Information exchange and behavior: A multi-method inquiry on online communities

Korfiatis, Nikolaos Theodoros Doctoral Thesis Information exchange and behavior: A multi-method inquiry on online communities PhD Series, No. 13.2009 Provided in Cooperation with: Copenhagen Business School (CBS) Suggested Citation: Korfiatis, Nikolaos Theodoros (2009) : Information exchange and behavior: A multi-method inquiry on online communities, PhD Series, No. 13.2009, ISBN 9788759383919, Copenhagen Business School (CBS), Frederiksberg, https://hdl.handle.net/10398/7797 This Version is available at: https://hdl.handle.net/10419/208721 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/ LIMAC PhD School Programme in Informatics PhD Series 13.2009 PhD Series 13.2009 Information Exchange and Behavior copenhagen business school handelshøjskolen solbjerg plads 3 dk-2000 frederiksberg danmark www.cbs.dk ISSN 0906-6934 ISBN 978-87-593-8391-9 Information Exchange and Behavior A Multi-method Inquiry on Online Communities Nikolaos Theodoros Korfiatis CBS PhD nr 13-2009 Nikolaos Korfiatis • A5 OMSLAG - GB.indd 1 27/05/09 9:40:14 Information Exchange and Behavior: A Multi-method Inquiry on Online Communities Dissertation submitted in partial fulfillments for the degree of Doctor of Philosophy by Nikolaos Theodoros Korfiatis LIMAC PhD School Programme in Informatics Copenhagen Business School, Department of Informatics Contents Abstract........................................... XI DanskResume....................................... XIII Foreword and Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . XV I Theoretical background 1 1 Introduction 2 1.1 Introduction to this dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1.1 Web enabled production and use of information . . . . . . . . . . . . . 6 1.1.2 Research background and motivation . . . . . . . . . . . . . . . . . . . 8 1.1.3 The problem of Motivating Contributions to virtual forms of social capital........................................ 27 1.2 Theresearchobjectives................................. 30 1.3 Research approach and methodology . . . . . . . . . . . . . . . . . . . . . . . . 34 1.3.1 Online communities and Information Systems research . . . . . . . . 35 1.3.2 The positivist view in IS research . . . . . . . . . . . . . . . . . . . . . . 36 1.3.3 Methodology.................................... 41 1.4 Structure of this dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1.4.1 Chapterstructure................................. 48 1.4.2 Datasets used in the empirical part . . . . . . . . . . . . . . . . . . . . . 52 2 Models and Theories for Understanding and Motivating Contributions in Online Communities 66 2.1 The social and economic cases for an online community . . . . . . . . . . . 66 I CONTENTS 2.2 Structuralapproaches.................................. 76 2.2.1 Social network analysis and online communities . . . . . . . . . . . . 76 2.2.2 The weak ties hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 2.2.3 Collective behavior and social loafing . . . . . . . . . . . . . . . . . . . 82 2.2.4 Reputationeffects ................................ 84 2.3 Social dilemmas and knowledge sharing on online communities . . . . . . 85 2.3.1 The argument of social preferences and the public goods dilemma 88 2.3.2 Social preferences and aversion to inequity . . . . . . . . . . . . . . . . 89 2.3.3 Fairness and reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 2.4 Models for understanding and enhancing activity in online communities . 93 2.4.1 The collective effort model . . . . . . . . . . . . . . . . . . . . . . . . . . 94 2.4.2 The Model of Whittaker et al. [1998]. . . . . . . . . . . . . . . . . . . . 96 2.4.3 The Model of Jones Ravid and Rafaeli . . . . . . . . . . . . . . . . . . . . 97 2.4.4 TheModelofButler ............................... 98 2.4.5 The Model of Wasko and Faraj . . . . . . . . . . . . . . . . . . . . . . . . . 99 2.5 Outlook to the empirical part of this dissertation . . . . . . . . . . . . . . . . . 100 II Empirical part 111 3 Behavioral Characteristics and Cooperation in Online Communities: An Experimental Investigation 112 3.1 Online communities and online cooperation . . . . . . . . . . . . . . . . . . . . 113 3.2 Motivation.......................................... 116 3.3 Cooperation and the public goods game . . . . . . . . . . . . . . . . . . . . . . 119 3.4 Experimental procedure and methods . . . . . . . . . . . . . . . . . . . . . . . 125 3.4.1 Experimental protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 3.4.2 Assignment to treatments . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 3.5 Data analysis and procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 3.5.1 Distribution to treatments and basic demographics . . . . . . . . . . 131 3.5.2 Defining online sociability . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 3.5.3 Online sociability and cooperation . . . . . . . . . . . . . . . . . . . . . . 144 3.6 Discussionandresults.................................. 153 3.7 Conclusion and further remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 4 Effort, Benefits and Commitment on Online Knowledge Communities: An empirical study on Yahoo!Answers 160 4.1 Introduction ........................................ 161 4.2 Benefit, effort and commitment on online communities . . . . . . . . . . . . 166 II CONTENTS 4.3 The Yahoo! Answers online service . . . . . . . . . . . . . . . . . . . . . . . . . 167 4.3.1 Dataset variables and description . . . . . . . . . . . . . . . . . . . . . . 169 4.3.2 Paneldatavariables............................... 174 4.3.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 4.4 Methodsandconstructs................................. 178 4.4.1 Constructssummary .............................. 178 4.4.2 Estimationresults ................................ 183 4.4.3 Relation between effort and benefit . . . . . . . . . . . . . . . . . . . . . 186 4.5 Discussion ......................................... 188 4.6 Conclusions......................................... 191 5 The impact of Extrinsic Rewards on Strategic Interaction in Online Communities: An analysis on Google!Answers 197 5.1 Introduction and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 5.2 Characteristics of tipping as an extrinsic reward . . . . . . . . . . . . . . . . . 204 5.3 The Google Answers online community . . . . . . . . . . . . . . . . . . . . . . . 209 5.3.1 Dataset and Variables Description . . . . . . . . . . . . . . . . . . . . . . 213 5.3.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 5.4 Methodsandconstructs................................. 219 5.4.1 Definition of constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 5.4.2 What drives tipping in general in Google Answers? . . . . . . . . . . 222 5.4.3 The effect of tipping to service quality . . . . . . . . . . . . . . . . . . . 230 5.5 Discussion ......................................... 236 5.6 Conclusions and further research . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 6 Evaluating Content Quality and Usefulness of Online Product Reviews245 6.1 Introduction ........................................ 246 6.2 A background on readability tests . . . . . . . . . . . . . . . . . . . . . . . . . . 250 6.2.1 The Gunning-Fog Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 6.2.2 The Flesch Reading Ease . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 6.2.3 The Automated Readability Index . . . . . . . . . . . . . . . . . . . . . . 254 6.2.4 The Coleman-Liau Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 6.3 Analysisandresults ................................... 255 6.3.1 Data collection and definition of variables . . . . . . . . . . . . . . . . . 255 6.3.2 Analysisandresults............................... 264 6.4 Discussion ......................................... 272 6.4.1 The usefulness of a review is affected by its positive or negative ratingvalue .................................... 272 III CONTENTS 6.4.2 The usefulness of a review is affected by its qualitative characteristics......................................... 273 6.4.3 The rating that the review provides is affected by its qualitative characteristics................................... 273 6.5 Conclusions and further remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 III Findings and Conclusions 278 7 Conclusions and retrospect 279 7.1 Discussion ......................................... 279 7.1.1 Importance of Signaling Mechanisms . . . . . . . . . . . . . . . . . . . 281 7.1.2 Identification of the Behavioral Characteristics . . . . . . . . . . . . . 283 7.1.3 Ability of the participants to interact strategically . . . . . . . . . . . 284 7.1.4 Importance of the quality evaluation mechanisms . . . . . . . . . . . 284 7.2 Conclusions......................................... 285 7.2.1 Retrospect ..................................... 285 7.2.2 Revisiting the general research question . . . . . . . . . . . . . . . . . 287 7.2.3 Summary of the findings and the implications of the empirical studies .......................................... 288 7.2.4 Additional contributions and discussion . . . . . . . . . . . . . . . . . . 292 7.3 Where do the findings apply? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 7.4 Limitations ......................................... 297 7.5 Topicsforfutureresearch................................ 299 IV List of Figures 1.1 The stages of positivist research in information systems research . . . . . 37 1.2 Confirmatory vs. Exploratory Research . . . . . . . . . . . . . . . . . . . . . . . 38 1.3 The general epistemological framework for the relation between theory andempiricalobservation ............................... 42 1.4 Two treatments under the ceteris paribus condition . . . . . . . . . . . . . . 44 1.5 Aspects of Validity for Experimental Results . . . . . . . . . . . . . . . . . . . 45 1.6 The structure of this dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.1 A sample representation of a community structure . . . . . . . . . . . . . . . 71 2.2 An interaction pattern representing a thread of an internet newsgroup . . 72 2.3 The set of different motifs formed in a triadic relation . . . . . . . . . . . . . 77 2.4 Types of connection degree in the network . . . . . . . . . . . . . . . . . . . . 79 2.5 TheWeakTiesHypothesis ............................... 80 2.6 Virtual community stimulation structure . . . . . . . . . . . . . . . . . . . . . . 94 2.7 The collective effort model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 2.8 A causal model of communication in a newsgroup . . . . . . . . . . . . . . . 97 2.9 Information overload and cognitive effort . . . . . . . . . . . . . . . . . . . . . 98 2.10Membership Size, Communication Activity and Sustainability . . . . . . . . 99 2.11The Wasko and Faraj model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 3.1 The login Interface of the Web application . . . . . . . . . . . . . . . . . . . . . 126 3.2 The evolution of the participation rate for the two waves that were used inthisexperiment .................................... 128 3.3 The experimental protocol used in this study. . . . . . . . . . . . . . . . . . . . 129 V LIST OF FIGURES 3.4 Distribution of genres representation in the experiment . . . . . . . . . . . . 134 3.5 Representation of genre among treatments . . . . . . . . . . . . . . . . . . . . 135 3.6 Distribution of subjects by treatments and age group . . . . . . . . . . . . . 136 3.7 Variation of the Education Level among the Subjects for First and Second waveoftheexperiment................................. 137 3.8 Computer usage among subjects of different age groups. . . . . . . . . . . 141 3.9 Participation and communication on online social networks for all the subjects participating on the experiment. . . . . . . . . . . . . . . . . . . . . . . . 142 3.10Communication Capacity, Structural Capacity and Online Sociability . . . 143 3.11Anticipated average contribution stated by the subject for the other membersofthegroup...................................... 147 3.12Average Conditional Contribution Schedules . . . . . . . . . . . . . . . . . . . 149 4.1 The Yahoo!Answers modus operandi . . . . . . . . . . . . . . . . . . . . . . . . 169 4.2 An example interface from the Yahoo!Answers online service . . . . . . . . 170 4.3 Number of Answers and Number of Questions Posted over one month period ............................................ 174 4.4 Relation between time to answer and the Number of Users . . . . . . . . . . 177 4.5 Number of Questions Answered with the total number of answers posted ontheonlineservice................................... 178 4.6 Service Benefit, Contribution Effort and Commitment . . . . . . . . . . . . . 179 4.7 Benefit-Effort elicitation for the user states . . . . . . . . . . . . . . . . . . . . 181 4.8 The results of the evaluation of our model. . . . . . . . . . . . . . . . . . . . . 185 4.9 Comparison between means for the number of answers across the two groups............................................. 189 4.10Comparison between means for the time to answer (minutes) across the twogroups.......................................... 189 5.1 Process Flow of interaction in Google Answers . . . . . . . . . . . . . . . . . . 210 5.2 Questions Posted per month to the webservice . . . . . . . . . . . . . . . . . 211 5.3 Number of unique askers and researchers per month . . . . . . . . . . . . . 212 5.4 Average Tip and Average Price per month . . . . . . . . . . . . . . . . . . . . . 213 5.5 DataCollectionMethod................................. 214 5.6 Generation of Panel Data variables . . . . . . . . . . . . . . . . . . . . . . . . . 220 5.7 Distribution of Tip relative to the quality index . . . . . . . . . . . . . . . . . . 231 5.8 An ideal case of reciprocation between tip and quality provided . . . . . . 233 5.9 An ideal case of tipping reward as a social norm . . . . . . . . . . . . . . . . . 234 VI Dansk Resume Denne afhandling undersøger informationsudveksling og deltageres adfærdskarakteristika i forbindelse med online fællesskaber. På baggrund af forskningsundersøgelser karakteriseres online fællesskaber som grupper af individer, der bruger computermedieret kommunikation til at lette interaktionen i forbindelse med fælles formål og / eller mål. Det hævdes at dette samspil skaber eksternaliteter, som i dette tilfælde er kodificerede oplysninger der kan anvendes af andre deltagere ved at udnytte søgefunktioner på nettet. Disse eksternaliteter etablerer en virtuel form for social kapital. Ved teoretisk at bestemme social kapital som en delt ressource, er forskningsmålsætningen med denne afhandling at adressere om forholdet mellem deltagernes adfærd er påvirket af måden hvorpå denne fælles ressource er formateret, administreret og delt. Afhandlingen består af to dele, en teoretisk del, hvor den empiriske baggrund og genstand for forskningsundersøgelsen er fremhævet, og en empirisk del, der består af fire empiriske undersøgelser foretaget i forbindelse med tre online fællesskaber nemlig Google Answers, Yahoo! Answers og Amazon Online Reviews. For at skabe en generel forståelsesramme begynder den empiriske del af denne afhandling med et kontrolleret forsøg på at efterligne et velkendt socialt dilemma, The Public Goods game. Denne undersøgelse bidrager med indsigt i, om deltagere i online fællesskaber ønsker at samarbejde eller ej. De næste to undersøgelser fokuserer på et særligt tilfælde af online fællesskaber, hvor deltagere stiller spørgsmål og andre deltagere svarer. Resultaterne af disse to undersøgelser bekræfter, at deltagerne i disse fællesskaber er interesserede i hinandens bidrag og udformer deres adfærd i overensstemmelse hermed. På Yahoo! Answers gør deltagerne en indsats for fællesskabet ved at svare på spørgsmål, men får samtidig gavn af den indsats, der leveres af andre deltagere. De empiriske resultater viser, at deltagere som yder en større indsats for onlinebrugere, ved at bidrage med svar på de øvrige deltagere spørgsmål, får svar hurtigere, mens dem der ikke yder en stor indsats i samfundet bliver sanktioneret XIII således, at der går længere tid, før de modtager et svar på deres spørgsmål. På Google Answers, hvor interaktion er baseret på monetære belønninger (snarere end sociale belønninger i form af omdømmeindeks som i Yahoo Answers) gør deltagerne brug af frivilligt tildelte udbetalinger (tips) sammen med explicitte belønninger, med henblik på at motivere dem, der kan give svaret (svarerne) til at levere bedre kvalitet i deres svar. Resultaterne af denne undersøgelse bekræfter den symmetriske virkning mellem monetære belønninger og kvalitet, men identificerer også andre tilfælde, hvor sociale normer kan have en betydelig virkning. Et særligt tilfælde er, når deltagerne søger at opnå bedre service med mindre indsats (målt i samlede omkostninger), ved at opbygge et omdømmeindeks, der bygger på deres tidligere interaktioner. Svarerne interesserer sig for vurderingen af deres svar, omdømme og historie og tilpasser deres adfærd i overensstemmelse hermed. Det sidste kapitel i den empiriske del fokuserer på en anden egenskab ved oplysninger som en fælles ressource: klarhed og forståelighed. Baggrunden for undersøgelsen, der benyttes i dette tilfælde, er online anmeldelser på Amazon.com. Resultaterne antyder at deltagerne er interesserede i klarheden af denne kodificerede form for erfaringer og belønner (med et hjælpsomhedsindeks) i overensstemmelse hermed. Afhandlingen konkluderer overordnet, at der er symmetrisk effekt mellem deltagelse i online communities og output for interaktion, men peger også på deltagernes evne til at interagere strategisk, idet de søger at minimere den indsats de yder for at finde de oplysninger de søger. Resultaterne understreger vigtigheden af signalog kvalitetsevalueringsmekanismer som modvægtskontrol der kan øge aktiviteten i online praksisfællesskaber. Acknowledgements This dissertation marks the “ithaca” to which I endeavored when I started working as an undergraduate research assistant for the Electronic Trading Research Unit (ELTRUN) at the Department of Management Science and Technology at Athens University of Economics and Business. I would like to thank Professor George Doukidis for taking the risk of offering to a second year undergraduate student at that time, the opportunity to become an active member of a fruitful research environment and participate in the research activities of the group. Angeliki Poulymenakou had been a great support during my introduction to the Information Systems research methodology working with the Organizational Information Systems group. Diomidis Spinellis had been a great mentor and teacher always pointing me to seek the practical and the efficient. My deepest gratitude goes to Miltiades Lytras for encouraging me to pursue research as well as for listening and supporting me in any personal and education related matter. All the time I spent as a student in Athens was a nice memory because of him. My research working experience with Ambjörn Naeve and his research group in Royal Institute of Technology (KTH) in Stockholm as well as Mathias Palmer of Uppsala Learning Lab in Uppsala University had been a breakthrough towards my decision to pursue a PhD degree. I would personally like to thank also all the members of the Information Engineering Research Unit in the University of Alcalá in Spain for being supportive during my research visits. I would like to specially thank my long term colleague and collaborator Miguel-Angel Sicilia as well as Daniel Rodríguez-García, Salvador Sánchez-Alonso and Elena García-Barriocanal for making my research and course visits in Spain a memorable experience. I would like to thank all the people of the Informatics Department at Copenhagen Business School: Jan Damsgaard for giving me the opportunity to be part of the PhD programme. Ioanna Constantiou for guiding my first steps in Denmark and giving her support in all the matters that might arise for a foreigner in a new country. The working experience in the department of informatics has been most enjoyable. I still rememXV ber the discussions and fruitful collaboration I had with Sudhanshu Rai, Mogens Kühn Pedersen and Rasmus Pedersen. I would like to thank Dorte Madsen and Niels Bjørn Andersen for offering me the opportunity to teach in the undergraduate programmes in Information Management and Business Administration and Computer Science. My special thanks to Karlheinz Kautz for the insightful hints that he has given me through the PhD process as well as to Martin Tong for being supportive in my technical inquiries during my employment period. It would have been rather impossible to finalize this dissertation without the guidance of my supervisor Volker Mahnke. Volker had been a great support all this years and i owe him a large portion of the maturity that a young PhD student can gain from the PhD process. I would also like to thank my friend and colleague Moshe Yonatany for devoting his time to read an early draft of this dissertation and provide useful comments. My great thanks to Anni Olesen for taking care of the thesis submission procedure and Jacob Nørbjerg as head of the department for providing me support to finalize the submission of the thesis. The writing of this thesis would have been impossible without the support of all those people that have been close to me, especially my family. I thank them for all these years of love and support. Copenhagen, May 2009 to my parents Theodoros and Athena for their love and support and to Katerina for always waiting me XVII Part I Theoretical background 1 CHAPTER 1 Introduction 1.1 Introduction to this dissertation One of the most influential economists of modern times and a Nobel Prize laureate, Hayek, suggested that a key problem of society is the coordination of dispersed knowledge — a problem a central planner would be unable to address [Hayek, 1945]. Many things have changed since then, including the fundamental ways information is transferred between individuals in a digital context. Nonetheless Hayek’s key perspective on that problem of society still remains central in a digital age, where the World Wide Web does increasingly facilitate collaboration between individuals. Wikipedia, the online encyclopedia, is a case in point: On many occasions it contains much more content than a centralized system can handle 1. If collaboration cannot rely on rules administered by a central planner, how then do people participating in online collaboration self-organize not only their interaction but also the incentives inducing online behavior? 1Although for some, the accuracy of Wikipedia still remains a controversial issue 2 CHAPTER 1. INTRODUCTION The thesis aims to contribute to the understanding of the motivational factors that affect participation in virtual/online communities, collectives of social interactions on the web, from a participant behavior point of view. For clarity reasons, we usually refer to the participants of online communities as users, since online communities are based on software that is accessible through the World Wide Web. Most of the current academic research on online communities (outlined in the next chapter) approaches the subject from rather a posterior perspective, treating it ex ante as a living body (e.g., a mailing list) rather than an ongoing formation with profound behavioral characteristics [Barak, 2008]. Inevitably, exchanging information plays a key role behind the motivation for participation of an individual user, especially because in online communities it is the primary medium of exchange and codified output. However, a key question that is tackled in this dissertation is what constitutes the nature and the driving forces that are behind this desire for information. Do users consider their desire for information as the key reason for participating in online communities? Is their participation affected by other characteristics that have to do with behavioral properties which are attached to this desire for information. In order to achieve this object of research inquiry, this dissertation encompasses a theoretical development outlined in this and the subsequent chapter (Part I) where the research context of user interactions in online communities is discussed together with the related theories. The empirical part of this dissertation (Part II) encompasses four studies that are grounded on user’s interactions captured in four datasets used in this dissertation. The overall goal of the empirical part is to demonstrate the issues discussed in this and subsequent chapters, as well explain the connection with theory presented on the following chapter using a mixed research design approach. The sequence of the studies presented on the empirical part follows a top down approach. While the third chapter studies the impact of the behavioral characteristics on a controlled environment (dictated by a quasi experimental design) the next three chapters provide an analysis on an online context. Chapters 4 and 5 provide an 3 1.1. INTRODUCTION TO THIS DISSERTATION analysis where cooperation is an important element of the study context (community managed question answering systems), while the Chapter 6 provides an analysis of an environment where self interest is evaluated by an online community mechanism. In particular, the studies presented in this dissertation are as follows:  The third chapter provides a study on cooperation in relation to online communities. Cooperation is an important factor for the sustainability of online communities since it affects the outcome of the interaction between the users. In this chapter the relation between cooperation and online social interaction characteristics is made clear using a public goods game. We first explain the public goods game and its game theoretical assumptions and then describe the experimental procedure. We use two distinct framings in relation to the presence of a subject in an online setting, where: (a) the game contributes to the common good (b) it receives benefits from it. The framing is distributed into two distinct treatments with an extra treatment acting as a control of offline participation. The subjects are then presented with a sequential version of the Public Goods Game where each decision (with the exemption of the unconditional choice) is given at once. Results indicate that participants in online communities indeed also show a high degree of cooperation both on the contributing and the benefiting framing, conditional on the contribution provided by the others.  The fourth chapter examines the effect of activity on service posture (measured by volume and time) as expressed by user contributed effort and user received benefit in an online community facilitated by users of the Yahoo!Answers service. Yahoo!Answers operates a question-answering community of users who post questions and receive answers on various topics. We describe how the ration of contributed effort and received benefit has an effect on service posture (volume and time). By programming a web crawler to store a random sample of questions posted over a period of one month, we use a set of time series, random effects and logistic regression models which confirm, to a large extent, our 4 CHAPTER 1. INTRODUCTION formed hypotheses. In particular, we find that users who contribute more effort in the community than received benefit get a question answered by more users in less time than users who receive more benefit than the effort they contribute to the community. The fifth chapter tests whether a particular type of voluntarily awarded monetary rewards (tips) are paid for strategic reasons in a quasi-experimental setting. The context of study is Google Answers. Google Answers was a marketplace of information inquiries in which any asker can post a question along with a price to be paid for a satisfactory answer. One researcher from a closed group of answerers answers the question, usually by providing reference to authoritative sources. Upon receiving an answer, the asker rates the quality of the answer obtained; if satisfied with the quality, the asker pays the price and additionally pays a voluntary tip. We investigate tipping behavior before (when strategic considerations can play a role) and after (when they cannot) the announcement of the shutdown of the answering service. To disentangle a motive of the strategic nature of tipping from other (reciprocal or norm-driven) motives of tipping, we analyze pre-announcement tipping behavior. The empirical results suggest that askers use tips to induce better (e.g., in terms of better promptness) service in the future, and that answerers respond to tipping by providing services more promptly to those with a better history. We particularly show that a class of users relies on repeated interaction in order to receive better service with less cost.  