University of Southern California Dissertations and Theses

Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

(USC Thesis Other)

Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i COMPETING ACROSS AND WITHIN PLATFORMS: ANTECEDENTS AND CONSEQUENCES OF MARKET ENTRIES BY MOBILE APP DEVELOPERS YONGZHI WANG University of Southern California Marshall School of Business Department of Management and Organization 701 Exposition Blvd – Hoffman Hall 431 Los Angeles, CA 90089-1424 yongzhiw@usc.edu A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (BUSINESS ADMINISTRATION) August 8, 2017 ii Copyright of the Dissertation is held by the Author Yongzhi Wang 2017 iii To my mother (Qixian Zhang), wife (Jiaojiao Liu), and beloved son (James Alex Wang). iv ACKNOWLEDGEMENTS It has been more than ten years since the moment that I stepped into a bookstore in Xi’an to seek my real calling besides an engineering undergraduate major. My journey as a strategic management scholar has not been an easy one, as I have stumbled along the way from time to time. Looking back from this point of ending one phase in that journey and starting a new one, I feel deeply indebted and thankful to several people who have graciously helped me along the way. I share the very same feeling with Anthea (Yan) Zhang, that meeting and studying with Professor Nandini Rajagopalan is the greatest, and most fortunate thing that could happen for a strategy doctoral student. The world does not lack talent, what is often missing is a mentor who can nurture such talent. Professor Rajagopalan possesses many virtues that we cherish but often find missing in academics and the broader society—selflessness, supportive, encouraging, and the courage to advocate for diversity and inclusion. I believe that, without her generous support and faith in me, I may have dropped out in my early years of the Ph.D. program due to emotional stress from other sources, not to mention going forward with publication and achieving research awards. I think what bonds people together is the fundamental beliefs that they share. Throughout the years of learning from my advisor, I have strengthened my beliefs in those virtues of being a scholar and being an individual in the society. With that regard and with all my respect, I would like to thank my mentor Nandini. I am also thankful to my dissertation committee members who have given me tremendous intellectual support and advice. Lori Yue is a brilliant and productive scholar who has demonstrated that, through hard work and perseverance, it is possible for a new graduate to soon become a leading researcher in one’s research field. Thank you so much, Lori, for patiently guiding me on how to make my research more interesting theoretically. I have also benefited v tremendously from Brian Wu’s insights and guidance. Brian is an outstanding strategy scholar whose achievements make us proud. Finally, I owe my methodological training and understanding to Professor Cheng Hsiao, a leading scholar in panel data analysis, and Florenta Teodoridis, a very rigorous empirical scholar. I really appreciate that your teaching has allowed me to improve the methodological rigor and empirical accuracy of my research. I would also like to thank several professors for whom I have deep respect, because they appeared at critical stages of my doctoral studies and helped me realize the promise of my doctoral training. Regarding this, I would like to first thank my former advisor at Xi’an Jiaotong University, Professor Yi Liu. You graciously gave me the opportunity to study in the U.S., then generously offered your strong support in my application to USC, and kindly helped with my job market. Besides my mentor Nandini, you are another prominent scholar whom I highly respect and wish to repay the favor through my academic excellence in my future research career. Finally, without the strong support and recommendation from Professor Yadong Luo and Professor Paul Adler, it would have been impossible for me to receive doctoral training at USC. I will always be grateful for this opportunity. For nearly every Ph.D. student, who has successfully gone through the process, the journey is emotionally stressful. It is particularly true for me as I am empathetic to powerless student colleagues with minority backgrounds (as me). I am grateful that my wife, Jiaojiao Liu, has been giving me emotional support since the first day I stepped into the U.S. to pursue my dream. We have together gone through many obstacles when seeking a good education, career, and building a family ground up in a foreign land. I fully understand the career sacrifices that you have made so that we could be together, now with our beloved little one James. I hope to share with you all the achievements that we have together obtained and that we will create. Also, vi thank you for the most wonderful gift in our lives—our son James. I love you both wholeheartedly and cherish every smile of yours and every moment when Jamie calls me “babaa~”. Your happiness means the world to me. A son is always the mother’s boy. My childhood consists of scattered plots, as I moved from one place to another with my mother’s life changes. It was not an easy life, but, mom, you are always the sunshine in my life. You are optimistic, strong in will, and liberal in your son’s education and life choices. I hope Jamie could learn these traits from you and be happy. You have done your best to raise and educate me, and I hope it will soon be the time that Jiaojiao and I start to provide you the support so you could enjoy a good life. Some friends are like family members, generously providing me with emotional support. Thank you, Wayne Chang and Sally Tsai, for being Jamie’s Buddha-Father and Buddha-Mother and for the many wise conversations throughout the years. It is truly our good fortune to have met you during our life in Los Angeles. I wish both of you the best in your career and life. Pablo Mondal, you were like an older brother to me while you were in the Ph.D. program. I hope you and your family are doing well after leaving LA. I will continue “liking” the amazing Instagram photography that you and your wife take of Ari and your little princess. Thank you, Master (oh, no, Dr.) Jake Grandy, for introducing me to martial arts. I hope it won’t take me another ten years to get a doctoral degree (i.e., black belt). Thank you for a fun job-market year. You were the best colleague I could have on the job market! Finally, I am very grateful for the generous financial support my dissertation has received from the Strategy Research Foundation of the Strategic Management Society and from my academic advisor, Professor Nandini Rajagopalan. Without this ongoing support, it would have been impossible to compile the proprietary big data for this research program. I would also like vii to thank the following research assistants for their painstaking help with the data collection efforts over several years: Christopher Liu (2013), Bobi Pu (2014–2015), Dharmik Locharam (2014–2015), Indu Mohanan (2015), Aamir Goriawala (2015–2016), Shitesh Saurev (2016– 2017), Sanjana Moody (2016), and Chaitanya Shah (2017). viii COMPETING ACROSS AND WITHIN PLATFORMS: ANTECEDENTS AND CONSEQUENCES OF MARKET ENTRIES BY MOBILE APP DEVELOPERS Author: Yongzhi Wang Advisor: Nandini Rajagopalan ABSTRACT Platforms have emerged as an important form of economy in the digital revolution age. Although existing research on platform-based competition has mainly focused on platform-level strategies, little empirical research exists on the strategic behaviors of platform complementors. In this dissertation, I investigate market entries by software application developers, acting as complementors of two dominant mobile platforms (iOS and Android), across and within the competing platforms, and the performance consequences of their strategic choices. I propose that developers’ market-entry decisions and post-entry performances are determined by both market structures and developer heterogeneities. The first empirical essay in the dissertation examines how developers’ cross-platform mobility shapes platform-level competition between iOS and Android. I argue that the market structure of the two competing platforms with different strategic emphases—one quality-driven (iOS), one quantity-driven (Android)—and app developers’ performance feedback jointly explain variance in developers’ mobility patterns across platforms. Although both high performers and low achievers tend to move to the competing platform, they are motivated by different incentives. High performers move across platforms in order to pursue greater network effects and the possibilities of “winner-take-all” in their niche markets; thus, they are more likely to adopt both platforms (i.e., multihome). In contrast, low performers on one platform move to a different platform because they are competitively crowded out and therefore engage in a problemistic search. Consequently, low performers are more prone to abandon the home ix platform and switch to the other side. The empirical evidence on the performance consequences of developers’ cross-platform mobility appears to suggest that the Android platform may have benefited more from developer migration than the iOS platform. The second essay focuses on within-platform segment entries. I submit that acquisitions, such as industry mega-events, generate shifts in the market structure, and thereby shape potential entrants’ segment-entry decisions. Although acquisition intensity in a segment could signal both potential benefits and costs, in general, my findings suggest that acquisition signals encourage subsequent segment entries. Responses toward acquisition signals, however, are heterogeneous across developers with different prior experiences. While acquisition intensity is positively related to entry likelihood by developers who have more platform experiences and whose experiences are more proximate and narrower in scope, it is negatively associated with the entry probability by others with fewer platform experiences and whose experiences are more distant and broader in scope. While developers’ prior experiences lead to heterogeneous incentives in making segment-entry decisions following acquisition signals ex ante, such experiences also contribute to their capabilities to capture opportunities and address market competition ex post. My findings on post-entry performance suggest that, among the first products that are introduced during a segment’s acquisition heydays, those by developers whose experiences are more abundant, proximate, and focused are less likely to fail. In sum, my dissertation’s empirical findings on the antecedents and performance consequences of complementors’ cross-platform mobility and within-platform segment entries significantly enhance our understanding of how complementors’ strategic choices can influence the competition for a dominant technological standard among platforms and the overall evolution of a platform market. x TABLE OF CONTENTS ACKNOWLEDGEMENTS ...................................................................................................................... iv ABSTRACT .............................................................................................................................................. viii LIST OF TABLES .................................................................................................................................... xii LIST OF FIGURES ................................................................................................................................. xiii APPENDICES .......................................................................................................................................... xiv CHAPTER 1: INTRODUCTION ...............................................................................................................1 CHAPTER 2: COMPETING ACROSS PLATFORMS: ANTECEDENTS AND CONSEQUENCES OF APP DEVELOPERS’ MOBILITY ....................................................................................................11 INTRODUCTION ...................................................................................................................................11 EMPIRICAL CONTEXT: TWO DOMINANT MOBILE PLATFORMS ..............................................16 PLATFORM-BASED COMPETITION .................................................................................................19 CROSS-PLATFORM MOBILITY .........................................................................................................22 Developer Heterogeneities in Performance Feedback ........................................................................24 Market Structure as Installed-Base Differences ..................................................................................32 Interaction Between Performance Feedback and Installed-Base Difference ......................................39 PLATFORM ABANDONMENT AFTER MOBILITY ..........................................................................41 PERFORMANCE CONSEQUENCES ...................................................................................................44 Multihoming Costs ..............................................................................................................................45 Multihoming Benefits .........................................................................................................................47 Comparing Costs and Benefits of Multihoming .................................................................................50 Performance on the Target Platform ...................................................................................................51 METHODOLOGY ..................................................................................................................................54 Empirical Context ...............................................................................................................................54 Data .....................................................................................................................................................56 Matching Developers Across Platforms ..............................................................................................59 Performance-Feedback Variables .......................................................................................................67 Installed-Base Difference Variables ...................................................................................................70 Controls ...............................................................................................................................................71 I. Hazard Models Estimating Platform Mobility .................................................................................72 II. Hazard Models Estimating Platform Abandonment .......................................................................78 III. The Effect of Mobility on Home-Platform Performance ..............................................................81 IV. The Effect of Mobility on Target-Platform Performance .............................................................83 DISCUSSION AND CONCLUSION .....................................................................................................85 Theoretical Contributions ....................................................................................................................88 Practical Implications ..........................................................................................................................92 Limitations and Future Research ........................................................................................................95 Conclusion ...........................................................................................................................................96 CHAPTER 3: FOOLS RUSH IN? ENTRY INTO A PLATFORM-BASED MARKET FOLLOWING ACQUISITIONS ...........................................................................................................119 INTRODUCTION .................................................................................................................................119 SEGMENT-ENTRY DECISIONS FOLLOWING ACQUISITIONS IN A PLATFORM-BASED MARKET ..............................................................................................................................................130 Segment Entries in a Platform Market ..............................................................................................130 Ambiguities of Acquisition Signals ..................................................................................................133 HOW EXPERIENCES SHAPE RESPONSES TOWARDS ACQUISITION SIGNALS .....................138 xi General Platform Experiences ...........................................................................................................139 Proximate versus Distant Experiences ..............................................................................................144 Narrow versus Broad Experiences ....................................................................................................147 CATEGORY HETEROGENEITIES IN NETWORK EFFECTS .........................................................149 POST-ENTRY PERFORMANCE OF FIRST PRODUCTS .................................................................151 Acquisitions as Signals of Market Opportunities ..............................................................................152 Acquisitions as Signals of Market Competition ................................................................................152 Moderating Effects of Experiences ...................................................................................................153 METHODOLOGY ................................................................................................................................158 Setting and Data ................................................................................................................................158 Firm Category-level Analyses Predicting Category Entry ................................................................160 Predicting Post-Entry App Failure ....................................................................................................170 DISCUSSION AND CONCLUSION ...................................................................................................175 Theoretical Contribution ...................................................................................................................179 Limitations and Future Research ......................................................................................................183 Conclusion .........................................................................................................................................187 CHAPTER 4: DISCUSSION AND CONCLUSION ............................................................................203 Summary of Main Findings ..............................................................................................................203 Contributions .....................................................................................................................................207 Practical Implications ........................................................................................................................211 Limitations and Directions for Future Research ...............................................................................212 Concluding Remarks .........................................................................................................................214 REFERENCES .........................................................................................................................................216 xii LIST OF TABLES TABLE 1: Descriptive Statistics of iOS Sample ........................................................................................98 TABLE 2A: Main Effects of Aspirations and Installed Bases on Mobility (iOS Sample) ........................99 TABLE 2B: Interactions between Aspirations and Relative Installed Bases (iOS Sample) ....................100 TABLE 3: Predicting Abandoning Home Platform (iOS) ........................................................................101 TABLE 4: iOS Movers’ Home-Platform Performance ............................................................................102 TABLE 5: iOS Movers’ Target-Platform Performance ............................................................................103 TABLE 6: Descriptive Statistics of Android Sample ...............................................................................104 TABLE 7A: Main Effects of Aspirations and Installed Bases on Mobility (Android Sample) ...............105 TABLE 7B: Interactions between Aspirations and Relative Installed Bases (Android Sample) .............106 TABLE 8: Predicting Abandoning Home Platform (Android) .................................................................107 TABLE 9: Android Movers’ Home-Platform Performance .....................................................................108 TABLE 10: Android Movers’ Target-Platform Performance ...................................................................109 TABLE 11: Descriptive Statistics of Variables Predicting Segment Entries ...........................................188 TABLE 12A: Conditional Logit Models Predicting Category-Entry Likelihood ....................................189 TABLE 12B: Conditional Logit Models Predicting Category-Entry Likelihood (Continued) ................190 TABLE 13: Pooled Logit Models (to Generate Interaction Graphs via Simulation) ...............................191 TABLE 14: Descriptive Statistics of Variables Predicting Post-Entry App Failures ...............................192 TABLE 15: Tabulate App Failure across Acquisition Intensity and Experience Variables .....................193 TABLE 16: App-level Logit Models Estimating Likelihood of App Failure ...........................................194 xiii LIST OF FIGURES FIGURES 1a and 1b: Histograms of Average Rating of Developer App Portfolio ................................110 FIGURE 2: Installed Bases of Apps Over Time ......................................................................................111 FIGURES 3a and 3b: iOS-to-Android Mobility .....................................................................................112 FIGURES 4a and 4b: Android-to-iOS Mobility .....................................................................................112 FIGURES 5a and 5b: Relative Smartphone Shipments ..........................................................................113 FIGURES 6a and 6b: Relative Installed Base of Developers .................................................................113 FIGURES 7a and 7b: Relative Installed Base of Apps ...........................................................................113 FIGURE 8a: Flow Charts of Matching Algorithm – The URL-matching Module ..................................114 FIGURE 8b: Flow Charts of Matching Algorithm – The App Portfolio Matching Module ...................115 FIGURE 8c: Flow Charts of Matching Algorithm – Overview ...............................................................116 FIGURE 9: Number of Acquisitions in the iOS Platform Over Time .....................................................195 FIGURE 10: Developers and Acquisitions across iOS Categories ..........................................................195 FIGURE 11: Interaction between Acquisition Intensity and Platform Incumbent ..................................196 FIGURE 12: Interaction between Acquisition Intensity and Platform Age .............................................196 FIGURE 13: Interaction between Acquisition Intensity and Product-development Experience .............196 FIGURE 14: Interaction between Acquisition Intensity and Time-depreciated Experiences ..................197 FIGURE 15: Interaction between Acquisition Intensity and Category Relatedness ................................197 FIGURE 16: Interaction between Acquisition Intensity and Number of Categories ...............................197 FIGURE 17: Interaction between Acquisition Intensity and Diversification Index ................................197 xiv APPENDICES APPENDICES 1a and 1b: Interaction Effect between Performance-Aspiration and Time ....................117 APPENDICES 2a and 2b: Interaction with Relative Smartphone Shipments ........................................118 APPENDICES 3a and 3b: Interaction with Relative Installed Base of Developers ...............................118 APPENDICES 4a and 4b: Interaction with Relative Installed Base of Apps .........................................118 APPENDIX 5: Conditional Logit Models Estimating Entry (Alternative Controls) ...............................198 APPENDIX 6: Conditional Logit Models Estimating Entry (Coarsened-Matched Sample) ...................199 APPENDIX 7: Conditional Logit Models Estimating Entry (Alternative Variables) ..............................200 APPENDIX 8: App-Level Developer-Random-Effects Logit Models Predicting App Failure ...............201 APPENDIX 9: App-level OLS Models Estimating App Rating ..............................................................202 1 CHAPTER 1 INTRODUCTION In the age of digital revolution, platforms have emerged as an important form of organization and marketplace for business transactions. Platforms can be defined as (digital) technologies that connect different groups of market participants so that participants on one side not only can transact with the platform directly, but also can interact with those on other sides (Eisenmann, Parker, & Van Alstyne, 2006; Hagiu, 2014; Parker, Van Alstyne, & Choudary, 2016). Platforms, as a form of organization, have been proliferating in modern economy. Companies that range from superstars (e.g., Apple, Alphabet, and Amazon) to nimble technology startups in Silicon Valley are platform-based. Prominent examples of platforms include mobile operating systems (e.g., Apple’s iOS and Google’s Android, which bridge software application developers, consumers, and occasionally hardware manufacturers), game consoles (e.g., Sony’s PlayStation and Microsoft’s Xbox, which connects game developers and players), credit card unions (e.g., VISA and MasterCard, which links merchandise and consumers), and e-commerce websites (e.g., Amazon and Alibaba, which brings together sellers and buyers). Of particular importance for the survival and success of platforms is the quality and quantity of platform complementors, which provide products and/or services that can be combined with platform offerings. A few examples of platform complmentors include application (app) developers for mobile operating systems, game developers for game consoles, and third-party sellers for e-commerce websites. Both academics and corporate practitioners have noted the importance of complementors in determining the fate of a technological platform in a standards war. Business history is filled with cases of how certain technologies (e.g., VCRs, the 1990s Mac operating system) lost the standards battle as a result of failure to attract 2 complementors. An impressive body of work has addressed (a) how platform-level strategies, such as exclusivity of content or services (Caillaud & Jullien, 2003), openness versus proprietarity (Boudreau, 2010; Economides & Katsamakas, 2006; Eisenmann, Parker, & Van Alstyne, 2008), bundling (Chao & Derdenger, 2013), and staged strategies (Hagiu & Eisenmann, 2007) explain the type of platform that could ultimately dominate; (b) how platforms compete with one another by dealing with challenges and capturing opportunities (e.g., Caillaud & Jullien, 2003; Cennamo & Santalo, 2013; Economides & Katsamakas, 2006; Zhu & Iansiti, 2012); (c) how platforms cooperate and compete with complementors (Eisenmann, Parker, & Van Alstyne, 2011; Majumdar & Venkataraman, 1998; Venkatraman & Lee, 2004; Zhu & Liu, 2017); and (d) whether firms decide to adopt a platform as a business model (e.g., Casadesus-Masanell & Zhu, 2013). Interestingly, very few studies have focused on complementors and examined their strategic decisions (Clements & Ohashi, 2005; Corts & Lederman, 2009). My dissertation, therefore, addresses this significant gap in extant literature by investigating how complementors make market-entry decisions both across competing platforms and within the product segments of one platform. Complementors usually make sequential adoption decisions across competing platforms, a phenomenon that I refer to as cross-platform mobility. In the literature on platform-based competition, scholars have mainly focused on explaining complementors’ initial adoption decisions. For example, prior research suggests that adoption decisions can be explained by both platform-level factors, such as the installed base of consumers (Majumdar & Venkataraman, 1998), platform access (Boudreau, 2010), platform owner’s network positions (Venkatraman & Lee, 2004), and complementor attributes, such as anti-expropriation mechanisms against platform owners (Huang, Ceccagnoli, Forman, & Wu, 2013), similarities with existing platform 3 adopters (Majumdar and Venkataraman, 1998), and their own performance uncertainties (Loch & Huberman, 1999). More recently, researchers have started to examine complementors’ choices among competing platforms—that is, which platform to choose, and whether to choose multiple platforms (e.g., Bresnahan, Orsini, & Yin, 2014; Liu, 2014). These studies have treated platform choices as static snapshots that happen instantly and simultaneously rather than as sequential decisions. Hence, such studies have not examined the importance of complementors’ sequential adoption decisions as strategic flows, which may explain greater variance in platforms’ long- term competitive advantages (Diericks & Cool, 1989). Realizing this significant limitation in prior literature, I therefore choose to focus on explaining why complementors, given their initial platform choices, make decisions to subsequently adopt a competing platform and in some cases abandon their home platform. I also study performance consequences as a result of such cross- platform mobility. In essence, examining how complementors make market-entry decisions across competing platforms provides useful insights into their influence on platforms’ competitive advantages. Within a single platform, complementors also frequently make market-entry decisions, as market segments of the platform are heterogeneous. For instance, in the App Store, the mobile operating system iOS platform market, product categories are heterogeneous in the sense that some categories (e.g., social networking) have stronger network effects than others (e.g., utilities). Given such segment heterogeneities, an intriguing and overarching question is why a complementor chooses to enter certain market segments but not others. Building on the theoretical arguments of prior research (Caves & Porter, 1977; Klepper, 1997), I suggest that industry events that happen in a segment, such as acquisitions, could influence potential entrants’ entry decisions. Prior research, however, does not provide a clear answer to the question of 4 whether acquisitions in a market segment encourage or deter entries. Also missing in prior literature is an examination of how entry responses towards acquisition signals from a market segment differ across heterogeneous