis across the customer life cycle 50

4.1 Illustrative benefits of combining traditional churn models with social

4.2 Illustrative results from prediction of churn influencers 74

x LIST OF FIGURES

7.1 Number of mobile phone subscribers in the UK (in thousands) 167

List of tables

4.1 Churn and iPhone uptake rate by number of ‘infected’ neighbours 81

5.6 Do you know which operator your friends/family/partner uses? 104

5.9 Mixing patterns between students from different nationalities 106

xii LIST OF TABLES

5.22 Handset choice by operator (expected values in brackets) 121

5.24 Do you know which operator your friends/family/partner use? 1255.25 Determinants of choosing the same operator (The Netherlands) 126

5.27 Operators chosen when respondents have multiple operators 130

5.29 Operator coordination between respondents and fathers (expected

5.34 Observed versus expected percentage of same operator dyads

LIST OF TABLES xiii

6.7 Coordination of operator choice by type of relationship 153

6.10 Predicted probabilities of operator choice (MNP model) 157

7.3 Expected shares (by volume of calls) second quarter 1999 171

A.1 Is your product/campaign suitable for word-of-mouth marketing? 190

To understand what influences consumers in their purchasing decisions has been atthe heart of marketing for decades Intuitively, everybody understands that purchasingdecisions are based on our own individual preferences and that we are at the same timeinfluenced by our friends and peers in what we do, how we behave and what products

we consume However, until recently, it was difficult to measure this interdependence,mainly because data on social networks were difficult to collect and not readilyavailable Nowadays, more and more companies, like mobile phone companies orsocial networking sites like Facebook, collect such data electronically There is,therefore, a strong academic and practitioner interest in measuring how consumersare influenced by their social network in the products they consume

This book uses the author’s unique experience in carrying out academic research

on consumer choice in social networks, starting up a company that successfullycommercialised these insights and working in a top-management consultancy advis-ing companies on Marketing and Sales It is relevant for both an academic and apractitioner audience:

r From an academic perspective, the book is most relevant to final year graduate, postgraduate and university researchers in industrial economics andconsumer marketing Each chapter uses different empirical studies demon-strating how consumption interdependences can be measured A number ofdifferent research techniques (primary and secondary surveys, electronic datacollection) and different statistical techniques (survival analysis, multinomiallogit, time-series statistics, permutation tests) are used The case studies andrelated questions can be used in the class room

under-r The book is also directly relevant for marketers interested in how to turnsocial network data into actionable insights and campaigns Based on theauthor’s experience working together with a large number of marketing andsales departments, each chapter starts with an executive summary of relevantaspects from a practitioner point of view Furthermore, each chapter is preceded

by a case study discussing practical implications of the research in areas such

as social network marketing, retention, pricing strategy and so on For example,Chapter 4 on how switching of mobile phone providers is influenced by one’speers, is preceded by a case study on how several mobile phone providersare using these insights to reduce customer churn among their subscribers

xvi PREFACE

Appendix A includes a discussion of the success factors for viral marketingcampaigns

This book mainly covers the following topics:

r Network effects and the analysis of social networks: Overview of the the art research

state-of-r Consumption interdependences between friends and peers: Who is influencingwhom through which channels and to what degree?

r Statistical methods and research techniques that can be used in the analysis ofsocial networks

r Social network analysis and its practical application for marketing purposes.This book contains an accompanying website Please visitwww.wiley.com/go/social_networks

I am very grateful to my wife Yundan and my children for coping with their husband/daddy locking himself in the office to write this book To them I am dedicating thisbook

This book has benefited from a number of people and I am very grateful for thishelp and support First and foremost I would like to thank my PhD advisor Peter Swannwho supported me from the first meetings at Manchester Business School, throughmeetings at Bridgewater Hall to the award of my PhD at Nottingham UniversityBusiness School, and since then as a very good friend I in particular enjoyed thestimulating discussions which helped me not only to write my PhD thesis, but tounderstand what is needed to become a good academic

Towards the end of my PhD in 2006 I started with Idiro Technologies, a softwarecompany specialising in analysing very large social networks in order to derivemarketing recommendations I had four fantastic years with Idiro and thoroughlyenjoyed being able to translate my PhD research into practical use and being able

to work with our customers on combining the model predictions with the otherelements of successful marketing campaigns I am in particular grateful to AidanConnolly, Brendan Casey and my team members A special thanks goes to SimonRees, Sales & Marketing Director of Idiro Technologies for his deep insights intothe mobile telecommunications industry (and many great nights in Istanbul!) Simonalso contributed the discussion of the success factors for viral marketing campaigns

in Appendix A which is a great reference resource for organizations who want torun a viral marketing campaign Thank you as well to Robert Walker from Ernst &Young’s Customer practice who enabled me to take a three months sabbatical to writethis book

I would also in particular like to thank John Belchamber, Ricardo Correia, PaulDavid, Chris Easingwood, Nicolas Economides, Koen Frenken, Sourafel Girma, Gau-tam Gowrisankaran, Francesco Lissoni, David Paton, Roy Radner, Paul Stoneman,Arun Sundararajan, Steve Thompson, Reinhilde Veugelers and many others who gave

me helpful comments and suggestions

This book also would not have been possible without the extensive access todata that I was able to gain from a number of sources I would like to thank IdiroTechnologies and two mobile phone companies that shall remain anonymous forproviding me access to the data for Chapters 3 and 4 By enabling and supporting

me to run surveys with their students at the University of Utrecht, University of

xviii ACKNOWLEDGEMENTS

Nottingham in Malaysia and at the University of Brescia, Koen Frenken, YoongHon Lee and Francesco Lissoni made the data collection for Chapter 5 possible.Ben Anderson from Chimera, the Institute for Social and Economic Research at theUniversity of Essex, the ESRC data archive and Nicoletta Corrocher helped me withdata for Chapter 6 Last but not least, I would like to thank Hilary Anderson fromOFCOM who granted me access to the data on which Chapter 7 is based