The sixth chapter investigates how users perceive interactions that affect their decision to buy and, in particular, whether their evaluations are related to communication issues as in the case of how readable the submitted reviews are in relation to the usefulness ratio that is attached each review. The unit of analysis in this chapter is online product reviews. Online product reviews are an important resource for consumers of experience goods in online marketplaces because they provide a useful source of support information during the purchase of a 5 1.1. INTRODUCTION TO THIS DISSERTATION can observe similar situations due to the fact that behavior in a societal structure is very much dependent on the context. Behavioral characteristics play a key role in defining what social capital is and its relation to social interactions. The fundamental axiom is that humans as social beings have the ability to manipulate their behavior conditional to the environment that they are in. This manipulation can happen either consciously or unconsciously, depending on the presence or absence of several factors which are axiomatically accepted to cause a change on the behavioral patterns of an individual. A very early study by Allport [1935] provided the first insight on the reasons why the change of a human subjects’ behavior is attributed to the presence of the others. In this and subsequent studies, behavior in terms of sociability is dealt with as a resultant of three basic elements: incentives (factors that push behavior to a certain direction), structure (the way behavior is affected by the presence of the others) and the setting that in which this behavior is observed (off-line or physical environment, online or virtual environment). According to Deci and Ryan [1985], incentives can be on the one hand extrinsic or exogenous in the form that the subject2receives a measurable compensation for his/her effort. On the other hand, incentives can be intrinsic, as subjects might also be motivated by intrinsic or endogenous means of motivation where the compensation is not measured with standard utility yardsticks. Coming back to the research context of this dissertation, as mentioned earlier, the research that has been undertaken for the development of the web has been extensively on the issues of technical realization and evolution of technical standards. Although the technical issues regarding the retrieval of information from web sources have been well addressed and well challenged by the information retrieval community by developing computerized methods for better information reference and retrieval [Brin and Page, 1998] there is an undermining of the social potential that the world 2With the term subject we characterize those social agents that participate in a type of interaction (social, economic etc) 12 CHAPTER 1. INTRODUCTION Definition Source Social Capital as the aggregate of the actual or potential resources which are linked to possession of a durable network of institutionalized relationships of mutual acquaintance and recognition Bourdieu(Bourdieu, 1986). Social Capital consists of a variety of entities with two elements in common: they all consist of some aspect of social structure and they facilitate certain actions of actors within the structure Coleman (Coleman, 1988). Social capital refers to the collective value of social structures /networks and the tendencies that arise from these networks in two perspectives: bonding (between homogeneous groups) and bridging (between heterogeneous groups) Putnam / World Bank [Putnam, 1995]. Social Capital as a combination of structural, relational and cognitive abilities in an organizational structure. Nahapiet and Ghoshal [1998] Table 1.1: Definitions of Social Capital 13 1.1. INTRODUCTION TO THIS DISSERTATION Argument Source Social capital does not resemble a standard form of capital in the way it can be transferred from one owner to the other. Arrow [2000] There is no evidence that social capital contributes to economic activity, especially if you compare studies across different societal structures. Solow [1995] Social capital is based on premature concepts encompassing several different aspects of social activity and therefore cannot be perceived as a distinct entity. Mondak [1998] Social Capital in terms of social structures, norms, trust and reciprocity cannot be theorized due to the context dependent nature of its value. Foley and Edwards [1999] Table 1.2: Arguments against social capital 14 CHAPTER 1. INTRODUCTION wide web offers for the production of information. This later trend known as Web 2.0 [O’Reilly, 2005] focuses on scenarios where individuals can use the web to find information, and then in cooperation with other individuals can structure and define this information better using the benefits of collective action. Concerning the later, web sites of collective action, such as Wikipedia3, provide an example of how the web can harness the collective wisdom of individuals and transform it to a dynamic artifact where the quality of resources becomes better and better. Virtual forms of social capital and online communities Thus far, the discussion provided in the previous section concerning social capital considers it to be a form of capital that takes place in offline settings where social interactions are formed in a physical form (either by affiliation e.g., participation in a club or a community group or spontaneous by context dependent settings such as the workplace). But how can social capital be addressed in a virtual form? Is there an infrastructure that permits the creation of social capital in a virtual setting? Can forms of social capital be found and studied on the Internet? In the perspective of this thesis, this is an important research question first from the conceptualization of social capital itself. This is because, as mentioned earlier, social capital addresses the importance of individual characteristics and their individual attributes over the access of shared social resources. However, an important issue is that the research literature that we discussed earlier approaches the offline definition of social capital. The Internet, however, is an environment that has well grounded social mechanisms; for example, sanctioning (in cases of antisocial behavior) is difficult to be imposed, thus making the adherence to social norms a difficult-to-moderate issue. None forbids an individual who participates in an online community to create a new alias and to behave in a similar manner as before (in case he gets sanctioned for antisocial behavior). However, the question still remains as to what the reaction 3http://www.wikipedia.org 15 1.1. INTRODUCTION TO THIS DISSERTATION of the other participants will be, that is, will they react the same? This is more or less related to whether the focus of analysis is the individual or the group. In that sense, it is important to study forms of social capital on the Internet to see whether there are any similarities with the theoretical arguments that have to do with offline settings. In this dissertation we approach social capital primarily from a collective action perspective. A particular case of this research issue is the case of collective action around information artifacts in the principles of the original model of the web [Bimber et al., 2005]. Groups of individuals are provided with a platform where information exchange between them can be facilitated [Turner et al., 2005]. This in fact can be seen as a mode of collective action. As Olson [1971] states in his fundamental work around collective action “groups of individuals with common interests are expected to act on behalf of their common interests” (Olson, The logic of Collective Action, pp: 149). Following Olson’s original definition, collective action is dependent on collective behavior. As aforementioned, collective behavior is a type of behavior that can be defined as coordinated action among a specified population. One characteristic of collective action as discussed in the literature is that it occurs as a result of temporal collective behavior under a specific context [Gurven and Winking, 2008]. For example the coordination of crowds in sporting events is an aggregate of the collective behavior of individuals that exists only during the context of the sporting event that they attend [Bartel and Saavedra, 2000]. Theoretical research around the characteristics of collective behavior can be generally classified into two theoretical perspectives. The first perspective occupies the view that collective behavior is a result of the social environment and its settings (e.g., already defined hierarchies and social structures); the second one advocates that behavior is a result of a context specific social action that acts for a specific outcome. According to this perspective offline cases of collective behavior such as rumor spreading can be explained due to the fact that the action has a specific outcome. 16 CHAPTER 1. INTRODUCTION Turner and Killian [1987] in their study of collective behavior among human subjects provided a classification of three types of collective behavior, namely the crowd, the public and the social movement. According to this study, this classification is very much based on the social setting in which the collective behavior takes place. This is due to the socio psychological perspective that collective behavior is not a pathological phenomenon, but is very much dependent on social change (e.g., the environment and the social norms that characterize it). Their model acknowledges several social properties that characterize collective behavior: (a) the existence of emergent norms, (b) feasibility of the action (c) timelines – the time setting in which the action occurs and (d) the preexisting groups and networks. Another well known research paradigm that is often adopted in studies of collective behavior is the one developed by Smelser [Smelser, 1962]. In this model Smelser summarizes a set of conditions that need to be present in order for collective action to occur. These conditions are classified in (a) Breakdown of social control, (b) Structural Conclusiveness and (c) precipitating incidents or triggers that occur before the emergence of collective behavior. In particular precipitating incidents are vital for an online community due to the asynchronous mode of communication among the participants [Ravid et al., 2004]. As will be discussed later, in principle, collective action as a result of collective behavior, occurs only under certain conditions, namely, Scope and Interests. From the perspective of the research question tackled in this thesis, collective action provides that individuals form groups which have as an objective function the addressing of the compilation of information sources either by doing it explicitly with a certain objective (e.g., a Wikipedia article or the development of a new software) or implicitly (by deliberately posting information in an Online Forum or in an Internet newsgroup). Olson [1971] takes this approach one step further, arguing that effective collective action (with no individuals taking advantage of the effort of other individuals, and thus becoming free riders) leads to a production of commodity that is known in the literature 17 1.1. INTRODUCTION TO THIS DISSERTATION as public good [Samuelson, 2000] Following this perspective, our perceived nature of information as a public good becomes that of a codified artifact, which under a context provides value for those that use it, though retaining its public nature and not losing its value. Nonetheless the view of information as an artifact is still limited around its consumption. Individuals tend to use it, consume it or produce it when their knowledge is limited on the domain that in which they are active. However an issue remains on how individuals communicate and how they contribute information? In that way the incremental adoption of the web as a communication channel has resulted in a broad variety of online communities which can be conceptualized as groups of individuals with a dense number of social interactions over the Internet. The later embellish a significant role into several application domains (e.g., opinion forums, online auctions, etc.) with potential applications in other areas, such as enhancing trust for electronic transactions. Particularly in this diversity of communities, there are cases where online social interactions are not only a way of communication, but act as an enabler of transactions (e.g., in the case of online auctions) where no contractual enforcement is present [Dellarocas, 2003]. Conversely, unless there is a formal protocol which defines how communication is facilitated, a significant problem of these online communities is the issue of participation, both in terms of membership and activity. Membership deals with the handling of participants in an online community and levels of functionality that the members can employ. For example, in communities where the content discussed is moderated, the structure of the members is not flat, but it employs a certain hierarchical structure. Furthermore, the membership has to be retained at all the stages of the community activity in order for it to become sustainable. Membership in online communities is also a limiting factor in cases where an online community might require participants to register their identity usually through a login system. Some online community systems, such as those provided on the discussion in 18 CHAPTER 1. INTRODUCTION online blogs, allow members to participate without registering but require them to use the same identity during their participation. Therefore, in this dissertation the notion of membership is not tied specifically to the registration policy but to the membership monitoring which is very much related to the identity management issue. Deciding whether to have a registration policy in an online community is also important for the attraction of members in the initial stages of development of the community since this might have a negative effect on the nurturing of the community [Preece, 2000] In particular, online virtual environments require a certain number of members or a critical mass in order to have some activity and thus retain their members. Activity acts as an incentive mechanism to the existing community members to participate and to outsiders to join the community. Nonetheless, although there are profound flexibilities to form interaction (e.g., related to time or space distance), this type of virtual communication is quite difficult to be formed in a non ad-hoc way. As Finholt and Sproull [1990] indicate, technical solutions that act as enablers of communication over the internet, address only the infrastructural solution to this problem. Preece [2000] adopts another perspective to this issue by addressing online community participation by using two pillars: the usability and sociability of the community mechanism. It’s commonly accepted that technology solutions per se cannot guarantee participation of individuals in order to assert an on-line (virtual) social activity. A particular need for understanding the social mechanisms that highlight the participation and social interactions on online communities arises as the potential of these communities has been addressed in the literature both from a theoretical [Quan-Haase and Wellman, 2004, Wellman et al., 1996] as well from a practical viewpoint[Godes et al., 2005]. The impact of incentives for this type of collective behavior is an important issue in online community research. In the literature there are several studies that try to outline what the incentives are for participation and thus explain the behavior of individuals that participate on these online social groups [Jones et al., 2004, Kollock, 19 1.1. INTRODUCTION TO THIS DISSERTATION 1999]. However as stated in the study by [Wang and Fesenmaier, 2003] there is little empirical evidence regarding the nature of the incentives that affect behavior in online communities and their contribution to the contagion. The study by [Ling et al., 2005] for example examined contributions from a collective perspective concluding that users will contribute more in an online community if they perceive their contributions as important for the group outcomes. The basic assumption taken in this dissertation is that incentives affect behavior to a way which is expressed with future action. We categorize incentives into two major groups, namely those related with social or psychological factors and those that have to do with economic behavior. The social and behavioral category of incentives deals with the cases where behavior is affected by endogenous social factors related with the social context and the position of the individual in it. Social incentives study the way group interaction patterns are formed by taking a holistic view of the interaction structure and behavior under certain viewpoints (e.g. contributed effort, activity and commitment). The other category of incentives studied in this dissertation, and particularly in Chapter 5, relates economic incentives in terms of compensation which can be either monetary rewards or elements of value which users consider to be important (e.g., non contractually stated rewards such as tips). In most cases, economic incentives or other extrinsic forms of motivation try to explain behavior by theorizing a rational agent model of the participant. That is, in the case where an individual’s objective function is to seek relevant information, in a way, it maximizes his/her utility by participating in a community. Nonetheless, empirical evidence may contradict this direction. One could argue that since members receive no profound compensation for their participation, they have a high opportunity cost. For example, an expert who participates in an online community (e.g., a forum of computer programmers) and devotes a significant amount of time for answering complex questions might have a high opportunity cost depending 20 CHAPTER 1. INTRODUCTION on his off-line activities and the compensation that he receives by doing them. Similar to the later, one of the much cited problems in the case of communication activity is the factor of the membership size [Butler, 2001]. As a club or a union, an online social structure, in order to operate appropriately, needs a critical mass of members. Related to this is the problem of activity. While due to design settings, people are obliged to become members of a virtual community in order to participate, there are several cases where activity that is not obligatory (e.g., in terms of a community facilitating transactions such as e-bay) is not directly affected by membership size. This phenomenon has been placed in computer mediated communication literature as lurking [Preece et al., 2004, Ravid et al., 2004]. Lurking characterizes the behavior of individuals who while participating formally in the community, are not active. An online community with a high number of lurkers has an activity problem which results in a low quality of social interactions between members. Furthermore, as Cummings et al. [2002] point out, a significant problem is the quality of those social interactions and the nature of the relational ties that are formed through them, with respect to the rest of the participating individuals. All these concepts have contributed to the viewpoint of the emergence of virtual forms of social capital taking place in the realm of online communities where individuals interact with each other by coordinating actions (e.g., online petitions) or by simply contributing information. While the growing focus that firms give to the cultivation of their social capital potential is evident in the management literature, online communities also provide a significant space for their interaction with customers. Armstrong and Hagel [2000] argue that a significant benefit for the nurturing of online communities by firms can be customer loyalty. This can be attributed to the network effect that might become evident when, for example, a community of customers of a specific product reaches the critical mass. As it is in offline settings where an adoption of a product in a market depends on a critical mass of consumers, so it is in online settings where the network 21 1.1. INTRODUCTION TO THIS DISSERTATION tation tackles. Kollock [1998] summarizes the problem of online cooperation in the context of a social dilemma. In general, social dilemmas represent cases where private interests collide with collective interests; this is because of the social nature of most participants who weigh their (short term) personal interests more heavily than the (long-term) interests of the group. Clearly, contributions in an online community can be seen as a social dilemma due to the fact that they have to be incentivized in order for some activity to occur, as well as pose impositions on the collective nature of the participation. In other areas of the research literature, the classical social dilemma is the case of public (non rival and non excludable) goods, where non participation might lead to the well known case of the tragedy of the commons [Hardin, 1968]. This is true in social dilemmas in situations where individuals who are driven by their own self interest may cause a decline of the collective value of a shared resource (e.g., low activity might lead to the fact that the information shared in an online community becomes less valuable). This example is contradicted by the long term interest of the individual, something which certainly is not to degrade the value of the online community, but to receive a benefit by becoming a member. A similar case with the tragedy of the commons can be seen in the example of lurking, where individuals observe but do not participate, and with this behavior contribute to the low activity of the community. Armstrong and Hagel [2000] argue that this is directly connected to the value that the community provides for its members. However, as the aforementioned incentives can be of exogenous, endogenous or combined form, their role is difficult to distinguish. On the other hand, how do the participants themselves consider their participation in an online community? What is the importance of these motivational factors? One might consider these questions to be of some sort of psychological nature. Indeed, the research problem is essentially a study of behavior. However, what makes it an interesting case from an information systems point of view is the context. In psychology and especially the area that is involved with the analysis of behavior in social 28 CHAPTER 1. INTRODUCTION Function Research Objective Value Expressive A,C Utilitarian A,C Social Adjustive B Knowledge Seeking B Table 1.4: Clusters of Functions and research objectives addressed using the framework of Snyder and Cantor [1998] groups, there is the dominant principle of the functionalist perspective which perceives behavioral outcomes to be dependent on different mental states and sensory inputs. In other words, individuals are expected to perform, based on the environment and the input that they receive from it, as well as the functions in which they are expected to perform [Block, 1980]. Snyder and Cantor [1998] position these functions into four different groups: (a) Value Expressive: where individuals express their values out of altruistic concern (b) Utilitarian: where individual seek rewards from the external environment (c) Social adjustive: where individuals do a certain thing in order to fit better with the group and (d) Knowledge Seeking: when an individual may be performing a task in order to get a new learning experience that he might use later. But can this functionalist perspective be fitted in the context of an online community? This is the research issue that we are interested to examine in this dissertation. In particular, what are the motivational factors that affect the participants in an online community and make them to become more engaged? Table 1.4 summarizes the above functions and the research objectives pursued. Taking this classification as a guide, we provide the three basic research objectives of this dissertation in the section that follows. 29 1.2. THE RESEARCH OBJECTIVES 1.2 The research objectives Having provided thus far the theoretical background and the motivation for the research issue that this thesis is targeting, we now frame a set of three research objectives based on the following arguments: 1. The research question should address the importance of social capital (in its virtual form) as an important factor for understanding participation in online communities. Therefore the research objectives should take into account both the behavioral properties of the participants, their interactions as well as the nature of the shared resource that these interactions are formed around. 2. The extrinsic and intrinsic form of the outcomes that the users expect to accomplish by participating in an online community. This is of particular importance if we take also into account the opportunity cost that a participant might have by participating in an online community and 3. The importance of social mechanisms that facilitate these types of online (or virtual) interaction should be highlighted among the individuals that form an online community. For the above reasons we chose the following three research objectives which are described in the following paragraphs. Research Objective A: What is the nature of the incentives that affect contribution in online communities? In Organizational Science there has been considerable literature that discusses the effectiveness of incentives as a mechanism for empowering individuals in several contexts, especially as a key component of agency theory in organizations [Eisenhardt, 1989]. Gibbons [1998],for example, has studied the impact of incentives in an 30 CHAPTER 1. INTRODUCTION organizational setting and provides evidence that incentives promote effort and performance. When it comes to the psychological interpretation of the term, incentives are seen as psychological stimuli or influence factors, responsible for modifying the behavior of the individual under a specific setting. Behavior is often operationalized as a blend of several sources of action which can be seen by the reaction of individuals in different settings (e.g., a buyer’s behavior might change as a reaction to the increase of the price of a good). In the framework presented by Snyder and Cantor [1998], this can fit with the Value Expressive and Utilitarian cluster of functions since this captures both extrinsic and intrinsic forms of incentives. In our case we study incentivized behavior by using a special model of social interactions that takes place in a class of social interaction structures called exchange networks[Cook and Whitmeyer, 1992]. Cook and Emerson [1987] approach the definition of an exchange network as a network structure where the directional relationships imply exchange of resources either of material nature (e.g. economic and business relations) or of a non economic relations such as power relations (where individual has an authority or power to influence another individual). According to Cook and Emerson [1978], exchange networks are formed by the mapping of exchange activity among different individuals into dyadic relations. Unlike standard network theory where individuals are concerned to interact with the whole set of social structure (e.g. the society or the market) an exchange network aims to analyze the reciprocal advantage that an individual might draw from the engagement of an individual to a form of transaction. As in any other type of network analysis [Borgatti and Foster, 2003], in exchange networks the fundamental unit of observation is the dyad (the reciprocal connection between the minimal actors required for an exchange). Two dyadic relations in an exchange network are connected in the situation where the one relation can be contingent on exchange (or no-exchange) on the other relation. The particular nature of this connection can be either positive or negative. A positive connection provides that the one exchange relation is contingent on exchange with the other 31 1.2. THE RESEARCH OBJECTIVES while a negative exchange relation is non-contingent on exchange with the other. In greater detail Chapter 3 provides a controlled study with a large set of participants where this particular research question is partially explored as well as in the case of Chapter 6 where apart from the incentivized contribution we look in to the nature of the contribution itself. Research Objective B: How do participants of online communities perceive the nature of their contribution? The second research objective of this thesis relates to the participants’ perceived nature of the contributions to the online community. In order to carry out this research, we consider behavior as a variable of latent nature, which implies that in order to consider behavior as a unit of observation (or the dependent variable), this has to be expressed as a set of other factors-variables which can be operationalized by methodical observation (either in a controlled setting or a field setting). We approach the case of contributions by using three preliminary types of contributions in an online community based on the approach of studies of collective action discussed in the previous sections. We categorize this approach into three different forms: (a) The case of contributions as contributions to the public good, (b) the case of contributions as a reciprocal action and (c) the case of contribution as a compliance with the social norms that characterize the group A contribution to the public good is a dominant form of perceived nature of contributions where collective action occurs also in offline settings. What makes this perspective interesting from a research point of view is how participants perceive the public good nature of the content of the online community. As will be argued in the next chapter, contributions in online communities can be seen as contributions to the public good since the consumption of the outcome is available to anyone; as long as it remains free, there is no rivalry among consumers for the consumption of the information (as in the form of a good in a tangible form). This resembles the characteristic of a 32 CHAPTER 1. INTRODUCTION public good as possessing the nature of non rivalry and non excludability [Samuelson, 2000]. On the other hand, some of the community participants are not incentivized by the connection of the community outcome as a public good due to the fact that they do not perceive the production of the collective action as a public good. Following the metaphor of exchange between two parties, reciprocity is in principle an exchange of the same or equivalent resources that the one party has given to the other. An exchange relation is reciprocal when participants perceive the value of an item to be exchanged to be of roughly equivalent value to the one received (in case of a dyadic exchange). When they are forced by other participants into the terms of the exchange, it means that, instead of seeing the contribution to the online community as a contribution to the public good, they are reciprocating interaction instantiated by other community participants. In such cases, there is usually a trigger event which can be in the form of a message around a particular subject that intrigues the participant to contribute (e.g., the case of an article in Wikipedia which contains no accurate information). There can also be the situation where community participants perceive their participation as something that is obligatory, and this is very much related to the cohesiveness of the online community, as in the offline settings [Gross and Martin, 1952]. This suggests a dichotomization of the factors related to the previous research question into those provided by the social environment or the consequences of social activity, and those provided by exogenous factors, as in the case of a contract enforcement mechanism which will be discussed in 4 and 5. Research Objective C: Do participants of online communities care about the contributions of others ? The discussion focusing on the case of collective action considers the cooperation of participants to be an important element that characterizes the outcome of this 33 1.3. RESEARCH APPROACH AND METHODOLOGY cooperative effort. However, there might be a situation where some participants rely on the effort of others in order to get the same benefit but with less effort. In other words, participants might consider the level of activity of others in order to participate or contribute effort in an online community. The purpose of this research objective is to evaluate whether participants do care about the level of activity of others or are agnostic in the way that they do not care about the activity level of the others (e.g., the contributions that they have already provided to the online community) or their status. In these environments, on the other hand, it is difficult for the participants to clearly observe this due to the high amount of interaction taking place. However, since this interaction is codified, participants have the ability to trace the activity of the other participants and compare it with their own. Therefore, we need to be able to answer whether such a social comparison process exists and what the output is of these processes. Having provided the problem formulation and the research objectives of this thesis, we present the methodological approach that was employed to study the above stated research objectives. 