platform complementors. Thus, based on such a motivation, the second empirical essay in my dissertation examines within-platform segment entries following acquisition signals, as well as related performance consequences. Studying how complementors make segment-entry decisions within a single platform has important implications for the evolution of the platform market. To systematically probe these phenomena, I integrate insights from the literature on platform-based competition (e.g., Eisenmann et al., 2006; Shapiro & Varian, 1999) with dominant theoretical perspectives of strategic management—i.e., the external view of industrial organization (Caves, 1992; Caves & Porter, 1977; Porter, 1981; Porter, 1985), the capability- based view (e.g., Barney, 1991; Dierickx & Cool, 1989; Levinthal & Wu, 2010), and the behavioral theory of the firm (Cyert and March, 1967; March, 2010). Industrial organization literature has strong roots in economics, therefore stressing profit maximization as business firms’ single motivation. The essence of the industrial organization perspective is the structure-conduct-performance framework, in which theorists propose that characteristics of market structure (such as concentration ratio) will determine firm conduct (such as product strategies), which in turn, determines performance of the market in its entirety as well as the performance of individual business firms (Caves, 1992; Porter, 1981). Hence, the industrial organization perspective provides the theoretical anchor of the dissertation, in which I use market structures in a platform-based market as determinants of platform complementors’ market-entry decisions. For instance, in the first empirical essay on cross-platform mobility, the market structure is reflected by two competing platforms with different strategic emphases—a 5 quality-driven platform and a quantity-driven platform. In the second empirical essay, I study how existing and potential complementors make segment-entry decisions within a platform and exploit industry mega-events, specifically acquisitions, as sources of shifts in the market structure (Gupta, Jain, & Sawhney, 1999; Haleblian, Devers, McNamara, Carpenter, & Davison, 2009). Because the power of the industrial organization perspective is limited by the fact that it can only explain inter-industry heterogeneity in firm behaviors and performance rather than firm heterogeneities, I therefore integrate with other more internally focused theories that are better equipped to explain heterogeneities across platform complementors. The first is the behavioral theory of the firm (Cyert and Marsh, 1967), from which I draw upon two major theoretical elements to theorize on heterogeneities across platform complementors—performance-aspiration differences (e.g., Greve, 1998) and prior experiences (e.g., March, 2010). Both elements have been well developed theoretically and empirically, thus providing a robust foundation to examine platform complementors’ market-entry decisions. I use the concept of performance-aspiration difference to differentiate complementors in terms of high versus low performers and to study why complementors move across platforms. The second theoretical element that I utilize is prior experiences, wherein I also draw upon insights from the capability-based view (e.g., Helfat & Lieberman, 2002; Nelson & Winter, 1982; Winter, 2003). I argue that prior experiences not only shape a complementor’s incentive in making segment-entry decisions within a platform ex ante, but also bestow upon the complementor capabilities to deal with market competition ex post. Consistent with Porter (1981) contention, a fundamental assumption of my dissertation is that the heterogeneities of firm conduct and performance cannot be explained by either side (of the external and internal views) alone. Hence, when I focus on the role of complementor 6 heterogeneities in explaining market-entry decisions and post-entry performance, I also account for the influence of market structures. As will be discussed in the following sections, my dissertation shows that the two aspects are indeed complementary. Building on this integrative theoretical framework, I situate the dissertation in one of the largest and fastest-growing platform-based markets: mobile platforms and their associated mobile app markets, specifically Apple’s iOS (and the App Store) and Google’s Android (and the Play Store). To study mobile app developers’ market entries, I built a comprehensive proprietary database that covers nearly the entire 2015 population of app developers on both platforms. I also conducted multiple field interviews in the San Francisco Bay Area and collected qualitative data to complement the archival data in order to have a more in-depth understanding of the empirical setting and phenomena. I systematically probe antecedents and consequences of app developers’ cross-platform mobility and within-platform segment entries in two empirical essays. I will briefly describe each of these essays and their key findings before delving more deeply into specific studies. Essay 1: Cross-platform Mobility The first essay, titled “Competing Across Platforms: Antecedents and Consequences of App Developers’ Mobility”, is motivated by a fascinating topic in strategy research—the competition between dominant technologies. A particularly interesting aspect is whether and how a technology ultimately wins the battle for dominance. In this research, I study a phenomenon that has direct implications for whether the battle between two technologies eventually witnesses a winner. The battle between two dominant mobile platforms—Apple’s iOS and Google’s Android—is to a large extent influenced by their competition for platform complementors, 7 particularly for software application (app) developers. While theorists and practitioners have long realized the importance of platform complementors, we have little understanding of the choices of technology adoption and how such choices affect technological competition. Particularly important is the sequence of complementors’ platform choices. Consequently, in this research I study (a) whether and how app developers, as complementors to the two dominant mobile platforms, decide to enter a competing platform, a behavior I refer to as cross-platform mobility, and (b) the performance consequences of such mobility. Drawing insights from the platform literature and the behavioral theory of the firm, I develop a theoretical framework that predicts the effects of market structure dimensions and firm heterogeneities on a developer’s platform-mobility decision and the resulting effects on the developer’s performance on both the home platform and the target platform. The market structure is conceptualized in terms of two competing platforms with different strategic emphases—one quality-driven (iOS), and the other quantity-driven (Android), thus making it possible for two directions of mobility. My empirical study highlights several key antecedents to cross-platform mobility and the performance consequences. First, both developers whose performance is “above aspirations” on their home platform (or higher performers) and those whose performance is “below aspirations” (or lower performers) are more likely to subsequently enter the competing (target) platform. While higher performers’ mobility appears to be motivated by the benefits of stronger positive network effects, lower performers’ mobility decisions seem to be guided by a problemistic search to deal with being competitively crowded out. Relatedly, above-aspiration developers are more likely to multihome, while below-aspiration developers are more likely to switch to the target platform and eventually abandon the home platform. Third, over time, as the quantity-driven platform gains more advantages in terms of installed bases (of 8 both hardware and software), developers on the quality-driven platform initially show an increasing probability of mobility; however, after a certain point, the likelihood of moving decreases. This inverted-U effect between installed-base differences and mobility likelihood can be explained by initial market growth and legitimacy of the quantity-driven platform, and, eventually by market maturity and intensified competition. The findings also suggest that this inverted-U effect varies in intensity for higher versus lower performers. Finally, I find that platform mobility in general hurts a developer’s short-term performance on the home platform, and that the negative impact is stronger for platform switchers than for multihomers. In addition, due to their different origins, multihomers, not switchers, outperform those that have only adopted and stayed on the target platform. In sum, by systematically studying the sequence of app developers’ strategic choices— namely, subsequently entering the other platform, abandoning the home platform—as well as performance consequences, the paper has implications for how complementors shape the relative competitive advantages of competing platforms. Essay 2: Within-platform Segment Entries In the second essay, titled “Fools Rush In? Entry into a Platform-based Market Following Acquisitions”, I study how acquisitions in an app category of the iOS mobile platform influence subsequent entries by app developers, and how entrants’ first products introduced in the new category fare. In this fast growing market filled with uncertainties, developers need to discern the correct entry timing following market signals to leverage network effects and successfully build installed bases of users, an important source of competitive advantage in platform markets. This study theoretically integrates insights from the platform literature with two major perspectives in 9 strategic management—industrial organization, and the behavioral theory of the firm—and empirically examines the influence of 1,009 acquisitions on category entries of more than 280,000 iOS developers, during 2008-2015. Empirical results support key theoretical predictions. First, although acquisitions present both benefits and costs for potential entrants, I found that, on average, acquisition intensity in a product category has a positive and significant effect on entry probability. Second, experiences moderate the effect of acquisition intensity on entry probability. While acquisition intensity is positively associated with the entry likelihood by inexperienced, distant, and broad-scope developers, it is negatively associated with entry probability by experienced, proximate, and narrow-scope developers. These differential effects of experiences may reflect the underlying heterogeneity in potential entrants’ interpretations of the effects of acquisitions on post-entry costs versus benefits, thereby affecting their incentives in making entry decisions. Network effects may be the common driving force for the perceived entry costs and benefits—While potential costs are in the form of stronger ex post competition from merging firms (because of stronger network effects from an enlarged installed base of customers, post- acquisition), potential benefits are in the form of opportunities, for entrants to take advantage of network effects (e.g., “the next big thing”). The finding, that developers’ opposing reactions to acquisitions are amplified in categories with stronger network effects, lends further support to this proposition. Third, I find that, in general, first products that enter a new category during heydays of acquisitions are less likely to fail. Finally, developer experiences moderate the relation such that the likelihood of failure is even lower if the first products, launched during heydays of acquisitions, are by developers who are more experienced, proximate, and narrow- scoped. In sum, the empirical findings indicate that category-entry decisions and post-entry product performance are shaped by market structures (as determined by distributions of 10 acquisitions across product categories), the underlying driving force of network effects, and developer heterogeneities (in prior experiences). 11 CHAPTER 2 COMPETING ACROSS PLATFORMS: ANTECEDENTS AND CONSEQUENCES OF APP DEVELOPERS’ MOBILITY INTRODUCTION Every once in a while a revolutionary product comes along that changes everything.... Today, we’re introducing three revolutionary products of this class. The first one is a widescreen iPod with touch controls. The second is a revolutionary mobile phone. And the third is a breakthrough Internet communications device…. Are you getting it? These are not three separate devices, this is one device, and we are calling it the iPhone. These words were spoken by Steve Jobs, the founder of Apple, after he took the stage at the January 2007 Macworld in San Francisco. This event marked the launching of the first iPhone, which went on to change the competitive landscape of the mobile industry. As remarkable as the hardware (iPhone) is its operating system—iOS. In this product-announcement event, iOS was referred to as a direct adaptation of the Mac operating system (OS X)—an adaptation so powerful that it dwarfed existing competitors (e.g., Palm OS). However, the true power of iOS was realized when it became open to third-party application developers in July 2008. 1 Since then, the mobile platform has witnessed historic growth in the amount of complementary products (i.e., apps) and complementors (i.e., app developers). In 2006, Google was also developing its own mobile operating system, later known as Android. The Android platform differs from iOS in that it is an open system to both hardware manufacturers and software (app) developers, a characteristic that contributed to its high growth potential and eventual takeover of iOS in terms of the quantity of complementary products. The business history is particularly interesting because, according to the Steve Jobs’ biography, Google’s then-CEO, Eric Schmidt, sat on Apple’s board during the development of the iPhone. 1 The announcement of iOS being open for third-party developers was in 2007, and the availability of apps by third- party developers in the App Store started in July 2008. 12 Although Schmidt recused himself in meetings regarding iPhone and iOS, in Jobs’ mind, Android was a copy of iOS, it had “ripped off” iOS, and consequently Jobs swore to mount a “thermonuclear war” against Android by spending “every penny of Apple’s [then] $40 billion in the bank, to right this wrong.” The history vividly illustrates how intense the rivalry had become between the two dominant mobile platforms. One arena particularly salient to this competition was the battle for app developers or platform complementors. Hence, in this research, I study whether and how app developers make subsequent entry decisions to a competing platform, a behavior that I refer to as platform mobility. A multi-sided platform, such as Android or iOS, can be defined as a technology that enables different groups of participants to directly interact (Hagiu, 2014; Hagiu & Wright, 2015; Parker et al., 2016). Other examples of platforms include credit card unions (e.g., VISA and MasterCard, which bridges merchants with consumers), game consoles (e.g., Xbox and Play Station, which bridges game developers and players) (Hagiu, 2014), and e-commerce platforms (e.g., Amazon.com and Alibaba.com, which bridges sellers and buyers) (Zhu & Liu, 2017). Platform complementors provide products or services (e.g., apps) that customers can use in combination with the platform offerings (e.g., mobile devices). Hence, to a large extent, a platform’s number and quality of complementors determine its competitive advantage (Hagiu, 2014). A complementor faces two strategic choices: which platform(s) to adopt, and, conditional on that decision, whether and where to move subsequently. A few recent studies have examined complementors’ initial platform adoption decisions (e.g., Bresnahan et al., 2014; Liu, 2014), but they have assumed that such adoption decisions are made simultaneously. In this research, I focus instead on how a complementor makes sequential moves across platforms. It is important 13 to apply a dynamic view and examine complementors’ sequential entry decisions rather than to adopt a static view of platform choices, because platform competition is a dynamic process. Complementors’ mobility can be viewed as a strategic flow while a platform’s total number of complementors can be viewed as its strategic stock (Dierickx & Cool, 1989). It is ultimately the strategic flow (or complementor mobility) that determines a platform’s competitive advantage in the long run. Furthermore, mobility is not just a single decision, but also a process that could potentially consist of a sequence of strategic choices. For instance, after entering the target platform, a developer may face the decision to abandon operations on its initial (home) platform, depending on the situation and performance. In such cases, developers’ post-mobility decisions can be categorized into at least two types: those who maintain operations on both competing platforms (namely, multihomers), and those who abandon the home platform (namely, switchers). Finally, a further issue for a complementor that subsequently enters a competing platform is whether and how the mobility decision will influence the complementor’s performance on its home platform, as well as on the target platform. Taken together, I examine the following research questions in this research: (1) What are the antecedents of app developers’ platform mobility decisions? (2) What explains developers’ strategic choices to abandon the home platform and switching to a competing platform? And, (3) what are the effects of platform mobility on a developer’s performance on the home platform and the target platform? To study these questions, I integrate insights from two literatures: the literature on platforms and network economies (Eisenmann et al., 2006; Hagiu, 2014; Hagiu & Wright, 2015; Parker et al., 2016; Shapiro & Varian, 1999), and performance feedback literature based in the behavioral theory of the firm (Cyert & March, 1963). The theoretical framework developed in 14 this research can be conceptualized in two major parts. The first part focuses on external market forces that influence app developers’ mobility decisions, specifically the market structure of two competing platforms with different strategic emphases—one quality-driven (iOS), which stresses customer satisfaction in order to improve customers’ willingness to pay, and the other quantity- driven platform (Android), which seeks to leverage positive same-side network effects for faster growth. The second conceptual foundation rests on a key internal force behind the platform mobility decision, namely, the performance feedback on developers’ home platform that categorizes developers into high performers (above aspiration) versus low performers (below aspiration). The setting for the study is the mobile app market, in which app developers develop products for the two major competing platforms—Apple’s iOS and Google’s Android. In July of 2008, the marketplace of Apple’s iOS platform, the App Store, started to support third-party apps. Shortly thereafter, Google introduced its own Android Market (the name was later changed to Play Store). For some time, iOS maintained a lead in terms of the total number of apps, but Android eventually surpassed it. As of late 2015 and early 2016, the iOS platform had approximately 1.5 million apps, while Android had approximately 2.5 million. However, for a platform to be successful, both the number and quality of complementary products matter, and the iOS continues to be viewed as more profitable for its third-party app developers (Hagiu, 2014). Therefore, one key assumption of this research is that, between the two platforms, Android is considered more quantity-driven while iOS is more quality-driven. This research makes several major contributions to the literature. First, the paper addresses two questions central to the strategy field: How do firms make strategic choices? And, what are the performance consequences of such strategic choices? I map these two general 15 questions onto specific research questions in regards to platform-based markets: How do app developers make platform-mobility choices? And, how does platform mobility affect developers’ performance? I specifically examine a sequence of developers’ strategic choices: platform mobility and abandonment of the home platform. I also examine how platform mobility affects a mover’s short-term performance on the home platform, as well as the ex-post target-platform performance. Second, the paper integrates insights from the platform literature and the behavioral theory of the firm to generate novel insights. Of particular importance is that both external market structures, in terms of installed-base differences of the two competing platforms (of both hardware and software) and internal performance feedback, affect developers’ strategic choices and performance. Hence, the paper offers compelling empirical evidence to support the complementary effects of external and internal factors in explaining the contents, as well as the consequences, of major strategic decisions. Finally, the paper contributes to the platform literature in the strategic management field in important ways. My study shifts the traditional focus on platforms to complementors, as the they are essential contributors to platform-level competitiveness. While the existing-platform literature focuses mainly on how platforms shape complementors’ behaviors, my research adopts a different perspective to provide insights into how complementors can, through their movements, shape platforms’ relative competitiveness. My findings of complementors’ flows across competing platforms shed light on platform-level competition, and specifically on the question of whether quality-driven and quantity-driven platforms can coexist over time, or whether one will eventually dominate. In other words, studying complementors’ platform mobility behaviors could potentially endogenize the emergence of a dominant technology (or 16 platform). In a sense, the study addresses the interaction between micro-level firm behaviors (i.e., complmentors’ mobility) and macro-level industry structure dynamics (i.e., the relative competitive advantage of competing platforms) (Porter, 1981). EMPIRICAL CONTEXT: TWO DOMINANT MOBILE PLATFORMS The setting in which I examine my theoretical predictions is the mobile application (app) market, in which developers create products for the two dominant platforms—Apple’s iOS and Google’s Android operating systems. In July of 2008, Apple’s App Store started to support third-party app developers. Shortly thereafter, Google introduced its own Android Market (later changed to the Play Store). Although iOS maintained a lead in terms of the total number of applications for some time, Android eventually surpassed it. As of 2015, iOS had approximately 1.5 million apps, 2 and Android had over two million. However, when determining whether a platform is successful, both the quantity and quality of complementary products matter. Although Android has a greater number of apps, app developers view iOS as more profitable. 3 I distinguish between two broad types of platform strategies in terms of growing complementary products: a quality-driven strategy that emphasizes customer satisfaction in order to improve its customers’ willingness to pay, and a quantity-driven platform that seeks to leverage positive network effects to compete for the dominant position. I propose that a tradeoff 2 Apple announced the size of its App Store at that time—1.5 million—at its 2015 World Wide Developer Conference (WWDC). 3 For example, in an article reviewing iOS and Android operating systems, the author argues that, “Apple [iOS] only sells premium products, which vastly limits its growth prospects” (http://pakwired.com/deep-dive-android-vs-ios- comparison-review/, accessed on April 10, 2017). In another article comparing educational apps on iOS and Android, the author suggests that iOS has superior apps to those of Android, a greater proportion of paid apps than Android, and a higher average price in this category. (http://www.marketwired.com/press-release/first-head-head- comparison-ios-vs-android-educational-apps-shows-ios-has-superior-apps-1855115.htm, accessed on April 10, 2014). 17 exists between these two strategies (Cennamo & Santalo, 2013; Hagiu, 2011, 2014). 4 Between the two platforms, iOS appears to place greater emphasis on quality. For example, Apple’s iOS operating system has more restrictive governance rules for app developers, 5 a fact that echoes Hagiu’s (2014) proposition that “[a]t a high level, a [platform’s] choice of tighter governance rules reflects a trade-off of quantity in favor of quality” (p.77). In fact, the App Store’s very restrictive rules have led to the media questioning the company’s tight control, as evident from Walter Issacson’s description of the company’s founder: Jobs had a tougher time navigating the controversies over Apple’s desire to keep tight control over which apps could be downloaded onto the iPhone and iPad. Guarding against apps that contained viruses or violated the user’s privacy made sense; preventing apps that took users to other websites to buy subscriptions, rather than doing it through the iTunes Store, at least had a business rationale. But Jobs and his team went further: They decided to ban any app that defamed people, [or] might be politically explosive… (p.516) The tight control over the App Store is consistent with Jobs’ belief in closed systems that help ensure seamlessness and simplicity of user experiences. To make his point, he questioned the division of labor between hardware and software providers in the PC industry, and clarified his objection towards adopting similar doctrines in the post-PC era. He said during a product announcement (Isaacson, 2011): Folks are rushing into this tablet market, and they’re looking at it as the next PC, in which the hardware and the software are done by different companies. Our experience, and every bone in our body, says that is not the right approach. These are post-PC devices that need to be even more intuitive and easier to use than a PC, and where the software and the hardware and the applications need to be intertwined in an even more seamless way than they are on a PC. We think we have the right architecture not just in silicon, but in our organization, to build these kinds of products. (p.527) Consequently, when viewing the iOS operating system as a platform, we can see that it is two-sided, bridging app developers and customers, but closed to potential device manufacturers. 4 For instance, Cennamo and Santalo (2013) suggest that pursuing both strategies simultaneously can negatively affect a platform’s performance. Zhao, Ishihara, and Jennigns (2013) have also documented that when game publishers develop games following after the introduction of killer apps, they often face a tradeoff between product quality and release speed (which contributes to greater quantity at the platform level). 5 This is evident from Apple’s App Store Review Guidelines (July 2016 version) to app developers, requiring them to “avoid piling on to a category that is already saturated,” which may lead to “removal from the Developer Program.” 18 The result is an integrated hardware and operation system that has been valued by app developers, as it eases their efforts in product development. iOS, being a closed system on the hardware side, protects user experiences, while being open to third-party app developers enables the platform to tap into the creativity of thousands of developers (Garud & Kumaraswamy, 1993). The ultimate goal is to ensure product quality. In contrast, Google’s Android operating system appears to stress quantity, 6 with a platform that is more open 7 to potential app developers through less-restrictive governance rules. Android is also open to the hardware side (i.e., device manufacturers). The platform’s open source code is available for any hardware makers, which is enabled by the Android Open Source Platform (Parker et al., 2016). 8 A key advantage of a platform being open is that it can leverage a vast pool of potential complementors and stimulate market competition among them, with the hope of eventually benefitting consumers. However, the strategy of being open to both ends of hardware and software may have sacrificed the Android platform’s quality, as reflected by developers’ complaints about difficulties of developing and maintaining software applications for the platform’s fragmented devices. Because both platforms (iOS and Android) are open to third-party app developers, the classic debate of the digital age—closed versus open systems—does not apply to the group of platform complementors, which is of central interest to my research. Hence, I build on the previously elaborated distinction between a quality-driven platform (iOS) and a quantity-driven 6 Android’s quantity of apps overtook that of iOS but with lower quality, because apps published on Android are not required to pass the rigorous quality checks applied to iOS apps. (http://www.techaheadcorp.com/blog/mobile/googles-android-quantity-quality-apps.php, accessed on July 25, 2014). 7 A formal definition of open platform can be found in Parker et al. (2016), in which the authors argued that “A platform is ‘open’ to the extent that (1) no restrictions are placed on participation in its development, commercialization, or use; or (2) any restrictions—for example, requirements to conform with technical standards or pay licensing fees—are reasonable and non-discriminatory, that is, applied uniformly to all potential platform participants” (p.130). 8 For instance, “[Android Open Source Platform] is the platform used by Amazon in its Kindle Fire and by China’s Xiaomi in its mobile phones.” (Parker et al., 2016: 140) 19 platform (Android) as my empirical context. PLATFORM-BASED COMPETITION A multi-sided platform has platform participants on different sides (who are customers of the platform) and enables direct interactions between the different sides (Armstrong, 2006; Hagiu, 2014; Parker et al., 2016; Rochet & Tirole, 2006). Platform complementors “are companies that provide complementary products to [the platform’s] customers such that [the platform owner’s] products and complementors’ products are used together in the customer’s application” (Kapoor, 2013: 8). Examples of platforms include online transaction websites (e.g., eBay), credit card unions (e.g., VISA and MasterCard), game consoles (e.g., Xbox and Play Station) and computer- and mobile-operating systems (Hagiu, 2014). When platforms are in the form of technological standards, such as operating systems, they compete for dominant position (Schilling, 2009). 9 A key question examined in this literature, therefore, is how the dominant technology emerges from competing standards (e.g., Schilling, 2002). The most widely studied mechanism to explain the crowning of the dominant platform is network effects (e.g., Boudreau, 2012; Lee, Song, & Yang, 2016; Majumdar & Venkataraman, 1998; Schilling, 2002), which refer to situations in which a user’s value of affiliating with a platform depends on the number of users with whom the user interacts (Eisenmann et al., 2011). The essential point of the network effects argument is that installed bases (of customers and/or complementors) matters. Recent research has also started to explore other explanatory factors of technological competition and the emergence of a dominant technology, such as quality and 9 I acknowledge that Gawer (2014) has differentiated competing platforms and dominant designs. I agree that the issue of competing for the position of a dominant design does not apply to all types of multi-sided platforms (e.g., night clubs). However, I suggest that when platforms are in the form of technological standards, the topic of dominant design becomes relevant. 