I would also like to gratefully acknowledge financial support from NottinghamUniversity Business School and the ESRC, which allowed me to focus on my researchduring my PhD years

Last but not least I would like to thank my publisher Wiley & Sons and their teamfor shepherding and guiding me through the publication process, in particular RichardDavies, Heather Kay, Debbie Jupe, Ilaria Meliconi, Paulina Shirley and Jo Taylor

Daniel Birke

The basic conjecture of this book is that consumers do not make decisions in tion, but are influenced by and influence other consumers with whom they interact.Everyday experience suggests that we are frequently influenced by others: we askour peers for restaurant tips, hear about new products from them, make joint con-sumption decisions for family cars within families, consume similar products to ourpeers in order to ‘keep up with the Joneses’ and use similar products as people weregard highly and aspire to These processes happen within social networks, which

isola-in this book means all social relationships between people In recent years, socialnetworks such as Facebook have become very popular Thinking about ones socialrelationships as a social network has consequently become very intuitive for manypeople – whether these relationships are maintained via Facebook, mobile phones orvia traditional offline channels

However, for a long time much of economics and marketing did not take theseinterrelationships into account There are a number of good reasons for this focus

on treating individuals as atomistic decision units: First, it was difficult to collect

Social Networks and their Economics: Influencing Consumer Choice, First Edition Daniel Birke.

© 2013 John Wiley & Sons, Ltd Published 2013 by John Wiley & Sons, Ltd.

2 SOCIAL NETWORKS AND THEIR ECONOMICS

appropriate network data, especially in the past when such data had to be gatheredwith the help of network surveys asking respondents to identify relationships theyhave with their peers (see also Chapter 5) Secondly, most traditional statisticalmethods assume that observations are independent, an assumption that is clearlyviolated in social networks and requires different statistical approaches Thirdly, datavolumes when analysing electronic social network datasets can be huge Large-scalesocial networks can have tens or hundreds of millions of users and for each user, therecan easily be 100–1000 communications per observation period, meaning that datavolumes can be 100–1000 times larger than for comparable individual-level data.Fourthly, it is more difficult to incorporate interdependences in social networks intotheoretical economic and marketing models

In the last couple of years, massive electronic social network datasets have becomeavailable and companies now have the computational capabilities to analyse themproperly, although this can still be a challenge when working with datasets coveringtens or hundreds of million users At the same time, researchers such as Snijders, van

de Bunt and Steglich (2010) have developed new statistical techniques and runningexperiments using electronic social networks have become increasingly popular (see

e.g Bakshy et al., 2012) There has also been some progress on theoretical models

(e.g Sundararajan, 2007), but in general there has been a shift in emphasis towardsmore empirical work, something which is also driven by the increased emphasis

of marketing practice on quantitative marketing and, therefore, greater demand forquantitative analystics skills

A particular challenge when analysing influence processes in social networks is toidentify the causality of events There can be a large number of reasons why decisions

of consumers in a social network are not independent (Manski, 1993) and one of thekey challenges of research into social networks is to tease such different effectsapart For example, individuals who interact with each other typically behave in asimilar way because they share the same environment, because they receive similarinformation and because of psychological factors like group pressure If we want touse the social network structure to achieve a certain outcome, say because we want topromote a certain product, then identifying the main cause(s) is key to pursuing theright marketing strategy If, for example, a lot of consumers who are likely to buy aproduct share the same context, then the company might want to target a certain social

or geographic segment For a wine retailer it is, for example, critically important tofind the right location if a high number of consumers is geographically concentrated

in a particular location as consumers will then buy their wine independent of the

social networks that exist in this location However, if a high number of customers of

an online wine retailer recommend this retailer to their friends, then it is important

to understand how customers interact with each other and which customers are mostlikely to successfully recommend the service

This book uses a variety of electronic and non-electronic datasets to study howconsumers influence each other, and establishes causality by using the time structure

of events occurring over the network and via cross-country case study research I alsouse social network analysis of very large electronic datasets, an approach that has thepotential to revolutionise marketing and can help extend our understanding of humanbehaviour

CONSUMER CHOICE IN SOCIAL NETWORKS 3

economics of social networks

A large part of the book is based on empirical analysis of the mobile nications industry Telecommunications networks are a prime example of marketswhere consumers influence each other and where this influence can be measured bydata Similar phenomena exist in other industries, such as social networking sitesand finance Furthermore, data on social network interactions are available to varyingdegrees in a number of other industries Even if there is no electronic data readilyavailable this book shows that social network surveys can be used in such cases (seeChapter 5) In mobile telecommunications, consumer demand is interdependent for

telecommu-a vtelecommu-ariety of retelecommu-asons:

r First, every subscriber to a mobile phone network benefits from other

sub-scribers also using mobile phones as it allows communication with a greater

number of users Network effects therefore influence the overall diffusionpattern of mobile phones The same is true for social networking sites likeFacebook, LinkedIn and so on Likewise, network expansion drives the usagevolume of people already using mobile phones The usage volume of exist-ing subscribers therefore increases with the total number of mobile telephonesubscribers

r Secondly, in mobile telecoms and other industries it is becoming increasingly

important to create product eco-systems While Apple, Google or Facebook

directly provide the basic functionality of their respective products, they alsocreate platforms and interfaces that allow other companies to offer their prod-ucts and services via their platforms This means that higher user numbersmake platforms more attractive for these third parties to develop their offeringfor a particular platform, and this in turn makes the platform more attractive

to end users as well Often, complementary services developed for a platformalso create direct network effects, such as Apple’s FaceTime application whichallows iPhone and iPad users to communicate with each other for free on theirmobile devices Such complementary services are, therefore, a good way ofmaking products stickier