1.3 Research approach and methodology When it comes to addressing the research questions defined above, we need to be able to clarify (a) the nature the research inquiry that we address (b) the research methodology that we are going to use in order to handle them and (c) ways to assess the external and internal validity of the findings that we will obtain by pursuing the stated research questions. 34 CHAPTER 1. INTRODUCTION 1.3.1 Online communities and Information Systems research In order to comply with the main elements of the scientific method, a research inquiry has first to be defined within the scope or the “epistemology” of a research field. The epistemology of Information Systems (IS) addresses the study of technology oriented phenomena and its relation between people and organizations. As a field, it is established around the socioeconomic implications of the use of Information Technology in both Micro (individual) and Macro (Organizational) levels. Thus, we need to clearly define the unit of observation, after which we will also define a set of constructs that will be used to address the research questions that we pursue. This is in contrast to the social science perspective which provides that the unit of observation is clearly the individual. The network based perspective focuses on the characteristics of the relations rather than the individuals themselves (such as personal or behavioral characteristics). Following the network perspective, we need to emphasize the assumption that actors of a social system form relational ties which can be interpreted differently under certain contexts. For example, in a social system there are often several different types of relations which have a different consequence for individuals that are part of them. For example, Padgett and Ansell [1993] studied the network of different relations among Florentine families in the early renaissance in order to explain the rise of the Medici family as rulers in renaissance Florence. This research example, although it can be considered to be distant from the research subject pursued in this thesis, it does have a common viewpoint on the research perspective: the network structure implies different access to resources for individuals that form relational ties. The authors’ research results provide an understanding of the relation between ties of different strengths (case.g., family ties through marriages and business ties). 35 1.3. RESEARCH APPROACH AND METHODOLOGY 1.3.2 The positivist view in IS research Methods of scientific enquiry start by making a basic distinction between two fundamental concepts, namely the epistemology and the methodology aspect of the research that is undertaken. On the one hand epistemological aspect of a research inquiry defines the process of how we come to know the properties of the research inquiry such as the extent to which the outcomes of the scientific inquire are still valid and are safe to be used to address a problem. In other words the epistemology of the fields defines the boundary between the beliefs and the scientific truths that can be extracted by addressing the research question formulated. On the other hand the methodological aspect of scientific inquiry deals with the more practical perspective of the epistemological view of the field. It is not only concerned with “how we come to know” which provides the background knowledge about it, but it focuses mainly on the specific ways of scientific inquiry-methods - that provide a tool for understanding and validating the background knowledge known about this scientific inquiry. rom this perspective, the positivist view of reasoning in IS research suggests that the researcher analyzes results concerning a technology oriented phenomenon based on statistical and formal reasoning, rather than on perspectives theorizing about individuals’ actions such as in action research [Baskerville and Wood-Harper, 1996]. However in order for a positivist research study to be able to provide a meaningful set of concluding remarks a set of steps need to be undertaken. Figure 1.1 provides a summary of the stages in a positivist research based inquiry in the field of information systems. The basic step in those procedures is the selection of the type of research that will be conducted. This can be either exploratory or classical/confirmatory. Exploratory research approaches data analysis from a perspective that it permits the researchers to employ an analysis which allows the data itself to reveal an underlying structure [Tukey, 1977]. This approach can be generally seen as an inductive view of the research inquiry that is not based on any preexisting model or theoreti36 CHAPTER 1. INTRODUCTION cal pathways in order to test or confirm a theory. By employing this approach, the researcher gets an overview of the data and the research issues that can be initially deducted from a preliminary data analysis. This is grounded in the assumption that the more a researcher knows about the nature of the data, the better these data can be used to construct a model in order to test or define a new theory. GeneralResearch T f CaseStudy Approach Etc. DataCollection T ypeo f  Research Exploratory Vs Confirmator y FieldExperiment LabExperiment TranscriptAnalysis ObjectiveMeasure y FieldStudy Interviews Vs Confirmatory Survey MDS Etc. DataAnalysisTechnique StructuralEquationModeling V FactorAnalysis MDS V s Confirmatory Regression Figure 1.1: The stages of positivist research in information systems research. Adopted from Boudreau et al. [2001] Exploratory data analysis employs a heavy use of descriptive statistics in order to provide some basic descriptive measures that can give researchers an idea about what the dataset describes and how the different variables are connected. For example, in the case of a dataset consisting of several variables, before the researcher can use a statistical technique such as regression analysis, it is always better to have an overview of the correlation between the variables for an overview of which dataset variables are useful for inclusion to the model or not. Visual representations of the 37 1.3. RESEARCH APPROACH AND METHODOLOGY dition that is imposed at any experimental stage (treatment). The ceteris paribus condition is important because we can use it to identify the consequences of each individual effect to the whole outcome of the experimental treatment. Let us consider a process X that takes inputs A and B and produces an outcome Y. Suppose that we want to measure what was the effect of B in the process X–>Y. To do that we consider two treatments: the first (t1)is the treatment with all the elements normal, the second (t2)considers an alternation of B to B* so the process becomes X->Y*. Figure 1.4: Two treatments under the ceteris paribus condition Under the ceteris paribus condition we will demand the following: A (t1)= A (t2) X (t1)= X (t2) This means that the setting is controlled under ceteris paribus condition if and only if under alternation of B in treatment 2 the input /variable A remains the same in both treatments and the process transformation is exactly the same. After that and in order to measure the effect we consider the Δ(B,B*) with Δ(Y, Y*). The ceteris paribus condition complements the two basic assumptions that are used in any empirical study. The first has to do with the hypothetical isolation or the case where we hypothesize that the factors or variables that we focus on in a study solely describe the unit of analysis that is defined. There might be other factors as well (endogenous or exogenous), however, this assumption permits us to study an effect of 44 CHAPTER 1. INTRODUCTION Figure 1.5: Aspects of Validity for Experimental Results one situation on another, as they were completely exogenous to the other. For example, the impact that the observed series of actions of a user will have an impact on the way the other participants perceive this user as an important participant in an online community or not. However, since empirical studies as the ones that are contained in the thesis also consider complex environments where the hypothetical isolation of factors can mislead to different results, another situation is when the ceteris paribus condition provided is the case of temporal isolation where we consider that other factors (e.g., the community size) change slowly during the research study. While ceteris paribus is not the only condition that a research study can assume, this is a safe way for the interpretation of the results that provide a discussion on the approximate effects, rather than safeguarding effects based solely on assumptions (something that can be dangerous for social studies in general). An experiment as every other empirical method needs to have some validity measures that can express how safe is to induce results and validate hypotheses. In principle experimental validity is an indication of the validity that the results of the experiment can advocate. There are two aspects of validity namely internal validity and external validity. Internal validity resembles how reliable the independent variables are to support the behavioral observation. A conclusion drawn from an experiment 45 1.3. RESEARCH APPROACH AND METHODOLOGY has verified internal validity when an independent replication of the same experiment provides the same likelihood of results. External Validity refers to the happenstance resemblance of the validity obtained by the experiment in the real world. In other words the external validity of an experiment describes how valid are the conclusions of the experiment to the real world. Although internal validity is observable and can be assembled by statistical techniques and tools, external validity is very difficult to obtain due to the fact that the settings facilitated in the experiment are not easy to apply to the real world. Apart from this there are several cases of experiments where the context is different than that of the lab where external validity is difficult to induce due to the high complexity from the environment. Criticism and validity For the aforementioned reasons the positivist perspective on information systems research provides a more sufficient ground for pursuing the research question framed on this chapter. However on the other hand, scholars often cite as a major criticism of the positivist perspective in information systems research the fact that it tries to account the unpredictable complexity of the human nature by employing the use of isolate models that incorporate a set of static variables [Myers, 1997]. Furthermore it is argued that positivists look for patterns around the case that they examine thus reducing the complexity that is provided by the human dynamics of the actors. To some extend this case deals with the operationalization of variables related with the research inquiry that is undertaken by the scholars. The employment of latent models for the description of variables is a standard practice that is used by behavioral scientists such as psychologists in order to comply with this criticism. Another major criticism on the perspective of positivism and positivist research in Information systems research is whether the instruments that are used for reasoning are reliable or not. Straub [1989] in an early study examined the use of various tech46 CHAPTER 1. INTRODUCTION niques of quantitative research in mainstream information systems research journals and reported that only the 17% of the articles reported reliability of the scales used. In other ways only few of these articles were validating the constructs that were used to assess the research questions posed. Boudreau et al. [2001] provided a follow up to this study which examined a broader collection of articles in information systems research journals. The results indicated an average of 55% of all the quantitative studies provided in this set of journals was reporting measures of validity both in preliminary phases (pilot test) and after. The findings suggested that positivist view is moving slowly to address the issues raised by Straub’s suggestions. Furthermore the development and validity of instruments based on structural equation modeling for example [Chin, 1998] has provided a better ground for positivist research in information systems. Structural instruments[Jobson, 1992] very often provide two perspectives on the way that they emphasize the output of the model of study. The first is the formalized or econometric perspective which is merely focused on prediction and the significance of the factors that characterize the model. The other perspective which is widely used in information systems is emphasized on psychometric analysis which models concepts as latent (unobserved) variables. The late provides that the units of study are not operationalized by a direct observation but are indirectly inferred from several observed variables which in the literature are often referenced as indicative or manifest variables. In the empirical part of this dissertation we make extensive use of estimation models based on ordinary least squares (OLS), as well as logistic regression models (TOBIT, PROBIT). This was done in order to provide (a) an evaluation of the significance level for the coefficients both from the perspectives of size and significance level, (b) an evaluation of the case where common measurement issues such as homoscedasticity might affect the significance level, and most importantly, (c) defining upper and lower limits for the dependent variable since in the studies we were theoretically informed 47 1.4. STRUCTURE OF THIS DISSERTATION about the actual range that the dependent variable was having. Furthermore, distribution assumptions, such as normality, might also violate the estimator’s trustworthiness if the dependent variables are not normally distributed. A logistic estimator can become independent of that issue and complement the case where the coefficient’s size is significant only upon this assumption. The same case applies to the comparison between groups since the assumption of normality entails the danger of statistical significance being present at a wrong level. This is avoided by testing for normality (using the Kolmogorov-Smirnov test), as well as using non parametric tests such as the Mann-Whitney U test. 1.4 Structure of this dissertation Having provided the research questions and method selection approach for this dissertation, we now supply an overview of how this dissertation is structured in order (a) to demonstrate the connection with the empirical part that covers the research questions provided above and (b) to provide an overview of the data used in the empirical part. 1.4.1 Chapter structure As can be seen in Figure 1.6 the dissertation consists of two parts: a theoretical part consisting of two chapters (this introductory Chapter and Chapter 2), where the research question is underlined and an empirical part (Chapters 3, 4, 5 and 6) where the connections with theory are highlighted in four different application settings. To this end, the empirical part provides a collection of four chapters structured according to the aforementioned research objectives to support conclusions related to the research questions formed in this and the subsequent chapter. Chapter 7 presents the conclusions of this dissertation and retrospectively to the research results summarized in the previous chapters, provides a more analytical view48 CHAPTER 1. INTRODUCTION point regarding the practical implications and contributions of this research. Figure 1.6: The structure of this dissertation To rest of this dissertation is structured as follows. Chapter 2: Models and Theories for Understanding and Motivating Contributions in Online Communities This chapter provides the background of studying online communities and the social interactions that characterize the activity of their members. In particular, an attempt is made to summarize the findings of resent research studies and methodological frameworks that are in relation to motivating contributions to online communities. The chapter approaches the research context and theories related to research on online communities by adopting a social interaction based approach in order to illustrate the rich social environment that is encapsulated in an online community. 49 1.4. STRUCTURE OF THIS DISSERTATION Chapter 3: Behavioral Characteristics and Cooperation in Online Communities: An Experimental Investigation In this chapter we present the results of a large Internet experiment that was run to evaluate whether factors such as trust and fairness play an intuitive role in the formation of attitudes towards contributions to online communities. The chapter describes the experimental procedure and the theoretical expected results and then continues by providing an analysis of the data and a discussion of the findings. This chapter is important for the empirical part since it presents a non-contextual study, such as the studies presented in the next three chapters, but evaluates attitudes and behavioral characteristics of subjects that already participate in online communities. The findings of the study presented in this chapter are also used in Chapter 7 to evaluate the findings from Chapters 4, 5 and 6 in order to provide the conclusions and contributions that can be drawn out of this thesis. Chapter 4: Effort, Benefits and Commitment on Online Knowledge Communities: An empirical study on Yahoo!Answers This chapter provides an evaluation of interactions taking place in a widely used online community system operated by Yahoo! and branded under the name: Yahoo!Answers. We provide an analysis of the interactions and the activity taking place in order to evaluate whether contributed effort as motivated by exclusively intrinsic forms of motivation is affected by the level of individual interaction of those participating in the platform. The chapter argues, based on empirical findings, in favor of a “cyber-public goods” theory taking place in the realm of the Yahoo!Answers service which is based on the perceived utility of the answers that users get from the service. Users then assess how much they participated in Yahoo!Answers based upon what they receive. The chapter argues in favor of the ??latter as a way of enforcing participation and motivating contributions with an effect into an increased user activity. 50 CHAPTER 1. INTRODUCTION Chapter 5: The impact of Extrinsic Rewards on Strategic Interaction in Online Communities: An analysis on Google!Answers Interaction in an online community system called Google Answers which has been operated by Google for a period of four years is explored in this chapter. We study extrinsic forms of motivation as depicted from the monetary incentives that the platform uses to motivate its exclusive list of participants to provide high quality answers to questions posted by users in the platform. This particular chapter is closely connected to Chapter 4 from an interaction context perspective, the basic difference being a concentration on the nature of incentives that characterize the participants. In the context of this chapter, participation is motivated by extrinsic (monetary based) and information demand side factors which means that the mode of operation is almost the same (online answers community). Chapter 6: Evaluating Content Quality and Usefulness of Online Product Reviews This chapter provides a different study context compared to the study contexts provided in the previous two chapters by addressing a case of online communities where the objectives of the users can contradict each other. An online community of reviews is formed around a specific product on the Amazon.co.uk website. In order to evaluate this approach, we use a set of two measures: an objective measure as measured by the aggregation of the individual preferences of the community members provided in the website, and a quantitative based evaluation approach based on content analysis. Chapter 7: Conclusions and Retrospect The final chapter provides the conclusions and the connection of the findings by connecting the theory with the results obtained from the empirical part. As aforementioned we address the findings from Chapters 4, 5 and 6 in relation with the findings 51 1.4. STRUCTURE OF THIS DISSERTATION from Chapter 3 and the theoretical background that we provided in this chapter and Chapter 2. The related contributions to theory and the practical implications of this dissertation are highlighted, and an outlook for future research in the subject is offered. We now provide an analytical overview as well the description of the datasets used in order for the four chapters of the empirical part to provide the analysis presented in each of these chapters. We summarize the dataset and the findings of each individual chapter in the following two sections. 1.4.2 Datasets used in the empirical part For the purpose of the analysis provided in the empirical part of this dissertation a set of four datasets were collected in various periods of the PhD process. With the exception of the Public Goods Experiment Dataset that was made accessible due to the researchers’ own involvement in the preparation of the experiment, the other three datasets summarized here and used in the studies presented in the Chapters 4, 5, and 6 which are a result of the programming effort of gathering the datasets. Each of the datasets is structured as a panel with an identifier attached to each individual interaction session. For the Public goods experiment the interaction section is linked to a user (subject) since the interaction data are unique for each subject. For the Yahoo!Answers, Google Answers and Amazon Reviews dataset the panel structure is keyed on the interaction section (Question in the case of Yahoo and Google Answers and review for the case of Amazon Dataset). Table 1.6 summarizes the qualitative characteristics of these datasets. The number of interactions is used as the main descriptive variable in order to demonstrate the volume of social activity that was recorded in each of these datasets. Bellow, we provide a description of each of the datasets used in the empirical part. 52 CHAPTER 1. INTRODUCTION Dataset Name Category Number of Interactions Dataset Description Public Goods Dataset Controlled Experiment 3200 Dataset from an online Internet Experiment Yahoo!Answers Field Study / Online Community 65000 Partial Data collected from the Yahoo!Answers online Service Google Answers Field Study / Online Community /Information Market 150000 Complete data from the GoogleAnswers online Service Amazon Reviews Field Study / Online Community 12000 Partial Data collected from the books section from the UK store of Amazon Table 1.6: Datasets compiled for the purpose of this dissertation and used in the Chapters 3, 4, 5, 6. 53 BIBLIOGRAPHY N. B. Ellison, C. Steinfield, and C. Lampe. The benefits of facebook friends: Social capital and college students’ use of online social network sites. Journal of ComputerMediated Communication, 12:1143–1168, 2007. B.H. Erickson. Good Networks and Good Jobs: The Value of Social Capital to Employers and Employees. Social Capital: Theory and Research, pages 127–158, 2001. T. Finholt and L. Sproull. Electronic groups at work. Organization Science, 1:41–64, 1990. M. W. Foley and B. Edwards. Is it time to disinvest in social capital? Journal of Public Policy, 19:141–173, 1999. D. Friedman and A. Cassar. Economics Lab: An Intensive Course in Experimental Economics. Routledge, 2004. D. Friedman and S. Sunder. Experimental Methods: A Primer for Economists. Cambridge University Press, 1994. R. Gibbons. Incentives in Organizations. Journal of Economic Perspectives, 12:115– 132, 1998. D. Godes, D. Mayzlin, Y. Chen, S. Das, C. Dellarocas, B. Pfeiffer, B. Libai, S. Sen, M. Shi, and P. Verlegh. The firm’s management of social interactions. Marketing Letters, 16: 415–428, 2005. N. Gross and W. E. Martin. On group cohesiveness. American Journal of Sociology, 57: 546, 1952. R. Gross and A. Acquisti. Information revelation and privacy in online social networks (the Facebook case). In Proceedings of the 2005 ACM workshop on Privacy in the Electronic Society, pages 71–80, 2005. M. Gurven and J. Winking. Collective action in action: Pro-social behavior in and out of the laboratory. American Anthropologist, 110, 2008. 60 BIBLIOGRAPHY G. Hardin. The tragedy of the commons. Science, 162:1243–1248, 1968. F. Hartwig and B.E. Dearing. Exploratory Data Analysis. Sage Publications, 1979. FA Hayek. The use of knowledge in society. The American economic review, pages 519–530, 1945. JD Jobson. Applied Multivariate Data Analysis. Springer-Verlag, 1992. Q. Jones, G. Ravid, and S. Rafaeli. Information overload and the message dynamics of online interaction spaces: A theoretical model and empirical exploration. Information Systems Research, 15:194–210, 2004. L. P. Robert Jr, A. R. Dennis, and M. K. Ahuja. Social capital and knowledge integration in digitally enabled teams. Information Systems Research, 19:314, 2008. P. Kollock. The economies of online cooperation. Routledge, London UK, 1999. P. Kollock. Social dilemmas: The anatomy of cooperation. Annual Reviews in Sociology, 24:183–214, 1998. P. Kotler and F. Bliemel. Marketing Management. Englewood Cliffs, 2000. C. Lampe and P. Resnick. Slash (dot) and burn: distributed moderation in a large online conversation space. In Proceedings of the SIGCHI conference on Human factors in computing systems, pages 543–550. ACM New York, NY, USA, 2004. J. Lave and E. Wenger. Situated Learning: Legitimate Peripheral Participation. Cambridge University Press, 1991. A.S. Lee. A scientific methodology for MIS case studies. MIS Quarterly, 13(1):33–50, 1989. A.S. Lee and R.L. Baskerville. Generalizing Generalizability in Information Systems Research. Information Systems Research, 14(3):221–243, 2003. 61 BIBLIOGRAPHY K. Ling, G. Beenen, P. Ludford, X. Wang, K. Chang, X. Li, D. Cosley, D. Frankowski, L. Terveen, and A. M. Rashid. Using social psychology to motivate contributions to online communities. Journal of Computer-Mediated Communication, 10, 2005. A. Majchrzak, A. Malhotra, and R. John. Perceived individual collaboration knowhow development through it-enabled contextualization: Evidence from distributed teams. Information Systems Research, 16:9˝ U27, 2005. D. H. McKnight, V. Choudhury, and C. Kacmar. Developing and validating trust measures for e-commerce: An integrative typology. Information Systems Research, 13: 334–359, 2003. J. Mondak. Psychological approaches to social capital, special issue. Political Psychology, 19:433–439, 1998. M. D. Myers. Qualitative research in information systems. Management Information Systems Quarterly, 21:241–242, 1997. E. D. Mynatt, A. Adler, M. Ito, and V. L. O’Day. Design for network communities. Proceedings of the SIGCHI conference on Human factors in computing systems, pages 210–217, 1997. J. Nahapiet and S. Ghoshal. Social capital, intellectual capital, and the organizational advantage. Academy of Management Review, 23:242–266, 1998. M. Olson. The Logic of Collective Action: Public Goods and the Theory of Groups. Harvard University Press, 1971. T. O’Reilly. What is Web 2.0. Design Patterns and Business Models for the Next Generation of Software, 30:2005, 2005. J. F. Padgett and C. K. Ansell. Robust action and the rise of the medici, 1400-1434. American Journal of Sociology, 98:1259, 1993. 