20 varieties of complementary products (Tellis, Yin, & Niraj, 2009), continuous investment in learning, and entry timing (e.g., Cennamo & Santalo, 2013; Schilling, 2002). A review of the platform competition literature suggests that an important limitation is the literature’s predominant emphasis on using platform strategies—such as platform quality (Zhu & Iansiti, 2012), exclusivity of content and services (Caillaud & Jullien, 2003), pricing, transaction fees, openness versus proprietarity (Boudreau, 2010; Economides & Katsamakas, 2006; Eisenmann et al., 2008), bundling (Chao & Derdenger, 2013), and staged strategies (Hagiu & Eisenmann, 2007)—to explain which platform(s) could ultimately dominate. Because the value-creation ability of a platform also depends on its complementary products and complementors (Cennamo & Santalo, 2013), a major source of predictive power should lie at the complementor level. The literature, however, assumes relatively passive roles for complementors; that is, their behaviors are primarily determined by platform policies. Indeed, if we focus solely on the relation between a platform and one complementary, the complementor is powerless and needs to find ways, such as intellectual property protection (e.g., Huang et al., 2013), to shield itself from potential competition from the platform, and, in particular, from the platform’s envelopment strategies (Eisenmann et al., 2011). However, this does not mean that complementors could not shape the platform’s competitiveness via their strategic choices. In some instances, a platform’s success is largely determined by a few complementors’ behaviors. 10 Indeed, most of the time, complementors, as a group, could determine platforms’ relative competitive advantages in a standard war. An important decision by complementors that could influence competing platforms’ prospects concerns platform adoption, which can be categorized 10 An example of the power of complementors to shape platform strategies is Microsoft’s attempt to add original equipment manufacturers (OEMs) of game consoles as another side of complementors to its XBox platform. However, the company faced rejections from potential complementors such as Dell, who argued that, in this particular industry, while the software (game) side is subsidized, consoles are the side that does not make money (Hagiu, 2014). Microsoft eventually decided to make game consoles in-house. 21 as (1) initial adoption, and (2) subsequent mobility. While the literature has focused mainly on initial adoption decisions (e.g., Bresnahan et al., 2014; Liu, 2014; Majumdar & Venkataraman, 1998), it has paid scarce attention to complementors’ mobility across platforms. Initial platform adoption. Prior research has examined complementors’ initial platform adoption decisions from the perspectives of platforms and complementors. From the platform perspective, it has been shown that having a large and diversified installed base of consumers (Majumdar & Venkataraman, 1998), granting complementors accesses to platforms (Boudreau, 2010), and network positions of platform owners (Venkatraman & Lee, 2004), all influence complementors’ adoption decisions. From the complementor’s perspective, firms tend to join a platform (or adopt a new technology) when there is less-competitive crowding (Boudreau, 2012), when preventive mechanisms (e.g., patents) are in place to deal with expropriations by platforms (Huang et al., 2013), when other similar firms also adopt the platform (Majumdar & Venkataraman, 1998), and when they expect positive externalities and face performance uncertainties (Loch & Huberman, 1999). Although recent research has started to study complementors’ choices among competing platforms, it has assumed that a complementor makes adoption decisions simultaneously (e.g., Bresnahan et al., 2014; Liu, 2014). This snapshot view of platform adoption decisions may have prevented scholars from realizing the importance of strategic flows in determining a platform’s long-term competitive advantage (Dierickx & Cool, 1989). Admittedly, the importance of studying complementors’ initial adoption decisions may be more relevant at the stage in which competing platforms are initially launched, because a key question during that initial period is how a platform obtains the critical mass of complementary products (and users) or a determination of the ignition strategy (Evans, 2009). However, when 22 two competing platforms have both surpassed the point of critical mass (after which network effects kick in), I argue that, to predict which platform will eventually dominate, a more important phenomenon to study is complementors’ subsequent mobility. My reason for this argument is that during this period, platforms will attempt to compete away a key source of the rival platform’s competitive advantage: its complementors and associated complementary products. While a fair amount of research exists on initial platform adoptions, there is very little research on platform participants moving across platforms 11 (e.g., Xu, Venkatesh, Tam, & Hong, 2010), and no research, to the best of my knowledge, on complementors’ mobility. Examining complementors’ mobility behaviors, to some extent, is even more important than their initial platform adoptions, because platform competition is a dynamic process shaped not only by initial adoptions, but also by the ongoing pattern of mobility. By studying complementors’ mobility across competing platforms, I conceptualize platform competition as a dynamic process. That is, a complementor’s mobility can be treated as a strategic flow, while a platform’s total number of complementors can be viewed as its stock (Dierickx & Cool, 1989). A platform’s long-term competitive advantage will be ultimately determined by the strategic flow (complementor mobility). CROSS-PLATFORM MOBILITY To deepen our understanding of complementors’ platform mobility, it is necessary to integrate the literature on platform competition with the broader literature on mobility. Although this literature is rather fragmented, the key theoretical arguments can be summarized as follows. First, 11 By assuming that multihomers adopt multiple platforms simultaneously, scholars (e.g., Bresnahan et al., 2014; Liu, 2014) have ignored adoption sequence and therefore have not focused on complementors’ mobility across platforms. 23 from an economics perspective, scholars have theorized that firms move to new markets by following others, a mechanism that is usually referred to as the herding effect. For example, Hsieh and Vermeulen (2013) have found that when a firm’s direct competitors enter a market, the focal firm tends to enter the particular market as well. Second, based on the network theory, researchers have explained mobility through the mechanism of embeddedness in a network with prior migrators (Rao, Davis, & Ward, 2000). For instance, Rao et al. (2000) studied why firms listed on NASDAQ migrate to the New York Stock Exchange and argued that social connections with prior migrators influence current firms’ moving decisions. Third, previous research has also examined mobility from the institutional theory. Specifically, studies grounded in the social categories literature suggest that firms move to other categories through various tactics including linguistic framing (Navis & Glynn, 2010) or through creating new labels (Pontikes, 2012). Institutional forces can also affect firms’ diversification into other categories (Lounsbury & Leblebici, 2004). Fourth, adopting a more technological view, scholars have suggested that complementarities across a platform’s different technological layers influence consumer migration from an older generation standard (GMS) to a newer generation (3G) (Xu et al., 2010). Finally, the literature on employee turnover suggests that employees’ mobility decisions are shaped by the local organizational environment, employees’ knowledge assets, and knowledge complementarity between the moving employee and the target firm (e.g., Campbell, Ganco, Franco, & Agarwal, 2012; Gambardella, Ganco, & Honoré, 2015). When it comes to understanding mobility decision, however, several limitations of prior literature are worth noting. First, extant literature does not address the issue of how changes in competing platforms’ composition of complementors affect their relative competitive advantages. Second, although the literature has examined only a unidirectional mobility—either moving 24 toward or away from an institution, category, or platform—both inflow and outflow of movement could exist. It is important, therefore, to study both directions of mobility, because it is the combined effects that determine the long-term prospects of a platform. In particular, it is valuable to clarify the causal mechanisms underlying different flows of migrators. Developer Heterogeneities in Performance Feedback An app developer’s decision to move from its home platform to a target platform is an important strategic decision. According to the behavioral theory of the firm, performance feedback—the extent to which a firm’s current performance deviates from its goal (or aspirations)—is a significant internal factor behind major organizational changes (Cyert & March, 1963). The concept of performance feedback, according to such a definition, therefore, consists of several important components. First and foremost is to have a clear understanding of the salient goal. According to the theory, organizations could have multiple goals. Making profit, perhaps the most obvious, is simply one of many goals that entrepreneurs pursue. As this process theory suggests, firms could shift their goals over time due to their experiences and the evolution of the market structure. Keeping this caveat in mind, I propose that the initial goal of an app developer, when joining a first (home) platform, is to have a good performance. 12 In the home-platform marketplace, performing well could mean multiple things, such as building a large customer base, starting network effects, being profitable, and attracting attention from potential customers. An important metric to gauge app developers’ performance is their ranking in the mobile platform marketplace 12 As will be subsequently shown, this initial goal changes as app developers consider and/or decide to move to the competing (target) platform. For instance, the goal for a developer that considers switching platforms may be to give up the current customer segment and try to have better performance with an alternative customer base on the target platform. 25 (e.g., Hallen, Davis, & Yin, 2017; Kapoor & Agarwal, 2017). Ranking lists are designed by platform owners and sponsors to facilitate matching between customer demands and app products. There are abundant ranking lists; some reflect revenue performance of apps (e.g., “grossing”) while others reflect market demands, such as downloads. Ultimately, however, a simple rule is that an app’s probability of being discovered by customers is seriously undermined if it is not listed on any of the top-ranking lists. Hence, I submit that a common goal for all app developers is to have their products listed on the top-ranking charts. Having established the goal-setting components of the performance feedback, the second component of the theory relates to aspirations, or “the borderline between perceived success and failure and the starting point of doubt and conflict in decision making” (Greve, 1998: 60). When the goal is defined as “performing well on the home platform,” aspirations are therefore related to a developer’s self-comparison and to similar others on the home platform. Consistent with the theory, aspirations consist of two elements: historical aspirations, which are determined by the performance history of the focal developer, and social aspirations, which are determined by “the past performance of other comparable” developers (Cyert & March, 1963: 162). The performance feedback literature can be categorized into studies that empirically differentiate between historical and social aspirations (e.g., Greve, 1998), and studies that combine the two elements for parsimony (e.g., Eggers & Kaul, 2017). Based on the observation that the two types of aspirations often exert a similar impact on organizational change outcomes (e.g., Baum, Rowley, Shipilov, & Chuang, 2005; Greve, 1998), I adopt the latter approach of aggregating the two elements and forming a unified concept of aspiration. The third component of the performance feedback concept therefore reflects a deviation from the aspiration level. Deviations could happen in both directions—above or below 26 aspirations. Above aspirations simply means that an organization’s current performance is above the anchor point of the aspiration level, while below aspirations indicates that performance is lower. The original view of the behavioral theory (Cyert & March, 1963), as well as the perspective of empirical research rooted in that theory (e.g., Eggers & Kaul, 2017; Greve, 1998; Greve, 2003), is that performance feedback is related to problemistic search—that is, when an organization notices that its performance is stymieing its original goal, it is motivated to seek alternatives for change. However, I believe that defining performance feedback in this way may constrain the power of the theory to a defensive or prevention-oriented perspective, while neglecting the more proactive alternative of a promotion-oriented approach to organizational change. Indeed, theorists have tried to modify the theory to incorporate this alternative—and more optimistic view—of organizations’ decision making. For instance, in their original work, Cyert and March (1963) intended to address this limitation by expanding the scope of performance feedback theory to include the effect of organizational slack. Their main argument, as indicated by the following quote, is that slack provides the driving force for a different type of innovation (and organizational change) rather than that spurred by problemistic search (i.e., below-aspiration performance feedback). At one level, it appears that our general theory – especially the concept of problemistic search – is of considerable relevance to the prediction of innovations…. Such a prediction is a legitimate derivation from the theory we have outlined where “innovation” means a new solution to a problem that currently faces the organization. Unfortunately, the evidence does not support the prediction for major technological changes…. It is possible that this means our theory cannot be used to predict innovations. We prefer initially, however, to modify the theory.... To do this we need to reconsider our discussion of organizational slack…. when we study the firms that have made specific significant technological improvements, we find that they were made by firms with substantial slack (and thus mostly successful firms). (Cyert and March, 1963: 188-190) Similarly, more recent theoretical work (Greve, 1998, 2003) and empirical research (e.g., Eggers & Kaul, 2017) have acknowledged this potential limitation and developed arguments on why above aspiration also leads to organizational change. For instance, Greve (1998) acknowledges 27 that his study’s empirical findings on the change-inducing effects of an above-aspiration performance could not be satisfactorily explained by the original theory that exclusively focused on problemistic search. In the limitations section of his paper, Greve (1998) wrote: As the relations of relative performance to change estimated here suggest, some stations performing well above their aspiration levels changed their formats. This cannot be completely explained by a theory of change as a response to social or historical aspiration levels, as there seems to be no reason for the probability of change not to drop to zero when the organization is performing highly. Perhaps social and historical aspiration levels are not the only goals that can be active in an organization. Up-ward-striving goals can also be activated, leading to high risk taking, even for organizations that are doing better than expected and better than their peers. The most important argument made here is that risk taking is guided by the performance relative to the goal currently active in the organization.” (p.82) Such a limitation is also acknowledged in other recent empirical research: Performance that exceeds external aspirations may trigger a different mechanism than we have discussed here, such as slack search (Baum et al., 2005), and a different mechanism may be at work when performance exceeds the internal social aspirations. Anecdotal evidence suggests that high-performing employees may take oversized risks; traders at J.P. Morgan and UBS who recently took excessive risks are one example…. Our findings do not support the idea that high relative internal social performance leads to more risk, but future research should consider this question in further detail. (Kacperczyk, Beckman, & Moliterno, 2015: 253) Consequently, I intend to advance the theory of performance feedback by probing the mechanisms behind why an above-aspiration performance could also lead to organizational change with the hope of contributing to the theory’s completeness, while maintaining its simplicity. As suggested by prior studies, performance feedback affects an organization’s motivation to change. The key to understanding why above aspiration drives an app developer’s platform- mobility decision, therefore, is to know its motivation. I argue that, in network economies such as the mobile app industry, an important motivator is the pursuit of stronger positive network effects, and the possibility of “winner-take-all.” The winner-take-all phenomenon describes situations under which a firm is able to dominant a market niche and wipe out market shares of (almost) all competitors. We have witnessed such a phenomenon frequently in high-tech 28 industries. For instance, in the U.S. search engine market, Google, an undoubtedly dominant player, has captured almost all the online search market. In the online social networking market, Facebook has beaten MySpace, capturing almost all the market share (customer base). In the online retailing sector, Amazon may be the winner that has captured a dominant market share. According to the theory of network economies (Shapiro & Varian, 1999), the winner-take-all phenomenon is driven by strong positive network effects that enable the winner to absorb most of the installed base of users, which, in turn, starts a positive feedback that makes the winner network more attractive compared with its rivals. Eventually, the rivals’ network economies become increasingly weaker and eventually disappear, which Shapiro and Varian (1999) label as the “tipping point.” I argue that winner-take-all situations are very likely to happen in the mobile app market, given that all the products in this marketplace are information-based. Information-based products rely on the ability to create and leverage a large, installed base of users for competitive advantage. In a specific market segment of the mobile app market, network effects are likely to create one (or a few) dominant players capable of occupying most of the market space. For instance, WeChat, a messaging app that was developed by the Chinese IT company Tencent, is apparently the winner in the social networking market segment of mainland China (and perhaps even the larger geographic region of East Asia). In fact, the dominant role of this social network platform has expanded to every area of people’s lives (e.g., online monetary transactions), and is the envy of Western firms. 13 However, a prerequisite to facilitating positive network effects, and thus the winner-take- all phenomenon, is to cover all possible customer segments. In the current mobile app market, customer segments are determined by the two dominant platforms—iOS and Android. 13 The Economist, August 6, 2016. 29 Customers’ adoption of mobile devices tends to be single-homing—that is, a typical customer tends to use devices of one platform. Consequently, we can view the entire customer base as being divided by dominant platforms. For an app developer, acting also as a platform complementor, adopting one platform means covering the part of the customer base that is served by the focal platform. Moreover, adopting both platforms—i.e., multihoming—means that the app developer could reach customer bases of both platforms. By moving to the competing platform, the focal app developer could therefore potentially reach customers of both platforms (at least in the short term, while the focal developer has not yet abandoned its position in either platform). Hence, platform mobility means that the mover fulfills the prerequisite of starting positive network effects that could potentially lead to a winner-take-all situation—that is, covering as large a customer base as possible. Several prominent examples of the mobile app market help illustrate this point more vividly. Airbnb is a business that has leveraged advantages of the sharing economy, allowing housing owners to share (part of) their property for short-term tenants. The company first released its app on the iOS platform, accumulated a user base, and then decided to move to the Android platform in response to its customers’ complaints that the app “was not on Android, which currently [2012] holds 48% of the entire world’s market share.” 14 Another example is Instagram, a mobile photo-sharing app. It was also initially launched on iOS and had accumulated 30 million users before subsequently joining Android for further expansion of its user base. 15 14 Source: https://techcrunch.com/2012/01/17/with-focus-on-international-expansion-airbnb-comes-to-android-and- revamps-mobile-web-offerings/, accessed on June 6, 2017. 15 The source is https://techcrunch.com/2012/04/03/instagram-android-demum/ (accessed on June 6, 2017). Incidentally, Instagram was acquired by Facebook soon after it joined Android (source: https://techcrunch.com/2012/04/09/facebook-to-acquire-instagram-for-1-billion/, accessed on June 6, 2017). Because of the acquisition, in my empirical analysis, Instagram is not treated as a separate app developer that made the platform mobility decision. A related issue is whether such acquisition deals comprise a large proportion of the sample so that they may affect the analysis and results. In the next essay of this dissertation, I systematically investigate acquisitions that have occurred in the marketplace of the iOS platform—App Store—and found 1,009 30 It is therefore natural to conclude, based on the abovementioned logic, that the motivation to have as large a customer base as possible tends to reside in app developers who are higher performers on their home platforms. Theoretically, this means that, when the current performance exceeds the aspiration level, the focal app developer is motivated by greater gains. Such gains are rooted in the desire to dominate the specific market niche of the focal developer (e.g., the mainland China market of social networking, as per WeChat). Note that this argument is different from the slack-driven search, proposed by the original behavioral theory of the firm (Cyert & March, 1963), in which the theorists attempted to address the limitation—that is, the inability to explain why above aspiration also drives organizational change—by expanding the scope of the performance feedback theory by incorporating organizational slack. The arguments of organizational slack, however, surround the availability of resources, and are therefore different from the logic of motivation. Acknowledging that moving to the competing platform, and consequently including the customer bases of both platforms, may require additional resources means that the availability of resources should not be a first-order issue. The first-order issue, instead, is whether the developer should and is capable of covering as large a customer base as possible, with the goal of becoming the dominant niche market player and enjoying the outcome of winner-take-all. Hence, the driving force behind the concept of above aspiration to explain a platform-mobility decision is also motivational (a parallel mechanism behind the logic of below-aspiration—e.g., Eggers and Kaul (2017)). The following hypothesis summarizes the previous arguments regarding the role of above aspiration in driving a platform-mobility decision. acquisitions during 2008–2015. Given that this number is very small compared with the total number of developers (and movers across platforms), I conclude that acquisitions in the mobile app market would not seriously affect my empirical analysis and findings. The more important point, based on the Instagram example, is that app developers move to the competing platform with the objective of capturing greater network effects. The fact that a dominant player, Facebook, acquired Instagram for its user base, testifies to the importance of the installed base of users in this marketplace. 31 Hypothesis 1.1a: Developers with above-aspiration performance feedback on the home platform will be more likely to move to the competing platform. Performance deviation on a certain goal dimension (e.g., top-ranked) from the aspiration level could also be negative, resulting in below aspiration. Below aspiration leads to problemistic search, an organizational process “that is stimulated by a problem…and is directed toward findings a solution to that problem” (Cyert & March, 1963). While below aspiration, similar to above-aspiration performance feedback, also leads to organizational change, the underlying mechanism or rationale motivating the change is different. As the theory suggests, one manifestation of the different effects of above versus below aspiration are the potential different outcomes of organizational change. When the change outcome in question is innovation, which is broadly defined as “a new solution to a problem that currently faces the organization,” (Cyert & March, 1963: 188) the theory predicts that different types of innovation will correspond to above versus below aspiration. While the type of innovation in response to below aspiration tends to be “justifiable in the short run and directly linked to the problem,” innovations due to above aspirations “tend to be difficult to justify in the short run and remotely related to any major organizational problem” (p.190). Consistent with the theory, I argue that the reasons driving lower performers (whose performance is lower than the aspiration level) to the competing platform are different from those that affect higher performers (that are above the aspiration level). Low-performing app developers make a moving decision because competitors crowd them out on their home platform. If the resource space, in the form of customer bases for a market niche, on the home platform can only support a certain number of (major) players, and if the growth of the resource space (i.e., customer base) is lower than the demand increases of firms occupying the resource space, then 32 competition intensifies and drives out weaker market players. Empirically, previous studies have found that when the number of platform complementors grows, weaker platform complementors are crowded out (e.g., Boudreau, 2010). Consequently, I predict that below-aspiration performance on the home platform will also drive a developer’s decision to move to the competing platform, though through a different mechanism that I will explore in a following section, which unpacks different types of platform mobility. Based on the arguments to this point, I propose the following hypothesis: Hypothesis 1.1b: Developers with below-aspiration performance feedback on the home platform will be more likely to move to the competing platform. Market Structure as Installed-Base Differences Shifting attention from firm heterogeneities to market structure, I now examine how the structure of two competing platforms with different strategic emphases provide market forces that drive platform complementors’ mobility decisions. Rooted in the theory of network effects (e.g., Eisenmann et al., 2006; Parker et al., 2016; Shapiro & Varian, 1999), I focus on one aspect that is essential for a platform-based market: installed bases of platform participants. I next develop the line of theoretical reasoning that will lead to the theoretical construct of installed-base difference, the focus of this section. Building on prior research (e.g., Zhu & Iansiti, 2012), I define, within the setting of two competing platforms (one quantity-driven, and the other quality- driven), installed-base difference to contrast the cumulative stocks of platform participants between the two competing platforms over a time period. I now unpack the definition of installed-base difference in the arguments that follow. First, for a single platform, the installed base of participants reflects the number that is active on the platform, a stock concept (as opposed to a flow concept) (Dierickx and Cool, 1989). 