r Thirdly, in mobile telecommunications, calls to the same network are typically

cheaper than calls to other networks, and it is therefore beneficial for consumers

to subscribe to the same network as the people they are calling Mobile phonenetworks from different companies are highly compatible with each other from

a technological point of view, but network effects are often induced by networkoperators through higher prices for off-net than for on-net calls, something

which Laffont et al (1998) termed tariff-mediated network effects

Tariff-mediated network effects can take the form of a general price discriminationbetween on- and off-net calls or can be created through discounts for certaintypes of on-net calls Probably the most famous example of such a scheme

is MCI’s Friends and Family plan, which was introduced at the beginning of

4 SOCIAL NETWORKS AND THEIR ECONOMICS

the 1990s and allowed MCI customers to call up to 20 other MCI customers

at a cheaper rate In most European countries such price differentiation iscommon place, but there are also exceptions, like the Netherlands, whereoperators charge the same prices for calls to the same network and calls toother networks

r Fourthly, the use of mobile phones is conspicuous and sends out social nals about the users Using an attractive handset, like for example an iPhone,enhances the social standing of its owners and peers might be influenced bytheir peer group in their choice of mobile phone, just as drivers of luxurycars influence others in their neighbourhood and peer group to buy similarprestigious car brands

sig-r Fifthly, users of relatively complex products such as mobile phones and the

services running on them benefit from information exchange with their peers.

Such information exchange can be about new services, about the advantagesand disadvantages of existing services or simply about how to use certainservices or functionalities

Variations of these factors will also be important in many other industries and can beobserved in electronic data, for example, in many online businesses such as onlinesocial networks While most of the empirical data for this book come from the mobiletelecommunications industry, the insights and methods are, therefore, more generallyapplicable

The book consists of six main chapters: one chapter reviewing the relevant priorresearch and five empirical chapters looking at various aspects of how consumersinfluence each other in a social network Each empirical chapter starts with an exec-utive summary and one or two case studies on how social network analysis can and

is used for marketing purposes

Chapter 2 starts with a short history of the relevant literature and research fromeconomics, marketing, sociology and physics One of the exciting aspects of study-ing how social networks influence consumer behaviour is that a number of verydifferent subject areas can contribute to our understanding The chapter reviews thekey relevant research strands in each area and shows how they add to our overallunderstanding of the underlying processes In general, the book draws from two mainbodies of literature that help us understand how consumers influence each other: eco-nomics/marketing and sociology/social network analysis The economics literature

in particular sheds light on aggregate phenomena, like overall competitive outcomes

in markets with network effects; whereas the social network analysis literature offers

a wide variety of lessons on how to influence behaviour at an individual level within asocial network Furthermore, Chapter 2 focuses on practical implications for compa-nies and discusses how causal relationships can be identified when studying dynamic

CONSUMER CHOICE IN SOCIAL NETWORKS 5

processes in social networks – something that is of particular importance for ing interventions

market-Chapters 3 and 4 study how the diffusion of a new product like the iPhone(Chapter 3) and switching decisions between rival networks (Chapter 4) are influenced

by social networks The analyses use call detail records from all subscribers of twolarge European mobile phone operators to construct a social network and trackproduct uptake and switching decisions over a period of four months Based onsurvival analysis models, the results show that the more network connections thathave taken up the iPhone, the more likely it is that the focal consumer also takes

up the iPhone Interestingly, this contagion effect decreases only slowly over timeafter an initial peak at product launch Likewise, one friend switching operators has

a strong impact on the switching decision of the focal consumer These two chaptersare particularly relevant for companies and researchers with access to large-scalesocial network data who would like to understand how to leverage the opportunitiesprovided by such data for both research and marketing Chapter 3 is accompanied bytwo case studies The first discusses how mobile phone companies can approach socialnetwork marketing for customer acquisition, product upsell/ cross-sell and customerretention The second discusses different ways in which social advertising is and can

be used on Facebook A third case study preceding Chapter 4 focuses specifically oncustomer retention, which is arguably the earliest and still most common application

of social network marketing in the mobile telecommunications industry

Chapter 5 demonstrates a different way of collecting social network data throughthe use of a social network questionnaire This approach is particularly useful ifelectronic social network data is not available at all or if researchers/marketers areinterested in particular individual-level data which are not available from electroni-cally collected datasets The chapter is based on primary survey data from a number

of university classes in Europe and Asia and uses a statistical permutation methodcalled Quadratic Assignment Procedure (QAP) to account for a correlation in errorterms for non-independent observations in a social network The results demonstratethat friends tend to choose the same mobile phone carrier and that this coordination

is stronger the closer the relationship Interestingly, using variations in the pricingstrategy between operators and countries, this chapter shows that this coordination

is caused by price differences, rather than by alternative potential causes such aspeer pressure or unobserved socio-demographic similarities among friends Besidesdeciding to use the same operator as their peers, consumers also react to the con-sumption decisions of their peers by choosing to be part of several networks at thesame time if their friends are on different networks The accompanying case studydiscusses how social network marketing can use homophily, the commonly observedtendency that similar people interact with each other more frequently, to identifyconsumers who are potentially interested in a particular product or to close gaps inthe knowledge of certain individual-level variables

Chapter 6 analyses how households – the core of most people’s social network –coordinate their consumer choice The chapter is based on a large traditional three-wave survey of British consumers and employs multinomial logit and probit models

to estimate the extent to which households coordinate their choice of mobile phone

6 SOCIAL NETWORKS AND THEIR ECONOMICS

operator Results show a very strong influence of household members on operatorchoice, with more than 50% of household members choosing the same operator.While there are market-level network effects, household effects are far stronger, withroughly 9.2 m subscribers to a network having the same impact as one additionalmember from the same household being on the same network The results also showthat parents coordinate more strongly than parents and kids and that older householdmembers also coordinate more strongly The accompanying case study discusseshow marketers can identify families/households in social networks and how suchknowledge can be used to improve marketing to these groups