62 BIBLIOGRAPHY A. Pinsonneault and K. L. Kraemer. The impact of information technology on middle managers. Management Information Systems Quarterly, 17:271–271, 1993. J. Preece. Online Communities: Designing Usability and Supporting Socialbilty. John Wiley & Sons, Inc. New York, NY, USA, 2000. J. Preece, B. Nonnecke, and D. Andrews. The top five reasons for lurking: improving community experiences for everyone. Computers in Human Behavior, 20:201–223, 2004. R. D. Putnam. Bowling alone: America’s declining social capital. Journal of Democracy, 6:65–65, 1995. A. Quan-Haase and B. Wellman. How does the internet affect social capital. Social Capital and Information Technology, page 113˝ U135, 2004. G. Ravid, S. Rafaeli, P. F. Marty, M. B. Twidale, D. Zeitlyn, F. Barone, A. Ciffolilli, and T. A. Maxwell. Asynchronous discussion groups as small world and scale free networks. First Monday, 9, 2004. U.D. Reips. Standards for Internet-based experimenting. Experimental Psychology, 49 (4):243–256, 2002. H. Rheingold. The Virtual Community: Homesteading on the Electronic Frontier. MIT Press, 2000. P. C. Rigby, D. M. German, and M. A. Storey. Open source software peer review practices: a case study of the apache server. In Proceedings of the 13th international conference on Software engineering, pages 541–550. ACM New York, NY, USA, 2008. P.A. Samuelson. The Pure Theory of Public Expenditure. Readings in Social Welfare: Theory and Policy, 2000. N.J. Smelser. Theory of Collective Behavior. Routledge & K. Paul, 1962. 63 BIBLIOGRAPHY V. L. Smith. Microeconomic systems as an experimental science. American Economic Review, 72:923–955, 1982. M. Snyder and N. Cantor. Understanding personality and social behavior: A functionalist strategy. The handbook of social psychology, 1:635˝ U679, 1998. R.M. Solow. But Verify. The New Republic, 11:36–39, 1995. D. W. Straub. Validating instruments in mis research. Management Information Systems Quarterly, 13:147–169, 1989. J.W. Tukey. Exploratory data analysis. Addison-Wesley Series in Behavioral Science: Quantitative Methods, Reading, Mass.: Addison-Wesley, 1977, 1977. R.H. Turner and L.M. Killian. Collective Behavior. Prentice-Hall, 1987. T. C. Turner, M. A. Smith, D. Fisher, and H. T. Welser. Picturing usenet: Mapping computer-mediated collective action. Journal of Computer-Mediated Communication, 10, 2005. A. Tversky and D. Kahneman. Judgment under uncertainty: Heuristics and biases. Science, 185:1124–1131, 1974. E. von Hippel. Horizontal innovation networks–by and for users. Industrial and Corporate Change, 2007. J.B. Walther. Computer-Mediated Communication: Impersonal, Interpersonal, and Hyperpersonal Interaction. Communication Research, 23(1):3, 1996. Y. Wang and D. R. Fesenmaier. Assessing motivation of contribution in online communities: An empirical investigation of an online travel community. Electronic Markets, 13:33–45, 2003. M. Wasko and S Faraj. Why should i share? examining social capital and knowledge contribution in electronic networks of practice. Management Information Systems Quarterly, 29:35–57, 2005. 64 BIBLIOGRAPHY B. Wellman, J. Salaff, D. Dimitrova, L. Garton, M. Gulia, and C. Haythornthwaite. Computer networks as social networks: Collaborative work, telework, and virtual community. Annual Reviews in Sociology, 22:213–238, 1996. D. B. Whittle. Cyberspace: The Human Dimension. WH Freeman, 1997. 65 CHAPTER 2 Models and Theories for Understanding and Motivating Contributions in Online Communities 2.1 The social and economic cases for an online community The previous chapter elaborated on the discussion of an online community at the level of sociability characteristics of its participants. However, we have not thus far provided much background theory related to the actual definition of an online community, and the social properties that an online community has. This is, in essence, the goal of this chapter. First, we explain what an online community is and then provide a discussion on the social properties and the known theory that is centered on the subject that this dissertation tackles. As aforementioned, online communities are of particular importance in organizational settings because they provide the ground for accumulation and dissemination of social capital [Blanchard and Horan, 1998]. It has also been argued that online communities resemble a virtual form of social cap66 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES ital in an organizational context, which, in turn, gives a strategic advantage for the firm that manages to exploit such a resource [Nahapiet and Ghoshal, 2005, Spender, 1996]. However, theory and research around virtual or online communities tackles this issue in a purpose oriented setting which can be independent of the production of social capital. This suggests that an online community can have several different forms, from an online review system to an Internet newsgroup, with its purpose and functionality being loosely coupled. Undoubtedly, the definition related to the concept of what an online community is varies in the literature. One of the earliest theoretical papers on the social aspects of online communities by Wellman et al. [1996] elaborated on an online community as an extension of computer networks to those networks where social interactions occur (social networks). This is mainly due to the fact that social theory and economics consider communities as a structure with collective properties where externalities are produced [Shaffer, 1989]. Normally, communities rely on interaction and communication between their members. In a virtual setting, an online community can be defined as an asynchronous communication channel where individuals exchange information following a set of rules which are defined in the communication protocol [Rheingold, 2000]. The communication protocol dictates the way in which the communication endpoints interact. For example, in the case of an Internet newsgroup, the communication protocol provides that a participant posts something and then another participant replies. Usually this type of communication takes place in an asynchronous mode since there is a considerable time window between the time point that someone posts a question (tpost)and the time he/she receives a reply (trepy). According to Koch and Wörndl [2001], a community mechanism facilitates the interaction between the individuals by providing the necessary tools for (a) Identity management, (b) Organization of activities and (c) Facilitation of interaction. Identity management refers to the infrastructural capability that the community 67 2.1. THE SOCIAL AND ECONOMIC CASES FOR AN ONLINE COMMUNITY mechanism offers to an individual in order to be able to manage his/her identity inside the community, as well as providing the capability for the community administrators to be able to manage the community memberships. For instance, Google Groups1 provides a subscription process which someone must follow in order to register on a specific group. While some communities allow open participation in terms of monitoring of activities (e.g. reading the discussions in a newsgroup), the majority of them require registration in order to actively participate (e.g. to start posting questions). Most often the participation takes place under an alias or a pseudonym by the community member. For privacy and anonymity reasons, some members prefer to hide their identity using pseudonyms - although in some communities the social properties such as trust are important and the use of pseudonyms might have a social cost at some point. The selection of the pseudonym has been proved that affects this kind of properties and especially trust [Friedman and Resnick, 2001]. For instance, in the case of online auctions (e.g., e-bay) where the primary interaction is transaction dependent, communication between the exchange parts is an important element of the transaction due to the fact that it is an enabler of trust between the sellers and the buyers. Communities of transactions pose also an interesting role in these cases since much of the literature considers them to be an important factor for the function of electronic markets [Dellarocas, 2003]. Exchange networks can be characterized as a subclass of social networks where actors (whether buyers or sellers) form relational ties that represent exchange relations [Cook and Emerson, 1978]. One particular aspect of the network properties in this type of network is the reputation that an actor possesses in the network. Reputation is a characteristic property in offline settings as well, which can attribute to better network position and thus better opportunities for exchange in the future to those that possess a high degree of reputation in the network of their contacts [Raub and Weesie, 1990]. 1http://groups.google.com 68 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES However, reputation is one side of the coin with the other side being trust. It can be argued that in this kind of relational quantification of social networks, trust imposes an important role in the formation of the relational ties. Actors form relational ties influenced by the trust they perceive from the other part of the dyad. In this case, social network metrics, such as centrality, provide a direct quantification of the reputation that the actor possesses in the overall network. For example, if we consider reputation as popularity, then a simple aggregation of the formed relational ties provides an indication of the reputation that the actor possesses in an exchange network (relational ties denote exchange relations). The actor’s perceived sense of trust towards another actor is described as benevolence. Benevolence is an essential element in the study of social relations, especially in the case where there is the possibility of exchange networks which might involve trading or institutionalized commitments [Pavlou, 2002]. A portion of the research literature on online communities focuses on the exploitation of online communities for the enhancement of the firm’s visibility as well as the improvement of exchange networks between the stakeholders of the firm [Williams and Cothrel, 2000].A particular emphasis is given in the transactions taking place on virtual settings since the Internet has transformed the traditional buyer-seller relationship, thus giving more options to consumers to interact with product vendors [Turban et al., 2000]. This has a significant effect on e-commerce activity through the establishment of exchange networks. Exchange networks of consumer-to-consumer transactions in particular (such as, for example, e-bay2), form a significant portion of e-commerce activity on the internet.Nonetheless, it is important for this activity to gain the same formal contractual commitments that can be established on offline settings as well. In this case there is a technological intervention based on community interactions between buyers and sellers which can be met in reputation mechanisms such as the one visible on e-bay. The importance of reputation mechanisms in environments where no 2http://www.ebay.com 69 2.2. STRUCTURAL APPROACHES given to the theories that are used in the studies discussed on the empirical part of this dissertation since the conclusions expand on the results obtained by these empirical studies in connection to theory. These theories tend to explain (a) the formation of relational ties on online communities (b) the group sense of the participants on these communities and (c) the expansion of the life cycle of the members’ activity as presented above. Section 2.3 provides background on the theory of social preferences, linked with the findings from Chapter 3 which is related with social preferences.We study their expression of behavior in particular cases such as inequity aversion, altruism and reciprocity. Section 2.4 presents a series of models that are well known in the literature in relation to understanding the contributions on online communities and how these communities evolve with time. Section 2.4 takes this one step further to provide a review of recent models for enhancing contributions on online communities. The chapter concludes with an outlook to the empirical part in relation to the theories and models presented. 2.2 Structural approaches 2.2.1 Social network analysis and online communities As aforementioned online communities facilitate social interactions that regardless the activity context, require a concrete methodological formalization. This is due to the fact that social interactions can be multiplex (e.g. communication, exchange etc) and such formalization may allow for inter-context studies of the behavior of the involved members [Uzzi, 1999]. One of the most suitable methodologies for observing social activity and modeling the interaction of individuals in a social group is Social Network Analysis (SNA). SNA has been established as a concrete methodology resulting from social psychology and communication studies [Wasserman and Faust, 1994]. The growing amount of SNA research done in organizational context [Borgatti and Foster, 2003] supports a broad positivistic view on the study of technology oriented phenom76 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES ena by using the formalism provided by SNA and Structural Analysis in general [Zack, 2000]. 12345 6789 10 11 12 13 Figure 2.3: The set of different motifs formed in a triadic relation One of the basic assumptions of social network analysis and in general the part of sociology that relates with structural analysis is that the topology of an individual (actor) on his/her network of contacts/relations has a profound effect to his/her behavior [Scott, 2000, Wasserman and Faust, 1994]. Actors with a better position (at the core) in their network are likely to get access to more resources than actors that stand in the periphery. In order to model social interactions in an online community as a social network, we first need to classify the two major types of network variables namely the structural and the compositional variables. Table 2.1 provides an example operationalization of these variables. Structural variables form the core of the network and contain a dyadic record of social interactions between two actors that belong on the same network. By drawing the set of the structural variables we have the complete network. Nonetheless structural variables describe only the relations and not the individual characteristics of the actor. This role is undertaken by the compositional variables which provide a way for expressing actor related attributes such as demographics etc. According to Carrington et al. [2005] when conducting Social network analysis and 77 2.2. STRUCTURAL APPROACHES Focus Operationalization Variables Focus on the Group Sets of triads Compositional Variables Focus on the Individual Dyads Structural Variables Table 2.1: Operationalization of Structural and Compositional Variables in Social Network Analysis relational analysis studies in general considers two major assumptions when presenting results:  That the relationships of the individuals studied, correspond accurately to the real context. Following this direction subconscious or illicit relationships are not represented or either not included as a subject of the sociometric study.  Group size is the optimal on the sense that the boundary is the optimal and includes those actors that influence directly at least another member of the network or the social group that is examined. Apart from the variable definition, In order to construct a social network, one has to define the unit of observation from which the structural variables of the social network will be constructed. Due to the fact that SNA tackles with the topological properties of the unit of study we have the flexibility to follow the same set of methods regardless the size or the nature of the unit. For instance we can seek for interaction patterns with the same analytical methods both for individuals and institutions. In relation with the unit of observation, Table 2.1 summarizes the relational quantification of the social network which defines the way the relational ties in the structural variables of the network are formed. Depending on the nature of the network a relational tie can be either directed (directed network) or reciprocal (undirected or symmetric network). The relational quantification is also subjective to the research question that is pursued. For instance if we are interested to examine information flow among individuals in online communities then the obvious way is to model it with as a directed network. However if the research question is broader for instance in the case of membership 78 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES Figure 2.4: Types of connection degree in the network then an un-directed network is more suitable. Relational quantification also depends on the network data available. Another approach on the analysis of social networks is the analysis of status [Raub and Weesie, 1990]. Usually in a sociological interpretation the concept of status denotes social power expressed in different contexts such as political or economical. Depending on the nature of the structural variables in the network status or power denotes the case of social support that the compositional variables of the network receive. The most basic theoretical implication of status is the availability of choices the entity receives in the network which gives the entity the advantage of negotiation over the others. For example let us consider the three different network topologies that are represented in Figure 2.4. It is obvious that the sociometric star (first diagram) is considered the one who has the highest degree (the number of lines adjacent to this particular actor connecting him/her with the other members of the group) thus is more popular. When there is the case of reciprocity in the social connections the node’s degree of influence can be used to provide an indication of prestige in the network. Quantifications of status employ techniques of graph theory such as the centrality index [Freeman, 1979] which have been adjusted to the various representations of ties in a network. In this dissertation the approach of social network analysis techniques 79 2.2. STRUCTURAL APPROACHES that we employ for the analysis in chapters 4 and 5 is limited to the modeling of dyadic ties and the portion that these ties represent to the overall network connectivity as depicted by the variation measures employed in Chapter 4. 2.2.2 The weak ties hypothesis Central to Social Network theory is the weak ties hypothesis that was first formalized by Rapoport [1963] and later theorized and discussed by Granovetter [1973] in the context of a job search network in a labor market [Granovetter, 1985, 1978, 1983]. Essential to the weak ties hypothesis is the idea that there might be cases that the structural variables considered do not actually provide a realistic view of the network structure. In particular, there might be contexts where the abstract formalism doesn’t consider different forms of strength regarding the relational ties which are formed between members of the network. In fact this formalization doesn’t distinguish between the strong and weakest forms of relational ties. Based on the particular strength or weakness of the relational ties there can be assumed cases of influence/obedience or isolation of the individual from the group and this effect to the social interactions that are already formed. B AC D E F G H Figure 2.5: The Weak Ties Hypothesis. According to the hypothesis, weak ties (dotted) act as bridges between sets of strong dyadic relations Nonetheless defining a strong or a weak relation in an online setting is a matter of great complexity. Several sources of interactions could be used to define weakness or 80 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES strength of a tie in a dyadic structure. Granovetter (1973, page:1361) in his original formalization argues that “the strength of a tie is a supposedly linear combination of the amount of time, the emotional intensity, intimacy and the reciprocal services which characterize the tie.” For example in the case of an email network we could consider the number of emails those two persons have exchanged as an indication of social activity between them, thus being positive correlated with the strength of their social connection as a longitudinal effect. A similar case can be for example in the case of chapter 5 where interpersonal communication is direct (one to one) and repeated interaction might lead to a stronger tie. According to Granoveter’s perspective, weak ties may act as bridges over time, between strong dyadic relations thus positioning an important role in the network structures by bridging strong dyadic neighborhoods of the network. In that case group cohesion (or group efficacy) might be a factor to consider. Weak ties affect the cohesion of the group which in turn affects the activity. It is more likely that a group with strong cohesion will be more active and energetic than a group with a weak one. This is because a weak tie is more likely to expose it’s weakness in an isolation from the other entities of the structure. However it often reestablishes itself with another destination. The possibility of this to happen depends on context dependent properties such as affiliation with a third party or a common activity. In that particular case there is the evidence of another structural property that is related to the transfer of network properties from one actor to another. According to Freeman et al. [1992], transitivity is a basic characteristic of a network structure which describes how resources are transferred from the one part of the dyad to the other. Depending on the setting where the network is studied transitivity may be operationalized with the amount of information transferred from the one actor to the other (e.g. in the case of an e-mail network) or the number of goods the one buyer has bought from the other (e.g. in the case of exchange networks). The case of transitive properties of a network becomes interesting to consider when we observe 81 2.2. STRUCTURAL APPROACHES transitivity among different levels of status or hierarchies (in an organizational context). In research context transitivity is treated as part of the unit of analysis (in that case the dyad). In a more formalized view a transitive relation considers a pair of two connected dyads and their influence to the establishment of a triad (set of three dyads) due to their transitive property. Let us consider a network of three actors (A, B, C) with the dyads (A−→ Band B−→ C)having a transitive property. In that case the chain that is formed between A, B and B,C forces a third dyad to be established between C and A thus resulting in a complete triad or cycle among these actors. For example if A has a good relation with B and this property is transitive on the relation between B and C then the probability that B and C will establish a contact is high. The later is referred to the literature as the triadic closure bias [Watts and Strogatz, 2006]. One particular approach on social network analysis when it comes to the individual level rather than the group is the use of egocentric networks [Marsden, 2002]. Egocentric networks represent an interesting focal point in social network analysis and especially in the analysis of communication patterns [Fisher, 2005] since they provide a direct focal point to the addressee of interaction, without demanding an analysis of the whole structure to induce facts about the structural activity (Sociocentric Approach).Following that approach in an online community and considering the relational ties as the interaction in terms of posting and answers the egocentric network of a user in the online community is modeled after the information flow between a user that has asked a question and the user that has posted an answer to this question. In this dissertation we use the egocentric network of the user in order to model the variables used in the empirical studies presented in the chapters 4 and 5 where we study interactions in a dyadic level. 2.2.3 Collective behavior and social loafing As aforementioned in the previous chapter, collective behavior is the case where the members of collective (e.g. a group) behave in a uniform manner [Ajzen and Fishbein, 82 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES 1980]. One of the earliest theoretical studies of collective behavior in an offline setting was presented early in the social psychology literature by LeBon [1897]. According to LeBon’s theory in a collective individuals imply to homogeneous behavior through basically three factors: (a) Anonymity (b) Contagion and (c) Suggestability. Although Lebon’s model intents to describe the production of antisocial or crowd behavior there are some useful observations that we can use from our own viewpoint on the context of an online community. The first derives from the case of anonymity that an online setting can assure. Although anonymity is against the factors related with signaling and reputation as we are going to discuss further, it’s a decisive factor for individuals behaving in a uniform matter due to the fact that they are not accountable in an offline setting for their action. The contagion contributes to the rapid spreading of these ideas and the way shifts are being made. A case of this example can be the setting of online forums where under specific conditions individuals do not contribute to the public good but direct their efforts to irrelevant activities which are underlined by subconscious social objective (suggestability). Social Loafing is a central theme on the study of performance of group members in accordance with their motivational incentive. Karau and Williams [2001] define social loafing as the characteristic reduction in motivation and individual effort that takes place where individuals work together with other group members on a collective task. According to theory, very often this individual effort in a group setting is less than in an individual setting. Social loafing is highly dependent on the importance of the group task that the collective has to address. In such cases where individual’s effort has a direct effect to the other group members the motivation to contribute becomes high so does the effort that the individual provides during the process of accomplishing the collective task. 83 2.2. STRUCTURAL APPROACHES 2.2.4 Reputation effects Reputation as a social incentive relates with the status of an individual in the community and the perceived importance that he/she processes by the activity inside the community. At that case an individual would want to sustain his/her reputation and thus behave accordingly. We highlight two types of reputation: group perceived reputation and individual status. Group perceived reputation relates with the reputation/status that the individual possess inside the group as a whole. Individual status relates with the reputation in the individuals ego centric network. That is the network of first and second degree acquaintances in the part of the group that he is active. For instance someone may be active in some topics in the newsgroup and thus reputable on these however he/she might not possess the same status in the whole. Considering reciprocal behavior we can hypothesize that an individual will answer a question in order to increase or maintain his/her status in the community while on the contrary he will answer a question if he considers that it will increase his/her visibility in the network. The later is very much connected with the concept of signaling first introduced by Spence [1973] on the context of a labor market. Signaling is a way to exchange meaningful information among two parties by having the interested party (agent) to communicate information about itself to the other party (principal). On the context of an online community signaling is taking place between two parties where an agent might be a peripheral member (random visitor) and the principal might possess a higher status. However signaling assumes the establishment of some form of direct contact between the principal and the agent which is not always possible. That’s due to the fact that the structure of interactions on most of the types of online communities (e.g. newsgroups) is not maintainable. Since everyone can participate there is no barrier on removing offtopic discussions which as in offline social interaction settings can distract the signaling parties. Another issue with signaling when it comes on online communities is the case of identity [Donath, 2007] where the communication 84 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES of information among the two parties might not be feasible due to the high level of cognitive effort required to infer and evaluate this information. 