33 It has been proposed that an installed base of participants is a platform’s source of competitive advantage, or its resource base, to borrow the terminology of the resource-based view (e.g., Sun & Tse, 2009). The effect of an installed base of platform participants on potential participants’ platform adoption decisions, however, depends on the nature of network effects. One way to categorize network effects is whether the effect is from the same side of platform participants (same-side network effects) or from a different side of the platform (cross-side network effects) ((Eisenmann et al., 2006; Nair, Chintagunta, & Dubé, 2004; Tanriverdı ̇ & Lee, 2008; Ye, Priem, & Alshwer, 2012). Second, I describe same- and cross-side network effects in a more concrete and clear way; that is, by integrating them with the specific platform context of the current research. Consistent with prior proposals on network effects in the economics and marketing research (e.g., Tellis et al., 2009; Zhu & Iansiti, 2012), the two typical sides of a technological platform are hardware and software. For instance, in studies of gaming platforms, the hardware side consists of the game consoles, while the software side consists of game packages. In studies on the computer industry, the hardware side refers to an operating system’s devices (e.g., Windows), while the software side is the available software package for the particular operating system. I therefore take into account both the installed bases of hardware and software in my current study. On the hardware side, my theory accounts for the installed bases of mobile devices for each of the two competing platforms, while on the software side, I account for the installed bases of mobile apps and app developers of both platforms. The final point establishing the market structure using the concept of installed-base difference is quality- versus quantity-driven platforms. As established in a previous section, the mobile market is dominated by two major platforms with different strategic emphases. The 34 quality-driven platform (iOS) has the attributes of being closed and integrated. I argue that being closed and integrated are sub-components of being quality-driven, because being closed is relative and only pertains to the hardware side of the platform, while integration is ultimately for the purpose of delivering a higher quality user experience. The quantity-driven platform (Android) on the other hand, has the attributes of being open, and consequently more fragmented. The implication for the installed bases is that the quantity-driven platform will have a larger installed base of participants, and stimulate a faster growth in its installed base. In contrast, the quality-driven platform will enjoy a smaller installed base of users, and the rate of base growth will be lower. Such a comparison applies to both the hardware side and the software side, and therefore can be distilled to an abstract single term of installed base. The comparison also suggests that separately considering the two competing platforms’ respective installed bases is not necessary, as both could be integrated into a single concept of an installed-base difference. This concept could capture the comparative essence between the two platforms: (1) the difference in installed bases between quantity- versus quality-driven platforms tends to be positive over time, and (2) this positive gap tends to widen as the market evolves. I therefore leverage the concept of installed-base difference (of quantity- and quality-driven platforms) to derive the next hypothesis regarding the market structure’s force in driving app developers’ mobility decisions. Mobility from quality- to quantity-driven platform. I first examine how the installed- base difference affects the quality-driven platform app developers’ market-entry decisions to a quantity-driven platform. To make the following arguments easier to track, I reiterate that the installed-base difference (between quantity- and quality-driven platforms) increases over time— that is, the gap between the cumulative platform participants on the quantity-driven platform and 35 that of the quality-driven platform widens as the market evolves. I argue that the attractiveness of the target (quantity-driven) platform is subject to two underlying market forces: market growth and market competition. Both forces operate at varying stages of the market evolution, albeit with different strengths. While market growth as a pulling force for potential entrants dominates in the early stage, market competition becomes the major driving force as the market reaches maturity. Consequently, the relationship between the installed-base difference and the likelihood of a developer on the quality-driven platform joining the quantity-driven platform will exhibit an inverted U-shape. I now unpack this argument in greater detail. In the early stage of the market evolution (when the installed-base difference is low), the quantity-driven platform does not yet depict an obvious competitive advantage. However, because of the openness of the platform, the quantity-driven platform’s ability to attract platform participants (on both the hardware and software sides) becomes apparent. For one, from an ecological point of view (Hannan & Freeman, 1977) a larger installed-base difference on the hardware side means that the platform increases in terms of customers (using the platform), which comprise the resource space to support more complementary products. The expanding resource space, therefore, could support a larger number of organizations to claim and occupy the space, making the platform attractive to potential entrants from the other side. For another, from an institutional perspective, a greater installed-base difference also means that the marketplace of the quantity-driven platform gradually gains legitimacy (Suchman, 1995), which means that stakeholders will lend support to the further growth of the platform. In other words, increases in the installed-base difference signal to potential entrants the legitimacy of the quantity-driven platform, thereby stimulating motivation to join the platform. Finally, although 36 the force of market competition is also at play because both incumbents and entrants on the quantity-driven platform attempt to claim certain shares of the resource space, the role of market competition is secondary compared with that of market growth, as the pace of resource growth outpaces what is needed to support existing players. However, as the installed-base difference further increases in magnitude, the marketplace of the quantity-driven market will eventually move closer to saturation. Consequently, market competition will become the major driving force affecting platform movers’ adoption decisions. From an ecological view, the growth in available resources (in the form of hardware installments for additional customers) slows to the point at which it can no longer sustain the growth in the number of new organizations in the local market. The growing ex-post competition translates into a potential platform mover’s costs when it makes the decision of whether to move to the quantity-driven platform. When the installed-base difference reaches very high levels, the fear of competition reduces a potential mover’s adoption decision. Admittedly, from an institutional perspective, a high installed-base difference might solidify the legitimacy of the marketplace of the quantity-driven platform. Yet, at a later stage of the market evolution, when both quality- and quantity-driven platforms have established strong footholds, the concerns over legitimacy of either marketplace gradually dissipates. It could also be possible that, due to the fragmentation issue associated with the quantity-driven platform, an increase in the installed-base difference may erode the legitimacy of the quantity-driven platform. This de-legitimization occurs because an open platform tends to have a greater variation in product quality, and as the platform’s advantage in installed bases increases, so does quality variation. Large amounts of complementary products with high levels of quality variation could potentially hamper the quantity-driven platform’s attractiveness to customers and high-quality complementors. 37 Consequently, as the installed-base difference increases beyond a certain inflection point (and as the market structure evolves towards maturity), the attractiveness of the quantity-driven platform for players on the other side decreases. In sum, Hypothesis 1.2a: The installed-base difference between the quantity-driven platform (Android) and the quality-driven platform (iOS) will have an inverted-U effect on the likelihood of a quality-driven-platform app developer moving to the competing platform. Mobility from quantity- to quality-driven platform. The other direction of mobility concerns movers from the quantity- (Android) to the quality-driven (iOS) platform. Since the market structure variable that I use to predict mobility is the installed-base difference (between the quantity- and quality-driven platforms), the overarching argument in this section is the reverse of the argument advanced in the prior section. That is, I argue that there will be a U- shaped effect of installed-base difference on quantity-driven-platform developers’ likelihood of moving to the competing platform. However, nuances exist in theoretically describing the U- shaped effect. In the earliest period of the market evolution (when the installed-base difference takes the lowest value), both platforms are not yet well established. Therefore, the likelihood of the market structure pulling developers away from their home platform to another would be very low, given that they have made their initial platform adoption decisions and are uncertain regarding the evolution of the other side of the market. I therefore predict that, when the installed-base difference is at its lowest value, there is a low starting point for the probability that a quantity- driven-platform developer will move to the other side. As the market structure evolves and the quantity-driven platform builds competitive advantages in terms of an installed base of platform participants (on both the hardware and software side), developers that initially adopted the platform will witness market expansion that 38 provides an expanding resource space to sustain their survival and growth, as well as legitimacy for the platform. Consequently, one could expect that in the short term, the attractiveness of the competing (quality-driven) platform will slightly decrease. In other words, as the installed-base difference increases in magnitude, the probability of a quantity-driven-platform developer moving to the other platform will decrease slightly. However, as the market structure evolves towards maturity and the installed-base difference of the quantity-driven platform becomes sufficiently large, market growth weakens while market competition becomes the major driving force. Given the gradually constraining resource space, developers of the quantity-driven platform seek alternative markets for further growth (or survival). The more lucrative market on the quality-driven platform becomes increasingly attractive at this stage of market evolution. Although meeting the more restrictive standards on the quality-driven platform would still be challenging, on average we would expect to observe a larger flow of movers from the quantity-driven-platform developers as the market matures and the gap between the installed bases of the two platforms increases further. Indeed, the push force due to the intensified market competition on the quantity-driven platform, and the pull force from the more lucrative (and thus attractive) quality-driven platform, tends to accelerate cross-platform mobility to a much higher level than the starting point at the earliest period of the market evolution. Taken together, I propose that the effect of the installed-base difference (between the quantity- and quality-driven platforms) on the probability of quantity-driven-platform developers moving to the competing platform will exhibit the following pattern. Namely, a low starting point of moving probability when the installed-base difference is at its lowest value—with a probability that decreases slightly as the market evolves—but a subsequent eventual move 39 upwards that ultimately surpasses the starting value of the moving probability. Accordingly, Hypothesis 1.2b: The installed-base difference between the quantity-driven platform (Android) and the quality-driven platform (iOS) will have a U-shaped effect on the likelihood of a quantity-driven-platform app developer moving to the competing platform. Interaction Between Performance Feedback and Installed-Base Difference I expect that market structures (as reflected in the installed-base difference of the two platforms) and firm heterogeneities (in performance feedback) will jointly affect app developers’ platform- mobility decisions. Such interaction effects between firm aspirations and external environment growth has been suggested by prior theory as reflected in the following quote from Cyert and March (1963): …consider what happens when the rate of improvement in the environment is great enough so that it outruns the upward adjustment of aspirations. In a general way, this seems to be the situation that faces business firms during strong boom periods. When the environment outruns aspiration- level adjustment, the organization secures, or at least has the potential of securing, resources in excess of its demands (p. 43). As I proposed in the previous section, the installed-base difference will have an inverted U-shaped effect on quality-driven-platform developers’ mobility decisions to move to the competing platform (H2a). My summarized arguments are as follows. As the installed-base difference (of the quantity-driven platform against the quality-driven platform) increases at the initial stage, market growth and increased legitimacy of the quantity-driven platform will be the major driving forces affecting quality-driven-platform developers’ moving decisions. However, as the installed-base difference increases beyond a certain inflection point, market competition and reduced appeal of the quantity-driven platform become the major driving forces that deter quality-driven-platform developers’ moving decisions. I submit that responses toward the market structure when making platform-mobility decisions tend to differ across developers with heterogeneous performance feedback. Initially, 40 when market growth of the quantity-driven platform is the main driving force, the developers on quality-driven platforms whose performance is above aspirations tend to realize earlier potential opportunities on the competing platform. As previously argued, above-aspiration developers decide to move mainly because they are motivated by the pursuit of stronger positive network effects (and winner-take-all) in their own niche markets—the desire to cover as many customer bases across platforms as possible. In comparison, below-aspiration developers on the quality- driven platform tend to respond more slowly to the signal of market growth on the quantity- driven platform. Since their search for organizational changes (e.g., moving across platforms) is driven by problems they currently face, the search for solutions tends to be narrower in scope. For example, within the organization they may ask, “How do we fix the problem of low performance on the home (quality-driven) platform by addressing issues on the platform”? As a result, the solution of launching (or porting) products on the target (quantity-driven) platform is more distant, and may be secondary to existing home-platform solutions. This means that below- aspiration developers on the quality-driven platform tend to respond slower to “positive” market signals of growth on the target platform. As the market structure evolves towards maturity and passes the inflection point, intensified market competition on the quantity-driven platform becomes the main driving force. Early responders toward the intensified competition tend to be below-aspiration developers on the quality-driven platform. Since the decision to move to the competing platform is motivated by a problemistic search and perhaps even involuntary (due to competitive crowding), they will be sensitive toward “negative” signals from the target platforms. Lower performance on the home platform also means fewer available resources to absorb the consequences of ex-post competition. As a result, as the installed-base difference advantage of the quantity-driven 41 platform reaches a high level, the intensified market competition will deter below-aspiration developers’ decisions to join the target (quantity-driven) platform. In contrast, above-aspiration app developers will be less deterred by the “negative” signals of potential competition on the target (quantity-driven) platform, not only because avoiding problems (or competition) is not their motivation in adopting another platform, but also because their higher performance on the home platform provides them with more resources to address the ex-post competition after entry into the quantity-driven target platform. In sum, I predict that although the installed-base difference will initially lead to earlier moving decisions, delayed decreases will emerge in the likelihood of moving by high performers (on the quality-driven platform). The theory, thus, suggests that the bell curve of higher performers will be flatter and envelope the curve of lower performers from above. Accordingly, Hypothesis 1.3: Performance-aspiration difference will moderate the inverted-U effect in H2a, such that for higher performers the inverted-U curve (1) will be flatter in slopes, and (2) will envelope the curve of lower performers from above. PLATFORM ABANDONMENT AFTER MOBILITY Platforms seek exclusive access to essential assets. They do this, in part, by developing rules, practices, and protocols that discourage multihoming…. Multihoming occurs when users engage in similar types of interactions on more than one platform…. Platform businesses seek to discourage multihoming, since it facilitates switching—when a user abandons one platform in favor of another. Limiting multihoming is a cardinal competitive tactic for platforms. (Parker et al., 2016: 213) As succinctly noted by Parker et al. (2016), moving across platforms, and thereby multihoming, is the first step in a developer’s strategic decisions that may lead to subsequent switching to the target platform and the abandonment of the home platform. By “standing in the shoes” of an app developer, we could imagine, from a behavioral perspective, the decision-making process that it follows (Cyert and March, 1963). When performance is above (or below) the aspiration level, the developer decides to move to the competing platform. Once the developer learns about 42 environments on both platforms, it faces the decision of whether to abandon one. The importance of platform complementors’ abandonment decisions rests on the potential effects of platforms’ competitive advantages, and the effects could go either way. Complementors’ abandonment decisions could potentially hurt a platform’s competitive advantage mainly because the amount of available complementary assets (Kapoor & Furr, 2014; Teece, 1986) in a platform’s ecosystem shrink, thus making the platform less appealing for potential platform participants, including customers. This is perhaps the most widely accepted wisdom. An opposing logic, however, is that if abandonment is concentrated on complementors that the platform deems less valuable or even as having negative effects on the ecosystem’s value creation, then complementors’ abandonment decisions may not be as detrimental, and at times, perhaps may even be beneficial. 16 The quality-driven platform is more likely to embrace this latter logic, while the quantity-driven platform may favor the former. It is, therefore, essential to analyze which type(s) of platform complementors are more likely to make abandonment decisions. Consistent with the main theme of the study, I propose that, among movers to the competing platform, it is the below-aspiration developers who are more likely to abandon the home platform earlier. Platform mobility for below-aspiration developers is a type of change that they seek as a result of a problemistic search. Although they may share with above-aspiration developers the same initial goal of performing well on the home platform, over time, because of performance feedback, their goal shifts toward avoiding competition on the home platform and the possibilities of better performance on the competing platform. Consequently, the initial 16 Hagiu (2014) gives an example of how a platform’s quantity-driven strategy, without the effort of filtering out lower quality complementors, leads to the collapse of the market. He suggests that, in the 1980s video game market, “opportunistic developers, wanting to take advantage of popularity of Atari’s console to make quick profits, flooded the market with poor-quality games. This, combined with a lack of information about game quality, led to a collapse of game and console prices” (Hagiu, 2014: 76). 43 mobility decisions by below-aspiration developers are more likely to result in their subsequent switching to the target platform. Below-aspiration movers’ choices of abandoning their home platform are rational, based on the logic of available resources and optimizing resource allocation. Being low performers on the home platforms suggests that they have limited resources. At some point, below-aspiration developers will realize that their limited resources cannot support the goal of performing well in both markets, because multihoming is very costly in the mobile app market. 17 Below-aspiration developers thus withdraw from the home platform so they can channel their limited resources to support operations on the target platform, and to meet the adjusted new goal of performing well only in the new environment (i.e., the target platform). In contrast, above-aspiration developers, due to their superior revenue returns from the home platform, are capable of sustaining operations on both platforms. That is, they are more likely to become multihomers. As described in a prior section, above-aspiration developers make platform-mobility decisions because they are motivated by covering as large a customer base as possible, and the subsequent possibility of winner-take-all. If their original goal (when making the initial platform-adoption decision) is to have good performance on the home platform (e.g., high ranking), their goal shifts when moving across platforms and toward becoming the dominant players in their market niche on both platforms. With this objective in mind, above- aspiration developers, after making platform-mobility decisions, tend to maintain businesses on both platforms (i.e., multihoming) and are less likely to abandon the home platform. Their multihoming decisions are also rational, based on the logic of available resources and optimizing resource allocation: that is, higher performers have more resources to cover multihoming costs. These arguments, taken together, lead to the following hypotheses. 17 I will develop this discussion in detail in a following section on post-mobility performance. 44 Hypothesis 1.4a: Conditional on multihoming, developers with above aspirations will be less likely to abandon the home platform. Hypothesis 1.4b: Conditional on multihoming, developers with below-aspirations will be more likely to abandon the home platform. PERFORMANCE CONSEQUENCES Mobility behavior essentially reflects a change in a complementor’s strategic position. I therefore expect that mobility will influence a complementor’s ex-post performance. After entering the competing platform, a complementor operates in two types of focused competitive environments—quality and quantity. The quality-driven and quantity-driven strategies at the platform (or environment) level tend to influence complementor-level strategic orientations. This implies that, after mobility, a complementor may span both the differentiation strategy (to meet the needs of a quality-driven platform) and the lower-cost strategy (to be successful in a quantity-driven environment). According to the traditional industrial organization literature, the two strategies are inherently incompatible at the productivity frontier. Companies therefore should make a trade-off. Those that follow both strategies will be stuck-in-the-middle (Porter, 1985), a situation wherein their competitive advantages are weakened by the competition from both low-cost pursuers or differentiators. The classic theory, consequently, implies a negative effect of mobility on ex-post performance. The core insight of the theory resides in its cost-benefit analysis when firms market further market-entry decisions (moving to the competing platform is one such type of decision). For instance, in a traditional industry setting where businesses have clear value-chain activities, the suggested benefits of entering an additional market are shared (value-chain) activities, and costs are in the form of coordination difficulties across different occupied markets (Porter, 1985). In a platform-based market, however, there is no clear value chain; instead there are different 45 groups of platform participants (and complementors) who could directly transact with the platform or with each other (Hagiu, 2014; Parker et al., 2016). 18 Thus, a cost-benefit analysis of entering a new platform market needs to take these differences into account. I therefore next integrate insights of the platform literature to describe costs and benefits of multihoming. I then draw conclusions regarding the effect of moving to the competing platform (thereby multihoming) on a developer’s performance on both the home and target platforms. Multihoming Costs “Multihoming takes place when users participate on more than one platform” (Parker et al., 2016: 224), and it has costs. From the industrial organization (IO) perspective that focuses on market structures, costs are created due to different business environments across competing platforms. The different market structures of one dominant quality-driven platform (iOS) and the other quantity-driven platform (Android) present multiple sources of costs for multihomers. First, the different platforms have distinct product-development environments. That is, they speak different (computer) languages and translation can be expensive. The quality-driven platform tends to be more closed with the intention of creating more coherent user experiences. Thus, the language for such a closed system is designed in a way to leverage the greatest potential of functionalities that the closed system could provide, from silicon (i.e., microprocess) to device interface. A counter example to this is Adobe Flash, a multimedia platform based on which developers could write programs that are compatible across multiple operating systems. Therefore, as a contrast, a quantity-driven platform’s language is designed in a way to facilitate 18 For instance, Hagiu (2014) differentiated multisided platforms with resellers, in which resellers deal with suppliers and then sell products to end buyers. A platform differs from a reseller because in the market of a reseller, suppliers do not directly transact with end buyers; in a platform market, different sides interact with each other directly. 46 compatibility. Compatibility ensures quantities of adopters, but it reduces the possible functionalities to “the minimum common denominator,” using Steve Jobs’ words. Consequently, for a multihomer it would be challenging to master both languages and to optimize product performance on both platforms. One way to address this challenge is to lower product quality on the less-demanding (quantity-driven) platform. As one of my field interviewees suggested, the company’s (game) apps on the iOS platform usually have a full set of features, while the same apps have only one- fourth to one-third of such features on the Android platform. Another way to address the heterogeneity in product-development environment is to adopt technologies to mitigate the difficulty of “translation.” Tools that function as mediators among different platforms (such as Unity, a cross-platform engine) could ease platform complementors’ product development across platforms, but at a cost. The cost is not only monetary, in that multihomers need to pay to use such cross-platform engines; it is also technical, in that learning and using such engines require tremendous resources and time. For instance, my field interviewees suggested that, before adopting Unity, it took the company approximately three to five months to release a new game app, but after using the engine, the company’s product-development cycle slowed down to eight months per app. In addition, 80% of the product-development personnel were allocated to develop new games using the cross-platform engine, while only 20% were left to maintain the old products. Even swapping existing products on the home platform to the target platform could be costly, because an iteration process is involved that requires engineers to conduct many tests across different devices (in particular, for the fragmented quantity-driven platform) and generate reports to debug. Such evidence suggests that multihoming costs can be high, even with cross- platform engines. 