Chapter 7 – the final empirical chapter – analyses how company pricing strategies

in markets with network effects can influence the market structure of the industry.Using price and subscriber number data from the UK telecoms regulator OFCOMand a time-series statistical model, the chapter demonstrates how higher charges forcalls to other networks than for calls to the same network have a very strong impact

on consumer choice Results show that consumers react to these pricing strategies bycoordinating operator choice and reducing calls to off-net numbers and that significantinertia maintains this coordination, even after the price differential decreases Thecase study discusses pricing strategies that companies can pursue when pricing digitalproducts which typically exhibit both strong network effects and economies of scale.The closing chapter discusses the impact that the increased availability of elec-tronic social network data has on academic research and marketing practice Foracademic research, it is increasingly more common to directly observe social net-work interactions and it is likely that future empirical work will use updated and newstatistical models to realistically account for these interactions rather than making thesimplifying assumption that individual consumer behaviour is independent It is alsolikely that many more population-level analyses will be carried out, which will lead

to a more frequent use of the research and statistical methods used in this book Formarketing practitioners, the abundant availability of electronic consumer data meansthat there is a continued shift towards using statistical and data mining approaches

in marketing In particular, as this book will show, it is now possible to measureand influence how consumers influence each other, which opens the way for newmarketing approaches Some of these approaches are discussed as part of the casestudies and I am sure that there will be many more in future as companies continue

to experiment with these new opportunities

Finally, Appendix A includes a discussion of the success factors for viral ing campaigns contributed by viral marketing expert Simon Rees

As a practitioner interested in marketing and big data:

The book contains several case studies demonstrating the importance of ing how consumers influence each other The case studies cover a range of topics,including social network marketing and social advertising (Sections 3.2 and 3.3), cus-tomer retention (Section 4.2), the use of the homophily principle for social network

understand-CONSUMER CHOICE IN SOCIAL NETWORKS 7

marketing (Section 5.2), identification of and marketing to families and nities (Section 6.2) and pricing strategies for digital services with network effects(Section 7.2)

commu-Of particular interest are also the first two empirical chapters which use two uniquesocial network datasets based on call records from two large European mobile phoneoperators These datasets make it possible to analyse a network containing millions

of individual consumers, which circumvents the problem of selecting adequatelybounded networks – a very common problem in social network analysis studies.Transactional electronic company data has rarely been used in the social sciencesand these two chapters show how such data can enrich our understanding of thecomplexities of consumer choice and, in particular, how such data can be used tounderstand the interdependences between the choices of a consumer and those ofher peers Chapter 5 demonstrates how the arsenal of a social network marketer can

be further enhanced by the use of network surveys, which can give a more in-depthunderstanding of particular social networks or can be deployed in industries where

no electronic social network data are available In the empirical analysis in this book

I show that students in a number of European countries coordinate their choice ofmobile carrier and highlight that pricing is the likely cause behind this coordination

As an academic interested in economics and marketing:

There is a rich body of literature on the economics of networks which is discussedand extended in this book This book in particular combines insights and methodsfrom the largely separate fields of industrial economics and social network analysis.Like empirical industrial economics, social network analysis is mainly quantitativeand, therefore, methods used in social network analysis can easily be applied totopics of interest to (industrial) economists The main benefit of doing so is theinherently structural and contextual perspective of social network analysis, whichenables the researcher to model interdependences between individual consumers.This reflects reality much better, as consumers are likely to make most economic

decisions based on their own individual characteristics and on decisions taken by

people with whom the individual interacts The approach can, therefore, be seen as

an attempt to reconcile insights from the two worlds of economics and sociology.Duesenberry (1960, 233) famously argued that ‘Economics is all about how peoplemake choices Sociology is all about why they don’t have any choice to make.’ Thisbook argues that the reality is best conceived and modelled as a combination ofboth paradigms Individuals do make their choices based on an economic rationalebut, for a number of reasons, are influenced in their decisions by the decisions ofother individuals The empirical chapters show different ways that this can be doneempirically by using survival analysis models (Chapters 3 and 4), a permutationtechnique called Quadratic Assignment Procedure (Chapter 5), multinomial probitand logit models (Chapter 6) and time-series econometrics (Chapter 7)

Furthermore, most of the economic network literature treats direct and indirectnetwork effects as equivalent (see for example Katz and Shapiro, 1985) However,this equivalence rests on the (strong) assumption that, in both cases, only theoverall number of network members matters, not which consumers are in the

8 SOCIAL NETWORKS AND THEIR ECONOMICS

network This assumption is plausible for markets with indirect network effects,but not for markets with direct network effects For an example of the latter –mobile telecommunications – results from Chapter 6 using multinomial choicemodels show that the influence of other household members is about 10 milliontimes more important than that of a random network member This is the firstempirical evidence that direct and indirect network effects can have very differentimplications in the real world Whereas consumers are shown to be influenced intheir operator choice by other household members, this is likely not to be the case forproducts with indirect network effects, as network effects in markets with indirectnetwork effects arise through product complementarities and not through consumerinteraction

Finally, Chapter 7 introduces a new method of testing for direct market-levelnetwork effects and uses a time-series approach to test the model Contrary to earlierempirical work on markets with direct network effects, this chapter focuses on tech-

nology usage instead of technology adoption and is, therefore, able to circumvent

some of the econometric difficulties that the researcher faces when analysing suchmarkets

As an academic interested in social network analysis:

Chapter 2 shows how the economics literature on network effects is related to socialnetwork analysis in that it departs from the standard economics assumption that eachconsumer’s choice is independent Chapters 3 and 4 use unique large-scale networksbased on mobile phone calling data to show how the iPhone diffuses through asocial network and how switching of carriers is correlated among peers Data forChapter 5 was specifically collected for this book and is used to causally identify thereason behind the coordination of mobile phone choice in a social network using across-country quantitative case study approach Identification of network effects asthe cause behind operator coordination is achieved through differences in consumercoordination across companies and countries that induce tariff-mediated networkeffects and companies and countries that do not Finally, Chapter 7 demonstrates away in which market-level outcomes of individual-level consumer choice in socialnetworks can be estimated