2.3 Social dilemmas and knowledge sharing on online communities According to Kollock [1998] social dilemmas are triggered in situations where individual rationality directs to collective irrationality. This provides that in a case of a group setting, reasonable behavior directs to a situation that all the members of the group are functioning worse that might have been if they were working alone. If we model social dilemmas from an equilibrium based viewpoint, a social dilemma can be classified as a situation where there exists at least one deficient equilibrium. That provides that in this equilibrium setting an individual is performing worst than it can otherwise. This type of social dilemmas is called N-Person dilemmas[Komorita, 1976]. There are two subclasses of social dilemmas that are of particular interest in the context of understanding an individuals’ contribution to an online community. The first class which is known in the literature as social trap [Platt, 1973, Rothstein, 2005] is where the individual is tempted with an immediate benefit that produces a cost shared by all. Such an example is the case of an individual posting a question in an online community, several other community members invest time and effort on trying to find an answer to this question, then the individual comes back and signals to the other members that he/she has found an answer to his/her inquiry but is not sharing the output with the others. The later imposes a cost to those that participated on helping this member to find an answer in terms of opportunity cost and lost benefit (which can be due to individual self-interest in that particular question). Social dilemmas have an output on externalities which in the concept of an online community is the situation where “a members’ behavior affects the situation of other persons without the explicit agreement of that person or persons” [Olson, 1971]. The 85 2.3. SOCIAL DILEMMAS AND KNOWLEDGE SHARING ON ONLINE COMMUNITIES behavioral topic that is often addressed with offline interactions [Swedberg, 2004]. As can be also observed in offline settings of group or individual behavior, individuals participating in an online community tend to reciprocate the behavior or service they received from another individual during their participation in the online community as a form of reciprocating the service [Fehr et al., 2003]. By receiving a conceivable good service by another individual, an individual feels a debt to that and has a tendency to reciprocate. This is mostly influenced by offline cultural settings [Miller and Bersoff, 1994] and the interpersonal communication factors characteristics that may affect it such as for example the case of the personal contact. In fact personal contact is an influential factor on reciprocation due to its imposition of reciprocity as a socially imposed behavior [Gouldner, 1960]. Reciprocity is an interesting phenomenon from a social perspective in an offline setting but how much different is the study of reciprocity in a virtual setting? What makes the case interesting to study reciprocation on these settings is the degree of anonymity that is provided under pseudonyms. Pseudonyms in fact represent informal contracts whose violation doesn’t lead to any consequences for the violating parties since even unacceptable behavior in the eyes of the others doesn’t lead to real social retaliations. Bad behavior in online settings is unlikely to lead to a reciprocal bad behavior by the others in an offline social mode. Drawing from that we can hypothesize that an online community member is more likely to answer to a question of another individual if that individual has provided an answer before. On the other hand this particular community member is more likely to not participate if he hasn’t had an interaction with that individual before. 92 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES 2.4 Models for understanding and enhancing activity in online communities So far we have reviewed models that essentially provide a framework on understanding the activity in the context of an online community. However what matters on an online community is the effort and the output that its members contribute. In this section we are going to review models that can enhance the contributions to an online community. While undoubtedly a particular way of enhancing contributions to online communities could be the case of monetary rewards, the distributed nature and the volume of membership would have made such a solution costly and in fact would have turned the collective into a market where interaction is strictly based on monetary terms and price premiums. Therefore the models that we are examining in this section of the chapter are based on socio-psychological factors. Koh et al. [2007] propose the model presented in Figure 2.6 as an overview of the factors that stimulate structure which is one of the most important characteristics that affects activity on an online community. The model itself relies in two basic pillars. The factors that drive the online community which have to do with (a) The leader’s involvement (b) The level of Offline interaction and (c) the apparent usefulness that the social capital created by the community has to its members. Leaders’ involvement and the level of offline interaction affects posting and viewing activity while the usefulness of the social capital affects the viewing activity and in return the attraction of the new members. To the above there is the controlling nature of the community size which is in direct connection with the critical mass required for the community to operate efficiently. The authors argue that with the improvement of the quality of the IT infrastructure the effect of the virtual community drivers will become more apparent. Although this approach is not a new idea it emphasized the effect that technology has on social activity. Rheingold [2002] takes that case one step further and approaches this case 93 2.4. MODELS FOR UNDERSTANDING AND ENHANCING ACTIVITY IN ONLINE COMMUNITIES Figure 2.6: Virtual community stimulation structure. Addopted from Koh et al. [2007] on the potential of mobile communities as new ways of organization where ad-hawk interaction can be facilitated using mobile devices. In the following sections we summarize models that tackle with the factors that affect contribution on online communities and their effects on community activity and sustainability. 2.4.1 The collective effort model The collective effort model is a meta-analytic model3that highlights the level of connection between individual motivation and group outcomes. The integrative nature of this model makes it more appropriate to use when addressing a social loafing condition than other models which are based mainly on the social impact of the effort, the evaluation potential of the group and the dispensability of the effort. The advantage 3A model that is build upon the results of several earlier studies 94 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES of the Collective effort model is that combine the motivational principles that derive from expectancy theory of work motivation [Pinder, 1992] together with the principles of self-evaluation theory [Ryan and Deci, 2000]. From that perspective the collective effort model is integrative on the sense that it combines those two theories to address the case of social loafing. In particular the original principle of the model is that individual motivation and effort in collective tasks will stay unaffected as long as a set of contingencies are satisfied. Individual Outcomes Individual Performance Individual Effort Contribution to group performance Group Performance Group Outcome Value of Group Outcome Value of Individual Outcomes Individual Motivation Figure 2.7: The collective effort model. Adapted from Karau and Williams [2001] The collective effort model builds on principles from expectancy theory [Goldman et al., 1987]. According to these principles, an individual needs to feel that his/her effort will lead to an individual level of performance. For example in an online community a participant might start posting and participating in the community activities if he/she believes that he will gain a higher status or visibility in the community. An interesting application of the collective effort model has been undertaken by Ling et al. [2005] where the contributions in terms of individual reviews and ratings were used by the MovieLens recommender system [Miller et al., 2003] in order to provide recommendations to other members of the community. In this case It is more likely that the individual effort will be dropped if with relation with the group outcome is 95 2.4. MODELS FOR UNDERSTANDING AND ENHANCING ACTIVITY IN ONLINE COMMUNITIES not significant to the overall group outcome. Unavoidably the collective effort model is a clear case of a social comparison process. It is expected that a similar case of social preferences where the mediated group factor will affect the persuasion of self interests in relation with the group interests [Bolton and Ockenfels, 2000]. 2.4.2 The Model of Whittaker et al. [1998]. One of the well known models which address activity in online communities (in that case internet newsgroups) is the one introduced by Whittaker et al. [1998]. Their work introduced a structural model which aimed to provide an overall understanding of newsgroup activity as a function of several properties related both with the newsgroup activities as well as the context and the purpose of the newsgroup discussion. The main latent variable in that model is the Thread Depth which characterizes how active the discussion is inside the newsgroup. As can be seen in Figure 8 a discussion thread is formed by several replies on a news item by the other members of the newsgroup. However during the thread discussion members do not only respond to the original poster of the news item but also they start discussing between them as well. The thread depth denotes how many sub-discussions have been formed under the original thread. It can be argued that high thread depth leads to high cognitive effort required by the members to follow the discussion and participate. A certain amount of criticism can be attributed to this model from our perspective mainly due to the selection of the “Thread Depth” as an expression of activity in a discussion group. Following in fact the original proposition of that study we fail to consider behavioral characteristics apart from the familiarity, which in fact might influence the behavior of the participants in the newsgroup (e.g. provocative content). Moderation is an important element in online communities since it has to do with the way participants behave and therefore express their activity through postings. This has an effect on the number of posters, the standard posts (FAQs), Cross postings (targeted messages) as well as the message length. Familiarity on the other hand 96 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES Moderation Number of Posters Familiarity FAQs Cross Posting Message Length Thread Depth Figure 2.8: A causal model of communication in a newsgroup. Adapted from Whittaker et al. [1998] considers the alternative case where a positive effect on familiarity will have a positive effect on cross posting and message length essentially due to the fact that familiarity is an interpersonal factor that affects communication characteristics as well [Krauss and Fussell, 1996]. 2.4.3 The Model of Jones Ravid and Rafaeli Jones et al. [2004] adopt a different approach on explaining contributions on online communities by evaluating the level of cognitive effort required by an individual member to participate and interact in an online community. The authors approach the output of the community as a public good where variable levels of cognitive effort is needed by the individuals to invest to (a) Infer and understand the messages posted in the particular group category that they are interested to participate (b) Formulate and post back an answer. The model is interesting to consider from the fact that it examines community interaction with a less set of rules imposed such as in the case of online newsgroups. Information overload by high activity can lead to a case of lurking [Preece et al., 2004] where individuals simply cannot follow the flow of the activity in the community. 97 2.4. MODELS FOR UNDERSTANDING AND ENHANCING ACTIVITY IN ONLINE COMMUNITIES Figure 2.9: Information overload and cognitive effort for the members of an online community 2.4.4 The Model of Butler The model of Butler [2001] adopts the perspective of the virtual community as a benefit creator for its members or as an aggregator of a benefit provision process that is provided through participation. As can be seen in Figure 10 the model contains three pillars that characterize the activity in a virtual community (it considers a Usenet group as an exemplar case) where individuals can become members or unsubscribe from the group at any time. Butler’s model is composed mainly of three factors: (a) Member attraction and retention (b) Resource Availability and (c) the benefit creation process facilitated by the community. Member Attraction and Retention encapsulates the community’s ability to attract and retain new members in order to keep the level of their interaction constant as well as to increase the size of the members. This in turn has an effect to the resource avail98 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES ability which can be defined as the volume of social capital available at the specific time point on the community. Increased resource availability leads to more interaction between the members in terms of a continuous communication activity, which defines a benefit creation process. Benefit Creation Process Communication Activity Resourc e Availability Membership Size Member Attraction & Retention Membership Loss Membership Growth ( + ) ( - ) Figure 2.10: Membership Size, Communication Activity and Sustainability in an online Community. Adapted from Butler [2001] 2.4.5 The Model of Wasko and Faraj Wasko and Faraj [2005] present a social capital based approach on evaluating contributions on an online knowledge community of law practitioners. As can be seen in Figure 2.11 the model is structured in four factor groupings namely (a) Individual Motivations (b) Structural Capital (c) Cognitive capital and (d) Relational Capital. Individual motivations consist of the notion of reputation by the community members as well as a general notion of altruism embedded in each community members’ social preferences. Structural capital evaluates the position of the member in the group which if we consider also the connection with the Karau-Williams model Karau and Williams 99 2.5. OUTLOOK TO THE EMPIRICAL PART OF THIS DISSERTATION [2001] which is a case of group mediation factor, it is expected that it will have some effect on the contribution of the individual members. Figure 2.11: The Wasko and Faraj model. Adapted from Wasko and Faraj [2005] The other two factors encompass both cognitive and relational or structural aspects of the position of the individual member in the online community. Cognitive capital relates with the ability of the member to contribute in terms of expertise and tenure with the field / topic that the community tackles. 2.5 Outlook to the empirical part of this dissertation The scope of this chapter was twofold. On the one hand, the intention was to introduce the reader to the theoretical concepts underlining the research approach that we adopted on the study of online communities and, in particular, the case of social dilemmas. Social dilemmas on online communities are essentially the reason why understanding and enhancing motivation in these rich social environments is a justified research question to pursue. On the other hand, the chapter aimed to elaborate on 100 CHAPTER 2. MODELS AND THEORIES FOR UNDERSTANDING AND MOTIVATING CONTRIBUTIONS IN ONLINE COMMUNITIES current theoretical development related to the problem of understanding and motivating contributions on online communities such as the models discussed in section 2.4. We reviewed both structural and behavioral approaches to the study of online communities and provided a modest literature review on the most important models related to cases of contributions to online knowledge communities. In the four empirical studies that follow this chapter we make use of several of the concepts presented here with each of the chapters representing a special research area discussed in this chapter. The case for social dilemmas can be examined further in the chapter that follows where we use a laboratory controlled setting to study behavior in social dilemmas; however, we do not study the inequity aversion part in depth due to the fact that there was no repeated interaction case taking place. This approach along with the case of strong and weak reciprocal ties (as presented in section 2 of this chapter) is set as a departure point in chapters 5 and 6. Both these chapters consider the same setting with a different breed of motivational factors being present (extrinsic vs. intrinsic form). These chapters are in connection with the research objectives B and C and consider as background theoretical ground the theory of social dilemmas combined with the theoretical concepts discussed in section 2.3. Chapter 6 is a special case of combination of the theories presented in this chapter and the previous. In particular the research approach that we adopt in that particular chapter is a bottom up approach where by examining the actual contributions of the participants on this online social system, we examine the effect that they have on a transaction activity (such as in the cases discussed in chapter 1) where the standard offline contractual commitments do not lead to any consequences for the participants that express their opinion. The rating mechanism discussed in this chapter has a unique characteristic which is the level of usefulness considered by the other participants and, in particular, the justification of this expression in relation to the individual’s notion of fairness (as expressed by the review provided for the particular book on that 101 BIBLIOGRAPHY P. A. Pavlou. Institution-based trust in interorganizational exchange relationships: the role of online b2b marketplaces on trust formation. Journal of Strategic Information Systems, 11:215–243, 2002. C. C. Pinder. Work motivation. Administrative Science Quarterly, 37:513–515, 1992. J. Platt. Social traps. American Psychologist, 28(9):641–651, 1973. J. Preece, B. Nonnecke, and D. Andrews. The top five reasons for lurking: improving community experiences for everyone. Computers in Human Behavior, 20:201–223, 2004. F. Radicchi. Defining and identifying communities in networks. Proceedings of the National Academy of Sciences, 101:2658–2663, 2004. A. Rapoport. Mathematical models of social interaction. Handbook of Mathematical Psychology, 2:493–579, 1963. W. Raub and J. Weesie. Reputation and Efficiency in Social Interactions: An Example of Network Effects. American Journal of Sociology, 96(3):626, 1990. P. Resnick and R. Zeckhauser. Trust among strangers in internet transactions: Empirical analysis of ebay’s reputation system. The Economics of the Internet and E-Commerce, 11:23–25, 2002. P. Resnick, R. Zeckhauser, J. Swanson, and K. Lockwood. The value of reputation on ebay: A controlled experiment. Experimental Economics, 9:79–101, 2006. H. Rheingold. Smart Mobs: The Next Social Revolution. Perseus Books Group, 2002. H. Rheingold. The Virtual Community: Homesteading on the Electronic Frontier. MIT Press, 2000. B. Rothstein. Social Traps and the Problem of Trust. Cambridge University Press, 2005. 108 BIBLIOGRAPHY R. M. Ryan and E. L. Deci. Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55:68–78, 2000. J. Scott. Social Network Analysis: A Handbook. Sage, 2000. R. Shaffer. Community Economics: Economic Structure and Change in Smaller Communities. Iowa State Pr, 1989. M. Spence. Job market signaling. Quarterly Journal of Economics, 87:355–374, 1973. J. C. Spender. Competitive advantage from tacit knowledge? unpacking the concept and its strategic implications. Organizational Learning and Competitive Advantage, pages 56–73, 1996. R. Swedberg. The economic sociology of capitalism: An introduction and agenda. Economic Sociology of Capitalism, 2004. R. L. Trivers. The evolution of reciprocal altruism. The Quarterly Review of Biology, 46: 35, 1971. E. Turban, J. Lee, D. King, and H. M. Chung. Electronic commerce: a managerial perspective. Prentice Hall, 2000. B. Uzzi. Embeddedness in the Making of Financial Capital: How Social Relations and Networks Benefit Firms Seeking Financing. AMERICAN SOCIOLOGICAL REVIEW, 64: 481–505, 1999. MR Ward and MJ Lee. Internet shopping, consumer search and product branding. Journal of Product and Brand Management, 9(1):6–20, 2000. M. Wasko and S Faraj. Why should i share? examining social capital and knowledge contribution in electronic networks of practice. Management Information Systems Quarterly, 29:35–57, 2005. 109 BIBLIOGRAPHY S. Wasserman and K. Faust. Social Network Analysis: Methods and Applications. Cambridge University Press, 1994. D. J. Watts and S. H. Strogatz. Collective dynamics of small-world networks. The Structure and Dynamics of Networks, 2006. B. Wellman, J. Salaff, D. Dimitrova, L. Garton, M. Gulia, and C. Haythornthwaite. Computer networks as social networks: Collaborative work, telework, and virtual community. Annual Reviews in Sociology, 22:213–238, 1996. S. Whittaker, L. Terveen, W. Hill, and L. Cherny. The dynamics of mass interaction. Proceedings of the 1998 ACM conference on Computer supported cooperative work, pages 257–264, 1998. R. L. Williams and J. Cothrel. Four smart ways to run online communities. Sloan Management Review, 41:81–92, 2000. F. Wu and B. A. Huberman. Finding communities in linear time: a physics approach. The European Physical Journal B-Condensed Matter, 38:331–338, 2004. M.H. Zack. Researching organizational systems using social network analysis. In Proceedings of the 33rd Hawaii International Conference on System Sciences, volume 7, page 7043, 2000. 110 Part II Empirical part 111 CHAPTER 3 Behavioral Characteristics and Cooperation in Online Communities: An Experimental Investigation Cooperation is an important factor for the sustainability of online communities because it affects the outcome of the interaction between the users. In this chapter the relation between cooperation and online social interaction characteristics is made using the public goods game. We first explain the public goods game and its game theoretical assumptions and then describe the experimental procedure. We use two distinct types of framing (contributing and benefiting), regarding the presence of a subject in an online setting where: (a) the subject contributes to the common good and (b) the subject benefits from it. The framing is distributed in two distinct treatments with an extra treatment acting as a control for offline participation. Empirical results show that the participants in online communities exhibit a high degree of cooperation conditional on the contribution provided by the others. 112 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION 3.1 Online communities and online cooperation Undoubtedly, one of the major elements that constitute social capital is the ability of individuals to cooperate and form structural relations within an organization or a society [Putnam, 1995].In the introductory chapter of this dissertation we emphasized the virtual forms of social capital as a factor that plays an important role to the sustainability and the evolution of social structures on the web such as online communities. This can be advocated by the fact that networks, norms and trust related issues do, in fact, interrelate with the theory of social capital and its application to the evaluation of sustainability and activity of a social structure, either virtual or offline. From this perspective of social capital in an online or virtual form, this ability is very much related to the case of online sociability. In other words, the ability of individuals to form social interactions in an online setting use the context of an online community in the same way that an offline community facilitates social interactions. A great deal of research on online sociability deals with the organization of social structures as a result of frequent computer mediated communication resulting in a form of a network connected type of communities known in the literature as Online Communities [Wellman et al., 1996].Online communities pose an interesting field of study on the Internet due to the fact that they facilitate the formation of ad-hoc social structures which permit the exchange of information and knowledge over a broad range of topics. Undoubtedly, the basic compound of these communities is the active participation of a number of individuals, thus providing information to other individuals who seek information. Undeniably, there is a need for a critical mass of participants in order for the online community to sustain and promote the communication activity that takes place in that setting [Marwell and Oliver, 1993]. Furthermore, for the community to be sustainable, this critical mass of individuals is expected to co-operate with another (not necessarily exclusive) set of other individuals who satisfy a particular need by participating in the online community. From a resource based perspective, an online community undoubtedly provides a public good which can be consumed by all com113 3.1. ONLINE COMMUNITIES AND ONLINE COOPERATION munity members. Modern economic theory has defined a public good as an economic good that has two major characteristics: (a) non-rivalry: providing that the consumption of this good by a consumer does not prevent another consumer to consume it as well (each consumer has access to the same type of provision of this good) and (b) non-excludability: where consumption does not impose any cost; in other words, the goods are free for everyone to consume [Samuelson, 2000]. Depending on its design, an online community might impose a cost of entry to a participant which, in general, is a cost attributed to social barriers and not necessarily on an actual monetary price (a community functions as a club where the participants who provide the public good might receive some compensation for doing so). In most of the cases, since participation is free, an important factor which affects the value of the information provided in an online community is undoubtedly the cooperation between an often large number of individuals. Further, cooperation plays an important role due to the fact that the production of the public good provided by an online community (in that case information) demands the active participation of two participant states: that of the consumer and that of the producer. These two states are not necessarily excludable. For example, someone who contributes information to an online community can also consume information that becomes available in an online community taking both the role of the information producer and the information consumer. However, studies have shown that there are some online community participants who prefer to consume more information than they actually produce, rather than giving back to the community, such as in social loafing, where participants are inactive and community activity is only dependent on the participation of a very low percentage of the actual participants. From a theoretical viewpoint, such a case is highly related to activity [Schoberth et al., 2003]. The assumption that we follow in this chapter is that low activity in an online community means less contribution to the public good (the overall information) resulting in “free-riding” where participants who do not contribute to the community get benefited in the same degree as those individuals who act more 114 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION altruistically in relation to the community. However, what are the characteristics of these contributors/consumers of online communities in socioeconomic terms? Are they actually cooperative persons? Do they have some form of altruism embedded in their behavior? Can we somehow disentangle these two aforementioned states? This study is stimulated by the above research question as departure hypothesis. Our initial goal is to seek whether participation levels in an online community are actually related to the behavior that these participants have shown by participating in an online public goods game carried out with actual monetary rewards and controlled by a random sample drawn from an actual population. Therefore, the context of this chapter is not to study the interaction process in an online community per se, but to elaborate on the social processes that take place during the interaction in an online community and, in particular, on the social dilemmas that arise during the participation of online community members. Literature has yielded that the public goods game is a standard method to test whether people are actually cooperative in their activities since cooperation is a behavioral characteristic which is embedded in the everyday behavior of an average person [Fehr and Gachter, 2000]. ]. By controlling for a set of different variables, such as demographics and online communication usage, we are able to define and evaluate our departure hypothesis by using a large pool of subjects that participated in our experiment. To this end, this chapter is structured as follows. Section 3.3 describes the public goods experiment that we conducted, the procedure behind its dominant strategy and its implications related with cooperation characteristics (such as altruism and cooperativeness). Section 3.4 provides the experimental protocol used, the framing used to disentangle the characteristics above, the methods and the procedure that was followed both for the standard (unconditional) public goods game and a variant public goods game based on interval-based conditional choice. Sections 3.5 and 3.6 report a thorough analysis of the experimental data across the different treatments 115 3.2. MOTIVATION and framing used, as well as a discussion of the results. The study concludes in section 3.7 with concluding remarks and issues for further investigation. 