47 According to the behavioral theory of the firm, which focuses on within-firm decision- making processes, a multihomer needs to make tough resource-allocation decisions, which can be costly. Both quality- and quantity-driven platforms can be resource-consuming, although in different ways. The quality-driven platform consumes a developer’s resources because it has higher standards of product quality and thus needs more resources to develop a product that can meet the minimum bar. In addition, such a platform takes more time to release and update products. Since the quality-driven platform also enables more available functionalities for a developer, another source of costs is maintaining the features of existing products. In contrast, the quantity-driven platform exerts huge costs for a developer mainly due to the platform’s fragmentation issue. Fragmentation on the hardware side means that developers need to adjust their programs to cater to different device specifications (such as screen size, whether a device has one or multiple memory chips, and so forth.) The fragmentation challenge is further compounded for developers when we also account for the software size—that is, due to generations of devices, multiple versions of (software) operating system coexist. For resource- constrained small app developers, allocation becomes a very serious issue, to such a degree that small companies may choose to service only part of the market—for instance, several major hardware devices. Multihoming Benefits Developers capable of maintaining operations on both platforms could enjoy potential rewards that the platform-based market could offer. The first benefit is a larger installed base of users with essentially similar tastes—even across platforms. This claim consists of two components. First, in the same market niche, customer tastes are the same across platforms (Bresnahan et al., 48 2014). Thus, across platforms within a market niche, such customer tastes ensure the functioning of the next point regarding network effects. By moving to the competing platform and operating in the developer’s existing market niche, a multihomer could have a larger installed base of users, which, in turn, enhances positive network effects. If customer tastes differ across platforms even within the same market niche, positive network effects will then be weaker. For instance, imagine the situation when different customers of the social networking app on iOS and Android have different emphases in functionalities of the app. If the iOS group focuses on picture sharing and image quality, and the Android group focuses on instant social interactions, they are less likely to form a common pool of users to encourage further adoptions of the app (i.e., they are attracting different new adopters and thus do not have positive network effects across platforms). Thus, the assumption that within a market niche customer tastes are the same even across platforms is essential. In addition, a multihomer who could successfully operate on both platforms enjoys the benefit of a larger customer base with same tastes. The second major benefit of multihoming is a better learning experience for the multihomer. Learning could be from the general platform environment or from direct contacts with customers. The quality- and quantity-driven platforms create opportunities for developers to improve their adaptation skills through different operating environments, and also enable developers to transfer best practices from one platform market to the other. For instance, according to one of my field interviewees, the frequency of refreshing ranking lists is different on iOS and Android. Ranking is also directly linked to a developer’s ability in user acquisition, with a greater likelihood of a ranking increase if a developer has a period of intense increases in the number of users. In addition, there are different approaches to enabling user acquisitions. One way is to directly purchase a certain amount of users (e.g., 1,000) with a fixed payment 49 amount, an approach the industry refers to as “charting.” Another way is through advertising. While the first user-acquisition method (charting) may provide the developer with observable and significant increases in user numbers, the product-customer fit may be low because the method targets users broadly. Instead, the second approach (advertising) may be more target- oriented (in that only customers who find a fit will adopt the product), but at the same time promises slower growth in the number of users. A developer that has learned how to use different user-acquisition tactics on one platform (with the intention of increasing ranking), could potentially leverage the knowledge learned, adapt to the (ranking) rules of the target platform, and apply the tactics accordingly. In other words, multihomers could apply knowledge that has worked on their home platform to enjoy returns from a larger installed base of customers, while at the same time pacing their continuous innovation activities (Brown & Eisenhardt, 1997). Learning could also occur as a result of direct contact with customers on different platform markets. Compared with traditional business environments (e.g., shoe manufacturing, as in Parker et al., 2016), one advantage of platform-based markets is the ability to obtain immediate customer feedback. If we view information products in a platform market as problems that are constantly seeking improvements and solutions, such solutions usually reside in customers (i.e., who knows better about user experiences than users themselves?). For instance, app developers obtain user feedback in the form of product reviews, not only reflected in customers’ one-to-five star ratings, but also in their comments on the app. From a search perspective of problems looking for solutions (Levinthal, 1997), a multihomer, compared with a single-homer, covers a larger landscape, conducts more exploratory searches, and is therefore more likely to end up at a higher performance point (if not the global optimal point). Hence, learning from a larger customer base through multihoming could benefit the developer in terms 50 of improved product performance. Comparing Costs and Benefits of Multihoming Given the aforementioned multihoming costs and benefits, it is interesting to examine whether, in general, costs overweigh benefits, and under what conditions the relative weight between the two vary. To probe this question, my first proposition is that, when the market structure is as clearly distinguished between quality-versus quantity-driven platforms, then mulithoming costs will outweigh the benefits. This is because the differences between the two platforms are so large that mulihomers must spend substantial resources on meeting the requirements. If satisfying high-quality standards and maintaining comprehensive features for the quality-driven platform requires one type of skill set, then making sure that minimum compatible features work across fragmented hardware devices on the quantity-driven platform may require a different set of skills. When multihomers are caught in the middle, they may have fewer benefits that a larger customer base could provide. Although I expect that the greater weight of mulithoming costs could potentially be reduced when the two platforms converge over time—that is, the quality-driven platform gradually becomes less restrictive and the quantity-driven platform starts to improve products’ minimum standards—in the short-term, costs will dominate benefits. This will be reflected in the mulithomer’s performance on its home platform. In other words, I expect that, because of moving to the competing platform, a multihomer’s home-platform performance will decrease in the short term. However, I also expect that the decrease in performance due to multihoming will differ for different types of movers. I therefore build on my previous distinction between muilthomers (who maintain operations on both platforms) and switchers (who eventually abandon the home 51 platform) and propose that decreases in short-term performance 19 will be weaker for multihomers (than for switchers). Tracing the logic back to the driving force of multihoming versus switching, we know that above-aspiration developers are more likely to multihome (because they are after the possibility of winner-take-all) while below-aspiration developers are more likely to switch (because they are competitively crowded out). Consequently, I expect that multihomers, who want the advantages of a larger installed base of customers, will invest in, and therefore benefit, from an expanded customer base across platforms. Switchers, in contrast, who want to avoid problems on the home platform, are less likely to reap the benefits granted by a larger customer base. In other words, the short-term, performance-reducing effects of moving to another platform will be stronger for switchers than for multihomers. In summary, the abovementioned arguments support the following hypotheses: Hypothesis 1.5a: Platform mobility will have a negative effect on the developer’s home- platform performance. Hypothesis 1.5b: The negative effect of mobility on the home-platform performance (H6a) will be stronger for switchers than for multihomers. Performance on the Target Platform As discussed in prior sections, mulithoming versus switching developers is driven by various incentives and correspond to different group of platform movers. While multihoming is more likely to occur among developers whose performance on their home platforms is above- aspiration, switching is more likely to occur among developers who performed below the aspiration level on their home platforms. As a result, when examining the effect of platform mobility on a mover’s performance on the target platform, I also address different performance consequences of mover types—that is, multihomers and switchers. 19 Note that short-term means within the time range during which switchers have not yet made the decisions to abandon the home platform, for instance, six months in the mobile app market. 52 The first step in drawing performance implications is to identify a meaningful comparison group (i.e., the control group). I choose to use “stayers” on the target platform as the comparison group for several reasons. First, comparing with stayers, rather than movers (including both multihomers and switchers) of the target platform, has the advantage of avoiding confounding effects due to the movers’ platform-mobility behaviors on the target platform. That is, when intending to capture one direction (e.g., iOS to Android) of mobility’s effect on movers’ performance on the target platform (e.g., Android), I want to avoid the confounding effects of target-platform movers’ changes in performance as a result of their mobility in the reverse direction (in this case, Android to iOS). In other words, stayers’ performance will be “clean” in the sense that it is not confounded by the effects from platform mobility, which is also a part of my theory. Second, the theoretical angle from which I choose to unpack firm heterogeneities is performance feedback, or the difference between a developer’s current performance and the aspiration level. Stayers on the target platform fit the theoretical framework in the sense that they tend to be a group that is neutral—i.e., neither above nor below the aspiration level. Thus, with this relative neutral group, I am able to compare different types of movers’ (multihomers versus switchers) performance on the target platform. With the comparison group (stayers) on the target platform established, I now examine the different groups of movers’ performance relative to this comparison group. My following theoretical arguments revolve around two (tightly linked) issues: resource allocation and available resources. On the one hand, I argue that, in terms of resource allocation, switchers should shoulder less than multihomers, because switchers (at a certain point) sacrificed all of their home platform business and dedicated its resources to target platform operations. Such a move enabled switchers to alleviate the significant multihoming costs, as described in the 53 previous section, because, after switching to the target platform, this group of movers can release resources needed to maintain operations on the home platform. Examples of such maintenance costs could be a team of engineers involved in live operations and business intelligence to address customer issues. Such costs could also be in the form of resources needed to upgrade an app to the next version. Admittedly, switchers also forgo the customer base and corresponding revenue source from their home platforms, leading to the next point. On the other hand, I submit that, between multihomers and movers, the former type of movers, on average, tend to have more available resources. One reason for these additional resources is because such movers tend to be above-aspiration developers on their home platforms. For these developers, above aspiration means that their superior performance provides extra resources (e.g., revenue, brand) for further allocation. In the mobile app market, above aspiration suggests that a developer has an above-average ranking that directly correlates with revenue and app downloads (Carare, 2012). Moreover, the correlation tends to follow an exponential function—that is, when moving from second to first rank, the amount of positive returns (e.g., revenue) will be much greater than if moving from rank 100 to 99. Therefore, I expect that multihomers, due to their better origins (i.e., above- aspiration), tend to have a greater amount of available resources to overcome the burden of multihoming costs. In contrast, switchers are more likely to be from humble origins (i.e., below aspirations), and consequently will be more resource constrained, a reason for forgoing operations on their home platforms (because their available resources from performance feedback cannot cover multihoming costs). Because of switching to the target platform, they have also paid the platform-switching costs during the process (e.g., learning the new computer language of the target platform, as well as the local environment). Hence, when comparing multihomers and switchers with the relative mutual group—stayers on the target platform—I 54 propose the following hypotheses: Hypothesis 1.6a: Compared to stayers on the target platform, multihomers will have better performance. Hypothesis 1.6b: Compared to stayers on the target platform, switchers will have worse performance. METHODOLOGY Empirical Context My empirical context includes the two dominant mobile platforms as mentioned: the App Store of Apple’s iOS platform, and the Play Store of Google’s Android platform. Although the mobile phone industry has existed for quite a long time, the two dominant firms’ entries into the industry have significantly reshaped the competitive landscape. In 2007, when launching Apple’s first mobile device, the iPhone, Steve Jobs maintained his belief in closed systems and only allowed applications that were developed by Apple to appear on it. The developer community voiced their complaints over such restrictions and found ways to bypass Apple’s technologies, “jailbreaking” the iPhone to use other applications (Miric, 2016). In the following year, Apple provided a Software Development Kit (SDK) for third-party developers to release their products on the iOS platform, and thus officially opened the App Store in July 2008 with the first set of such apps. Since then, the App Store has grown significantly, as acknowledged by Apple’s senior vice president, Phil Schiller, who said, “as that accumulative effect appeared, then we all started to realize: oh my goodness, this is bigger than any of us imagined.” 20 Soon after the opening of the App Store to third-party developers, Google’s Android platform, which started as an open platform (to both software developers and hardware manufacturers), also allowed third-party developers to release products on the platform in early 20 Source: 2015 Apple WWDC. 55 2009. The platform’s openness on both the hardware and software ends, as well as its less restrictive rules for third-party app developers, has enabled Android to grow much faster in its installed base of apps and developers. 21 Since launching, both platforms have been competing head-to-head while dominating almost the entire mobile phone market share. According to Statista, an online statistics portal, by 2016, Android and iOS had almost completed occupied the smartphone operating systems market, with 84.8% and 14.4% shares, respectively, based on unit sales. 22 The intense rivalry between the two platforms is reflected in a story behind the two platforms’ creators, Apple’s founder Steve Jobs, and Google’s previous CEO, Eric Schmidt. The latter was on Apple’s board of directors when the iPhone and the iOS operating system were still in development. Although Schmidt recused himself in meetings when Apple’s board discussed issues related to the development of the iPhone, Jobs still believed that Android was a copy of iOS. In Jobs’ own words, Android had “ripped off” iOS and he was going to mount a “thermonuclear war” against Android “with every penny of Apple’s [then] $40 billion in the bank.” The bitter rivalry between the two companies continues to this day. The two platforms differ profoundly in many ways. In this research, I propose a key distinction is that iOS is more quality-driven while Android more quantity-driven. To support this point, in addition to the reasoning described in the theory section, I further illustrate it with empirical data. In Figures 1A and 1B in which histograms of average developer product ratings are provided, we see that the distribution of iOS developers lean more toward high ratings than Android developers. This empirical evidence supports the assumption that iOS is more quality 21 Many markets exist for apps of Android devices, including those for modified versions of Android by different device manufacturers. The scope of my research focuses only on the official marketplace of the Android platform: Google’s Play Store. 22 Source: https://www.statista.com/chart/4112/smartphone-platform-market-share/, accessed on May 10, 2017. 56 driven. Further, Figure 2 presents a cumulative number of apps available on each of the two platforms. As can be seen, Android has grown substantially faster and has accumulated a greater number of apps, thus providing evidence to support the assumption that Android is more quantity-driven. ----------------Figures 1A, 1B, and 2 about here---------------- Data The major database used to analyze app developers’ mobility across platforms includes (nearly) the population of iOS developers and apps in late 2015, and (close to) the population of Android developers and apps in early 2016. The iOS developer and apps data were compiled from laborious web scraping from multiple sources (e.g., the App Store, Bumblebee—a mobile analytics company that granted me the right to obtain and use its data, and Optimus—another analytics company that specializes in Android app data and granted me proprietary access to their data), as well as a purchased proprietary data set from Megatron, a data company, resulting in a comprehensive 1.44 million apps. 23 This is comparable to the number (1.5 million) of apps announced by Apple at its 2015 Worldwide Developers Conference. The Android app data were also compiled from web scraping and from the aforementioned data aggregator and analytics company in the Bay area (Optimus), resulting in a total of approximately 2.5 million apps as of early 2016. In particular, a part of my Android data is proprietary and thus off-the-shelf, because I worked with several Optimus data engineers to export release-date information of the earliest version of all Android apps from the gigantic Android database they have accumulated over the years. Given that Optimus is a pioneer in comprehensively compiling Android data in this 23 Due to formal or informal non-disclosure agreements with the data companies, the company names are disguised with well-known cartoon characters. 57 industry, I used the date of the earliest Android app version as the app release date. 24 I also collected platform-specific information including: (1) hardware data of the two platforms (revenue, price, and shipments) obtained from Bloomberg terminals, and (2) platform major events (operating system version releases, updates, and developer conferences), hand- collected from platform sponsors’ (e.g., Apple) websites and media websites. The purpose of including hardware sales variables is to capture of cross-side network effects (Eisenmann et al., 2006). By controlling for platform events, I could account for shocks induced by such events (Dranove & Gandal, 2003). To have a sense of the data used in this research, I now describe substantive information by following the platform-mobility phenomena that I investigate. The first step in determining whether a developer made a platform-mobility decision was to identify whether a developer of one platform (e.g., iOS) in my data also appeared in the data of the competing platform (e.g., Android). This relates to the task of matching developers across platforms, a highly complex task that I will describe in detail in a subsequent section. The second important issue was to identify a developer’s entire app portfolio (on both platforms if the developer has spanned both platforms). The earliest released app on a platform, and its release-date information, provide the date of the developer entering the platform. For a developer that had ever adopted both platforms, by comparing the entry dates I was able to learn the sequence of platform entry decisions, and the time when the developer made the platform- mobility decision. That is, I used the initial release-date information of all the developer’s apps on both platforms to determine platform-mobility decisions. 24 One may argue that in some cases the earliest version may not be the initial version of an Android app. However, I suggest that, in spite of such a limitation, this is the best a researcher in this area can do because the company is likely to be one of the first in the industry with a systematic compilation of Android app data. Unless Google is willing to directly share such information to individual researchers, to the best of my knowledge, Optimus’ proprietary information of Android’s app version history is the best available data. 58 Third, I used developer-specific information on performance. Performance on a platform is based on apps’ ranking information. In the data used for analyses, I have daily ranking information for iOS apps during the period from May 3, 2012 through May 14, 2015 and from August 22, 2013 through March 18, 2015 for Android apps. Because of the different types of rankings (e.g., by categories, based on downloads or revenues, and so forth.), I chose to use “grossing” ranking across all categories, as this type reflects apps’ revenue performance. Thus, performance measures and aspiration variables were coded based on such app ranking information. I also aggregated other app-level attributes to the developer-level and used them as controls, such as information on total app numbers and distribution across different app categories. Another major part of the data used to predict platform mobility included platforms’ installed bases. Although theoretically I have integrated both the hardware and software sides (referring to them as the installed-base difference in general), empirically I used data on both hardware and software. Hardware-installed base variables were based on smartphone shipments of the two platforms. 25 For the software side, I considered both the installed bases of complementary products (apps) and the installed bases of complementors (developers). Fourth, to code different types of movement across platforms—multihome or switching—as well as platform-abandonment decisions, I relied on information that indicated whether and when a developer withdraws from its home platform. For iOS apps, I leveraged information of app version history. Specifically, for all a developer’s apps, I gathered the entire historical information of the apps—that is, the date of each new version. I then made the (reasonable) assumption that, if a developer did not update an app for a long time (e.g., 12 25 Admittedly, in reality, hardware also includes other types of devices besides smartphones, such as tablets. Because of the scope of this research and data constraints, I only included the dominant devices of the two platforms for the hardware side. 59 months), it may well be the case that the developer had given up the product, which I referred to as app failure. Further, if the developer had given up all its apps on a platform (e.g., iOS), it meant that the developer had withdrawn from the platform. Consequently, I used the final date of a developer’s last product version, plus a wide time range that was used to define an app failure (e.g., 12 months), to indicate when the developer abandoned the home platform. As a result, whether a developer eventually abandoned its home platform enabled me to identify whether the mover was a switcher. Accordingly, note that the time of switching (when the developer launched on the target platform) is different from the time of platform abandonment (when the developer updated its last failed app, plus the time range until the app’s failure). In the case of Android apps, because I did not have the version history of all apps, I used a different type of information to code multihoming versus switching, and for platform abandonment. The information I used was whether an app still existed in the Play Store (by the time of my data collection), which was provided in the original Optimus data. The procedure was similar to one that I used when coding iOS platform abandonment. First, whether a developer had withdrawn from Android as indicated by whether all its apps had failed (or did not exist in the Play Store); and second, the time at which an Android developer withdrew from the platform as indicated by the time it launched its last app, plus a certain time range (e.g., 12 months). Finally, to predict ex-post performance, I coded performance variables based on raw app- ranking information. Matching Developers Across Platforms Because the two platforms use different templates to assign app identification numbers (IDs) and developer IDs, I needed to match developers across platforms to distinguish the following types: 60 iOS-only, Android-only, and iOS-and-Android. The matching was done through Python programming and by leveraging information on apps’ supporting web URLs (Uniform Resource Locators), developer names, app names, app categories, and so forth. I now describe in greater detail the goal of the matching algorithm, the different modules, and the output of matching. Goal of the algorithm. The goal was as simple as a “yes” or “no” decision, when comparing an iOS developer (and its corresponding app portfolio) and an Android developer (and its app portfolio), to determine whether they were the same developer across the two platforms. This seemingly simple task became very challenging when I began to dive into details with technical support from computer science research assistants. The first challenge was learning that correct matches required comprehensive data on both platforms. That is, if data on one platform was not comprehensive, one could not reach the conclusion that an iOS developer, for instance, ever appeared on the Android platform, or if the developer was single-homing. Such a conclusion would be false because of data constraint. I was able to eventually overcome this challenge through the laborious data collection effort, as described above, and I am confident that the iOS data approximate the 2015 population and the Android data approximate the population of early 2016. The second challenge arose from the complexity of the industry setting, as I next describe. Matching could not be done solely based on developer or app names, because the same developer could change names (slightly or significantly) across platforms. Moreover, matching could not rely solely on URLs, because the same developer could have had multiple support web URLs (for instance, when it has multiple country divisions), or because different developers use the same root URLs as supporting links (for instance, when individual developers registered on popular websites such as Google or Facebook). Also, multiple developers IDs may belong to the 61 same organization, as different divisions launched products on their own, resulting in the need to combine developer ID’s with a more universal ID in such cases. 26 Hence, to address these major challenges of matching developers across platforms, the matching algorithm that I designed, with the technical support from the research assistants, included two major modules. The first module, “matching by supporting URLs,” relied mainly on apps’ URL information and produced matches with higher quality; thus these matches were given priority when drawing the conclusion (of yes or no). The second module, “matching by app-portfolio information,” was an algorithm that integrated information of product categories, developer names, and app names. Because of the lower matching quality of this second module, it was given less priority when drawing the matching conclusion. I now describe each module in detail. Matching by supporting URLs. Mobile applications (either on iOS or Android) often provide a supporting URL that directs the user to the developer’s webpage. Most of the time URLs are in the form of the developer’s own websites, but there are situations in which URLs are associated with major websites (e.g., Google, Yahoo, Facebook). For example, if a developer is an individual and does not want to host her/his own website, but is instead porting on a major website. In addition, the same company may have different divisions that release apps on their own using different sub-domains (e.g., countries) of the company’s major domains. The algorithm that we developed generated output that addresses the aforementioned issues. As is shown in the flow chart of this module (Figure 8A), the first step was to determine whether to use the entire URL or just the URL’s domain-subdomain. The decision rule was that, if the URL 26 In fact, company affiliations of app developers are so complex that app analytics companies such as AppAnnie has a dedicated project to map the complex relations among developer IDs and companies, a project the analytics company refers to as “App DNA.” 62 was based on a major website, 27 we would then use the entire URL as the key for matching—the rationale being that the developer registered an individual account (or link) under the major website. If the URL was not one of the major websites, we would then extract the domain- subdomain as the matching key, 28 which discarded the differences in web-link variations (due to different countries/regions, for instance). The second step was to match two developers based on the (extracted) URL-key. The decision rule was that, if a match were not found, we would then conclude that the two were not the same developer. If we found a match, then a field would be coded (“match-key”) to indicate a match by URL-key, which would be combined with other criteria to reach the final conclusion of the matching algorithm, “yes,” In aggregate, the ULR- matching module outputs a table of “matched” developers, 29 which includes several fields for final decision making: (1) match-key-type (whether the whole URL was used or an extracted domain-subdomain), (2) match-key (the key used), (3) repeated-site-count (i.e., the number of times the domain-subdomain of the URL was used by developers to indicate whether the URL is from a popular/major website). ----------------Figures 8A about here---------------- Matching by app-portfolio information. Because not all apps have URL information, we also leveraged other available information about a developer’s app portfolio, to complete the cross-platform matching. This matching module used three types of information: developer names, app categories, and app names. The flow chart in Figure 8B shows the procedure of this matching module. At the developer level, we treated a developer’s name as a string and 27 Major websites were identified based on the fact that such websites appeared many time across different developers and thus were included as exceptions in the Python code written by research assistants. 28 This was done using the Python package tldextract. 29 The matching results of the table should be viewed as a mid-process, because whether two developers (in a row) is a match needs to be decided with other criteria combined. 63 conducted string-matching 30 between two developer names across the iOS and Android platforms. This matching procedure generated a field—developer-similarity-score—that can be used as a subsequent decision rule. A shortcoming of matching by using only developer names, however, was the easily committed error of (1) different developers sharing similar names (false positives), and (2) the same developer using different names (false negatives). Consequently, we leveraged app-portfolio information for matching as well. Given the big data that we were handling, the possible combinations of developer pairs were enormous, 31 and we improved the efficiency for this next step by selecting developer pairs whose developer-similarity-score was above 0.5, while discarding the rest in the next step of matching based on app portfolio. ----------------Figures 8B about here---------------- Matching by app portfolio included two steps. To prepare the data, we pooled two developers’ apps as a matrix, with the rows representing apps on one platform and as columns on the other platform. Each cell therefore could be used to compare the two products. The first step was to determine the overall similarity of a pair of developers’ app names. We used a modified bi-gram string matching technique and generated a score after comparing strings of two app names. In a matrix of two app portfolios, as an example shown in the following figure, we used the smaller dimension to compare app similarities. Assuming that the iOS developer has three products and the Android developer has six products, we used, in this case, the iOS developer’s side of the matrix to generate maximum similarity scores. From the first row (of App A), therefore, the maximum similarity score is 0.8 (when the category is also matched). Similar to the other two rows, we obtained the maximum similarity scores. We then took the average of the 30 The string-matching method that we use is a modified version of matching by bi-grams, described in detail in the following web link: http://www.catalysoft.com/articles/StrikeAMatch.html (accessed on April 30, 2017) 31 To be more precise, !"#$%& 400,000 +%&"#,& &-.-/#0-"1 × !"#$%& 300,000 ,45 &-.-/#0-"1 2 possible combinations. 64 (3) maximum similarity scores and reported it as a decision rule of app-similarity-score, which reflects the overall similarity of the two developers’ app names. Android Developer iOS Developer App 1 App 2 App 3 App 4 App 5 App 6 App A 0.8 0 0.5 0.2 0.6 0.1 App B App C The second step, an advanced version, was to integrate information of app categories. We wanted to determine whether two apps were in the same category. For this, we developed a schema that mapped iOS (first-level) categories to Android (first-level) categories. 