As somebody interested in regulatory policy:

The importance of local social networks in markets with direct network effects alsohas implications for regulatory and anti-trust policy For example, results from this

book suggest that the high price of off-net calls can not only be a result of market power, but also can be a significant source of market power, which can especially

be used to pre-empt entry by new competitors On the other hand, when networkeffects are local in nature, multiple networks can more easily co-exist than in caseswhere only overall network size matters If only overall network size matters, thenmarkets with network effects indeed have a very strong tendency to be concentrated.However, for markets with direct network effects, this assumption is not supported

by the results from this book and consequently the case for regulatory interventions

in such markets is much weaker

CONSUMER CHOICE IN SOCIAL NETWORKS 9

References

Bakshy, E., Eckles, D., Yan, R and Rosenn, I (2012) Social Influence in Social Advertising:Evidence from Field Experiments Proceedings of the 13th ACM Conference on ElectronicCommerce, pp 146–161

Duesenberry, J.S (1960) Comment on ‘An economic analysis of fertility’ by Gary S Becker,

in Demographic and Economic Change in Developing Countries - A Conference of the

Universities - National Bureau Committee for Economic Research, Princeton University

Press, Princeton

Katz, M.L and Shapiro, C (1985) Network externalities, competition, and compatibility

American Economic Review, 75 (3), 424–440.

Laffont, J.J., Rey, P and Tirole, J (1998) Network competition: II Price discrimination Rand

Journal of Economics, 29 (1), 38–56.

Manski, C.F (1993) Identification of endogenous social effects – the reflection problem Review

of Economic Studies, 60 (3), 531–542.

Snijders, T.A., van de Bunt, G.G and Steglich, C.E (2010) Introduction to stochastic

actor-based models for network dynamics Social Networks, 32 (1), 44–60.

Sundararajan, A (2007) Local network effects and complex network structure The B.E.

Journal of Theoretical Economics, 7 (1), Article 46.

2.5 Empirical research on social networks and network effects 32

2.5.3 Challenges when identifying causal relationships in social

2.5.4 Empirical strategies to identifying causal effects in social

Social Networks and their Economics: Influencing Consumer Choice, First Edition Daniel Birke.

© 2013 John Wiley & Sons, Ltd Published 2013 by John Wiley & Sons, Ltd.

12 SOCIAL NETWORKS AND THEIR ECONOMICS

2.5.5 Empirical challenges and advances in the economics of

mar-The economics literature on network effects, starting with (Rohlfs, 1974) andKatz and Shapiro (1985)/Farrell and Saloner (1985), analyses how one consumer’sproduct choice can have a positive (or more rarely negative) impact on anotherperson consuming the product This research has a strong focus on aggregate leveloutcomes – for example, how these consumption externalities lead to a strong marketconcentration and the choice of one dominant standard, like VHS for video recorders.The structure of social networks is typically not taken into account, but consumersare assumed to be, on average, uniformly influenced by other consumers

The focus on network structure is strongest in sociology where many of the basicmethodologies for social network analysis were developed Duesenberry (1960, 233)famously quipped that ‘Economics is all about how people make choices Sociology

is all about why they don’t have any choice to make.’ Social network analysis has, inrecent years, spread from its beginnings in sociology and is now broadly used acrossdisciplines

Physicists became seriously interested in the properties of complex networks ticularly after the studies by Watts and Strogatz (1998) on small world networks andBarabasi and Albert (1999) on scale-free networks They were particularly interested

par-in the properties of large-scale networks from a variety of sources (medicpar-ine, ogy, infrastructure networks etc.) with a particular emphasis on how properties of theoverall network can be explained by micro-level processes

biol-The last decade has seen a strong growth in empirical research using verylarge social network datasets in a variety of fields This research typically usescustomer-level data and, in recent years, has used electronic social network data fromtelecommunications or Internet companies to analyse how consumers influence eachother, and to some extent how companies can make use of these influence processesfor marketing purposes The latest research is particularly interested in the identifica-tion of causal effects (friends influencing each other) as opposed to correlated effects(friends are similar to each other and therefore behave in a similar way) Teasingthese two effects apart is critical for choosing the right marketing/research strategy

1 See also Birke (2009) in which part of the material from this chapter has been published and which

in particular provides further material on the network effects literature in economics Reproduced by permission of Wiley.

RESEARCH INTO SOCIAL NETWORKS 13

This chapter is organised as follows: Section 2.2 reviews the key findings fromthe economics and marketing literature and their implication for company strategies.Section 2.3 focuses on the sociology literature on social network analysis, whileSection 2.4 discusses the key findings from the literature on complex networks Inline with the emphasis of this book on empirical research, Section 2.5 reviews thelatest developments in the empirical literature on social networks, with a particularemphasis on social network analysis as a technique to derive value from very large-scale social network datasets and on identifying causality

economics and marketing

In economics there is a large body of literature on so-called network effects Networkeffects exist if a consumer’s benefits from using a product increase with the number of

other consumers using the product Network effects are direct if the benefits are driven

by the availability of more interaction partners, as is the case in telecommunicationswhere users are interested in being able to communicate with other users However,contrary to the standard assumption in this body of literature, in markets with directnetwork effects consumer utility is typically only strongly affected by adoptiondecisions of people with whom they interact, which typically is only a small fraction of

the overall network members For indirect network effects, the interest of the consumer

is typically not in the direct interaction with his peers, but rather in the availability

of complementary services Buying a mobile phone, consumers are interested in theoverall number of software apps available for their operating system, but by and largenot in who is using the app.2

Although there has been earlier work on consumption interdependences (seefor example Leibenstein (1950) and the references therein), research in the morenarrowly and more precisely defined area of network effects started in the 1970swith Rohlfs’ seminal article (Rohlfs, 1974).3 Rohlfs notes that the utility of a user

of a communications service increases as other consumers use the service as well.The paper focuses on an equilibrium analysis and on how to overcome the start-

up problem when nobody uses a particular service Contrary to most of the laterliterature, Rohlfs (1974) actively discusses the assumption of a uniform interaction

of users and argues that the existence of small self-sufficient user sets greatly reducesthe start-up problem for a new service

2 The exceptions are apps that allow users to directly interact with each other (like WhatsApp, Skype etc.) which exhibit direct network effects.