3.2 Motivation Our motivation to study the problem of contributions to online communities from the prism of the public good game was intrigued by the fact that online communities do resemble a social space where the production of a public good takes place. This is evident from the following:  the digital nature of information as a public good since the production and consumption of information by an individual does not directly affect the utility that another individual will receive by performing the same actions;  the way we perceive an online community as a space where everyone (depending on the membership policy of the community mechanism) can participate, contribute information and be benefited by the information that is already available;  the social dilemma that is framed by the temptation of an online community member to free-ride on the effort contributed by the other members; thus, if everyone tries to free-ride, then the public good produced by the community will suffer and gradually dissolve [Kollock, 1999]. Related to the case of the social dilemma in a public goods game (or the public goods dilemma as it is known in the literature) provided in the context of an online community, let us provide some examples for why is important to address the problem of contribution in an online community from the perspective of the public goods game. Let us consider the following scenario. In an online community such as the Yahoo!Answers online communityYahoo!Answers online community1a user posts a 1a more analytical study on the case of contributions in Yahoo!Answers is presented in the next chapter 116 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION question related to a topic. Another user responds to this question by providing an answer. The user receives this answer and a piece of information is produced out of this interaction. This information becomes available to every other member of the Yahoo!Answers online community which he/she can use in case there is a similar information need. Here the public goods dilemma can be seen as follows: If a member of the Yahoo!Answers community only posts questions and receives answers in decent time by the other community members without contributing on other open questions (that he/she possesses some expertise to answer), then the information available in the community (which in our case is the public good) will stop evolving and become less useful. On the other hand, if all members give back to the community by taking some time to answer questions in which they have some expertise, then the online community will increase the value of the public good that it provides to its members. But how can we distinguish these cases and assess what makes members of an online community contribute more or equal to that which they take from the community? Another interesting example, as highlighted by Lerner and Tirole [2002], is the case of open source software. Positioned on successful open source software projects, such as the Apache HTTP server (which is the standard solution for providing access to web pages on the Internet), a very large number of users is benefited by it (e.g., network systems administrators, web developers, etc.), while few people contribute to its development, leading to a case of asymmetry between contributors and users that receive high benefit from it (e.g., large companies providing web hosting services). While contribution is difficult to measure in such cases (e.g., a contribution can be the case of pointing out a software bug or providing some documentation or a patch of source code to enhance a function of the product), the case remains the same. A large number of users are benefited by the effort of a lesser number of contributors who also receive a benefit from the software project that they develop. In such cases if everyone uses the software without contributing, then the software will gradually lose its value and dissolve (no participation will affect the development effort of the 117 3.3. COOPERATION AND THE PUBLIC GOODS GAME Each subject starts with zero Danish crowns Each participants takes an amount between 0 and 50 crowns from the group account (pot) The remaining amount in the pot is doubled The resulted amount is divided in four equal parts Each participant receives its share The share is added to the subjects’ initial amount Table 3.4: The same numerical example configured with a TAKE framing for the public goods game 124 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION and retention of new members. In order to control for psychological effects imposed by the method of contribution, a second configuration of the public goods game was available to the subjects as a treatment. In that treatment the subjects were required instead of contributing to a common pot, to take from an already filled one. On that case instead of contribution the subjects were able to take from the already doubled pot with a limit Pe=50 DKK. The rest were once more doubled and divided equally among the members of the group and distributed. Again the dominant strategy is to take the most out of the pot. If the members were fully cooperating for the common good and deciding not to take out any of the initial amount then this full cooperation would result to the highest available return rate of 100 DKK per subject. Having summarized our experimental configuration we proceed with the description of the protocol and the assignment of the subjects to treatments representing the different types of framing discussed above. 3.4 Experimental procedure and methods The experiment was conducted online using a web application that was programmed for that purpose. Figure 3.1 shows the login interface of this application. The selection of the subjects was conducted with the collaboration of the National Statistics Bureau which drew a random sample of the Danish population around the country. Due to the fact that citizens in Denmark need to be associated with a registration number for social security purposes the statistics office was able to obtain the factual mail addresses of the subjects and send a recruitment letter. The recruitment letters were divided in three types: Letter type A, Letter type Band Letter type C. In letter types Aand Cit was clearly stated that there will be a monetary reward for their participation while in letter type B it was generally written that their participation will help research in Denmark. The letters contained a personal identification code and the 125 3.4. EXPERIMENTAL PROCEDURE AND METHODS web address of the web application. The personal identification code was randomly generated and was used as an identifier to associate data for a subject for different parts of the experiment. Each set of letters was assigned in a set of two waves (First wave and Second Wave). The first wave was constructed mainly for participation rate identification purposes while the second wave was the main wave of the experiment. Although not stated in the letter the subjects were informed by the platform that they could login on a time window which was roughly 7days from the time they received the letter in their mail box. After that date the subjects were receiving a message that they were not able to login with this specific pincode because it has expired. Figure 3.1: The login Interface of the Web application For the evaluation of different frames the subjects were randomly assigned to one out of a set of three treatments. The two basic treatments were to distinguish the framing in the public goods game as follows 126 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION  Treatment 1: Public goods game with give configuration  Treatment 3: Public goods game with take configuration Treatment 2 was a public goods experiment with give configuration as well; however it was not giving an actual monetary reward to its participants (economic incentives were absent). Figure 3.2 depicts the participation rate for the experiment for the two waves since the letters was sent to the subjects. The first wave was an preliminary one to asses the participation level of the invited subjects (2500 invited subjects) and the second wave was the main wave of the experiment (19.500 invited subjects). Both waves have a similar participation rate which averages 15% of the invited subjects. A 16 hours time window was added to control for postage delays. It is clear that there is faster response rate to the subjects for the fourth wave which can be slightly explained by the fact that the time window for participation on the experiment for the third wave included the weekend where participation was expected to be low. An issue when comparing the two waves might be the demographic or geographical scattering of the recipients of the letters (e.g. lower participation rates would be expected for older people in rural areas). However the participation rates converge at the end of the time window therefore the actual response rate within the time window of the experiment doesn’t impose any selection effect on the wave comparison. 3.4.1 Experimental protocol Figure 3.3 shows the protocol of the experiment. As aforementioned subjects were recruited from the standard population using online invitations. The configuration for the give and take types of framing was a sequence of a set of instructions and decision screens where subjects had to read and then input their decision allonge with other data. The sequence of the decision states was as follows: 127 3.4. EXPERIMENTAL PROCEDURE AND METHODS Figure 3.2: The evolution of the participation rate for the two waves that were used in this experiment (Wave 1: green line , Wave 2: red line 128 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION 1. The introductory and demographics state where subjects were asked to give their demographic details read the instructions of the public goods game and answer to a set of control questions to test whether they understood the questions provided by the public goods game. (introduction) 2. The decision about how much of the endowed amount they will contribute to the common goal (unconditional choice) 3. An estimation of how much did they thing that the other members of the group contributed on the common goal. 4. The decision about how much of the endowed amount they will contribute to the common goal if they know how much the other members of the group contributed on average (conditional choice). 5. The online community participation characteristics (online community) where subjects had to answer to a set of questions related with online sociability characteristics such as for example how often do they communicate etc. Figure 3.3: The experimental protocol used in this study. The Take framing had exactly the same configuration with the contribution to be 129 3.4. EXPERIMENTAL PROCEDURE AND METHODS presented as Pg−Pcso the maximum amount that a subject would take from the pot would be 50 Danish Crowns. After the experiment was over the subjects were asked to login again to the platform in order to see how much they earned and enter their banking details in order to receive the amount they earned. The next section discusses the assignment of the subjects to treatments. 3.4.2 Assignment to treatments Treatment assignment was based on two parameters: the Letter type and a randomly generated two point decimal variable (between 0 and 1) generated once a subject accepted the invitation and logged in to the system. Table 3.5 depicts the procedure of the assignment to the treatments. Treatments Letter Type Treatment 1 (Give: Economic Incentive Stated) Treatment 2 (Give: No Economic Incentive Stated) Treatment 3 (Take: Economic Incentive Stated) A RN <0.66 - RN ≥0.66 B RN <0.5 RN ≥0.5C RN <0.5 RN ≥0.5Table 3.5: Randomization procedure for the assignment of subjects to treatments according to the letter type. RN denotes the random variable value generated by the system for the treatment assignment The give framing was divided in two treatments to observe where an a priori economic incentive might lead to different results in initially participating on the experiment. 130 CHAPTER 3. BEHAVIORAL CHARACTERISTICS AND COOPERATION IN ONLINE COMMUNITIES: AN EXPERIMENTAL INVESTIGATION 3.5 Data analysis and procedures Having described the procedure of recruiting the subjects and assigning them to treatments we describe the basic demographic characteristics of the subjects and we provide an intra-treatment comparison for the variables that we are interested to analyze. 3.5.1 Distribution to treatments and basic demographics As aforementioned the data were collected by running the experiment in two separate waves (wave 1, wave 2). Table 3.6 depicts the distribution of the subjects to the waves along with the letter types included. Wave Letter Type (Completed/Assigned) First Wave Second Wave A 153/1503(10.1%) 1889/16524(11.4%) B 24/517(4.6%) 91/1483(6.1%) C 36/483(7.4%) 98/1517(6.4%) Total Subjects Completed the Experiment 213(9.3%) 2078 (10.6%) Subjects originally assigned 2503 19524 Table 3.6: Distribution of Letter Types by waves. As can be seen in the table the assignment in separate waves (First Wave, Second Wave) was done highly asymmetrically. That was due to the fact that the first wave acted as an identification wave in order to estimate participation rates for the second wave. This is depicted also on the number of subjects completing the experiment. For the first wave the full response rate was around 8.5% while for the second wave the actual response rate was a little bit higher ∼10.6%. As aforementioned this might be attributed to the fact that the first wave includes the weekend of the week 20 (2008), 131 3.5. DATA ANALYSIS AND PROCEDURES so physical absence of the subjects from their official residence might have affected the participation rates. Table 3.7 depicts the distribution of the subjects to the waves and the subsequent treatments. An interesting observation from the participation rate comes from the comparison between treatments 1 and 2 for the letter types B and C where the participation ratio was not significantly different although there was the controlled absence for an economic reward in both letter types for the second treatment. An important aspect of an experimental procedure is the representation of demographics within and between treatments. Such an important demographic for every experimental procedure, can be the case of gender representation on treatments [Eckel and Grossman, 1998]. As can be seen in Figure 3.4 both genres are represented equally in the two waves of the experiment. In particular for the first wave we had 104 male and 109 female subjects participating while in the second wave we had a slight difference of with 1078 male and 1000 female subjects participating. The equality between the genre representations can be also attributed to the sample selection procedure which was facilitated by a random sample of the Danish population. Genre representation can be also seen in Figure 3.5 where the distribution by treatment displays also no significant difference between genres. In particular for Treatment 1 we have 770 males (51%) and 728 (48%) females, for Treatment 2 we have 55 males (47%) and 62 (52%) females and for Treatment 3 we have 357 males (52%) and 319 females (47%). Although genre is one important demographical factor for a population another important factor is the distribution of the age among treatments. We defined four basic groupings for the age variable as: Age group 1 (Age≤30), Age Group 2 (Age>30 and Age ≤40), Age group 3 (Age>40 and Age≤50) and Age Group 4 (Age >50). Figure 3.6 depicts the assignment of subjects in age groups according to our categorization. As can be seen also in Table 3.8 there is an over-representation of the age group of subjects that are older than 50 years old while on the contrary we have 132 Letter Type (Completed/Assigned) Treatment 1 (Give: Economic Incentive Stated) Treatment 2 (Give: No Economic Incentive Stated) Treatment 3 (Take: Economic Incentive Stated) A1366 - 676 B47 68 - C85 49 - Total Subjects Completed the Experiment(Treatments) 1498 117 676 Table 3.7: Distribution of Letter Types by framing and treatment 5.5. DISCUSSION Dependent Variable Variable(Controlling for frequency of use) dy/dx P > |z|Significant Tip Qi .454 0.000 YES QI Tip -.0137951 0.000 YES Table 5.10: Marginal Effects the TIP and Quality on the strategic interaction case (N=892) magnitude perspective. In particular increasing TIP by one unit this will increase quality by 0.45 units which is a result of the intrinsic nature of the incentive provided. The other important result that can be extracted from the estimator model is the case that increase in the past quality will result to a slightly lower tip which is highly significant for the post effects estimator. In the three cases that we have discussed on this section we controlled for the frequency of use and the price as factors that might affect the decision to give as well as the size of the tip. We summarize these findings as well as connections with the literature in the discussion section that follows. 5.5 Discussion Summary of findings Returning back to the theoretical discussion on Section 5.2 we revisit the three cases of tipping as it has been described for offline environments and used in our case: (a) Reciprocation of the Service Provided, (b) Adherence to moral codes in order to attain social approval (or avoid social disapproval) even in an online anonymized environment and (c) Tip as an incentive for future service. 236 CHAPTER 5. THE IMPACT OF EXTRINSIC REWARDS ON STRATEGIC INTERACTION IN ONLINE COMMUNITIES: AN ANALYSIS ON GOOGLE!ANSWERS Evidence of Tipping used as a reciprocation of the service in an online environment The results presented in the previous section concerning the case of reciprocity provided that in 71.30% of our subjects (8538) has tipped due to reciprocity. The marginal effect of the constructed quality index to the size of the tip was 0.6178. While the significance obtained from the ex-post estimator is not on the highest level it still can explain the cases falling into the 99.5% confidence interval. This can be attributed to the fact the initial clustering of the three cases made was not exact due to the fact that membership on the first cluster was not exclusive. This was done in order to keep cases where relatively high quality and high tip could also be included. Number of Cases Percentage Use of Tipping 8538 71.30% Reciprocity 2544 21.2% Social Norm 892 7.4% Strategic Interaction Table 5.11: Summary of the Cases identified in our dataset Essentially this was due to the fact that we needed to control for other factors that affect tip in general as we saw in Section 3.2 and in particular the effect of the price and previous history both in service use and individual interaction. This provides a ground for the importance of extrinsic rewards as a motivation factor for enhancing interaction on online communities. The prospect of tipping in that case works as a mechanism for improving quality which is highly evident on the cases that we analyzed in this study. That makes the contribution of this study solid with the findings of Benabou and Tirole [2003] from the perspective that explicit stated rewards can have a negative effect on the long run while a non explicit stated reward (as the tip in our case) can have a positive effect on performance which in that case is indicated by the quality index. 237 5.5. DISCUSSION Existence of social norms in online environments What is interesting from this finding is that social norms typically seen in offline environments do exist on a highly anonymized setting as the one in Google Answers. In fact in our study we found that 21.2 % of the cases adhere to a social norm where tip is given regardless the quality of the answer given. One probable explanation can be the cultural background of the users. In US for example for every service transaction customers are expected to provide a tip as a percentage of the price of the food. Surprisingly if we revisit the numbers from Table 5.4 then the average price/tip ratio is 11.68% and accounting for the case of reciprocation and strategic interaction the ratio is well around 10% making it almost the same with the reported price/tip ratio that is used in every restaurant in the US. This is a result that might need additional exploration in online environments and it might be that in such cultural background social norms are so strong that become internalized especially in collective settings [Adler and Kwon, 2002]. Tipping as an incentive to future service In that case the initially formulated argument that an individual tips strategically in order to maximize the benefit that he/she receives from the service founds solid ground with the analysis that we conducted. We were able both to justify such a case as well as to show with the empirical data gathered that such a behavior exists on an online and anonymized setting where we control for other influential factors of tipping such as the price and the frequency of use. In particular the ex-post marginal effects estimators used for that case provided that gradually decreasing the tip by 0.013 units will have an increase of 0.45 units in the service quality (as measured by the users judgment and the time to get an answer). 238 CHAPTER 5. THE IMPACT OF EXTRINSIC REWARDS ON STRATEGIC INTERACTION IN ONLINE COMMUNITIES: AN ANALYSIS ON GOOGLE!ANSWERS Connections with the literature The context of GoogleAnswers has also been studied by other researchers in the information systems literature. The first paper that appeared with a study of social interactions in GoogleAnswers is the one by Edelman and Draft [2004] where the relationship between earnings and ratings is examined. Rafaeli et al. [2005] and Rafaeli et al. [2007] have published a series of papers where they tackle the effect of price and social incentives (such as previous participation) as a factor that affects the outcome of the interaction and the sustainability of activity. Regner [2004] approaches GoogleAnswers from an economic perspective and in particular on the way such a standard principal-agent relation can induce evidence of social or other regarding preferences in such an online environment. Regarding the service model papers from the information and library science literature as in [Cahill, 2007] and [von Retzlaff, 2006] have addressed possible implications for the use of librarians and library users as possible agents where libraries could provide the facilitator by connecting librarians and users on such type of environment. Our study is positioned differently in the literature from the fact that it tackles with (a) User tactics and behavior – why users in such settings provide tips ? and (b) what are the implications for online communities (from a service viewpoint) where such an incentive reward is used. To this end the study presented on this chapter contributes to the exploitation of such a model from a user perspective as to how he/she can facilitate better service by strategically motivating servers (researchers that are willing to carry out the task on a payment basis).From a service design perspective the use of such mechanisms can enhance the activity as it has also been discussed in [Rafaeli et al., 2007]. From a contribution to the service literature, undoubtedly most of it deals with cases of restaurant tipping. What this study contributes is that is studying tipping on an online environment where the possibility to disentangle the reasons why someone will tip is much better than in field settings. 239 5.6. CONCLUSIONS AND FURTHER RESEARCH 5.6 Conclusions and further research This chapter connects with the research objective of this dissertation in multiple ways. First, it presents a complex environment where social capital contributions are made only by one category of users who, in turn, can be incentivized by both explicit (price) and implicit (tip) monetary rewards. The effect of extrinsic rewards can be seen on the high quality (both perceived and observed) received by the customers who make the outcome of the interaction more valuable. Such an environment is interesting to study, both from an economics and social interaction perspective, since the trading of the economic commodity in this case is information produced by searching the vast index of a popular search engine. One particular extension of the study presented here could be on understanding the users’ desire or willingness to pay for information contextualizing his/her preferences by category. In the analysis we have seen the general factors that affect tipping when in the context of information, it can play a role (e.g., both tipping and price were higher in categories such as business and money and computers and internet in contrast with the other general categories). Another possible extension would be to study the other’s judgment of the quality of the contributions provided and the way in which the communication code affects it. Such a study is the one presented in the following chapter where regardless of the incentive mechanism used for deriving the contributions from the users/contributors, other users can evaluate the usefulness of such contributions and the way they affect their own choice decision in an environment where the commodity is codified information (books) which are context independent. 240 Bibliography P. S. Adler and S. W. Kwon. Social capital: Prospects for a new concept. Academy of Management Review, 27:17–40, 2002. T. Amemiya. Tobit models: A survey. Journal of Econometrics, 24:3–61, 1984. A. Ardichvili, V. Page, and T. Wentling. Motivation and barriers to participation in virtual knowledge-sharing communities of practice. Journal of Knowledge Management, 7: 64–77, 2003. K. J. Arrow. The economics of information: An exposition. Empirica, 23:119–128, 1996. R. Axelrod and W. D. Hamilton. The evolution of cooperation. Science, 211:1390–1396, 1981. O. H. Azar. The history of tipping˚ Ufrom sixteenth-century england to united states in the 1910s. Journal of Socio-Economics, 33:745–764, 2004a. O. H. Azar. The implications of tipping for economics and management. Economics, 30:1084–1094, 2003. O. H. Azar. The social norm of tipping: A review. Journal of Applied Social Psychology, 37:380–402, 2007. 241 BIBLIOGRAPHY O. H. Azar. What sustains social norms and how they evolve? the case of tipping. Journal of Economic Behavior and Organization, 54:49–64, 2004b. R. Benabou and J. Tirole. Intrinsic and extrinsic motivation. Review of Economic Studies, 70:489–520, 2003. G. E. Bolton and A. Ockenfels. Erc: A theory of equity, reciprocity, and competition. American Economic Review, 90:166–193, 2000. B. Butler, L. Sproull, S. Kiesler, and R. Kraut. Community effort in online groups: Who does the work and why? Leadership at a Distance: Research in TechnologicallySupported Work, 2007. K. Cahill. Worth the price? virtual reference, global knowledge forums, and the demise of google answers. Journal of Library Administration, 46:73, 2007. R. B. Cialdini and M. R. Trost. Social influence: Social norms, conformity, and compliance. The handbook of social psychology, 2:151˝ U92, 1998. B. Edelman and P. Draft. Earnings and ratings at google answers. Unpublished Manuscript, 2004. E. Fehr and U. Fischbacher. The nature of human altruism. Nature, 425:785–791, 2003. L. Festinger. A theory of social comparison processes. Human Relations, 7:117, 1954. H. Gintis. The hitchhiker’s guide to altruism: Gene-culture coevolution, and the internalization of norms. Journal of Theoretical Biology, 220:407–418, 2003. B. Gu, P. Konana, B. Rajagopalan, and H. W. M. Chen. Competition among virtual communities and user valuation: The case of investing-related communities. Information Systems Research, 18:68, 2007. KHEE Lee and CB Hatcher. Willingness to pay for information: An analyst’s guide. Journal of Consumer Affairs, 35:120–140, 2001. 242 BIBLIOGRAPHY M. Lynn and A. Grassman. Restaurant tipping: an examination of three rational explanations. Journal of Economic Psychology, 11:169–181, 1990. M. Lynn and B. Latane. The psychology of restaurant tipping. Journal of Applied Social Psychology, 14:549–561, 1984. M. Lynn and M. McCall. Gratitude and gratuity: a meta-analysis of research on the service-tipping relationship. Journal of Socio-Economics, 29:203–214, 2000. M. Lynn and K. Mynier. Effect of server posture on restaurant tipping. Journal of Applied Social Psychology, 23:678–685, 1993. R. E. Nisbett and T. D. Wilson. Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84:231–259, 1977. J. Preece. Online Communities: Designing Usability and Supporting Socialbilty. John Wiley & Sons, Inc. New York, NY, USA, 2000. J. Preece, B. Nonnecke, and D. Andrews. The top five reasons for lurking: improving community experiences for everyone. Computers in Human Behavior, 20:201–223, 2004. S. Rafaeli and DR Raban. Experimental investigation of the subjective value of information in trading. Journal of the Association for Information Systems, 4:119–139, 2003. S. Rafaeli, G. Ravid, and V. Soroka. De-lurking in virtual communities: a social communication network approach to measuring the effects of social and cultural capital. System Sciences, 2004. Proceedings of the 37th Annual Hawaii International Conference on, page 10, 2004. S. Rafaeli, D. Raban, and G. Ravid. Social and economic incentives in google answers. In ACM Workshop Sustaining Community: The role and design of incentive mechanisms in online systems, Sanibel Island, FL USA, 2005. 