32 Using this category-mapping schema we were able to assess whether two apps were in the same category. When the decision of the first step was “yes” (in the same category), we moved to the next step of matching two apps by their name strings. Therefore, continuing with the previous example, we now needed to first decide whether two apps were in the same category (in each cell), while only retaining the string comparison score when the answer was “yes.” This part of the algorithm generated another decision rule that we referred to as category-app-similarity-score. Android Developer iOS Developer App 1 App 2 App 3 App 4 App 5 App 6 App A Same category? Yes 0.8 Same category? no Same category? Yes 0.5 Same category? Yes 0.2 Same category? no Same category? No App B App C 32 This category-mapping across platforms is a challenge, as some categories are clear matches across platforms (e.g., first-level game category), while in the case of others, it is difficult to find a one-on-one match. When designing the mapping schema, we sometimes mapped multiple Android categories onto a single iOS category (as Android has more first-order categories), with the intention to strike a balance between (1) comprehensiveness (broad enough so that we did not miss an Android app to the corresponding iOS category), and (2) accuracy (narrow enough so that not all categories of Android apps are mapped to a few iOS categories). The mapping schema could be provided upon request. 65 Finally, to draw conclusions based on this app-portfolio matching algorithm, we used multiple decision rules. The first was: if the category matching was “yes,” and if the app-name string matching found a perfect match (i.e., score = 1), then we jumped directly to the conclusion that the two developers were the same. This decision rule was an exception rather than a general one. The second, and more general rule was to use the two scores to filter out developer pairs with very high values (e.g., 0.8). Figure 8C presents the flow chart of the algorithm that integrates both modules. ----------------Figures 8C about here---------------- Decision rules from the matching algorithm. In sum, the different modules of the matching algorithm generated different fields that I could use as decision rules to determine whether two developers (on iOS and Android, respectively) are, in fact, the same one developer. From the URL-matching module, the decision rules included: (1) match-key-type, (2) match-key, and (3) repeated-site-count. From the second matching module that leverages app-portfolio information, the output decision rules included: (4) app-similarity-score (which reflects the similarity of app names), and (5) category-app-similarity-score (which calculates app-name similarities only when also matched by categories). The procedure that I used to identify developers who appear on both platforms is to use the decision rules to select matched pairs (of developers). A general pattern was that, if I moved the decision rules along the continuum of very restrictive to relatively loose, I would end up with a different sample size of matched developer pairs, such that more restrictive rules would lead to smaller, but more accurate, matches. Because the URL-matching module was likely to be more accurate, the corresponding decision rules were given higher priority. For instance, the most restrictive criteria that I used were: (1) to only rely on the URL-matching module, (2) limited to 66 cases when a supporting website was used only once by developers on iOS and Android (i.e., repeated-site-count = 1). This resulted in a sample of 77,870 matched developer pairs 33 that were used as the analyses sample and reported results in the current version of the manuscript. Admittedly, such restrictive criteria sacrificed comprehensiveness for matching accuracy. By gradually loosening the decision rules, I could trade off matching accuracy for comprehensiveness. In empirical tests, I intend to address this concern by testing my theoretical predictions using different samples of matched developer pairs. If results are consistent across different sample sizes, we can then be more confident in the validity of the empirical conclusions. I now turn to a description of my approach to the matched developer pairs and the aforementioned data to code variables, as well as to test my theoretical predictions. Because in this research I use multiple dependent variables and different types of modeling techniques, the following sections will be structured as follows. I first describe how I coded major explanatory variables that I used as predictors across models that include performance-feedback variables and installed-base variables. I then introduce a common set of control variables that I used across different models. Finally, I turn to the hypothesis-testing sections. In each section, I will describe the dependent variable(s) of the corresponding set of hypotheses and econometric model(s) that I have used, and the results of regression models. I provide descriptive statistics and pairwise correlations of all the variables that will be used for analyses in Table 1 (for iOS developers) and Table 6 (for Android developers). ----------------Table 1 and Table 6 about here---------------- 33 Note that the matched developer pairs included repeated cases—that is, the same iOS developer could be matched onto multiple Android developers, and vice versa. Therefore, I combined such same-developer matches and created a universal ID to uniquely identify a (combined) developer. Consequently, the sample size of developers spanning both platforms was smaller than this number. 67 Performance-Feedback Variables As described in the theory, the performance metric used was apps’ ranking information on both platforms. I specifically chose to use the “grossing” ranking because this type captures most appropriately apps’ revenue performance (Kapoor & Agarwal, 2017). To construct the aspiration variables, I first coded the performance measure at the developer level. Developer performance. An app developer’s monthly performance was aggregated from its app portfolio. For each app, the grossing ranking data indicated the days and categories in which the app was ranked in any of the top lists. 34 I therefore aggregated an app’s total number of list times on any of the top grossing ranking lists to reflect its monthly performance. At the developer level, I aggregated these scores across all of its apps to reflect the developer’s performance in the month (on one platform). In other words, the performance measure, 7-"8#"9!%:- ;< , reflects the average number of times that =-.-/#0-" ; ’s products are listed in any of the top grossing ranking lists in >#%?ℎ < on one platform. Building on this foundational performance measure, I was able to construct aspiration variables, as well as the eventual performance-aspiration difference. Historical aspiration. Following prior literature (e.g., Eggers & Kaul, 2017; Greve, 1998, 2003), historical aspiration can be expressed as the following equation: A,1?#",:!/_+10,"!?,#% ;< = D∙7-"8#"9!%:- ;,<FG +(1−D)∙A,1?#",:/_+10,"!?,#% ;,<FG in which the firm’s historical aspiration at >#%?ℎ < (A,1?#",:!/_+10,"!?,#% ;< ) is equal to a 34 In the data that I use, top-ranking lists range from the top 1 to the maximum of top 300. This range of ranking reflects how platform owners/sponsors, such as Apple, provide the top ranking lists for customers to ease their searches of apps. For instance, in the early App Store years, Apple provided customers with top 500 apps, per AppAnnie, but later changed to only the top 200 apps, perhaps because a typical customer would stop after scrolling down the list to certain degree. Furthermore, many categories of top ranking lists exist in which an app could potentially land. For instance, for the data I used, the App Store has 23 primary categories and 18 game subcategories, and the Play Store has 36 primary categories and 18 game subcategories. Finally, ranking lists also vary across countries/regions. Thus, to reduce the level of empirical complexity, I chose to only include the U.S. app market. Therefore, my analyses will not face potential confounding effects due to variations across institutional and cultural environments. 68 combination of its performance in the previous month (7-"8#"9!%:- ;,<FG ) and its historical performance in that month (A,1?#",:/_+10,"!?,#% ;,<FG ). D is between 0 and 1 and reflects the relative weights given to prior performance and prior historical aspiration. I used 0.5 for the value of D, consistent with prior research (e.g., Greve, 2003). Social aspiration. According to the performance feedback theory and previous empirical research, social aspiration can be measured by the average performance of similar others in a marketplace. I therefore averaged all the other developers’ (beside the focal developer) performance to reflect social aspiration. That is, for =-.-/#0-" ; , its social aspiration in >#%?ℎ < can be expressed as: 5#:,!/_+10,"!?,#% ;< = 7-"8#"9!%:- M< M∈℘ MP; . Performance-feedback. Prior literature has two different perspectives, with one side supporting the separation of historical versus social aspirations and the other unifying the two types of aspiration. In the interests of theoretical and empirical parsimony, I adopted the former approach (e.g., Greve, 2003) as follows: +10,"!?,#% ;< =Q∙A,1?#",!/_+10,"!?,#% ;< + 1−Q ∙5#:,!/ RST;UV<;WX ;< , where Q is a factor that ranges from 0 to 1 and it gives relative weight to historical versus social aspirations. I followed prior literature (Greve, 2003; Eggers and Kaul, 2017) and used 0.3 to weigh the two types of aspirations. 35 Based on the performance variable (7-"8#"9!%:- ;< ) and the above-aspiration variable (+10,"!?,#% ;< ), performance feedback can be measured in different ways. The first type is the simple mathematical difference between the two (i.e., 7-"8#"9!%:- ;< −+10,"!?,#% ;< ). 35 Incidentally, the decision rule that previous studies have used (Greve, 2003) when choosing the value of Q is to use the value that could eventually produce the highest model fit (in terms of log likelihood, for instance). Results of some studies (e.g., Eggers and Kaul, 2017) have suggested that the optimal model fit was achieved when Q equals to 0.3. 69 Although this measurement has not been widely used in the previous literature, I will use it in later sections when examining the interaction effects between performance feedback and installed-base difference, for reasons of simplicity. The second way is to follow previous studies (e.g., Greve 1998; 2003; Eggers and Kaul, 2017) and code above- versus below aspirations using spline functions. Specifically, if performance is above aspiration, then +Y#.-_+10,"!?,#% ;< is equal to ( 7-"8#"9!%:- ;< −+10,"!?,#% ;< ); if performance is below aspiration, then Z-/#[ _+10,"!?,#% ;< is the absolute difference between the two (i.e., 7-"8#"9 !%:- ;< − +10,"!?,#% ;< ). One potential concern is that historical aspiration- and social-aspiration levels would define high versus low performers differently. That is, the same developer may be a low performer based on social comparison, but a high performer based on historical comparison. To address this concern, I tabulated the sample along the aforementioned two dimensions. The statistics suggest that only a very small percentage of the sampled developers are in such misfit categories. For instance, in the iOS sample, I found that only 0.7% of developers are above historical aspiration but below social aspiration, and only 1.6% are above social aspiration but below historical aspiration. Consequently, such a small sample percentage would not qualitatively change the findings of the study. Another potential concern is that the above- and below-aspiration variables may have excessive zero values. I therefore calculated percentages of none-zero values for each of the two variables. The developer-month panel data suggest that 33% of the observations have non-zero values for the above-aspiration variable, and 69% for below aspiration. Therefore, I conclude that the data do not have excessive zero problems. 70 Installed-Base Difference Variables The relative installed-base variables were intended to capture the difference between the two platforms’ installed bases (of hardware or software). Because Android eventually had a larger installed base during the time investigated, I therefore subtracted the corresponding iOS installed-base variable from the Android installed base. Based on the hardware shipments, and available apps and developers on both platforms, I generated the following three relative installed-base measures. Installed-base difference in smartphone shipments. Because the hardware data that I obtained from the Bloomberg Terminal were at either the yearly or quarterly level, I broke the data down to the monthly level, assuming that within such quarters (or years) shipments are evenly distributed. 36 For each platform in a certain month, therefore, I was able to calculate the cumulated smartphone shipments until the month, which I referred to as the platform’s installed base of smartphones. If denoting the installed base (IB) of iPhones (of the iOS platform) in >#%?ℎ < as \Z_59!"?0ℎ#%- ;]^ ,< , and that of the Android platform as \Z_59!"?0ℎ#%- RX_UW;_,< , then the installed-base difference in smartphone shipments is equal to \Z_59!"?0ℎ#%- RX_UW;_,< −\Z_59!"?0ℎ#%- ;]^ ,< . Installed-base difference in developers. Similarly, to code a measure of installed-base differences in developers, I first calculated the total number of developers on each platform in a month, and then subtracted the cumulative iOS developers (\Z_=-.-/#0-"1 ;]^ ,< ) from the cumulative value of Android (\Z_=-.-/#0-"1 RX_UW;_,< ). That is, this relative installed-base 36 In my hardware shipments data, these are the time ranges that I have quarterly information (2012Q3–2013Q4, 2015Q2–2016Q3), as well as the time ranges for which I only have yearly information (2008–2015). Admittedly, assuming that shipments are evenly distributed within such time ranges does not fully reflect the reality, but, given the data constraints, I have to make such an assumption. However, as will be shown, the results based on hardware- installed-base variables are also consistent with the results when using installed bases of apps and developers (which are more fine-grained). The consistent results across all three types of installed-base measures thus partially overcome the weakness of the hardware data and bolster the confidence in the empirical conclusions. 71 measure equals to \Z_=-.- /#0-"1 RX_UW;_,< −\Z_=-.-/#0-"1 ;]^ ,< . Installed-base difference in apps. Finally, on the software side, I coded another measure that reflected the two platforms’ difference in cumulative apps, which can be expressed mathematically as \Z_+001 RX_UW;_,< −\Z_+001 ;]^ ,< . Controls I also identified a common set of controls that could potentially affect a developer’s decisions to move across platforms, to abandon the home platform, or the performance on the home- and target platform ex post. I categorized these controls as developer attributes and platform attributes. I controlled for a developer’s number of apps, portfolio size, in the focal platform, which included two variables: the number of apps in iOS and the number of apps in Android. These variables were used in the two sets of regressions testing iOS-to-Android mobility and Android-to-iOS mobility. To reflect the extent to which a developer’s products are diversified across different product categories, I coded the variable diversification index, which is a modified version of the Herfindhal Index and can be expressed as 1− θ a b (where θ a is the developer’s percentage of products in category i) (Teodoridis, 2016). App developers’ capabilities are generally unobservable. Product characteristics can be appropriate indicators of firm capabilities, because product and firm capabilities are two sides of the same coin (Wernerfelt, 1984). I therefore coded several product-related indicators to reflect app publishers’ capabilities. 37 Downloads are an important indicator of app performance. I therefore coded the 37 Several developer attributes are reported in the descriptive statistics tables (Table 1 and Table 6) but I did not control them in analyses, because missing values of those variables take a large percentage of the sample, resulting in the concern of sample selection bias. Nevertheless, I hereby briefly describe those variables. The first is average app rating (ranging from 1 to 5). High ratings indicate a developer’s capability to develop products that elicit consumers’ higher willingness to pay. The second indicator is average price of paid apps. Higher prices indicate greater value that can be delivered to customers. The third indicator is the average installed base of app users. A 72 variable average downloads of a developer’s portfolio apps. 38 I used platform attributes to reflect platform attractiveness and to control for platform effects on developers’ migration decisions. When a platform owner releases an update of its operating system, it could either cause disruptions (Kapoor and Argarwal, 2017), or introduce novel opportunities, for app developers. Updates could be major or minor. Android (or iOS) major release was coded as 1 if, in the month, Google (or Apple) released a major Android (or iOS) version (e.g., Android 4.4 KitKat), and 0 otherwise. Android (or iOS) version updates was coded as 1 if, in the month, Google (or Apple) released an update of the Android (or iOS) operating system (e.g., 4.4.1) or released a major version, and 0 otherwise. Again, descriptive statistics and pairwise correlations for all the variables can be found in Table 1 (for the iOS sample) and Table 6 (for the Android sample). I. Hazard Models Estimating Platform Mobility To test platform-mobility decisions, I compiled a sample for each platform as consisting of (1) developers who eventually moved to the other platform (i.e., movers), and (2) developers who initially adopted the platform and have since stayed on the platform (i.e., stayers) until the time of observation. Admittedly, there are cases in which developers simultaneously adopted both platforms in the same month, a group that was only a small percentage of the entire population (e.g., less than 0.2% of iOS developers) so I chose to exclude it from the analysis. It was important to determine whether and when an app developer of the focal platform decided to also large installed base of users reflects an app’s appeal and quality, as well as the publisher’s capability in developing quality products. I do not have specific information on an app’s installed base of users so I followed prior research (Davis, Muzyrya, and Yin, 2013) by using the total number of app ratings to proxy for the lower bound of the installed base. 38 I only have app downloads information for the Android data, which provide categorical downloads information in terms of 0, 1, 5, 10, 50, 100, until 1.00e+09. Therefore, the average downloads variable will only appear in analyses of the Android sample. 73 adopt the other platform, which I captured with a dummy variable—platform mobility. Dependent variable. The dependent variable, platform mobility, was generated based on the results of matching developers across platforms. Platform mobility is a dummy variable indicating whether a developer shifts to the competing platform. Therefore, in the risk set of developers that initially adopted iOS, for instance, the variable platform mobility equals to 0 if the developer stayed with iOS, and changed to 1 in the month the developer also adopted (i.e., moved) to Android. Analysis. I applied proportional hazard models to estimate the hazard that an app developer moved to the competing platform, treating the hazard rate as continuous and in the following form (Hsiao, 2014): d ;< = lim ∆<→i 7"#Y[? ≤ = ; < ?+∆?|= ; ≥ ?] ∆? , where = ; is the time spent by the developer in the state of not moving. $ ;< is the hazard function of the duration variable = ; . The hazard rate gives the instantaneous conditional probability of the app developer leaving the state of not moving and thereby entering the competing platform. I assumed the hazard rate d ;< as a function of platform- and developer attributes—p q —that are specific to the focal developer. For instance, the vector includes the proposed antecedents that explain platform mobility—that is, performance-feedback variables (i.e., above- and below aspirations) and installed-base differences. The proportional hazard model takes the form of d ;< = d(?)exp (p q u v), where d(?) is the baseline hazard function. In a Cox proportional hazard model, the baseline hazard does not need to be estimated, because we only need to estimate the part of the likelihood function that contains the coefficient vector v. An app developer is in the risk set of moving to the competing (target) platform, starting from the month that it adopts its first (home) platform. The developer exits the 74 risk set (1) when it makes the mobility decision, or (2) when it reaches the right censoring month (the end of data collection). The unit of analysis is a developer and the time interval of the panel structure is a month. In all the models, I controlled for developer- and platform attributes, as well as time trend, which could potentially affect a developer’s moving decision. At the developer level, I controlled for (1) the focal developer’s app portfolio size on the home platform (because big players may have more resources), and (2) the extent to which the developers’ products are diversified in multiple product categories—diversification index (because generalists may have a greater tendency of pursuing further diversification by moving to the other platform market). At the platform level, I controlled for (1) iOS version updates, and (2) Android version updates (because operation system updates could make a platform more or less attractive to developers) (Kapoor & Agarwal, 2017). Finally, I included a continuous time variable, which equals to 1 for the first month (July 2008) when third-party apps first appeared in the App Store and cumulates over time. This variable helps to account for the effects of natural time trend on developers’ platform-mobility decisions. Results. As an overview of the developers’ mobility patterns in both directions (i.e., iOS- to-Android, and Android-to-iOS), Figures 3A and 3B present survival curve and cumulative hazards of iOS developers’ platform-mobility decisions (Figures 4A and 4B for the Android sample). As can be seen, around 15% of iOS developers, and 3.5% of Android developers, eventually chose to move to the other platform (i.e., further adopting the competing platform). Because the hypothesis testing leverages both the iOS and Android samples, and I am examining both directions of mobility, I will mainly present results based on the iOS sample in regards to the iOS-to-Android direction of platform mobility. If the results of the hypothesis testing 75 regarding the other direction of mobility (Android-to-iOS) are inconsistent, I will highlight such results as footnotes, but the comprehensive results are presented as tables and figures. ----------------Figures 3A, 3B, 4A, and 4B about here---------------- Tables 2A and 2B present the results of proportional hazard models estimating iOS developers’ likelihood of moving to the Android platform. Model 1 of Table 2A includes only control variables; Models 2 to 4 test aspiration-related hypotheses; and Models 5 to 7 test the hypothesis regarding installed-base difference. In Hypothesis 1.1a (H1.1a), I predicted that above aspiration would be positively related to a developer’s likelihood of moving to the competing platform. In both Model 2 and Model 4, Hypothesis 1.1a is supported. In Model 2, where the main effect of above aspiration enters (besides control variables), above aspiration has a positive and significant effect on the likelihood of the developer moving to Android (in Model 2, w = 0.016,0 < 0.001); while in Model 4, where both above aspiration and below aspiration are included, above aspiration also has a significant positive effect on the likelihood of moving (w = 0.017,0 < 0.001). Similarly, Hypothesis 1.1b (H1.1b) predicts that below aspiration will be positively related to the likelihood of moving to Android (because of competitive crowding). As can be seen, in both Model 3 (where only the main effect of below aspiration enters) and Model 4 (where both above- and below aspirations are entered) the variable below aspiration has positive and significant effects (in Model 3, w = 0.197,0 < 0.001; in Model 4, w = 0.198,0 < 0.001) on the mobility likelihood. H1.1b is thus supported as well. With respect to effect sizes, results in Model 4 suggest that a 1 standard deviation increase in above aspiration (i.e., 2.81) is associated with a hazard ratio of 1.049 (i.e., - i.iG}∗b.G ), and an increase below aspiration by 1 standard deviation (i.e., 0.31) is associated with a hazard ratio of 1.063 (i.e., - i.GÄ∗i.ÅG ). ----------------Tables 2A and 2B about here---------------- 76 Hypotheses 1.2a (H1.2a) and 1.2b (H1.2b) concern the effects of installed-base differences (between Android and iOS) on the likelihood of platform mobility. H1.2a predicts that an installed-base difference will have an inverted-U effect on the likelihood that an iOS developer moves to Android. The hypothesis is tested with three measures of installed-base differences: (1) in smartphone shipments, (2) in app developers, and (3) in apps. I include the first- and second-order terms of each of the three variables in Models 5 to 7 (in Table 2A). To support H1.2a, the first-order term of an installed-base difference variable needs to be positive and significant while the second-order term of that variable should be negative. As can be seen in Table 2A, the results generally support the predicted pattern. That is, in Model 5, the coefficient of the installed-base difference in smartphone shipments is positive (w ÇÉÑV<;ÖÉ_^ÜVU<T áWXÉ_^á;TÜÉX<S = 1.653) and significant (0 < 0.001), and the squared term is negative (w ÇÉÑV<;ÖÉ_^ÜVU<T áWXÉ_^á;TÜÉX<S â = −1.552) and significant (0 < 0.001); in Model 6, w ÇÉÑV<; ÖÉ_äã_åÉÖÉÑWTÉUS = 4.089 ( 0 < 0.001 ), and w ÇÉÑV<;ÖÉ_äã_åÉÖÉÑWTÉUS â = −116.947 ( 0 < 0.001 ); and in Model 7, w ÇÉÑV<;ÖÉ_äã_RTTS = −0.092 (not statistically significant), and w ÇÉÑV<;ÖÉ_äã_RTTS â = −4.329 (0 < 0.001 ). The inverted-U effects of relative installed-base variables on the moving likelihood are further shown in Figures 5A, 6A, and 7A. These figures were generated based on results of logistic regressions, which produced qualitatively similar results, as I needed to account for intercepts when drawing the figures. Thus, based on the hazard model results and the figures of interaction effects, I conclude that H1.2a is supported. ----------------Figures 5A, 6A, 7A about here---------------- Contrary to H1.2a, H1.2b predicts that the relationship between installed-base differences and the likelihood an Android developer moves to iOS will exhibit a U-shaped pattern. This hypothesis was tested with the Android sample, with results presented in Table 7A. As can be 77 seen in Models 5–7 in the table, the results also support an inverted-U effect of installed-base differences on the Android-to-iOS mobility likelihood, thus against the prediction in H1.2b. I will discuss this contradictory finding in the Discussion section. ----------------Tables 7A and 7B about here---------------- Hypothesis 1.3 predicts that performance feedback and installed-base differences of the two platforms would have interaction effects in influencing the likelihood of platform mobility, such that the curve of high performers (when the performance-aspiration difference is high) will be flatter and will envelope the curve of low performers (when the performance-aspiration difference is low). To support such a pattern, not only do the first- and second-order terms of installed-base difference need to exhibit the pattern as described in H1.2a, but we also need two additional empirical patterns. First, the direct effect of the continuous variable, performance- aspiration, needs to be positive (so that the curve of higher performers will possibly envelope that of lower performers). Second, the performance-aspiration variable’s interaction effect with the installed-base difference should be negative, while its interaction effect with the squared installed-base difference should be positive (so that the slopes of lower performers will be steeper than those of higher performers). Table 2B tests H1.3. For simplicity, in Models 1–4, I used a continuous variable of performance-aspiration difference to differentiate high versus low performers and interacted that variable with the installed-base difference variables. In Models 5–13, I split the performance- aspiration variable to above- and below aspirations and tested how they interacted with different measures of installed-base difference variables. As can be seen in Models 2–4, the direct effects of the continuous performance-aspiration variable are positive and significant (in Model 2, w = 0.029,0 < 0.001; in Model 3, w = 0.015,0 < 0.001; and in Model 4, w = 0.011,0 < 0.001). 78 Performance-aspiration interactions with the installed-base difference are significantly negative (in Model 2, w = −0.042,0 < 0.001; in Model 3, w = −0.258,0 < 0.001; in Model 4, w = −0.044,0 < 0.001); and its interaction effects with the squared installed-base differences are positive and significant (in Model 2, w = 0.019,0 < 0.001; in Model 3, w = 1.548,0 < 0.001; in Model 4, w = 0.069,0 < 0.001). Based on splitting the performance-aspiration variable into above- versus below aspirations, the results from Model 5–13 suggest that H1.3 is always supported when the interaction effects are with above aspiration, but are relatively weaker and sometimes inconsistent across models when the interaction effects are with below aspiration. I then graphed the interaction effects (with performance-aspiration) using logistic regression results that are provided in Figures 5B, 6B, and 7B. 39 The figures support the predicted pattern. Thus, combining the results and figures, I conclude that H1.3 is generally supported (although not across all models). ----------------Figures 5B, 6B, 7B about here---------------- II. Hazard Models Estimating Platform Abandonment To further probe the different motivations behind developers’ platform mobility decisions, I focused on the group of movers and examined which members are more likely to abandon the home platform. The sample for this analysis is therefore restricted to platform movers, consisting of multihomers and switchers. 39 Because time trend might be driving the force behind installed-base-difference variable, in robustness check with the purpose of teasing out the effect of time trend, I have also applied residual inclusion models (Terza, Basu, and Rathouz, 2008), in which I regressed each installed-base-difference variable on a continue time variable at the first stage, then added the residual as a regressor at the second stage. Appendices 1a-1b graph the effect of time trend on mobility likelihood and its interaction with performance-aspiration difference. Results of the effects of installed- base-difference variables on mobility likelihood and their interactions with performance-aspiration difference are presented in Appendices (2a-2b, 3a-3b, and 4a-4b) as graphs. As can be seen, the patterns generally support prediction of H1.3. 