3Rohlfs’ article uses the term consumption interdependences and only after Katz and Shapiro (1985) did the term network effects become widely popular Nowadays the latter term is also commonplace in

related disciplines, like marketing.

14 SOCIAL NETWORKS AND THEIR ECONOMICS

In the 1970s and early 1980s, research on demand externalities similar to thephenomena later coined as network effects was mainly conducted in the context of thetelecommunications industry and, as the industry was organised as state monopolies

at that time, the setting for research was of a monopoly context (Rohlfs, 1974; Orenand Smith, 1981) A particular focus of interest was the optimal pricing strategy toensure that product diffusion is supported and a critical mass of users is achieved.Dybvig and Spatt (1983), for example, develop a model of adoption externalities in avery broad sense The main focus of this paper is on how government intervention can

be a remedy to public good problems4 due to such adoption externalities, as initialadopters might incur losses, but have a strong positive impact on later adopters Theproblems discussed are the loss incurred by initial adopters if product diffusion isnot widespread, diffusion inertia and uninternalised positive externalities of earlyadopters on later adopters In 1985, the seminal papers by Katz and Shapiro (1985)and Farrell and Saloner (1985) started an avalanche of theoretical work on networkeffects Since the mid-1990s, a variety of industries have also been covered byempirical studies on network effects5 Overviews and discussions of the networkeffects literature can be found in Shy (2001), Shapiro and Varian (1999) and Rohlfs(2001) Shy (2001) has a stronger focus on the theoretical literature, while Shapiroand Varian (1999) and Rohlfs (2001) focus on practical business implications of theexistence of network effects

Katz and Shapiro (1985, 424) characterise network products as ‘ products forwhich the utility that a user derives from consumption of the good increases with thenumber of other agents consuming the good.’

There is a close relationship with economies of scale on the supply-side andnetwork effects are sometimes also referred to as demand-side economies of scale.However, in contrast to supply-side economies of scale, network effects are typicallynot limited to one firm but include the whole compatible network of a technology.Network effects can thus be a source of externalities, as coordination problems arisethat are not present for economies of scale

4 Public goods problems are closely related to the ‘free-rider’ problem or the tragedy of the commons,

in which people not paying for the good may continue to access it.

5 For example, Fax-machines (Economides and Himmelberg, 1995), software (Brynjolfsson and Kemerer, 1996; Gr¨ohn, 1999), wired telecommunications (Majumdar and Venkataraman, 1998), digital

TV (Gupta, Jain and Sawhney, 1999), CD-players (Gandal, Kende and Rob (2000)), computers (Goolsbee and Klenow, 2002), DVD-players (Dranove and Gandal, 2003), antiulcer drugs (Berndt, Pindyck and Azoulay (2003)), video cassette recorders (Ohashi, 2003; Park, 2004), home video games (Shankar and

Bayus, 2003; Clements and Ohashi, 2005), P2P music sharing (Asvanund et al., 2004), personal digital

assistants (Nair, Chintagunta and Dub´e, 2004), yellow pages (Rysman, 2004), electronic payment systems (Gowrisankaran and Stavins, 2004; Tucker, 2005; Ackerberg and Gowrisankaran, 2006), payment cards (Rysman, 2007) and electronic medical records (Miller and Tucker, 2009) See also Shurmer and Swann (1995) for a simulation approach.

RESEARCH INTO SOCIAL NETWORKS 15

Network effects do not necessarily arise only from current sales, but may beaffected by expectations of future sales (Farrell and Saloner, 1986) This is becausenetwork products are durable goods and are, therefore, used over a considerableperiod of time Furthermore, component purchases for a single system are spreadover time Purchasing a mobile phone handset, the consumer is not only interested inthe number of other users in the network now, but also in the number of users in thenetwork over the mobile phone’s life cycle

Direct network effects are present if the quality of a good is directly linked to thenumber of other consumers of the same good In economics terms: an individual’sutility function is not independent of other individuals’ consumption choices Instead,utility increases with the number of other individuals present in the same network.The classic example of a direct network effect is a telecommunications network, inwhich the utility is zero if one is the only user of a particular technology Being theonly person using fax or e-mail is obviously of little value and value is only created

if there are other people using the same technology

Network effects do not only arise in physical networks like telecommunicationsnetworks, but are also present in many virtual networks In financial markets, a highernumber of traders increases the utility for other traders in ensuring that a minimumliquidity in the market is present Here, the relevant network is the virtual network

of people trading in a specific market Other examples of virtual networks includeonline markets like e-bay and betting exchanges, as the value these markets are able

to create for their customers depends directly on the number of people taking part in

a transaction/auction Likewise, a virtual network can be established by institutional

or informal rules and conventions, for example, rules on which side of the road todrive or a language network In particular for driving rules, it is immediately clearthat every driver benefits from other drivers using the same road side Furthermore,the value of a sporting event or a music concert is influenced by the aggregate size

of the audience and value is created by the presence of other spectators (Economidesand Flyer, 1997)