243 BIBLIOGRAPHY S. Rafaeli, D. R. Raban, and G. Ravid. How social motivation enhances economic activity and incentives in the google answers knowledge sharing market. International Journal of Knowledge and Learning, 3:1–11, 2007. T. Regner. Why voluntary contributions? google answers! 2004. RR Reno, RB Cialdini, and CA Kallgren. The transsituational influence of social norms. Journal of personality and social psychology, 64:104–112, 1993. B. J. Ruffle. Gift giving with emotions. Journal of Economic Behavior and Organization, 39:399–420, 1999. T. S. H. Teo, V. K. G. Lim, and R. Y. C. Lai. Intrinsic and extrinsic motivation in internet usage. Omega, 27:25–37, 1999. H. R. Varian. Pricing information goods. Coyle’s Information Highway Handbook: A Practical File on the New Information Order, 1997. L. von Retzlaff. E-commerce for library promotion and sustainability: how library technicians can market themselves and their library’s services online. Australian Library Journal, 55:102, 2006. E. Walster, G. W. Walster, E. Berscheid, W. Austin, J. Traupmann, and K. Mary. Equity: Theory and Research. Allyn and Bacon, 1978. S. Whittaker, L. Terveen, W. Hill, and L. Cherny. The dynamics of mass interaction. Proceedings of the 1998 ACM conference on Computer supported cooperative work, pages 257–264, 1998. 244 CHAPTER 6 Evaluating Content Quality and Usefulness of Online Product Reviews Online product reviews are an important resource for consumers in online marketplaces because they provide a useful source of support information during the purchase of goods. Furthermore, in some online marketplaces consumers have the opportunity to assess the usefulness of a review by using a dichotomous evaluation form provided by the online marketplace. These evaluations produce a usefulness score, which can be calculated as a fraction of helpful votes out of the total votes that a review has received. This enables testing hypotheses regarding the factors that affect the usefulness of reviews which may, in turn, be used as metrics if evidence supporting the connection between the factor and the usefulness is found. This chapter reports an empirical study test applied to a large dataset of reviews collected from the United Kingdom section of the popular online marketplace, Amazon, to explore the connections between readability and usefulness. The results of these evaluations point out that usefulness is significantly affected by the qualitative characteristics of the review as measured by readability. 245 6.2. A BACKGROUND ON READABILITY TESTS the result of a readability formula cannot tell us whether the content of the review expresses personal views on the product or contains some gender, class, or even cultural bias. Furthermore, to avoid cultural background variance, and to a large extent language proficiency, we collected the reviews only from the United Kingdom store of the online marketplace in order to have only native English speakers and as much geographical concentration of the population as possible (population from one country)3. Readability tests to study qualitative characteristics of several types of texts have been applied to several areas in information science, and a large set of readability indexes has been developed over the years [Paasche-Orlow et al., 2003]. For our study we selected four major readability texts which individuals on various educational levels have used extensively to evaluate the readability of a piece of text. Table 6.1 lists the readability tests that we used in our study. These are the Gunning-Fog Index, the Flesch/Kincaid Reading Ease, the Automated Readability Index (ARI) and the Coleman-Liau Index. The major reason for selecting these readability metrics is the availability of software to provide reliable measurement of this indexes (The GNU style command). All four tests evaluate the readability of a text by consistently decomposing the text to its basic structural elements, which are then combined using the empirical regression formula. An important issue of a readability test is that it can be used to evaluate only texts of a certain length since a reader’s ability to comprehend a text also involves cognitive properties that are beyond the scope of this study. The logic behind the calculation and the norms of these instruments is described in the sections below. 6.2.1 The Gunning-Fog Index The Gunning-Fog index [Gunning, 1969] provides a measure of how well an individual with an average high school education is able to comprehend the evaluated piece of 3Furthermore, the Readability tests used in the analysis section of this study have been developed only for the English language, and language proficiency undoubtedly affects their validity. 252 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS text.The approach to compute this index was the following:  For each review we calculated the average number of words per review sentence on a 100+ word review passage. This gives as the average sentence length (L).  We then obtained the number of difficult words (D)-that is words that have more than three letters-by excluding proper nouns, compound words, and common suffixes.  We finally added the average sentence length to the number of the difficult words. The following equation describes the empirical relation in the Fog Index. Fog =0.4×Words Sentence +100 ×N(compe_ords) N(ords) An obvious difficulty in measuring the Fog index for a given text is the evaluation of the number of complex words. In our analysis we considered a word as complex if it had more than two syllables. 6.2.2 The Flesch Reading Ease The Flesch Reading Ease index [Flesch, 1951, Kincaid et al., 1975] is a readability test that uses as a core linguistic measure the number of syllables per word and the number of words per sentence in a given text. The Flesch test is used to evaluate the complexity of the text in order determine the number of years of education needed for someone to understand it. The following equation describes the calculation of the Flesch-Kincaid score for a given text: FK =0.39 ×tot_ords tot_sentences+11.8×tot_sybes tot_ords −15.59 The variables total_words, total_sentences and total_syllables denote the total number of words, sentences, and syllables, respectively, found in the text. For calculating 253 6.2. A BACKGROUND ON READABILITY TESTS the Flesch score of a particular review we decomposed the text into sentences, then words, and finally into syllables, which were combined using the constants presented in the formula above. It can be easily inferred from the mathematical expression that the sorter is the number of words per sentence—the fewer words per sentence, the better the readability score that the Flesch test will provide. 6.2.3 The Automated Readability Index The Automated Readability Index (ARI) differs from the Gunning-Fog and the FleschKincaid tests in that it uses simpler metrics to evaluate the readability of a typical English language text. In order to calculate the ARI for a given review we first calculated the total number of characters (excluding standard punctuation such as hyphens and semicolons) and the total number of words. AR =4.71 ×chrcters ords +0.5×ords sentence−21.43 The calculations for the ARI involved the same steps as for the Fog and Flesch indexes where additionally the number of characters, that is the review length, had to be calculated as well. The ARI can provide an indication of the impact of the review length on the readability of the review. 6.2.4 The Coleman-Liau Index The Coleman-Liau Index [Coleman and Liau, 1975] is similar to the Automated Readability Index, the only difference being that the second part of the formula considers a more careful selection of the textual characteristics of the evaluated piece of text. The CL index has been developed specifically for machine-based scoring, thus the calculations that it involves are quite tiresome to do by hand. The following formula 254 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS describes the Coleman-Liau index. CL =5.89 ×chrcters ords −0.3×sentences ord −15.8 The calculation of the Index considers fragments of sentences of 100 words multiplied by a constant (0.3). 6.3 Analysis and results Having provided a background on the readability tests that we are going to use, we continue to the analysis of the reviews in our dataset in order to test whether the readability tests can actually give us an indication how the qualitative characteristics of a review influence its usefulness for a consumer. 6.3.1 Data collection and definition of variables In order to apply the readability tests that we discussed in the section above, we developed a web crawler to capture the content of the book section of Amazon UK. The crawler consisted of two parts: (a) A web client to randomly pick items from the front page of the bookstore and (b) a client to the web-service interface provided by Amazon (AWS) where the data for the particular item were collected4. The list of books was stored in a relational database which we used for further processing of the reviews expressed in each individual product page. We omitted from the database those books for which the publication date was older than 6 months or had no rating. Furthermore, we excluded books at special offers or discounts to control for price or bargain effects. The reason that we picked Amazon UK in order to obtain the dataset used in this study is the degree of language homogeneity among reviewers and consumers which 4The Amazon Web Services API is provided by Amazon to developers and resellers and is publicly accessible at http://developer.amazonwebservices.com/. The API version that was used was the 2008-04-07 255 6.3. ANALYSIS AND RESULTS Variable Code Variable description Productid The id of the product that this review is written for. It is used to control for the publication date and other product characteristics Summary The summary / title of the review Content The actual content of the review. To be used for content analysis. revieworder The order that the review appears on the product review page. Reviewpage The page that the review appears (default setting is five reviews per page). rating The rating that this review justifies, measured on a 1-5 Likert scale. totalvotes The number of total votes that have been given to this review. helpfulvotes The number of votes that consider this review helpful. reviewerid The id of the customer used to control if the customer is a professional reviewer or not. Table 6.2: The main variables of the initial dataset collected by using the web crawler might play a role in the comprehension of a text. This is important because readability tests are useless if a reader is not a native speaker of the language in which the text is written. This is due to the fact that many languages differ in syntactical form, and the style of the language in the review might be totally different from the reader’s native language. Table 6.2 provides a description of the variables of our dataset. These variables can be categorized into two groups: the numerical expressions of the review (rating, totalvotes, helpfulvotes), and the textual or qualitative characteristics (summary, 256 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS content) including the identifiers and exposure (productid, reviewerid, revieworder, reviewpage). A particular issue with the dataset that we collected was that of by-passing promotionbacked items such as bestsellers. Since these items are more accessible to the visitors to the online bookstore, there is always a selection bias towards the more visible items. This placement may result in a high exposure of recent product reviews in contrast to older ones. In order to avoid that bias, the web crawler kept a list of the frequency of the items that were displayed in the front page and randomly chose items listed by categories. Our dataset contains in total seven variables and two identifiers. The reviewerid actually provides the id of the customer in the online bookstore’s central database. By using this identifier we can group the reviews by customer since a customer may have submitted reviews for more than one product, in this case books. The productid is the unique product identifier provided for a product. With this identifier we can group the reviews by product and check for variances between products of different categories. Figure 6.2: Distribution of the rating values among the items in our dataset. (Total of reviewed items/books: Nb=7320 We define the usefulness ratio of a review (UR) as the fraction of the votes that 257 6.3. ANALYSIS AND RESULTS Figure 6.3: The distribution of usefulness scores on our dataset plotted by density (N=37221) considered this review helpful (helpfulvotes) divided by the total number of readers that evaluated the usefulness of the review (totalvotes). Thus we have the dependent variable for our analysis defined as: UR =hepƒotes tototes The usefulness ratio is in fact a measure of the quality of the review as considered by the readers themselves. From that definition it is easy to infer that the bigger the number of helpful votes a review receives from those that evaluated the review, the higher will be the usefulness ratio. However, since the number of total votes that a review has (that is the minimum number of readers) may affect the consistency of the metric, we need to keep control for the exposure of this review since some reviews at a certain period of time receive more exposure than others. Typically, this exposure is affected by time since the 258 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS Figure 6.4: Histograms of the distributions of the four readability tests used on our dataset. system displays first the most recent reviews for a product (in this case the book). In our study the particular exposure of a review was measured by keeping a set of two variables for the pagination results. In particular, the variable reviewpage indicates whether this review was at the first, second, or third page at the time the review was retrieved. The same applies to the revieworder, which controls the display order for a particular review on a particular page. Combining the two variables (reviewpage, revieworder) into a new compositional variable, we are able to control for the review exposure on the website during the time the review was posted. For example, if a review appears on page 2 and was ordered as third in the page then the exposure value is 23, and so forth. It is generally assumed that reviews which appear on the top of a page get much higher exposure than a review that appears at the bottom since visitors’ attention is captured by elements that are displayed in the beginning of the space under the product description. On the other hand, we don’t have a variable that justifies the actual exposure of a review and, in particular, the number of people that read the review. However, in 259 6.3. ANALYSIS AND RESULTS Figure 6.5: Scatter plot matrix of the usefulness ratio and the qualitative characteristics of the text of an online review. 260 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS order to hold our analysis to an acceptable level we assume that the total number of people that evaluated the usefulness of a particular review is the minimum number of the readers who read it. In that way we get an indication whether a review has been read by a high number of visitors since it is assumed that those two numbers are positively correlated. The dataset consists of 38,366 reviews where the total votes (totalvotes) were greater than zero, which means that the reviews on our dataset have been evaluated for their usefulness at least once. Figure 6.3 provides an overview of the distribution of usefulness scores on our dataset. It is interesting to note that around 47% (total of: 17,695) of the reviews have received a perfect score from the readers, which provides that around half of the reviews were very highly acclaimed by their readers. The result is that for this group of particular reviews, the number of helpful votes is the same as the number of potential buyers that have read the reviews. On the other hand, we find that approximately 9% of the votes (total of: 3,292) found the reviews to be totally non-useful for their readers, receiving an absolute 0 of helpful votes. As can be observed in Figure 6.3, much of the variance in the usefulness score happens between the 0.8 and perfect (1). In total, 7,320 books are covered by the reviews. Figure 6.6 depicts the distribution of the average usefulness score per rating scale value for the items covered by our dataset. Figure 6.2 presents the distribution of the rating scores that were given to all reviews on our dataset. It is interesting to note that more than 70% of the reviews are highly positive (rating>=3). This might be explained by the fact that most of the reviewers were more than satisfied with the books they read and therefore provided a review. It is likely that the unsatisfied customers were not willing to report their experience with that particular book. The variable we used to apply the readability tests was the review’s content as presented on the website, encapsulated by the content and summary variables re261 6.3. ANALYSIS AND RESULTS Since splitting the data into two groups gives us two independent samples from the same dataset, we are able to use a statistical test to find how significant the difference is between the means of the characteristic in which we are interested in these groups. The usefulness of a review is affected by the rating that the review has received Having split our dataset into two grouping variables, we are able to test the relation between the usefulness ratio of a review (UR) and the rating that this review has received. In particular, we are interested in identifying—by comparing the means of the ratings—whether the usefulness ratio has any relation to the rating that a review has received. Figure 6.7 shows the distribution of the usefulness ratio for each of the values of the rating scale (1-5). Figure 6.7: The distribution of usefulness ratio for each of the values of the rating scale that a particular item was evaluated 268 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS As aforementioned, we selected a nonparametric test to test whether the mean value of the rating is the same across the groups that contain high and low UR. For the grouping variable groupid-A the groups are split equally by the UR (groupid-A =1 if the UR is less than 0.5 and groupid-A =2 otherwise). We selected the Mann-Whitney test to compare the mean value of the rating between the two groups. The reason for choosing this particular test is that the Mann-Whitney test is a nonparametric statistical test that, unlike parametric tests, (t-test) does not rely on the assumption of normality (the distribution of the rating among the groups follows the normal distribution). For the grouping category Groupid-A we ran the Mann-Whitney test for a total of N= 34002 observations. The Z value that we obtained from the test was Z =-39.407, which results in a P value of P=0.000, which is highly significant at three degrees of freedom, so we can reject the hypothesis that the mean of the rating is the same for reviews that have a high and low usefulness ratio. In order to verify the above result in case the rating plays no role in determining the different values of the UR also in the case of the upper and lower limit (since most of the UR is concentrated on values of 0 and 1), we used the second grouping (Groupid-B) which splits the dataset in two parts, which in fact are the first and fourth quintile of the UR values. We ran again the Mann-Whitney test for a total of N= 26884. The Z value obtained from the test was Z=-35.547 which corresponds to a P value of P=0.000 providing that the hypothesis that the rating is the same between the two groups is rejected. Both results affirm that, indeed, the rating is affected by the usefulness ratio of the review. In fact, as can be seen from Figure 6.8 the value of the UR is increased depending on the rating scale. 269 6.3. ANALYSIS AND RESULTS The usefulness of a review is affected by the qualitative characteristics of the review In order to test whether the usefulness of a particular review is affected by the qualitative characteristics of the review text, we followed the same procedure for both grouping categories (Groupid-A and Groupid-B), which we used for testing the relation between the usefulness ratio and the rating. Running the same test for the first grouping category (Groupid-A), we obtained a Z value of Z= -27.433, which provides a P value of P=0.000. Again, this value is highly significant at three degrees of freedom. The hypothesis that the mean of the review length is the same when the UR is high or low is rejected providing that the length of the review text also affects the usefulness of the review. Figure 6.8: Average word length comparison between the two groups (groupid-B=2 if r > 0.75) For the second grouping variable, the same test gives us a P value of P=0.000 (Z= -37.050) which also rejects the null hypothesis. The results of the tests confirm that the difference displayed in Figure 6.8 is significant and provides that the qualitative 270 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS characteristics of the review (in that case the review text6) also affect the usefulness ratio of the review and are positively correlated (higher review usefulness implies that the review text will be longer). The rating of a review is affected by the qualitative characteristics of the review Running the Mann-Whitney test for the qualitative characteristics of the review text and using the third grouping variable (Groupid-C), we obtained a Z value of Z = -3.097 resulting to a P value of P=0.0020. This provides that we can reject the null hypothesis and confirm the significance of the relation between the qualitative characteristics of the review text and the actual rating value that a particular review has. Figure 6.9: Distribution of the average length of the review text following the rating of the review In fact, Figure 6.9 can be used to compare the tendency of the review text across the values of the rating scale that is used when a review is submitted (1-5). It is 6The results of the Mann-Whitney test for the Fog, Flesch, and ARI indexes also provide the same results (P=0.000 for Z values of -14.903, -12.547 and -3.728) 271 6.4. DISCUSSION clear that reviews with positive ratings tend to contain more text (as depicted by the number of words contained in the review text). 6.4 Discussion The results of the tests, along with the interpretation of the regression coefficients that we obtained in Section 6.3.2, provide us with some interesting insights into the relation between the usefulness of an online review as considered by the visitors to the online marketplace and the actual qualitative characteristics of that review. By our hypothesis testing of the data we were able to verify the following: 6.4.1 The usefulness of a review is affected by its positive or negative rating value Going back to Figure 6.7, we see a significant (as confirmed by the tests) tendency of the usefulness ratio towards reviews with higher rating. That might be explained by the fact that consumers (as visitors of an online information resource) tend to read appraisals of a product first (the fact that a review is marked with 5 stars also increases attention from a usability point of view). From the definition of the usefulness ratio, the higher the rating value of a review, the higher the number of helpful votes that a (future) customer will give it. At least in our dataset, reviews with a rating value above three had a higher amount of usefulness ratio with a perfect score resulting in the fact that indeed the higher the number of helpful votes the review received, the higher was its usefulness ratio. This finding implies that customers react to positive and negative reviews differently, which also confirms results from the study done by Hu et al. [2008]. 272 CHAPTER 6. EVALUATING CONTENT QUALITY AND USEFULNESS OF ONLINE PRODUCT REVIEWS 6.4.2 The usefulness of a review is affected by its qualitative characteristics The word length and the readability scores (as a result) have confirmed that the style of the text in a review also provides an indicator of why a consumer considers a review to be highly useful. This can be explained by the fact that consumers evaluate a critique by how well it is justified and whether it provides them with as much information as possible in order to form their own views about the quality of a particular product (in our case a book) and reduce the uncertainty about its quality. 6.4.3 The rating that the review provides is affected by its qualitative characteristics Our results indicate that there is a clear relation between the value of the rating that a review provides and its qualitative characteristics in terms of review length. From these results we can imply that consumers who are satisfied with the book they read want to express more of their personal opinions in their reviews, which makes the standard case from word-of-mouth scenarios that excited customers are often willing to provide more information about their experience and reflect their excitement in their judgment of the product or service they have consumed. 6.5 Conclusions and further remarks The main result of our study is that when a particular review is considered useful to the potential buyers of a product or a service, this has something to do with the qualitative characteristics of the review justification as a piece of text. By employing readability formulas we were able to analyze the reviews in our dataset and provide a set of results in connection with the usefulness of the particular review. In this study we have focused on the content-specific characteristics of the review text. However, one of the limitations of this approach is that we were not able to as273 6.5. CONCLUSIONS AND FURTHER REMARKS sess whether a review was written in a way that expressed a personal opinion about a product or a service. This limits the study because we know from the marketing literature that potential consumers tend to associate themselves with other consumers who express a more personal experience about the product that might influence the potential consumers’ choice process [Bettman and Park, 1980]. Another limitation of this study is its inability to check the actual reliability of the readability tests by cross validating whether the tests actually measure the readability of a review written on a website, since the readability tests do not take into account usability factors (e.g., the position of the text on the screen, etc.). The empirical results of this study also contribute to the ever-growing literature on the importance of online reviews as an advantage to online marketplaces over traditional markets where the codification of information related to the products or services can actually help future buyers to evaluate the quality of an experience good (in our case books) by reading the judgments provided by other customers. In relation to the research question pursued in the context of this dissertation, we showed that the qualitative characteristics of online reviews communities and online communities in general are an important element that affects the perceived usefulness of the online community output. The study also shows that the qualitative characteristics of online reviews are a rich source of information toward understanding the way consumers evaluate information about products in an online marketplace, which can be combined with evidence coming from studies regarding quantitative aspects as perception based on accumulation of negative reviews [Lee, Park and Han, 2008]. The results of this study represent a point of departure to extend the analysis further by incorporating cognitive characteristics of consumers as captured by their reviews. 274 Bibliography J. R. Bettman and C. W. Park. Effects of prior knowledge and experience and phase of the choice process on consumer decision processes: A protocol analysis. Journal of Consumer Research, 7:234, 1980. S. Borenstein and G. Saloner. Economics and electronic commerce. Journal of Economic Perspectives, 15:3–12, 2001. J. J. Brown and P. H. Reingen. Social ties and word-of-mouth referral behavior. Journal of Consumer Research, 14:350, 1987. J. A. Chevalier and D. Mayzlin. The effect of word of mouth on sales: Online book reviews. Journal of Marketing Research, 43:345–354, 2006. E. K. Clemons, G. G. Gao, and L. M. Hitt. When online reviews meet hyperdifferentiation: A study of the craft beer industry. Journal of Management Information Systems, 23:149–171, 2006. M. Coleman and T. L. Liau. A computer readability formula designed for machine scoring. Journal of Applied Psychology, 60:283–284, 1975. 275 BIBLIOGRAPHY L. J. Cronbach. Coefficient alpha and the internal structure of tests. Psychometrika, 16:297–334, 1951. C. Dellarocas. The digitization of word of mouth: Promise and challenges of online feedback mechanisms. Management Science, 49, 2003. R. F. Flesch. How to Test Readability. Harper, 1951. R. Gunning. The fog index after twenty years. Journal of Business Communication, 6: 3, 1969. M. Hu and B. Liu. Mining and summarizing customer reviews. Proceedings of the 2004 ACM SIGKDD international conference on Knowledge discovery and data mining, pages 168–177, 2004. N. Hu, P. A. Pavlou, and J. Zhang. Can online reviews reveal a product’s true quality?: empirical findings and analytical modeling of online word-of-mouth communication. Proceedings of the 7th ACM conference on Electronic commerce, pages 324–330, 2006. Nan Hu, Ling Liu, and Jie Zhang. Do online reviews affect product sales? the role of reviewer characteristics and temporal effects. Information Technology and Management, 9, 2008. J. P. Kincaid, R. P. Fishburne Jr, R. L. Rogers, and B. S. Chissom. Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel. 1975. F. Lehner. Quality control in software documentation based on measurement of text comprehension and text comprehensibility. Information Processing and Management, 29:551–68, 1993. G. L. Lohse and P. Spiller. Electronic shopping. Communications of the ACM, 41:81–87, 1998. 276 BIBLIOGRAPHY T. W. Malone, J. Yates, and R. I. Benjamin. Electronic markets and electronic hierarchies. Communications of the ACM, 30:484–497, 1987. P. Nelson. Information and consumer behavior. Journal of Political Economy, 78:311, 1970. J. C. Nunnally. Psychometric theory. McGraw-Hill New York, 1978. M. K. Paasche-Orlow, H. A. Taylor, and F. L. Brancati. Readability standards for informed-consent forms as compared with actual readability. New England Journal of Medicine, 348:721, 2003. P. A. Pavlou and A. Dimoka. The nature and role of feedback text comments in online marketplaces: Implications for trust building, price premiums, and seller differentiation. Information Systems Research, 17:392–414, 2006. M. L. Richins. Negative word-of-mouth by dissatisfied consumers: A pilot study. Journal of Marketing, 47:68–78, 1983. G. J. Stigler. The economics of information. The Journal of Political Economy, 69:213, 1961. B. L. Zakaluk and S. J. Samuels. Readability: Its Past, Present, and Future. International Reading Association, Newark, 1988. 277 7.1. DISCUSSION 7.1.3 Ability of the participants to interact strategically The practical implication of interacting strategically essentially confirms the empirical evidence from the study of online social networks where status seeking participants seek to increase their status by accumulating as many connections as they can. In this case their strategic motive is to have better access to resources than other fellow participants, and in order to achieve that, they devise a strategy. This can depend on the importance of the nature of the social resource to each participant. If the social resource is, for example, access to the labor market (as in the case of Linkedin.com) or the internet music scene (e.g., MySpace), participants tend to act more strategically in the way that they interact with their fellow participants. Therefore, it is important for the community software mechanism to provide the ability for the participants to track the activity of others in order to compete with different ways to gain status or go higher in the hierarchy of the online community. In relation to incentivized actions, this can be either in an intrinsic mode (e.g., receiving the status of an expert in a high profile community of programmers) or by getting an extrinsic reward by answering a question (e.g., in the case of Yahoo!Answers). Furthermore, the community participants who receive more benefits, that is, those that make a greater effort in providing material to the community, would be able to give some extrinsic forms of rewards (e.g., standard payments or gift coupons) to those who would like to undertake their task (e.g., question about a specific problem). 7.1.4 Importance of the quality evaluation mechanisms Quality evaluation is another important factor when it comes to motivating activity in online communities. With the growing amount of unsolicited information posted to online fora by automated software agents (spambots), an important issue is the evaluation of the quality of the material available in a community. For example, much discussion has been tackled in the online community literature regarding the importance of moderation mechanisms. A special case of a moderation mechanism can be 284 CHAPTER 7. CONCLUSIONS AND RETROSPECT the case of user assisted moderation, where users can collaboratively filter out any non relevant or non important information, thus making the material available for the online community participants valuable. In the final chapter of part II of this dissertation we discussed the case of online reviews and whether the usefulness score that was assigned to them by other participants had anything to do with their qualitative characteristics. The main practical implication of this finding relates to the importance of a quality evaluation mechanism for an online community. The empirical findings of Chapter 6 confirm a symmetric relation between high quality reviews and good qualitative measures, thus proving that the results from the social filtering mechanism provide a evidence of the actual quality of the reviews (in terms of readiness), as measured by the readability indexes that were constructed for that purpose. 7.2 Conclusions Having provided a discussion of the implications of this dissertation, we revisit the research question formed in the introduction and we summarize the limitations and the future research in the following sections. 7.2.1 Retrospect This dissertation encompassed four empirical cases related to the study of behavioral characteristics from the perspective of motivating contributions in online communities. In connection with the research question developed in the introductory part of this thesis, our aim was to understand the nature of contributions in online communities and identify the factors that enhance and sustain them. In order to better understand these behavioral factors, we adopted a bottom up approach. First, we studied the nature of the contributions in an online setting using a controlled environment with declared extrinsic rewards (payoffs) positioned in the case of the cooperative contri285 7.2. CONCLUSIONS bution mechanism implemented by the public goods game. Having identified some of the behavioral characteristics that sustain contributions in an online community, such as communication and social interaction, we continued to study an application of the cooperative contribution mechanism in the context of Yahoo Answers. Here we analyzed how past activity (in terms of contributed effort and perceived benefits) had an influence on the realm of a purpose oriented online community such as the one in Yahoo Answers. In Yahoo Answers those that contribute effort, however, are not compensated in an extrinsic form, but receive intrinsic forms of motivation, such as reputation signals, etc. Extrinsic rewards might have an effect in that context and that was the purpose of studying such interactions in a different environment Extrinsic rewards might have an effect in that context and this was the purpose of studying such interactions in a different environment which was operated by Google called Google Answers. In Google Answers those who were contributing effort were compensated using extrinsic rewards and, in particular, the fixed price that someone was willing to pay if the question that was submitted received an acceptable (by an agent) answer. This is, in fact, a standard Principal-Agent mode of operation where the principal has a pre-declared price for a task and the agent takes on the task or not [Regner, 2004]. Furthermore, if the principal was more than satisfied with the agent’s performance, he/she was awarded a further reward in the form of a tip. We studied the factors that affected tipping in the GoogleAnswer’s platform in order to find out whether extrinsic forms of motivation are efficient for sustaining activity and increase the overall turnover of the community, both in terms of volume (in that case the volume of answers produced) and participation levels. Contributions might also be affected by the (perceived) value of the information available and the way it is expressed. This was the focus of the fourth empirical study presented in the previous chapter. This approach relates content quality with the perceived usefulness that a contribution might have in the context of an online com286 CHAPTER 7. CONCLUSIONS AND RETROSPECT munity that is formed around information goods (in the case of the previous chapter: books). Chapter 5, on the other hand, adopts a market perspective which is dictated by the context of study. Participants in GoogleAnswers (askers) do have a strong willingness to pay, due to the fact that processing of information available for retrieval requires high levels of cognitive ability. This is also related to the standard problem of query formulation from the information retrieval perspective [Aula, 2003]. 7.2.2 Revisiting the general research question After summarizing the key findings and the implications of the four empirical studies we are in a position to revisit the general research question: What are the main driving factors that affect contribution in Online Knowledge Communities? The obvious conclusion reached is that behavioral characteristics of the users are a key element affecting participation; further, the facilitation of social interactions through the community mechanism is an important element for the success in terms of the sustainability and evolution of an online community. The research presented in this dissertation has examined the nature of the motives that affect participation and, in particular, the effect of extrinsic and intrinsic rewards as an important factor that drives this participation. Another issue with the research question framed above is the case of the unit of analysis. In this thesis the unit of analysis for the research question pursued was the characteristics of the individuals since we intended to analyze interaction between individuals and not collective characteristics of an online community (which is evident in other approaches on how communities sustain and evolve) [Boccaletti et al., 2006, Hansen, 2002, Jackson, 2003]. This dictated the approach to the research question to be of behavioral nature since we were interested in the individual motivational characteristics and not in the group properties that might be formed during the formation of an online community. 287 7.2. CONCLUSIONS Part I of this thesis offered previous theoretical and empirical work concerning the issue of participation in online communities. The research summarized in this part essentially provided the ground for the four empirical studies presented in Part II of this dissertation, where each particular case was addressed in the setting of an online community. Chapter 3, however, was not a study that took place in the context of an online community; rather, it was a controlled reconstruction of the dilemma of participating or not in an online community, thus using two concrete framings (GIVE and TAKE) as a model of the participation (contribute to the community or get benefited by it). The findings suggest that users who participate more frequently in an online community tend to give more than those that have a more spontaneous participation rate. The research findings also have more context specific implications to the research question framed above. In connection with the theoretical background presented in Part I of this dissertation, we provide the related argumentation as to the contributions and the empirical findings of this thesis. 7.2.3 Summary of the findings and the implications of the empirical studies As discussed in Chapter 1 of this dissertation, we position the findings of this thesis by revisiting the framework of Snyder and Cantor [1998] in relation to the motivational factors that affect participation in online communities. Value expressiveness is positioned as a way of expressing values about other’s actions and concerns, something that was evident in the studies presented in chapters 3 and 4. On the other hand utilitarian functions were related to the study of Chapter 5 where we provided empirical evidence that participants also think strategically when they interact with other participants in order to receive the maximum benefit with the least effort. The function of a participant being socially adjusti??ve, on the other hand, relates 288 CHAPTER 7. CONCLUSIONS AND RETROSPECT Function Supporting Evidence on Research Findings Value Expressive 1,3 Utilitarian 1,3 Social Adjustive 2 Knowledge Seeking 2 Table 7.2: Motivational Factors and the supporting evidence provided by this dissertation very much to the existence of social norms, which was evident in the study of Chapter 5, where participantsˇ S tension to comply with an offline social norm also had an influence in their online social behavior. Research Finding 1: Participants on online communities do care when other participants participate or not The study presented in Chapter 4, having as a context of study the realm of Yahoo!Answers presents empirical evidence that participants do care about the contributions of other participants (and provide an answer). That is in essence a confirmation in online settings of the general literature of social preferences, where participants do feel envy about other participants receiving more benefit with less effort, and therefore are not willing to contribute. Social preferences also relate with the so called group mediation factor as the one tackled in the Collective Effort Model [Karau and Williams, 2001]. This might lead to an increase of expectations of contributions by those that have already contributed a level of effort that is higher than the average contributed effort. For example, in the study of chapter 4, the empirical evidence suggests that participants tend to decide to reciprocate by the level of previous effort made by those who ask for an answer to a posted question, that is, whether to participate in the thread or not. Although this is not observable by users in a direct way, the ability of the software 289 7.2. CONCLUSIONS to provide information (in the user’s profile page) about how much a user has benefited by the community provides the ground for a discussion whether the users’ do actually care about the actions of the other participants in that particular online community. In the time period that we undertook this analysis, the results suggested that (a) a high level of contribution resulted in a shorter time to get an answer and (b) a low level of contribution resulted in a higher time to get an answer. This confirms the argument by Kollock [1999] as to the change of the production function of an individual in an online setting, based on the perceived benefit or effort that this individual will conceive by his/her participation on the online community. Research Finding 2: A High degree of social Interaction leads to higher contributions in the online community The study presented in Chapter 3 has provided empirical evidence which suggests the existence of symmetric effects between cooperation and participation frequency in online communication activities. The more someone participates in an online social activity the more cooperative he/she becomes. This, in essence verifies findings from the social capital literature and, in particular, the literature related to quantifications of social capital with respect to social and organizational activities [van der Gaag, 2005, Quan-Haase and Wellman, 2004, Wellman et al., 2001]. This finding opens up the question of whether social norms are sustained on a highly anonymized setting such as on the internet. On such terms, although social interaction is anonymous from the perspective of personal interchange, structural relations are sustained and the hiding behind pseudonyms still provides an identification of the actions of each individual in the context of a group. This finding suggests that exploiting social interaction in the realm of an online community will have a positive effect on the activity of the community and the resulted contributions, since "normative" social influence will become an important factor in dictating increased participation.Dholakia et al. [2004] report a case as a result 290 CHAPTER 7. CONCLUSIONS AND RETROSPECT of episodes of social interactions in the context of a virtual community of consumers. Such a case of social influence relies on the provision and further support of mechanisms (such as those that are provided in the realm of Yahoo!Answers)where participants will be able to trace other participantsˇ S activities, as the first finding suggests. Research Finding 3: Content quality is an important factor for the perceived value that the community contributes to a participant This research finding supports the knowledge seeking perspective from Snyder and Cantors’ framework. In particular, the results from the study presented in 6 to some extent, the significance of the answer length variable in the case of Chapter 4) support this behavioral characteristic in the direction that the quality of content enhances participation and enhancement of contributions. This is also evident in Wikipedia where contribution of content has an effect on the participant’s reputation, as the study by Ciffolilli [2003] argues. This, in fact, can be related, to some extent, to the reputation of the individual contributors. It is expected that highly reputable members of an online community will contribute high quality content and have an extra incentive to contribute in order to maintain their status index. This is also connected with the volume of social interactions that take place in the community. A high level of social interactions and/or a high number of participants makes the importance of a status index significant. An individual then becomes self motivated to contribute high quality content to maintain this status index, which also results in a continuous cycle of participation, since the empirical evidence from chapter 6 and 3 suggests that high quality content makes the perceived value / benefit for an individual important. The findings of Chapter 6 are also particularly connected with a study by Rashid et al. [2006] where it was shown that the perceived value of information displayed had an effect on the participation by community members. These findings hold in the context of an online experience sharing community however to some extend it can be generalized that the quality of content has an effect on the personal attitudes of the 291 7.2. CONCLUSIONS users towards the online community [Curien et al., 2006]. 7.2.4 Additional contributions and discussion This thesis also contributes empirically. First, this is done by providing empirical evidence as to what social mechanisms support cooperation in the context of an online community. This can be used as a factor to consider the design of more effective online community software that will allow for the creation of sustainable and evolving online communities. A particular issue in online communities, as has been highlighted in the related literature, is the phenomenon of lurking or participation in absentia. Online communities seem to suffer from that factor, since interaction is low, regardless of the number of registered participants. Empirical evidence that this dissertation provides is that the provision of direct (user-to user) social interaction mechanisms can enhance the activity in the online community. While this seems to be obvious, from a design perspective this thesis provides empirical evidence for the support of this direction to the design of online community software. Furthermore, the ability of participants to rate the quality of the content provided in the online community is another factor that affects the participation in an online activity as it does not pose any interpersonal barrier, such as the lack of expertise. This partially enhances the level of social activity since it does not involve a direct contribution but an indirect contribution as to the improvement of the existing content status and quality. The classification of activities, as presented by Koch and Wörndl [2001] and discussed in the introductory part of this dissertation, connects empirically with the findings of this dissertation since the empirical evidence that we presented highlights the importance of the facilitation of interaction as an important element of an online community. The support for this case has been extended to the way that dyadic (and thus no hierarchical) form of interaction can be enhanced with the presence of social cooperation mechanisms (e.g., the display of an index highlighting the contributed posts as to highlight contributed effort). 292 CHAPTER 7. CONCLUSIONS AND RETROSPECT The use of experimental methods in information systems research Although the core focus of this dissertation has been to examine the relation between information exchange and behavior in the context of an online community, it also uses a broader methodological paradigm related to the use of experimental methods in information systems research. The use of such a detailed methodology also addresses the recent critiques of the relation of information systems research to the current practice, as well as the call for research in the area of information systems to focus more directly on the technological artifact [Orlikowski and Iacono, 2001, Benbasat and Zmud, 2003]. The significant potential of experimental methods in contrast to other information systems research methodologies is the prescriptive (rather than descriptive) nature of the research inquiry which provides explicit guidance to information systems developers as to what will make a technology useful (rather than focusing on whether it is useful or not). For example, the celebrated TAM model [Davis et al., 1989, Venkatesh et al., 2003], even though it identifies predictors for technology adoption in an organizational setting (e.g., utility, ease of use etc), it does not provide guidance on what technological features will make a technology easy to use and increase its usefulness. Other information systems research theories such as adaptive structuration theory [DeSanctis and Poole, 1994] or fragmented institutionalism [Lamb and Kling, 2003] are only useful to retrospectively explain why a technology was successful or not because it provides little practical guidance in either designing or managing new technological interventions. An additional important case related to this dissertation is the switch between the organizational to a user oriented setting, where the technology provided here is mainly used as a platform and not as a stand for accomplishing a specific task in an organizational setting. From this perspective, the empirical studies presented here consider the organizational affiliation as an enabler to engage in communication with other users and not as a boundary which limits their interaction with actors from the same 293 7.5. TOPICS FOR FUTURE RESEARCH network. Peer to peer networks also suffer from the problem of lurking where security concerns have to be addressed, as pointed out by Davidson [2003]. The study of lurking behavior is a research stream that is becoming more and more important in relation to online communities [Preece et al., 2004]. Possible extensions of the research presented in this thesis should also consider lurking as an important factor for the impact of the motivational factors addressed in this dissertation. 300 Bibliography P. Antoniadis, C. Courcoubetis, and R. Mason. Comparing economic incentives in peerto-peer networks. Computer Networks, 46(1):133–146, 2004. A. Aula. Query Formulation in Web Information Search. In Proceedings of IADIS International Conference WWW/Internet, volume 1, pages 403–410, 2003. I. Benbasat and R.W. Zmud. The identity crisis within the IS discipline: defining and communicating the discipline’s core properties. Management Information Systems Quarterly, pages 183–194, 2003. R. Blood. How blogging software reshapes the online community. Communications of the ACM, 47:53–55, 2004. S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D. U. Hwang. Complex networks: Structure and dynamics. Physics Reports, 424:175–308, 2006. C. M. Bonjean. Community leadership: A case study and conceptual refinement. American Journal of Sociology, 68:672, 1963. R.S. Burt. Brokerage and Closure: An Introduction to Social Capital. Oxford University Press, 2005. 301 BIBLIOGRAPHY B. Butler, L. Sproull, S. Kiesler, and R. Kraut. Community effort in online groups: Who does the work and why? Leadership at a Distance: Research in TechnologicallySupported Work, 2007. A. Cabrera and E. F. Cabrera. Knowledge-sharing dilemmas. Organization Studies, 23: 687–710, 2002. R. B. Cialdini and M. R. Trost. Social influence: Social norms, conformity, and compliance. The handbook of social psychology, 2:151˝ U92, 1998. A. Ciffolilli. Phantom authority, self–selective recruitment and retention of members in virtual communities: The case of Wikipedia. First Monday, 8(12):1396–0466, 2003. N. Curien, E. Fauchart, G. Laffond, and F. Moreau. Online consumer communities: escaping the tragedy of the digital commons. Internet and Digital Economics, Cambridge University Press, Cambridge, page 201˝ U219, 2006. A. Davidson. Peer-to-Peer File Sharing Privacy and Security. Center for Democracy and technology, 2003. F. D. Davis, R. P. Bagozzi, and P. R. Warshaw. User acceptance of computer technology: a comparison of two theoretical models. Management Science, 35:982–1003, 1989. E. L. Deci, R. Koestner, and R. M. Ryan. A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin, 125:627–668, 1999. G. DeSanctis and M.S. Poole. Capturing the complexity in advanced technology use: Adaptive structuration theory. Organization Science, pages 121–147, 1994. U.M. Dholakia, R.P. Bagozzi, and L.K. Pearo. A social influence model of consumer participation in network-and small-group-based virtual communities. International Journal of Research in Marketing, 21(3):241–263, 2004. 302 BIBLIOGRAPHY M. T. Hansen. Knowledge networks: Explaining effective knowledge sharing in multiunit companies. Organization Science, 13:232, 2002. M. H. Hsu, T. L. Ju, C. H. Yen, and C. M. Chang. Knowledge sharing behavior in virtual communities: The relationship between trust, self-efficacy, and outcome expectations. International Journal of Human-Computer Studies, 65:153–169, 2007. M. Huysman and V. Wulf. IT to support knowledge sharing in communities, towards a social capital analysis. JOURNAL OF INFORMATION TECHNOLOGY, 21(1):40, 2006. M. O. Jackson. A survey of models of network formation: Stability and efficiency. Group Formation in Economics: Networks, Clubs and Coalitions, 2003. S. J. Karau and K. D. Williams. Understanding individual motivation in groups: The collective effort model. Groups at work: Theory and research, pages 113–141, 2001. A. Kavanaugh, J.M. Carroll, M.B. Rosson, T.T. Zin, and D.D. Reese. Community Networks: Where Offline Communities Meet Online. Journal of Computer-Mediated Communication, 10(4), 2005. M. Koch and W. Wörndl. Community support and identity management. Proceedings of the seventh conference on European Conference on Computer Supported Cooperative Work, pages 319–338, 2001. Joon Koh, Young-Gul Kim, Brian Butler, and Gee-Woo Bock. Encouraging participation in virtual communities. Communications of the ACM, 50:68–73, 2007. P. Kollock. The economies of online cooperation. Routledge, London UK, 1999. P. Kollock. Social dilemmas: The anatomy of cooperation. Annual Reviews in Sociology, 24:183–214, 1998. R. Lamb and R. Kling. Reconceptualizing users as social actors in information systems research. Management Information Systems Quarterly, pages 197–236, 2003. 303 BIBLIOGRAPHY C. Lampe and P. Resnick. Slash (dot) and burn: distributed moderation in a large online conversation space. In Proceedings of the SIGCHI conference on Human factors in computing systems, pages 543–550. ACM New York, NY, USA, 2004. C.A.C. Lampe, N. Ellison, and C. Steinfield. A familiar face (book): profile elements as signals in an online social network. In Proceedings of the SIGCHI conference on Human factors in computing systems, pages 435–444. ACM Press New York, NY, USA, 2007. PV Marsden and NE Friedklin. Network studies of social influence. Sociological Methods & Research, 22:127, 1993. W.J. Orlikowski and C.S. Iacono. Research commentary: desperately seeking the" IT" in IT research-A call to theorizing the IT artifact. Information Systems Research, 12 (2):121–134, 2001. N. Poor. Mechanisms of an online public sphere: The website slashdot. Journal of Computer-Mediated Communication, 10, 2005. J. Preece, B. Nonnecke, and D. Andrews. The top five reasons for lurking: improving community experiences for everyone. Computers in Human Behavior, 20:201–223, 2004. A. Quan-Haase and B. Wellman. How does the internet affect social capital. Social Capital and Information Technology, page 113˝ U135, 2004. Al M. Rashid, Kimberly Ling, Regina D. Tassone, Paul Resnick, Robert Kraut, and John Riedl. Motivating participation by displaying the value of contribution. In Proceedings of the SIGCHI conference on Human Factors in computing systems, pages 955– 958, Montreal, Quebec, Canada, 2006. ACM. T. Regner. Why voluntary contributions? google answers! 2004. 304 BIBLIOGRAPHY M. Snyder and N. Cantor. Understanding personality and social behavior: A functionalist strategy. The handbook of social psychology, 1:635˝ U679, 1998. S.J.J. Tedjamulia, D.L. Dean, D.R. Olsen, and C.C. Albrecht. Motivating Content Contributions to Online Communities: Toward a More Comprehensive Theory. In Proceedings of the 38th Hawaii International Conference on System Sciences, volume 193, 2005. R. M. Titmuss. The gift relationship: From human blood to social policy. London and New York, 1971. M.P.J. van der Gaag. Measurement of Individual Social Capital. s. n, 2005. V. Venkatesh, M.G. Morris, G.B. Davis, and F.D. Davis. User acceptance of information technology: Toward a unified view. Management Information Systems Quarterly, pages 425–478, 2003. B. Wellman, AQ Haase, J. Witte, and K. Hampton. Does the internet increase, decrease, or supplement social capital?: Social networks, participation, and community commitment. American Behavioral Scientist, 45:436, 2001. 305