THREE ESSAYS ON NETWORK DEVELOPMENT, SOCIAL INFLUENCE, AND USER ENGAGEMENT IN ONLINE COMMUNITIES by Mina Ameri APPROVED BY SUPERVISORY COMMITTEE: Ying Xie, Chair Elisabeth Honka, Co-Chair Dmitri Kuksov Xiaolin Li Brian T. Ratchford Copyright c 2018 Mina Ameri All rights reserved Dedicated to my biggest supporter, my best friend, my love, and my husband, Mahdi. THREE ESSAYS ON NETWORK DEVELOPMENT, SOCIAL INFLUENCE, AND USER ENGAGEMENT IN ONLINE COMMUNITIES by MINA AMERI, BS, MS DISSERTATION Presented to the Faculty of The University of Texas at Dallas in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY IN MANAGEMENT SCIENCE THE UNIVERSITY OF TEXAS AT DALLAS May 2018 ACKNOWLEDGMENTS Writing this dissertation was only possible because of the help and support of my amazing advisors, Drs. Ying Xie and Elisabeth Honka. I would like to also thank my other committee members, Drs. Dmitri Kuksov, Xiaolin Li, and Brian T. Ratchford for their consistent encouragement and support. March 2018 v THREE ESSAYS ON NETWORK DEVELOPMENT, SOCIAL INFLUENCE, AND USER ENGAGEMENT IN ONLINE COMMUNITIES Mina Ameri, PhD The University of Texas at Dallas, 2018 Supervising Professors: Ying Xie, Chair Elisabeth Honka, Co-Chair I study influential factors on users’ activities in online communities in three chapters. Using unique data coming from an online anime (Japanese cartoon) platform containing data on users’ friendship networks, anime adoptions, and generated content over time, I look into consumers’ decisions in the context of new media such as online streaming. In the first chapter of this dissertation, I study the e↵ects of observational learning and word- of-mouth (volume and valence) on consumers’ product adoptions. Understanding whether these two social learning devices provide di↵erent and unique information or whether one is redundant in the presence of the other is crucial for companies’ information provision strate- gies. I di↵erentiate between the e↵ects of word-of-mouth and observational learning from friends (“personal network”) and the e↵ects of word-of-mouth and observational learning from the whole community (“community network”). The relative importance of word-of- mouth and observational learning at each network level provides guidance for companies’ platform design. My results reveal that both word-of-mouth and observational learning from both the community and personal networks have significant and positive e↵ects on individual users adoption decisions. I find that word-of-mouth valence from the community vi network is the largest adoption driver among the social learning forces I study. Lastly, I test for asymmetric observational learning from positive and negative actions and observational learning creating product awareness versus transferring unobserved quality information. In the second chapter, I quantify the e↵ects of binge-watching on consumers’ engagement with media franchises in two areas: interactive and personal engagement. The former involves auser’scontentgenerationrelatedtoafocalmediaproductandthelatterconcernsthe adoption of franchise extensions (sequels and other extensions). I find that binge-watching has a negative e↵ect on the production of user-generated content. The e↵ect of binge- watching on the adoption of franchise extensions critically depends on both the availability of franchise extensions at the time of watching the focal anime and the extension type: if it is available, binge-watching increases (decreases) the probability that a user watches the sequel (other franchise extensions). In the third chapter, I develop a structural model for the co-evolution of individuals’ friend- ship tie formations and their concurrent online activities (product adoptions and production of user-generated content) within a social network. Explicitly modeling the endogenous for- mation of the network and accounting for the interdependence between decisions in these two areas (friendship formations and online activities) provides a clean identification of peer e↵ects and of important drivers of individuals’ friendship decisions. My results reveal that, compared to a potential friend’s product adoptions and content generation activities, total number of friends and number of common friends this potential friend has with the focal individual are the most important drivers of friendship formation. Further, while having more friends does not make a person more active, having more active friends increases a user’s activity levels in terms of both product adoptions and content generation through peer e↵ects. Via counterfactuals I assess the e↵ectiveness of various seeding strategies in increasing website traffic while taking the endogenous network formation into account. vii TABLE OF CONTENTS ACKNOWLEDGMENTS ................................. v ABSTRACT ........................................ vi LISTOFFIGURES .................................... xi LISTOFTABLES..................................... xii CHAPTER 1 WORD-OF-MOUTH, OBSERVATIONAL LEARNING, AND PROD- UCT ADOPTION: EVIDENCE FROM AN ANIME PLATFORM . 1 1.1 Introduction . 1 1.2 Relevant Literature . 6 1.3 Data . 9 1.3.1 Data Collection, Cleaning, and (Re-)Construction . 13 1.3.2 Data Description . 17 1.4 Model and Estimation . 21 1.4.1 Challenges . 21 1.4.2 Model Description . 23 1.5 ResultsandDiscussion .............................. 24 1.5.1 E↵ects of Word-Of-Mouth and Observational Learning . 24 1.5.2 PositiveandNegativeObservationalLearning . 28 1.5.3 Awareness versus Learning about Unobserved Quality . 29 1.6 Limitations and Future Research . 29 1.7 Conclusion . 30 CHAPTER 2 THE EFFECTS OF BINGE-WATCHING ON MEDIA FRANCHISE ENGAGEMENT .................................... 33 2.1 Introduction . 33 2.2 Relevant Literature . 37 2.2.1 UserEngagement............................. 37 2.2.2 Binge-Watching.............................. 39 2.2.3 Online Movie Streaming . 41 2.3 Data . 42 viii 2.3.1 Data Cleaning . 44 2.3.2 Engagement . 45 2.3.3 Binge-Watching.............................. 47 2.3.4 Data Description . 48 2.4 Model and Estimation . 54 2.5 Results . 57 2.6 Robustness Checks . 61 2.7 Limitations and Future Research . 63 2.8 Conclusion . 63 CHAPTER 3 A MODEL OF NETWORK DYNAMICS: TIE FORMATION, PROD- UCT ADOPTION, AND CONTENT GENERATION . 67 3.1 Introduction . 67 3.2 Relevant Literature . 73 3.2.1 NetworkFormation............................ 74 3.2.2 Peer E↵ects . 76 3.2.3 Seeding . 77 3.3 Data . 79 3.3.1 Estimation Sample . 80 3.3.2 Data Description . 84 3.4 Model . 89 3.4.1 Tie Formation . 90 3.4.2 ProductAdoptionandContentGeneration. 92 3.4.3 IntegratingAllActions. 92 3.4.4 UtilitySpecifications . 93 3.5 Estimation . 96 3.6 Identification ................................... 100 3.7 Results . 101 3.8 Counterfactual . 107 3.8.1 How to Increase In-Site Activities through Platform-Wide Stimulation System? . 109 ix 3.8.2 How to Increase In-Site Activities through Seeding? . 109 3.9 Limitations and Future Research . 113 3.10 Conclusion . 114 APPENDIX A ORIGINAL AND FINAL DATA . 116 APPENDIX B VARIABLE (RE-)CONSTRUCTION . 119 APPENDIX C ROBUSTNESS CHECKS . 122 APPENDIXD PROBITRESULTS ........................... 123 APPENDIX E COMPLETE RESULTS FOR OTHER CONTINENTS . 125 APPENDIX F ALTERNATIVE CLASSIFICATIONS OF BINGE-WATCHING . 129 APPENDIXG LIKELIHOODDERIVATION . 131 REFERENCES....................................... 133 BIOGRAPHICALSKETCH................................ 141 CURRICULUM VITAE x LIST OF FIGURES 1.1 ExampleofanAnimePage ............................. 10 1.2 ExampleofaUserPage ............................... 10 1.3 ExampleofaUserWatchList............................ 12 1.4 DatesUsersJoinedMyAnimeList.Net. 14 1.5 Percentage of Friends Added During First 2 Years After Joining MyAnimeList.net (grouped by length of membership) . 15 1.6 Average Number of News Articles (shaded area denotes 5th and 95th percentiles) 16 1.7 Histograms of the Number of Friends and of Descriptives Related to Adoption . 18 1.8 WOMandOLfromPersonalandCommunityNetworks . 19 2.1 DatesUsersJoinedMyAnimeList.Net. 43 2.2 Number of Days After Release of First or Final Episode in a Season That Animes WereWatched..................................... 46 2.3 Watch Period Distribution (truncated at 200 days) . 48 2.4 NumberofHoursWatchedPerDay. 49 2.5 PercentageofAUser’sWatchListThatIsBinge-Watched . 50 2.6 Binge-WatchingvsNon-Binge-WatchingAcrossTime . 51 2.7 DistributionofRatingsforBingedvs. Non-BingedCases . 54 2.8 ExamplesofReleaseDatesonNetflix. 65 3.1 DatesUsersJoinedMyAnimeList.Net. 80 3.2 NumberofDaysBetweenActivities. 82 3.3 UserSamplingStrategy ............................... 83 3.4 Network Co-Evolution Over Time Lines Between Nodes Indicate Friendship Ties. Node Size Increases with More Animes Watched. Node Color Darkens with More PostsWritten...................................... 85 3.5 Average Activity Levels Over Time Since Joining (New Users) . 87 3.6 Percentage of Observations with Certain Activities Conditional on Performing at leastoneActivity................................... 88 3.7 NumberofActivitiesinEachAreaPerDay . 89 3.8 Number of Activities and Active Users Per Day Under Di↵erent Recommendation Strategies (Colors Represent Seeding in Indicated Areas of Activities.) . 110 xi LIST OF TABLES 1.1 Descriptive Statistics . 20 1.2 Results . 25 2.1 NumberofEpisodesinandDurationofaSeason . 45 2.2 ProbabilityofEngagementAction. 51 2.3 TimingofEngagementActions . 53 2.4 Weekday Frequencies of Start Dates . 57 2.5 Results-NorthAmerica ..............................