79 Dependent variables. Intuitively, platform abandonment refers to a developer forgoing all business operations on the home platform. I operationalized the foregone businesses on the home platform as the time at which all the developer’s products on the platform fail. Ideally, failures should include all products (1) that were withdrawn from the market, as well as (2) those products left unattended in the marketplace. In my data, I only observed iOS apps that survived until 2015, thus I used the criterion of whether an iOS app was left unattended to define app failure. Specifically, if an iOS app had not been updated for a long time (e.g., 12 months) then I coded the app as failed. Further, if all the developer’s products on the iOS platform failed, then the developer abandoned the platform, and the time of abandonment was coded as 12 months since the developer’s most recent update(s) of its products. This coding relied on the time-variant version history of iOS apps in my database. Hence, for the iOS sample, the dependent variable, abandonment, was coded as 0 from the developer’s initial adoption of the home platform, and changed to 1 in the month when a long time (e.g., 12 months) had passed since the developer’s last update of its product(s) on the iOS platform, conditional on the fact that all the developers’ iOS apps eventually failed (not updated for 12 months since the most recent update). For the Android data, I did not have complete version-history information. Therefore, I relied on another variable that indicates whether the app still exists in the Play Store to code whether an app failed. Platform abandonment, in this case, refers to the situation under which all apps of an Android developer were withdrawn from the Play Store (i.e., no longer existed at the time of my data collection—early 2016). Timing of abandonment, however, would be difficult to determine because of data constraints. As a result, I could only use the release date of the last released app, plus a long time range, to proxy the time of platform abandonment, assuming that 80 since the last app’s release it had not been updated (a weak assumption due to data limitation). To summarize, for the Android sample of developers, the dependent variable, abandonment, was coded as 0 from initial adoption of the platform, and changed to 1 in the month when a long period of time (e.g., 12 months) had passed since the developer’s release of its last product, conditional on the fact that all the developers’ Android apps eventually failed (i.e., withdrawn from the Play Store). Analysis. Restricting the sample to platform movers, I used proportional hazard models to estimate the mover’s likelihood of abandoning the home platform. Therefore, a developer in such a sample enters the risk set when it first joins its (home) platform and exits the risk set when it abandons the platform, or when it reaches the right censoring month (of data collection). 40 Results. Table 3 presents results with the iOS sample to estimate the likelihood of iOS movers abandoning the platform. 41 My hypotheses predict that above aspiration will be negatively (H1.4a), while below aspiration positively (H1.4b), related to the likelihood of platform abandonment. As can be seen from the table, the coefficients of above aspiration are negative and significant across models (in Model 1, w = −0.035,0 < 0.001; in Model 3, w = −0.029,0 < 0.001), thus consistent with the theory and supporting the hypothesis (H1.4a); additionally, the coefficients of below aspiration are positive and statistically significant (in Model 2, w = 0.294,0 < 0.001; in Model 3, w = 0.188,0 < 0.05), which also support the hypothesis (H1.4b). ----------------Table 3 about here---------------- 40 For the Android sample, however, I used the month that a developer initially adopted Android as the starting time in hazard models. The reason is that, because of data constraint, coding abandonment based on release time of the last app was inaccurate, to the extent that a large percentage of Android movers abandoned the home platform before their mobility decision to iOS. To deal with the potential sample-selection issue that is caused by hazard models dropping such cases, I thus changed the starting time to the month of Android platform adoption. 41 Corresponding results for the Android sample is provided in Table 8, which shows the same directions of effects by above and below aspiration, although statistically insignificant. 81 III. The Effect of Mobility on Home-Platform Performance Ex post, I care about how a developer’s platform-mobility decision influences its performance, both on the home platform, as well as on the target platform. Since I was able to observe a mover’s before and after performance on its home platform, I built a panel data structure in such a way that covered six months before, and six months after, the mobility decision. Because I intended to investigate within-developer effects as a result of mobility, the sample was thus constrained to developers of a platform (e.g., iOS) who eventually made the decision to move to the competing platform. The final iOS sample that was used for the analysis included a total of 2,860 movers of which 2,404 were multihomers and 456 were switchers. Dependent variable. I used a developer’s performance in grossing ranking lists as the outcome measure. As described in a previous section, I coded the performance measure, 7-"8#"9!%:- ;< , as the average number of times that =-.-/#0-" ; ’s products were listed in any of the top grossing ranking lists in >#%?ℎ < on the home platform. Therefore, the variable was a count measure. Analysis. I chose to use Negative Binomial panel models instead of Poisson models, because the dependent variable is a count measure and the variable is over-dispersed (i.e., the standard deviation is much greater than the mean, as is shown in Table 1 descriptive statistics). 42 I added developer fixed-effects specifications to wipe out confounding effects from time- invariant developer-level observables and non-observables and also controlled for developer attributes (including app portfolio size, diversification index), platform attributes (iOS- and Android operation system updates, installed-base difference of developers), and a continuous time variable to capture time trends. 42 An assumption of the Poisson distribution is that the mean equals the standard deviation. 82 In addition to control variables, two main explanatory variables were used to predict home-platform performance. The first is a dummy variable that indicates whether before or after the mobility decision, denoted as +8?-"_>#Y,/,?ç ;< , which equals to 0 if =-.-/#0-" ; has not yet made the moving decision in >#%?ℎ < and changes to 1 for the following months since its moving decision. The second is a time-invariant dummy variable that indicates whether a developer is a switcher or a multihomer. Thus, the variable, 5[,?: ℎ-" ; , equals to 1 if the developer eventually abandons the home platform, and 0 if the developer remains multihoming on both platforms after moving. Results. Table 4 presents the results of analyses on home-platform performance. In Hypothesis 1.5a (H1.5a), I predicted that platform mobility would have a negative effect on a mover’s performance on its home platform. In Model 1, only the time-variant dummy variable +8?-"_>#Y,/,?ç ;< is included (besides controls and fixed effects). The coefficients on this variable are negative (w = −0.039) and significant (0 < 0.05), suggesting that a mover’s performance on the home platform in the post-mobility periods is lower than its pre-mobility periods. This empirical evidence therefore supports H1.5a. In Hypothesis 1.5b I further predicted that the negative effect of platform mobility on the developer’s home-platform performance would be stronger if the developer is a switcher. This hypothesis is tested via the interaction effects between the time-variant dummy +8?-"_>#Y,/,?ç ;< , and the time-invariant dummy 5[,?: ℎ-" ; , indicating developer type. As is shown in Model 2 of Table 4, the interaction effect is negative (w = −0.163) and statistically significant (0 < 0.001), providing strong support for H6b. ----------------Table 4 about here---------------- I also conducted subsample analyses by examining mobility’s effects on multihomers 83 versus switchers. Results are presented in Models 3 and 4 of Table 4. We can see, in Model 3, that, although the effect of mobility on multihomers’ home-platform is negative, therefore in the predicted direction, it is not statistically significant. However, mobility’s effect on home- platform performance is negative and significant when based on the subsample of switchers (Model 4). These results suggest that, consistent with the theory, mobility mainly exerts a negative effect on performance for switchers. 43 IV. The Effect of Mobility on Target-Platform Performance The final set of analyses examines how movers perform on the target platform in the post- mobility time period. Because I could only observe platform movers’ performance after they had moved to the target platform, the sample and method of this set of analyses are different from the previous section on home-platform performance. Building on my theoretical framework, I categorized and used a sample of developers on the target platform that consists of five types. I illustrated with one direction of movement: iOS-to-Android. The first two types are movers from the home platform (iOS) to the target platform (Android), and these include (1) iOS-to-Android multihomers, and (2) iOS-to-Android switchers. For developers on the target platform I then identify corresponding similar groups: (3) Android-to-iOS multihomers, and (4) Android-to-iOS switchers. The final set of developers, also the group that I chose to be the baseline comparison, consists of (5) Android stayers, or developers who initially adopted Android and remained on the platform ever since. The dependent variable used for the analysis is the same as that for the home-platform- performance analysis, where 7-"8#"9!%:- ;< reflects the average number of times that 43 Results for Android movers are different. As can be seen from the results in Table 9, I found that for Android movers, mobility has a positive effect on performance on the home platform (Android), but the positive influence is weaker for switchers. 84 =-.-/#0-" ; ’s products are listed in any of the top grossing ranking lists in >#%?ℎ < on the target platform. I used Poisson random-effect models 44 to test the hypotheses not only because the variable is a count measure, but also to account for developer-specific time-invariant unobservables. Results of the iOS sample are presented in Table 5. As can be seen, coefficients of multihomer variables are all positive and significant, suggesting that both (1) multihomers who moved from iOS (iOS-to-Android multihomers, w = 0.932,0 < 0.001), and (2) mutihomers who moved away from Android outperform those who stay on the target platform (Android-to-iOS multihomers, w = 0.884,0 < 0.001). Thus, Hypothesis 1.6a, which predicts that multihomers would outperform stayers on the target platform, is supported. Hypothesis 1.6b, in contrast, predicts that switchers tend to perform worse than those who stay on the target platform. As can be seen from the results of Table 5, the coefficient of iOS-to-Android switcher is positive w = 0.072, though not significant, suggesting that such movers do not necessarily underperform stayers on the target platform and, as such, not in the predicted direction. Thus, H1.6b is not supported. Interestingly, switchers who move away from the target platform (Android-to-iOS switchers) perform worse than stayers w = −0.452,0 < 0.1, indicating that such movers tend to have lower performance, which is consistent with my theoretical prediction. A final interesting observation is that, for the same type of movers (either multihomers or switchers), developers who started with iOS generally outperform those who started with Android. This is consistent with the platform market assumption that iOS is quality-driven and therefore more selective with respect to its initial adopters, while Android is quantity-driven and thus less selective when it 44 I do not use developer fixed-effects models to avoid dropping cases when developers do not have within- developer variations of the outcome variable. Such developers tend to be lower performers, thus dropping them may create a sample selection issue. 85 comes to its initial adopters. 45 ----------------Table 5 about here---------------- DISCUSSION AND CONCLUSION The goal of my study is to examine how organizational and environmental factors influence app developers’ cross-platform-mobility decisions, as well as their performance consequences. In this research I systematically investigate the platform-mobility phenomena through a set of related research questions: (1) What internal attributes and market forces explain app developers’ decisions to move to a competing platform? (2) What explains developers’ strategic choices of abandoning their home platform? (3) What are the performance consequences of their strategic choices, on both their home- and target platforms? To probe the first question of platform-mobility decisions, I study mobility as a general strategic choice that developers face. Anchored in the performance feedback component of the behavioral theory of the firm, I found that both above and below aspirations lead to developers’ decisions to subsequently enter a competing platform. I further argued that the underlying mechanisms that explain higher performers’ (i.e., above aspirations) platform-mobility decisions and that of lower performers (i.e., below aspirations) are different. While lower performers move because of competitive crowding, which is a type of problemistic search (Cyert and March, 1963), higher performers move because they are after the possibilities of winner-take-all (due to positive network effects) in their niche markets. 45 Results of the analysis when iOS was the target platform suggest more nuanced patterns. As can be seen in Table 10, where the baseline group is iOS stayers, both groups of movers from Android to iOS—multihomers and switchers—have worse performance than local stayers. Such findings also suggest that only one of the two hypotheses—H1.6b but not H1.6a—is supported. Consistent with the theory section on predicting mobility, the results in this table suggest that iOS-to-Android multihomers outperform those who stay on iOS, while iOS-to- Android switchers underperform. 86 In the first part of my study that focused on mobility decisions, I also found that a central environmental factor—installed-base difference between the quantity- and quality-driven platforms—influences developers’ cross-platform-mobility decisions. Specifically, I found that, as the quantity-driven platform (Android) gains advantages in installed bases of both hardware and software (i.e., increases in installed-base difference), the likelihood of a developer on the quality-driven platform (iOS) moving to the other platform initially increases, then after a point, decreases, suggesting an inverted-U relationship between an installed-base difference and mobility probability. I proposed that the increasing part of the inverted-U reflects the market growth and legitimization of the quantity-driven platform, while the decreasing part reflects market maturity and intensified competition on this target platform. However, contrary to the prediction in H2b, I did not find a mirror effect of installed-base difference on the reverse direction of mobility (i.e., U-shape). Instead, that relationship also exhibited an inverted-U shape, suggesting that market forces other than Android’s growth and saturation may have also shaped the flow of Android-to-iOS movers. Finally, I found interesting interaction effects between installed-base difference and performance-aspiration differences on the likelihood of platform mobility. For the direction of mobility from quality- (iOS) to quantity-driven platform (Android), I found that the inverted-U effect of installed-base difference on mobility likelihood of higher performers is flatter than and envelops the inverted-U curve of lower performers. The rationale for such action, I argue, is that higher performers realize market opportunities earlier, while at the same time, such performers have more resources as a buffer against effects from external environments (e.g., market competition on the target platform). To answer the second question regarding platform-abandonment decisions, I study which 87 developers are more likely to abandon the home platform subsequent to mobility. Studying this helped me to differentiate two types of movers: multihomers versus switchers. Multihomers are those who keep operations on both home- and target platforms after mobility, while switchers refer to those who abandon their home platform at a certain time after the mobility decision. I found that above-aspiration developers are less likely to abandon their home platform (thereby more likely to multihome), while below-aspiration developers are more likely to abandon their home platform ex post (thus becoming switchers). Taken together, such findings provide further evidence of the different motivations of above- versus below-aspiration developers to move across platforms. Finally, I examine how platform mobility affects a mover’s performance on the home- and target platforms. With respect to home-platform performance, I found mixed findings for iOS movers and Android movers. Using the iOS sample, I found that platform mobility, in general, has a negative effect on the developer’s performance in the short term, and that the negative effect is more pronounced for switchers than for multihomers. These findings support the argument that, in the mobile app market of two dominant platforms (iOS and Android), multihoming costs, on average, outweigh benefits. Moreover, because of their humble origins on their home platform (i.e., below aspirations), switchers are less capable of handling the costs of multihoming in the short term. For Android movers, however, I found that mobility instead increases their short-term performance, and, consistent with findings about iOS movers, switchers enjoy less of the positive performance returns. Turning to target-platform performance, the findings also differ for iOS movers and Android movers. Findings about iOS movers suggest that, compared to stayers on the target platform (Android), multihomers perform better, while switchers do not have significant 88 performance difference with local (Android) stayers. However, both groups of Android movers (multihomers and switchers) to iOS perform worse than local (iOS) stayers. The contrasting findings on the two platforms further support my theoretical assumption of quality-driven (iOS) versus quantity-driven platforms (Android), in which the initial platform selection sorts developers into different markets. Theoretical Contributions The first and foremost contribution of this research is that it addresses two major questions of concern to strategic management scholars: How do firms make strategic choices? And, what are the performance consequences of such strategic choices? In a platform-based market in which the costs of adopting multiple platforms are high, such as the mobile app market of the two dominant mobile platforms (iOS and Android), a critical strategic choice for a complementor (an app developer in this context) is whether to subsequently enter a competing platform. Such cross-platform, market-entry decisions are essential and costly. A mover to the target platform needs to adapt to the new business environment of the new business, where platform governance rules are often different. A mover will also face potentially high multihoming costs, such as “translating” codes in order to port products on the other platform via either third-party engines (e.g., Unity) or developing its own, ensure products’ compatibilities across platforms, and interact with different audiences (such as developer communities and consumers of the target platform). Consequently, it is important to know which developers are willing to take on such costs and make platform-mobility decisions. Thus, I map the first major topic (about strategic choices) onto my first research question: What are the antecedents that predict a developer’s subsequent platform-mobility decision? Anchored in the behavioral theory of the firm, my 89 findings suggest that developers whose performance deviates from aspirations are likely to be movers, but higher performers (above aspiration) and lower performers have different motivations. Integrating insights from the platform literature on network effects, my findings also suggest that installed-base differences between the quantity- and quality-driven platforms also matter in shaping a developer’s mobility decision, and that the effects from such external market forces are complemented by the internal forces from performance feedback. Finally, the idea that platform mobility is not a one-time strategic choice, but rather, a sequence of (possible) choices owes its theoretical origin to the key theme of the behavioral theory of the firm—that organizations set, and then, adjust their goals, based on which to make (sequential) decisions to achieve such goals (Cyert and March, 1963). I further investigate another research question to reflect the performance-variance issue that concerns strategic management scholars: What are the effects of platform-mobility on a developer’s performance on the home platform and on the target platform? The importance of answering this question lays in the implications for a developer (as a complementor of the platform), as well as for the platform. At the developer level, performance implications due to movers’ platform-mobility decisions contribute to our understanding of firm performance heterogeneities in a platform market. However, more importantly, at the platform level, the performance increases or decreases that result from complementors moving across platforms has implications for how the market structure of two competing platforms will evolve over time. If compelementors are the source of competitive advantages of a platform, my findings suggest that in the battle for complementors, the quality-driven platform (iOS) has been losing competitive advantages to the quantity-driven platform (Android). I will unpack this argument in terms of complementors’ outflows and inflows of each platform. The impact from outflow can be 90 reflected by findings on how mobility affects performance on the home platform. I found that when iOS developers move to the competing platform, their home-platform performance deteriorates, and such erosion in performance is more severe for those who eventually abandoned iOS (i.e., switchers). The negative performance consequences thus suggest that iOS loses the competitive advantage due to outflow. The impact from inflows can be reflected by findings on how movers from a competing platform (Android) perform relative to local stayers. I found that Android-to-iOS multihomers, on average, perform worse than those who stayed on iOS, and that Android-to-iOS switchers have an even worse performance. In other words, the inflow from Android not only does not contribute to the competitive advantages of the iOS platform, but the action could potentially bring about competitive disadvantages. Conversely, on the Android platform, the outflow of complementors (both multihomers and switchers) enjoys better performance on the home platform due to cross-platform mobility. The inflow of complementors from iOS also tends to enhance its competitive advantages because iOS-to-Android multihomers tend to outperform Android stayers and that even iOS-to-Android switchers’ performance is on par with local stayers. In essence, the dynamics of complementor flows help to explain how iOS has been losing competitive advantages (of complementors) to Android over time. This trend also echoes Job’s earlier fear that Android might “rip off” iOS. A related point concerns the ways different types of movers (i.e., multihomers and switchers) affect a platform’s competitive advantages. One argument is that platforms need to discourage multihoming, as it could lead to switching platforms and thereby a loss of inimitable resources (i.e., valuable complementors) to competing platforms (Parker et al., 2016). My research suggests that such a preventive mechanism may not always be necessary and that its usage should depend on the particular group of movers. What could potentially hurt a platform’s 91 competitive advantage is when switchers, who tend to be lower performers but perhaps still valuable to the platform, make abandonment decisions. Questions for the two competing platforms then become: Who will lose more switchers to the other side? And, is it worthwhile for the platform to implement mechanisms in regards to such switchers to prevent their switching decisions (because, generally, they tend to be lower performers)? Complementors who are most valuable to a platform are higher performers, and the good news is that they tend to multihome intead of switching platforms. However, for platform-level competition, a question remains: Which platform could attract more attention (and resources) from multihomers? A multihomer’s relative investments on two competing platforms could indicate how they may “bet” on the future of the two platforms, and simultaneously indicate their platform-switching-decision directions if they face such strategic choices during a certain stage of market evolution. The second main contribution of this research is the theoretical integration of insights from the platform literature and the behavioral theory of the firm. The main theoretical argument that I draw from the platform literature is that of network effects, both cross-side (e.g., from customers to complementors) and same-side network effects (e.g., from complementors to other complementors). From behavioral theory, I borrow the ideas regarding performance feedback, goal setting (and switching), and sequence of strategic choices (Cyert and March, 1963). Such an integration is consistent with Porter’s (1981) observation that market structures and firm heterogeneities jointly determine firms’ strategic choices and performance. In my study, the market structure, according to the platform literature, is reflected in the installed-base difference (of software and hardware) between the quantity- and quality-driven platforms, and, firm heterogeneities, grounded in the behavioral theory of the firm, are reflected by performance feedback—above versus below aspirations—and the consequent two types of movers— 92 multihomers versus switchers. Third, the paper contributes to the platform literature in the following ways. First, while prior research in this literature has largely focused on platforms, I put platform complementors in the forefront and study an important strategic choice that they face: whether to subsequently adopt the competing platform. The second and related point is that by studying complementor flows across platforms, I can also derive implications for the platforms’ (relative) competitive advantages. This point essentially reflects two interrelated levels of the analysis: competition at the complementor level and at the platform level. Third, by differentiating quality-driven versus quantity-driven platforms, and by differentiating complementors (i.e., app developers in my context) in terms of higher (above aspiration) and lower (below aspiration) performers, the study sheds light on the issue of whether the two competing platforms will converge or diverge over time. Finally, by understanding whether the two competing platforms are converging or diverging over time, we can further predict whether the two platforms will coexist (if the two diverge) or whether we will witness the crowning of a single dominant platform (if the two converge). In other words, by studying complementors’ platform-mobility decisions, and thus flows of movers across platforms, my paper attempts to endogenize the emergence of a dominant technology (platform). Practical Implications This research has implications for platform complementors, such as app developers in the mobile app market. First, whether to move sequentially across competing platforms is a major strategic decision. A platform complementor needs to evaluate the potential costs of mulithoming (or switching). In some platform markets, such as the mobile app market, multihoming (and 93 switching) costs can be high. My field interviews suggest that, even when third-party tools to facilitate multihoming exist, companies still treat mobility as a major decision and, when they make the decision, will pay the costs of learning to use such tools. Hence, a complementor needs to weigh the potential benefits (of accessing different customer bases) versus the potential costs (of multihoming or switching) before making platform-mobility decisions. The second implication for complementors is that they need to understand their positions on the home platform, and, if they decide to move to the competing platform, what performance consequences to expect. My findings suggest that the implications for higher performers are different than those for lower performers, and are different for iOS-to-Android movers versus Android-to-iOS movers. When iOS higher performers move, they are likely to end up multihoming; they must prepare for a short-term performance loss on the home platform; but they are also likely to perform well on the target platform. For iOS lower performers, however, once they have made mobility decisions, they are likely to go down the path of suffering severe performance loss on the home platform: namely, switching to the target platform while paying high switching costs during the process, having lower than average performance on the target platform, and the eventual possibility of fully withdrawing from the platform market. Android movers, however, tend to enjoy performance increases on their home platform as a result of moving to the competing platform, perhaps because the learning they accrue in the more demanding iOS environment. Further, both Android multihomers and switchers are not expected to outperformance average players on iOS. My research also has practical implications for platform owners and sponsors. First, given the mobility patterns of high versus low performers, it seems that platform owners do not need to worry about the possibilities of significantly losing competitive advantages. However, 94 when we pay attention to performance consqeuences due to platform mobility, we could notice that, over time, the quantity-driven platform (Android) has been accumulating more competitive advantages from developer migration, mainly through learning. Finally, complementors’ mobility across platforms, the eventual success of multihomers, and the failure of switchers suggest a self-adjusting process of the market—that is, the retention of high-quality complementors while eliminating lower performers. Although platform developers do not need to worry much about switchers, they do need to pay attention to multihomers, because their relative resource allocations, to some extent, will affect the long-term evolution of the market structure. Since no one (complementor) wants to be on the wrong side of history, platforms need to interpret from complementors’ relative investments to foresee their long-term prospects and make strategic moves accordingly. Finally, the quality- versus quantity-driven strategy at the platform level has respective merits and shortfalls. The quality-driven strategy could ensure customer satisfaction, but at the same time it faces the danger of losing its competitiveness because the strong network effect from the competing platform may attract the source of its competitive advantages (high-quality complementors). In contrast, the quantity-driven strategy could attract a larger number of complementors, but at the same time face the danger of losing its appeal to high-quality complementors if the diversity of complementary products dilutes its