For software products, direct network effects arise from the exchange of files andlearning spillovers between users Using Microsoft Word software, the utility directlyincreases with the number of other people also working with Microsoft Word, as easyand convenient file exchange is possible The strength of this network effect positivelycorrelates with the frequency of file exchange It is high for those types of softwarewhich require a frequent exchange of files (like word processing), while it is lowerfor other types, like one-player computer games

Finally, network effects can be introduced by companies for strategic reasons Forexample, mobile phone carriers often charge higher prices for off-net than for on-netcalls, something which Laffont, Rey and Tirole (1998) termed tariff-mediated networkeffects In most European countries such price differentiation is commonplace, but

16 SOCIAL NETWORKS AND THEIR ECONOMICS

there are also exceptions like the Netherlands where operators charge the same pricesfor calls to the same network and calls to other networks.6

With direct network effects, the utility function of a user is typically modelled asincreasing monotonically in the number of other users in the network In mathematical

notation, U i = U i (x i ,x j) and δU i /δU j > 0 for at least one j = i, where u i the

utility of user i positively correlates with the number of other users

x j in thenetwork More generally, the utility function increases with network size: that is,

the first derivative of the utility function is positive (U(N ) > 0).

However, as I argue and show throughout this book, the traditional assumptionthat overall network size matters is often unrealistic for markets with direct networkeffects, as many users are most interested in a small subset of people they know and

to a lesser degree in the option to interact with the large rest of the network Tucker(2006) therefore distinguishes between interactive and non-interactive networkeffects Interactive network effects arise from direct interaction between users of

a technology Users of mobile phones, for example, benefit from other users ofmobile phones by calling, texting or otherwise interacting with them through theirmobile phones Tucker (2006) argues that the traditional assumption of the solerelevance of total network size can be justified by so called option-value or dominonetwork effects, which are both non-interactive network effects Option-value

network effects describe the utility that a user derives from the potential interaction

with other users Calling partners constantly change and overall network size isthe relevant size if consumers are completely myopic Furthermore, access to someservices, like the emergency services, can be valued by users even without any directinteraction

Another reason why overall network size might matter for markets with directnetwork effects is so-called domino network effects, which describe the impact thatother users might have on the probability of direct contacts adopting a technology

In other words, overall network size can have an indirect influence on the adoptionprobability of users that the consumer actually wants to interact with Another reasonwhy overall network size can be important to consumers is the likely correlationbetween network size and the success probability of a technology, something that allusers are interested in, regardless of their specific interaction patterns There are manynetwork technologies which have lost out to competitors (think Beta or MySpace)and therefore have had a dramatically decreased value for the end consumer.Sundararajan (2007) introduces a game theoretic model of network effects in asocial network The assumptions are similar to Tucker’s domino network effects inthat every consumer indirectly affects the equilibrium outcome of adoption decisions

of every other consumer Although every consumer is connected to a different set ofindividuals, these sets are overlapping and, therefore, interdependent In Figure 2.1,individuals A and B both have a direct connection with each other, have a commonfriend C and friends who they do not have in common However, these non-common

6 We will come back to these differences in pricing strategy between countries and operators in Chapter 5 as they help to establish causality.

RESEARCH INTO SOCIAL NETWORKS 17

Figure 2.1 Overlapping social networks.

Source: Adapted from Sundararajan (2007).

friends have an indirect impact on adoption probabilities Individuals L to O influencethe adoption decision of individual A and, therefore, also indirectly the adoptiondecision of individual B

2.2.4 Indirect network effects

Indirect network effects are generated by complementary relations between goods.Such complementary goods share many essential features with goods which exhibitdirect network effects Thum (1995) identifies three complementary relations:

r Consumption interdependences between complementary products: many

prod-ucts have little or no value in isolation, but generate value when combined withother complementary products Computer hardware for example is only ofvalue, if computer software is available at the same time Likewise, many con-sumer electronics products like DVD-players require the availability of contentfor their systems

r Learning effects and informational spillovers: these are present for any

tech-nology requiring specific training (Katz and Shapiro, 1986b) The training here

is a complementary product to the technology The most famous example isthat of the QWERTY-keyboard David (1985) argues that the alternative DSK(Dvorak Simplified Keyboard) is superior and that the only reason why todaynobody uses DSK is that, due to network effects with complementary learn-ing how to use the keyboard, the market has already locked into the inferiorQWERTY-standard.7

r Uncertainty: In durable product markets where products are used for a long

time and where network size increases the likelihood of a service network or

7 These claims from David (1985) have been disputed by Liebowitz and Margolis (1990).

18 SOCIAL NETWORKS AND THEIR ECONOMICS

complementary products being available during the product life cycle, tainty favours products which are widely in use Spare printer cartridges, forexample, are more likely to be available for printers made by a leading printermanufacturer than from rather obscure manufacturers Likewise, it is easier tofind a garage capable of servicing a Ford than a garage capable of servicing aLamborghini

uncer-In the literature, consumption interdependences between complementary productsare the most commonly discussed source of indirect network effects, typically underthe analogy of hardware (network) product and software (complementary) product(Church and Gandal, 1992; Katz and Shapiro, 1994) An indirect effect arises forproducts where hardware sales influence software sales and vice versa The positivenetwork effect for hardware sales stems either from economies of scale in softwareproduction or from benefits arising from the availability of increased software variety.The feedback circle can have salient features where increased hardware sales lead

to an increase in software variety, which in turns leads to increased hardware sales.Conversely, the feedback circle can also result in a ‘chicken and egg’-problem, that is,

no hardware is supplied because no software is available and no software is suppliedbecause no hardware is available

In the telecommunications industry such coordination problems have traditionallybeen solved by vertically integrated national monopolies Network externalities arethen internalised and take the form of economies of scope (Economides, 1994) Withthe liberalisation of these markets, vertical integration has been lowered and hardwareand software supply now has to be coordinated both within companies and throughthe market Besides the widely conceived positive effects of market liberalisation,this also might bring along new problems In software production, economies of scaleare often present, which can result in inefficient technology adoption, if developmentcosts for software are lower for one (inferior) hardware technology (Church andGandal, 1993)

In practice, products often exhibit both direct and indirect network effects atthe same time Mobile telecommunications networks increasingly do not only offerconnectivity to other users (direct network effect), but also complementary services(indirect network effects), such as mobile Internet applications Often, these effectsare also of opposite direction In air transport, the consumer benefits from more(direct) travel destinations, if there are more people using air travel (indirect networkeffect) On the other hand, this might lead to congestion or reduce the exclusivity ofthis transport mode (direct network effect)

Markets with network effects exhibit a number of special characteristics which havefar-reaching implications for company strategies, especially in areas such as the nature

of competition in the market, technology standardisation, technology diffusion, thetransition between different technology generations, pricing strategy and anti-trust.These characteristics are discussed in turn in the following sections

RESEARCH INTO SOCIAL NETWORKS 19

2.2.5.1 Competition in the market is replaced by competition for

the market

Network markets can have strong winner-takes-all tendencies As network marketstend to tip, the winning technology in a standardisation battle promises to yield highpayoffs However, rents in later periods are partially dissipated in earlier periods:

competition in the market is replaced by competition for the market This competition

for the market can be very intense and often Schumpeterian competition races formarket dominance can be observed, as in the dot.com competition in 1999–2000(Economides, 2004)

Interestingly, on the one hand companies are forced to cooperate in order toincrease overall market size and the ever-growing importance of standardisationissues mirrors this need for cooperation and coordination (Grindley, 1995) On theother hand, there is stiff competition after some level of standardisation has beenreached Ray Noorda, founder of the networking software company Novell, inventedthe neologism ‘coopetition’ to describe this tension between cooperation and compe-tition in markets with network effects (Brandenburger and Nalebuff, 1996) Whethercompanies should pursue a more aggressive winner-takes-all strategy or a strategyfocussed more on cooperation is a question of the strength of network effects, regu-lation in the market and of timing

Case study: The VHS vs beta standards battle

For a number of years after the introduction of Beta in 1976, VHS and Betahad a fierce battle for market domination Ohashi (2003) studied this battle

in the US video cassette recorder (VCR) market between 1976 when Sonyfirst launched Beta and the mid-eighties when VHS commanded a marketshare of over 90% However, according to results from empirical work byOhashi (2003), the format battle was all but over in 1981–1982 By thistime, VHS had already overcome Beta’s initial first mover advantage andnetwork effects were strongly working in favour of VHS Interestingly,Ohashi’s results show that network effects initially were unimportant, asVCR were mainly used for recording purposes Network effects only grewstronger with the development of a rental market of pre-recorded videos atthe beginning of the 1980s These pre-recorded videos are complementaryproducts to the VCR machines and the availability of these products istherefore important to consumers when making purchasing decisions forVCR machines Availability of pre-recorded video titles is again interde-pendent with the installed base of VCR machines, as it is more beneficialfor content providers and video rental stores to use a more widely adoptedsystem

By the time Sony realised that it had fallen behind, market positionswere already entrenched and its new strategy to move into medium andhigh-value VCR in 1983 was not able to change this As Ohashi’s resultshows, this was the case although Beta’s product characteristics itself were

20 SOCIAL NETWORKS AND THEIR ECONOMICS

regarded favourably by consumers after this up-market move and weresuperior to VHS’s In a counterfactual simulation, Ohashi finds that Sonywould have been able to win the standard battle if it had used strategicpricing in the crucial years 1978–1981 and had incurred initial losses togrow its installed base of consumers and corner the market

2.2.5.2 Standardisation is more beneficial, but harder to achieve

One basic finding of the network effect literature is that social and private incentives

to standardise often do not coincide (Katz and Shapiro, 1985; Katz and Shapiro,1986a) Companies that do have a good market position because of a bigger networksize are found to have lower incentives to standardise.8 Standardisation issues areparticularly interesting in a dynamic context The self-propelling growth results in

a tendency for network markets to ‘tip’ in favour of one standard As Arthur (1989,116) points out, ‘ a technology that by chance gains an early lead in adoptionmay eventually “corner the market” of potential adopters, with the other technologiesbecoming locked out’ The notion of path dependence is, therefore, at odds with thecommonly held view that a natural selection process results in the survival of thefittest technology by adoption of the superior technology Case studies of inferiorstandardisation include Besen and Johnson (1986) on AM radio, Cowan (1990) onthe nuclear power industry, Postrel (1990) on quadraphonic sound and Cowan andGunby (1996) on pest control strategies Foray (1997) gives an excellent review ofthe empirical literature on path dependence and lock-in

That path dependence is not ubiquitous, stresses the significance of nication, planning, property rights and other market institutions to overcome suchsituations (Liebowitz and Margolis, 1995) In reality, a variety of markets that exhibitnetwork effects nevertheless have more than one product surviving Apple, for exam-ple, was able to cater to a niche in the computer market for a long time, beforerelaunching itself as the most successful consumer electronics company The AppleMac example is of particular interest with regard to the argument that network struc-ture matters, as it is widely used in some user communities (for example creativeindustries and media), but hardly at all in others (Suarez, 2005) Another reason forsuch findings can be found in heterogeneity between adopters of a technology There

commu-is a dcommu-istribution of preferences for a product and consumers might have a demand forvariety that cannot be satisfied by a single standard/product

The degree of openness of a standard has also been discussed as a key variable

to influence the success of a standard (Shapiro and Varian, 1999) An open policy

is likely to attract more producers of complementary products and customers, asthey do not want to become dependent on a single company For anyone launching

a new technology, there is a basic trade-off between offering a proprietary product,which faces more difficulties in market development, and an ‘open’ product

8 This result carries over to other aspects of network effects Network effects, like economies of scale, favour size – either at the standard/product-level or at the company-level.