value. My conclusion is that, although the quality- versus quantity-driven strategy is at odds, platforms could sequentially emphasize one of the two as the market evolves. That is, a quantity-driven platform could start to emphasize quality in order to mitigate the threat that quality diversity may drive out high-quality complementors. For instance, the key sponsor of the quantity-driven Android platform, Google, has started to address the hardware fragmentation issue by introducing its own version of a 95 smartphone (Pixel). Similarly, a quality-driven platform can gradually relax its restrictive rules so that it could better retain its complementors and prevent being on the losing side of a “winner- take-all” situation. With respect to this, Apple announced at its 2017 WWDC that it will reduce the timeframe of reviewing app submissions. Limitations and Future Research My study has several limitations. First, I used proprietary data from the analytics company Optimus for Android apps’ release dates that are proxied by the date of the earliest version of an Android app. A reasonable concern is that the earliest version recorded in the company’s database may not be an Android app’s first version. That is, the earliest version date may not always be an app’s release date. Indeed, this is a limitation of the research, but one that is unlikely to be fully addressed, because the analytics company from which I obtained the Android app data is likely to be one of the industry pioneers in such comprehensively collected data. From a researcher’s standpoint, unless he or she has access to Android app data directly from Google, this data limitation cannot be fully addressed. This limitation will affect my empirical tests in the following way. Because it is likely that I have coded a developer’s entry timing to the Android platform in a delayed manner, I may have over-counted the number of movers from iOS to Android, while at the same time under-counted movers from Android to iOS. However, a general norm in the mobile app market is that developers tend to adopt iOS first and then move onto Android. This phenomenon is specific to the setting, and thus somewhat mitigates the seriousness of the limitation. The second limitation relates to matching developers across iOS and Android. Specifically, the set of developers for which I found matches on both platforms are not likely to 96 be 100% accurate or 100% comprehensive. As I have described in detail in the section on matching algorithms, the matching approach attempts to strike a balance between comprehensiveness (if I use the least-restrictive decision rules when selecting matching developer pairs) and accuracy (if I constrain the sample to the most-restrictive decision rules). Although such a limitation is unavoidable, per app analytics company experts, the theoretical predictions of this paper have validity as long as the findings hold across samples selected based on different decision rules. Finally, another limitation of the paper is the big data used in this study. Mobile apps cover almost all aspects of our daily life and span many industry settings. As a result, not all apps are created with the same purpose. Consistent with my focus on a strategy audience, I focus on profitability as a major goal of app developers. However, as the behavioral theory of the firm suggests, there could be many other goals when an organization or individual releases applications. Consequently, the performance metric used in this study—ranking in the grossing categories—may not be generalizable to all sorts of goals. This provides an opportunity for future research. An obvious and direct remedy to this limitation is to use all types of app rankings instead of just the grossing category. Nevertheless, although worth pursing, that approach would face the challenge of sacrificing the focus on a single motivation for the purpose of generalizability. Conclusion In this study, I attempt to address a central topic in strategy and management research: competition between dominant technologies. The intense rivalry between two major mobile platforms of the recent business history—iOS and Android—provides an ideal setting to examine 97 this question. To do that, I integrate insights from the literature on platform-based competition and from the performance feedback component of the behavioral theory of the firm. My research focused on the important role of platform complementors (or adopters of technologies) in shaping the competition between two dominant platforms (technologies) with different strategic emphases—one quality-driven (iOS), the other quantity-driven (Android). Key empirical insights are that platform complementors move to competing platforms because of different motivations, and that platform mobility does influence their ex-post performance. An implication for platform-level competition is that, although platforms do not need to worry about complementors’ mobility in the short term (because higher performers tend to multihome, while lower performers tend to switch), they need to pay more attention to how multihomers could shape their long-term competitive advantages in the battle to become the dominant technology. 98 TABLE 1: Descriptive Statistics of iOS Sample Mean S.D. Min Max 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 1 Mobility (1/0) 0 0.05 0 1 1.00 2 Abandon Platform (1/0) 0 0.02 0 1 0.09 1.00 3 Average # Days Listed on Grossing Ranking in the Month 0.39 3.9 0 303 0.01 0.00 1.00 4 Above Aspiration 0.26 2.81 0 292.02 0.01 0.00 0.99 1.00 5 Below Aspiration 0.26 0.31 0 59.39 0.03 0.02 -0.08 -0.08 1.00 6 iOS Cumulative Smartphone Shipments 5.18E+08 2.23E+08 1.35E+07 8.41E+08 -0.03 -0.02 -0.04 -0.04 -0.28 1.00 7 Android Cumulative Smartphone Shipments 1960000000 1100000000 1220000 3.63E+09 -0.03 -0.02 -0.04 -0.04 -0.34 1.00 1.00 8 iOS Installed Base of Apps 779933.39 371680.98 562 1350000 -0.03 -0.02 -0.03 -0.03 -0.21 1.00 1.00 1.00 9 iOS Installed Base of Developers 1.92E+05 88153.65 385 3.24E+05 -0.02 -0.02 -0.03 -0.03 -0.19 1.00 1.00 1.00 1.00 10 Android Installed Base of Apps 1330000 797442.74 1 2390000 -0.03 -0.02 -0.04 -0.04 -0.33 0.99 0.99 0.99 0.99 1.00 11 Android Installed Base of Developers 327250.73 182418.5 1 582392 -0.03 -0.02 -0.04 -0.04 -0.32 1.00 1.00 1.00 1.00 1.00 1.00 12 Installed-Base Difference in Shipments in Billions (Android-iOS) 1.44 0.88 -0.03 2.79 -0.03 -0.02 -0.04 -0.04 -0.35 0.99 1.00 1.00 1.00 0.99 1.00 1.00 13 Installed-Base Difference in Apps in Millions (Android-iOS) 0.44 0.4 -0.13 0.88 -0.03 -0.02 -0.03 -0.04 -0.27 0.96 0.98 0.97 0.96 0.99 0.98 0.98 1.00 14 Installed-Base Difference in Developers in Millions (Android-iOS) 0.11 0.08 -0.02 0.21 -0.03 -0.02 -0.03 -0.03 -0.25 0.98 0.99 0.98 0.98 1.00 1.00 0.99 0.99 1.00 15 iOS Major OS Update (1/0) 0.1 0.3 0 1 0.00 0.00 0.00 0.00 -0.03 0.07 0.08 -0.07 -0.07 0.07 0.07 0.08 -0.05 -0.06 1.00 16 iOS Version Update (1/0) 0.64 0.48 0 1 -0.01 0.00 0.00 0.00 -0.02 0.20 0.21 0.09 0.10 0.21 0.21 0.21 0.13 0.12 0.25 1.00 17 Android Major OS Update (1/0) 0.07 0.26 0 1 -0.01 -0.01 -0.02 -0.02 -0.16 0.18 0.19 0.02 0.02 0.18 0.18 0.20 0.06 0.04 -0.01 0.21 1.00 18 Android Version Update (1/0) 0.41 0.49 0 1 0.00 0.00 -0.01 -0.01 -0.09 -0.09 -0.08 -0.08 -0.08 -0.07 -0.08 -0.08 -0.04 -0.06 -0.05 0.16 0.34 1.00 19 Portfolio Size 4.06 14.65 1 3095 0.00 0.00 0.00 0.00 -0.02 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 1.00 20 App Rating 3.67 1.12 0 5.00E+00 0.00 0.00 0.02 0.02 0.02 0.16 0.16 0.16 0.17 0.15 0.16 0.15 0.14 0.15 0.01 0.03 0.01 -0.02 -0.06 1.00 21 Price of Paid Apps 4.57 23.81 0.99 999.99 0.00 0.00 0.03 0.02 -0.01 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.00 0.00 0.00 0.00 -0.01 0.01 1.00 22 Number of Ratings 404.09 7504.73 1 1710000 0.00 0.00 0.11 0.11 0.01 -0.03 -0.02 -0.03 -0.03 -0.02 -0.02 -0.02 -0.02 -0.02 0.00 0.00 0.00 0.00 -0.01 0.00 0.00 1.00 23 Diversification Index (Herfindahl Index) 0.22 0.13 0 0.69 0.00 0.00 0.02 0.02 0.02 -0.10 -0.10 -0.09 -0.09 -0.10 -0.10 -0.10 -0.09 -0.09 -0.01 -0.02 -0.01 0.01 0.01 -0.02 0.00 0.00 1.00 99 TABLE 2A: Main Effects of Aspirations and Installed Bases on the Likelihood of iOS-to-Android Mobility (iOS Sample) (Dependent Variable: Mobility, 1/0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Above Aspiration (grossing rank) 0.016*** 0.017*** 0.011*** 0.011*** 0.012*** (0.00) (0.00) (0.00) (0.00) (0.00) Below Aspiration (grossing rank) 0.197*** 0.198*** 0.003 0.025 0.124*** (0.00) (0.00) (0.03) (0.03) (0.01) Installed-Base Difference in Shipments (Android-iOS) 1.653*** 1.660*** (0.14) (0.14) Installed-Base Difference in Shipments 2 -1.552*** -1.550*** (0.04) (0.04) Installed-Base Difference in Developers (Android-iOS) 4.089*** 4.205*** (0.59) (0.59) Installed-Base Difference in Developers 2 -116.947*** -116.336*** (2.31) (2.37) Installed-Base Difference in Apps (Android-iOS) -0.092 0.003 (0.09) (0.09) Installed-Base Difference in Apps 2 -4.329*** -4.256*** (0.10) (0.10) Continuous Time from 2008-07 (Period 1) -0.009*** -0.009*** -0.010*** -0.010*** 0.046*** 0.056*** 0.053*** 0.045*** 0.054*** 0.049*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) App Portfolio Size 0.000 0.000 0.000 0.000 0.001 0.001 0.001 0.001 0.001 0.001 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Diversification Index 0.084 0.078 0.116+ 0.111+ 0.024 0.029 0.033 0.020 0.024 0.033 (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) iOS Version Update (1/0) -0.065*** -0.065*** -0.074*** -0.075*** -0.058*** -0.076*** -0.068*** -0.058*** -0.077*** -0.072*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Android Version Update (1/0) -0.323*** -0.321*** -0.302*** -0.299*** -0.120*** -0.145*** -0.171*** -0.119*** -0.143*** -0.162*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Developer-Month Observations 5.6e+06 5.6e+06 5.6e+06 5.6e+06 5.6e+06 5.5e+06 5.5e+06 5.6e+06 5.5e+06 5.5e+06 Developers 270422 270422 270422 270422 270422 251622 251622 270422 251622 251622 iOS-to-Android Movers 17577 17577 17577 17577 17577 17577 17577 17577 17577 17577 Pseudo-R-squared .0019 .0023 .0038 .0042 .026 .022 .020 .027 .023 .020 Log Likelihood -208143 -208070 -207749 -207667 -203022 -203326 -203839 -202992 -203295 -203791 Notes: (1) Robust standard errors, clustered at developers, are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include: iOS-to-Android movers identified based on url-matching and iOS stayers. 100 TABLE 2B: Interaction Effects between Aspirations and Relative Installed Bases (iOS Sample) (Dependent Variable: Mobility, 1/0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Above Aspiration (grossing rank) 0.029*** 0.030*** 0.015*** 0.014*** 0.011*** 0.011*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Below Aspiration (grossing rank) 0.109 0.112 -0.124** -0.075+ -0.030 0.017 (0.09) (0.08) (0.04) (0.04) (0.03) (0.03) Difference in Smartphone Shipments (Android-iOS) 1.675*** 1.702*** 1.845*** 1.844*** (0.14) (0.14) (0.19) (0.19) Difference in Smartphone Shipments 2 -1.558*** -1.564*** -1.673*** -1.653*** (0.04) (0.04) (0.06) (0.06) Installed-Base Difference in Developers (Android -iOS) 4.212*** 4.372*** 4.259*** 4.661*** (0.59) (0.59) (0.89) (0.86) Installed-Base Difference in Developers 2 -117.366*** -117.801*** -125.869*** -125.493*** (2.31) (2.32) (3.58) (3.39) Installed-Base Difference in Apps (Android -iOS) -0.074 -0.043 0.293* 0.389** (0.09) (0.09) (0.15) (0.14) Installed-Base Difference in Apps 2 -4.347*** -4.373*** -4.952*** -4.972*** (0.10) (0.10) (0.16) (0.15) Above Aspiration X Difference in Shipments -0.044*** -0.046*** (0.01) (0.01) Above Aspiration X Difference in Shipments 2 0.020*** 0.021*** (0.00) (0.00) Below Aspiration X Difference in Shipments -0.610** -0.486** (0.20) (0.18) Below Aspiration X Difference in Shipments 2 0.349*** 0.292*** (0.08) (0.07) Above Aspiration X Installed-Base Difference in Developers -0.267*** -0.285*** (0.05) (0.06) Above Aspiration X Installed-Base Difference in Developers 2 1.635*** 1.785*** (0.33) (0.33) Below Aspiration X Installed-Base Difference in Developers -2.479* -2.311* (1.15) (1.04) Below Aspiration X Installed-Base Difference in Developers 2 26.546*** 24.500*** (5.60) (5.08) Above Aspiration X Installed-Base Difference in Apps -0.047*** -0.053*** (0.01) (0.01) Above Aspiration X Installed-Base Difference in Apps 2 0.075*** 0.084*** (0.01) (0.01) Below Aspiration X Installed-Base Difference in Apps -0.918*** -0.920*** (0.21) (0.18) Below Aspiration X Installed-Base Difference in Apps 2 1.725*** 1.665*** (0.26) (0.23) Performance-Aspiration 0.013*** 0.029*** 0.015*** 0.011*** (0.00) (0.00) (0.00) (0.00) Performance-Aspiration X Installed-Base Difference in Shipment -0.042*** (0.01) Performance-Aspiration X Installed-Base Difference in Shipment 2 0.019*** (0.00) Performance-Aspiration X Installed-Base Difference in Developers -0.258*** (0.05) Performance-Aspiration X Installed-Base Difference in Developers 2 1.548*** (0.33) Performance-Aspiration X Installed-Base Difference in Apps -0.044*** (0.01) Performance-Aspiration X Installed-Base Difference in Apps 2 0.069*** (0.01) Continuous Time from 2008-07 (Period 1) -0.009*** 0.045*** 0.055*** 0.053*** 0.045*** 0.046*** 0.044*** 0.055*** 0.060*** 0.057*** 0.052*** 0.052*** 0.050*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) App Portfolio Size 0.000 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Diversification Index 0.079 0.020 0.024 0.029 0.020 0.021 0.019 0.025 0.026 0.024 0.029 0.035 0.033 (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) iOS Version Update (1/0) -0.065*** -0.057*** -0.076*** -0.068*** -0.058*** -0.054*** -0.056*** -0.076*** -0.072*** -0.073*** -0.068*** -0.071*** -0.072*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Android Version Update (1/0) -0.323*** -0.119*** -0.144*** -0.170*** -0.118*** -0.122*** -0.119*** -0.143*** -0.151*** -0.146*** -0.169*** -0.164*** -0.159*** (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Developer-Month Observations 5.6e+06 5.6e+06 5.5e+06 5.5e+06 5.6e+06 5.6e+06 5.6e+06 5.5e+06 5.5e+06 5.5e+06 5.5e+06 5.5e+06 5.5e+06 Developers 270422 270422 251622 251622 270422 270422 270422 251622 251622 251622 251622 251622 251622 iOS-to-Android Movers 17577 17577 17577 17577 17577 17577 17577 17577 17577 17577 17577 17577 17577 Pseudo-R-squared .0021 .027 .023 .02 .027 .026 .027 .023 .023 .023 .020 .020 .020 Log Likelihood -208106 -202985 -203286 -203798 -202984 -203005 -202971 -203284 -203293 -203256 -203794 -203787 -203739 Notes: (1) Robust standard errors, clustered at developers, are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include: iOS-to-Android movers identified based on URL-matching and iOS stayers. (4) All models are stratified at developer-level (i.e. baseline hazard is developer-specific). (5) Models 1-4 test interaction effects between the performance-aspiration variable and installed-base variables. (6) Models 5-7 test aspiration variables’ interactions with Difference in Smartphone Shipments; Models 8-10 test aspiration variables’ interactions with Installed-base Difference in Developers; Models 11-13 test aspirations’ interactions with Installed-base Difference in Apps. 101 TABLE 3: Hazard Models Predicting Abandoning Home (iOS) Platform, Conditional on Multihoming (iOS Sample) i.e., Prob(Abandonment | Multihome) (Dependent Variable = Abandon Platform, 1/0) (1) (2) (3) Above Aspiration -0.035*** -0.029*** (0.01) (0.01) Below Aspiration 0.294*** 0.188* (0.06) (0.08) Continuous Time from 2008-07 (Period 1) -0.059 -0.064 -0.061 (0.06) (0.06) (0.06) Installed-Base Difference in Developers (Android-iOS) 43.864*** 45.392*** 44.602*** (9.46) (9.52) (9.49) Installed-Base Difference in Developers 2 -342.071*** -343.622*** -342.525*** (14.61) (14.65) (14.63) App Portfolio Size -0.013*** -0.012*** -0.013*** (0.00) (0.00) (0.00) Diversification Index -0.210 -0.231 -0.206 (0.17) (0.17) (0.17) iOS Version Update (1/0) 0.105* 0.106* 0.106* (0.05) (0.05) (0.05) Android Version Update (1/0) 0.430*** 0.431*** 0.431*** (0.05) (0.05) (0.05) Developer-Month Observations 327179 327179 327179 Developers 18037 18037 18037 Switchers 2214 2214 2214 Pseudo-R-squared .073 .073 .073 Log Likelihood -19262 -19271 -19260 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) The sample includes: (a) multihomers who do not abandon platform and (b) switchers who eventually abandon home platform. 102 TABLE 4: Negative Binomial Models, with Developer Fixed-Effects, Estimating iOS Movers’ Performance on the Home Platform (iOS) (Dependent Variable = Monthly Average Times Listed on Top Grossing Ranking) (1) (2) (3) (4) All Movers All Movers Multihomers Switchers After Mobility (1/0) -0.039* -0.017 -0.021 -0.132* (0.02) (0.02) (0.02) (0.05) After Mobility X Switcher -0.163*** (0.04) Continuous Time from 2008-07 (Period 1) 0.052*** 0.051*** 0.059*** 0.019 (0.01) (0.01) (0.01) (0.03) Installed-Base Difference in Developers (Android-iOS) -9.080*** -8.958*** -9.879*** -7.217 (1.55) (1.55) (1.65) (4.65) Installed-Base Difference in Developers 2 -14.176*** -15.253*** -16.434*** 5.210 (2.92) (2.93) (3.04) (11.39) App Portfolio Size 0.034*** 0.033*** 0.031*** 0.136*** (0.00) (0.00) (0.00) (0.02) Diversification Index 0.307** 0.302** 0.169 1.350*** (0.10) (0.10) (0.11) (0.35) iOS Version Update (1/0) 0.003 0.003 0.009 -0.036 (0.01) (0.01) (0.01) (0.04) Android Version Update (1/0) 0.009 0.009 0.001 0.057 (0.01) (0.01) (0.01) (0.04) Developer-Month Observations 30547 30547 25991 4556 Developers 2860 2860 2404 456 Log Likelihood -43410 -43399 -37840 -5511 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) The sample includes: (a) multihomers who do not abandon platform and (b) switchers who eventually abandon home platform. (4) Models 1-2 include all multihomers and switchers while Models 3-4 split the sample into each of the two groups. 103 TABLE 5: Poisson Models, with Developer Random-Effects, Estimating iOS Movers’ Performance on the Target Platform (Android) (Dependent Variable = Monthly Average Times Listed on Top Grossing Ranking) (1) iOS-to-Android multihomers 0.932*** (0.10) iOS-to-Android switchers 0.072 (0.22) Android-to-iOS multihomers 0.884*** (0.12) Android-to-iOS switchers -0.452+ (0.25) Month Fixed Effects Yes Developer-Month Observations 5.9e+06 Developers 400088 Log Likelihood -680940 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include developers on the target platform (Android) and were categorized into 5 groups as indicated in the regression model. The group—Android Stayers—is set as the baseline for comparison. (4) For iOS-to-Android movers, including both multihomers and switchers, I only used 6 months of their observations in the post-mobility period to account for short-term performance change. 104 TABLE 6: Descriptive Statistics of Android Sample Mean S.D. Min Max 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 1 Mobility 0 0.03 0 1 1.00 2 Abandon Platform (1/0) 0 0.01 0 1 0.15 1.00 3 Average # Days Listed on Grossing Ranking in the Month 0.19 1.97 0 143 0.00 0.00 1.00 4 Above Aspiration 0.14 1.49 0 126.91 0.00 0.00 0.99 1.00 5 Below Aspiration 0.15 0.1 0 19.84 0.00 0.00 -0.14 -0.14 1.00 6 iOS Cumulative Smartphone Shipments 5.62E+08 1.85E+08 1.35E+07 9.08E+08 -0.02 -0.01 0.00 -0.01 -0.01 1.00 7 Android Cumulative Smartphone Shipments 2160000000 947000000 1220000 3.94E+09 -0.02 -0.01 0.00 -0.01 -0.01 1.00 1.00 8 iOS Installed Base of Apps 864233.71 314267.66 1946 1350000 -0.01 -0.01 0.00 -0.01 -0.01 1.00 1.00 1.00 9 iOS Installed Base of Developers 2.13E+05 73614.69 1057 3.24E+05 -0.01 -0.01 0.00 -0.01 0.00 1.00 1.00 1.00 1.00 10 Android Installed Base of Apps 1490000 690591.3 1 2440000 -0.02 -0.01 0.00 -0.01 0.01 0.99 0.99 0.99 0.99 1.00 11 Android Installed Base of Developers 362335.54 155692.98 1 595956 -0.02 -0.01 0.00 -0.01 0.00 1.00 1.00 1.00 1.00 1.00 1.00 12 Installed-Base Difference in Shipments in Billions (Android -iOS) 1.6 0.76 -0.03 3.03 -0.02 -0.01 0.00 -0.01 0.00 1.00 1.00 1.00 1.00 0.99 1.00 1.00 13 Installed Base Difference in Apps in Millions (Android -iOS) 0.53 0.35 -0.13 0.88 -0.01 -0.01 0.00 0.00 0.02 0.97 0.97 0.97 0.97 0.99 0.98 0.97 1.00 14 Installed Base Difference in Developers in Millions (Android -iOS) 0.13 0.07 -0.02 0.21 -0.01 -0.01 0.00 -0.01 0.01 0.99 0.99 0.99 0.99 1.00 1.00 0.99 0.99 1.00 15 iOS Major OS update (1/0) 0.1 0.3 0 1 0.00 0.00 0.01 0.01 0.09 0.06 0.07 -0.07 -0.07 0.06 0.07 0.08 -0.04 -0.06 1.00 16 iOS OS update (1/0) 0.66 0.47 0 1 0.00 0.00 0.00 0.01 0.03 0.18 0.18 0.07 0.07 0.18 0.18 0.18 0.11 0.10 0.24 1.00 17 GP Major OS update (1/0) 0.07 0.26 0 1 -0.01 0.00 0.00 -0.01 -0.06 0.21 0.23 0.07 0.07 0.21 0.21 0.23 0.11 0.09 -0.05 0.20 1.00 18 GP OS update (1/0) 0.4 0.49 0 1 0.00 0.00 0.00 -0.01 -0.03 -0.03 -0.02 -0.01 -0.01 0.00 -0.01 -0.02 0.03 0.01 -0.01 0.24 0.34 1.00 19 Portfolio Size 4.22 20.22 1 3170 0.00 0.00 0.00 0.00 -0.01 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.00 0.01 0.01 0.00 1.00 20 App Rating 3.92 0.84 1 5 0.00 0.00 0.01 0.01 0.00 0.11 0.11 0.11 0.11 0.11 0.11 0.10 0.11 0.11 0.01 0.02 0.02 0.00 0.00 1.00 21 Highest App Rank 1050000 642448.79 5 2.55E+06 -0.01 0.00 -0.09 -0.09 0.00 -0.01 0.00 -0.02 -0.02 -0.01 -0.01 0.00 -0.03 -0.02 0.01 0.00 0.01 0.01 -0.01 -0.19 1.00 22 App Downloads 45348.03 2660000 0 1E+09 0.00 0.00 0.06 0.06 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.03 1.00 105 TABLE 7A: Main Effects of Aspirations and Installed Bases on the Likelihood of Android-to-iOS Mobility (Android Sample) (Dependent Variable: Mobility, 1/0) (1) (2) (3) (4) (5) (6) (7) Above Aspiration (grossing rank) 0.051*** 0.052*** (0.01) (0.01) Below Aspiration (grossing rank) 0.264** 0.291*** (0.10) (0.08) Installed-Base Difference in Shipments (Android-iOS) 2.589*** (0.34) Installed-Base Difference in Shipments 2 -0.680*** (0.09) Installed-Base Difference in Developers (Android -iOS) 6.681*** (1.58) Installed-Base Difference in Developers 2 -17.850*** (5.17) Installed-Base Difference in Apps (Android -iOS) 1.152*** (0.24) Installed-Base Difference in Apps 2 -0.580** (0.22) Continuous Time from 2008-07 (Period 1) 0.018*** -0.000 -0.001 -0.001 -0.042*** -0.004 -0.004 (0.00) (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) App Portfolio Size -0.002 -0.007* -0.007* -0.007* -0.002 -0.002 -0.002 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Average Downloads 0.000 -0.000 0.000 -0.000 0.000 0.000 0.000 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) iOS Version Update (1/0) -0.012 -0.070 -0.071+ -0.073+ -0.030 -0.020 -0.015 (0.03) (0.04) (0.04) (0.04) (0.03) (0.03) (0.03) Android Version Update (1/0) -0.204*** -0.162*** -0.159*** -0.157*** -0.106** -0.166*** -0.185*** (0.03) (0.04) (0.04) (0.04) (0.04) (0.03) (0.03) Developer-Month observations 6.5e+06 4.4e+06 4.4e+06 4.4e+06 6.5e+06 6.5e+06 6.5e+06 Developers 358503 357057 357057 357057 358503 358503 358503 Android-to-iOS Movers 4409 2963 2963 2963 4409 4409 4409 Pseudo-R-squared .0018 .0009 .00051 .00098 .0024 .002 .0021 Log Likelihood -54573 -34952 -34966 -34950 -54545 -54565 -54559 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include: Android-to-iOS movers identified based on URL-matching and Android stayers. 106 TABLE 7B: Interaction Effects between Aspirations and Relative Installed Bases (Android Sample) (Dependent Variable: Mobility, 1/0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Above Aspiration (grossing rank) -0.104 -0.041 -0.054 -0.030 -0.039 -0.064 (0.19) (0.19) (0.12) (0.12) (0.08) (0.08) Below Aspiration (grossing rank) 16.300* 12.285* 3.995 3.736+ -5.952 -3.945 (7.14) (6.17) (2.52) (2.13) (4.17) (3.95) Installed-Base Difference in Shipments (Android-iOS) 62.274*** 61.796*** 68.232*** 64.408*** (5.55) (5.55) (6.48) (6.25) Installed-Base Difference in Shipments 2 -5.633*** -5.601*** -7.043*** -6.570*** (0.37) (0.37) (0.66) (0.61) Installed-Base Difference in Developers (Android-iOS) 98.820*** 98.595*** 108.814*** 107.658*** (7.97) (7.98) (10.42) (9.80) Installed-Base Difference in Developers 2 23.420 23.644 -20.821 -18.843 (41.24) (41.28) (49.64) (47.41) Installed-Base Difference in Apps (Android-iOS) 2.315* 2.296* -1.687 -0.356 (0.95) (0.95) (2.96) (2.85) Installed-Base Difference in Apps 2 17.568*** 17.538*** 21.337*** 19.907*** (1.61) (1.61) (3.15) (3.05) Above Aspiration X Difference in Shipments 0.234 0.151 (0.26) (0.26) Above Aspiration X Difference in Shipments 2 -0.086 -0.060 (0.09) (0.09) Below Aspiration X Difference in Shipments -22.270* -16.491* (9.20) (8.00) Below Aspiration X Difference in Shipments 2 7.116* 5.259* (2.80) (2.45) Above Aspiration X Difference in Developers 1.876 1.558 (1.89) (1.85) Above Aspiration X Difference in Developers 2 -7.770 -6.679 (7.25) (7.12) Below Aspiration X Difference in Developers -49.038 -44.498 (38.33) (32.07) Below Aspiration X Difference in Developers 2 156.664 140.410 (137.97) (115.97) Above Aspiration X Difference in Apps 0.380 0.452 (0.29) (0.31) Above Aspiration X Difference in Apps 2 -0.362 -0.412 (0.27) (0.28) Below Aspiration X Difference in Apps 18.074 12.307 (13.23) (12.27) Below Aspiration X Difference in Apps 2 -13.222 -9.020 (10.49) (9.57) Performance-Aspiration 0.050*** -0.122 -0.063 -0.040 (0.01) (0.19) (0.12) (0.08) Performance-Aspiration X Difference in Shipments 0.260 (0.26) Performance-Aspiration X Difference in Shipments 2 -0.095 (0.09) Performance-Aspiration X Difference in Developers 1.990 (1.94) Performance-Aspiration X Difference in Developers 2 -8.178 (7.41) Performance-Aspiration X Difference in Apps 0.381 (0.29) Performance-Aspiration X Difference in Apps 2 -0.365 (0.27) Continuous Time from 2008-07 (Period 1) -0.000 -3.238*** -0.817*** -0.941*** -3.211*** -3.348*** -3.180*** -0.815*** -0.794*** -0.789*** -0.938*** -0.976*** -0.956*** (0.01) (0.34) (0.11) (0.06) (0.34) (0.39) (0.38) (0.11) (0.11) (0.11) (0.06) (0.07) (0.07) App Portfolio Size -0.007* -0.007* -0.007* -0.007* -0.007* -0.007** -0.007* -0.007* -0.007* -0.007* -0.007* -0.007* -0.007* (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Average Downloads -0.000 -0.000 -0.000 -0.000 -0.000 0.000 -0.000 -0.000 0.000 -0.000 -0.000 0.000 -0.000 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) iOS Version Update (1/0) -0.069 -0.006 -0.103* -0.012 -0.006 -0.004 -0.008 -0.104* -0.111* -0.113** -0.013 0.009 -0.002 (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05) (0.05) GP Version Update (1/0) -0.163*** 0.176*** 0.195*** 0.239*** 0.177*** 0.184*** 0.183*** 0.196*** 0.206*** 0.208*** 0.240*** 0.222*** 0.230*** (0.04) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) Developer-Month Observations 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 4.4e+06 Developers 357057 357057 357057 357057 357057 357057 357057 357057 357057 357057 357057 357057 357057 Android-to-iOS Movers 2963 2963 2963 2963 2963 2963 2963 2963 2963 2963 2963 2963 2963 Pseudo-R-squared .00088 .0046 .0032 .0054 .0046 .0043 .0047 .0033 .0029 .0034 .0054 .005 .0054 Log Likelihood -34953 -34822 -34870 -34795 -34822 -34833 -34819 -34870 -34882 -34865 -34795 -34809 -34794 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include: Android-to-iOS movers identified based on URL-matching and Android stayers. (4) Models 1-4 test interaction effects between the performance-aspiration variable and installed-base variables. (5) Models 5-7 test aspiration variables’ interactions with Installed-Base Difference in Smartphone Shipments; Models 8-10 test aspiration variables’ interactions with Installed-Base Difference in Developers; Models 11-13 test aspirations’ interactions with Installed-Base Difference in Apps. 107 TABLE 8: Hazard Models Predicting Abandoning Home Platform (Android), Conditional on Multihoming (Android Sample); i.e., Prob(Abandonment | Multihome) (Dependent Variable = Abandon Platform, 1/0) (1) (2) (3) Above Aspiration (grossing rank) -0.033 -0.032 (0.03) (0.03) Below Aspiration (grossing rank) 0.173 0.085 (0.33) (0.39) Continuous Time from 2008-07 (Period 1) -0.552*** -0.522*** -0.540*** (0.15) (0.16) (0.16) Installed-Base Difference in Developers (Android-iOS) 103.205*** 101.118*** 102.335*** (15.18) (15.51) (15.67) Installed-Base Difference in Developers 2 -140.632** -147.327** -143.371** (47.93) (49.08) (49.51) App Portfolio Size 0.003*** 0.003*** 0.003*** (0.00) (0.00) (0.00) iOS Version Update (1/0) -0.216** -0.213** -0.215** (0.08) (0.08) (0.08) Android Version Update (1/0) 0.085 0.080 0.084 (0.09) (0.09) (0.09) Developer-Month Observations 112615 112615 112615 Developers 9189 9189 9189 Switchers 1133 1133 1133 Pseudo-R-squared .0054 .0053 .0054 Log Likelihood -9461 -9462 -9461 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) The sample includes: (a) multihomers who do not abandon home platform and (b) switchers who eventually abandon platform. 108 TABLE 9: OLS Models, with Developer Fixed-Effects, Estimating Performance on the Home Platform (Android) (Dependent Variable = Logged Monthly Average Times Listed on Top Grossing Ranking) (1) (2) (3) (4) All Movers All Movers Multihomers Switchers After Mobility (1/0) 0.009** 0.015*** 0.010* 0.009+ (0.00) (0.00) (0.00) (0.01) After Mobility X Switcher -0.022*** (0.01) Continuous Time from 2008-07 (Period 1) -0.032*** -0.032*** -0.039*** -0.008 (0.00) (0.00) (0.00) (0.01) Installed-Base Difference in Developers (Android-iOS) 2.068*** 2.086*** 2.540*** 0.677 (0.33) (0.33) (0.40) (0.49) Installed-Base Difference in Developers 2 8.661*** 8.687*** 11.269*** 0.227 (1.59) (1.59) (1.96) (2.27) App Portfolio Size -0.000 -0.000 -0.000 0.000 (0.00) (0.00) (0.00) (0.00) Average Downloads 0.000*** 0.000*** 0.000*** -0.000*** (0.00) (0.00) (0.00) (0.00) iOS Version Update (1/0) 0.013*** 0.013*** 0.016*** 0.004 (0.00) (0.00) (0.00) (0.00) Android Version Update (1/0) -0.011*** -0.011*** -0.013*** -0.003 (0.00) (0.00) (0.00) (0.00) Developer-Month Observations 43062 43062 33190 9872 Developers 6634 6634 5186 1448 Log Likelihood 18771 18781 12282 7553 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) The sample includes: (a) multihomers who do not abandon platform and (b) switchers who eventually abandon platform. (4) Models 1-2 include all multihomers and switchers while Models 3-4 split the sample into each of the two groups. 109 TABLE 10: Poisson Models, with Developer Random-Effects, Estimating Android Movers’ Performance on the Target Platform (iOS) (Dependent Variable = Monthly Average Times Listed on Top Grossing Ranking) (1) Android-to-iOS multihomers -0.170* (0.07) Android-to-iOS switchers -0.686*** (0.14) iOS-to-Android multihomers 0.803*** (0.05) iOS-to-Android switchers -0.194* (0.09) Month Fixed Effects Yes Developer-Month Observations 6.4e+06 Developers 295551 Log Likelihood -1.9e+06 Notes: (1) Standard errors are in parentheses. (2) + p<0.10 * p<0.05 ** p<0.01 *** p<0.001. (3) Sample include developers on the target platform (Android) and were categorized into 5 groups as indicated in the regression model. The group—Android Stayers—is set as the baseline for comparison. (4) For Android-to-iOS movers, including both multihomers and switchers, I only used 6 months of their observations in the post-mobility period to account for short-term performance change. 110 FIGURES 1a and 1b: Histogram of Average Rating of App Portfolio (iOS is more quality-driven) 0 .2 .4 .6 Density 0 1 2 3 4 5 portfolio_rating Histogram of iOS Developers' App-Portfolio Rating 0 .1 .2 .3 .4 .5 Density 1 2 3 4 5 portfolio_rating Histogram of GP Developers' App-Portfolio Rating 111 FIGURE 2: Installed Bases of Apps Over Time (Android is more quantity-driven) 112 FIGURES 3a and 3b: iOS-to-Android Mobility FIGURES 4a and 4b: Android-to-iOS Mobility 113 Graphs of Interaction Effects Between Performance-Aspiration and Relative Installed Base (iOS Sample) FIGURES 5a and 5b: Relative Smartphone Shipments (Android-iOS, in billions) FIGURES 6a and 6b: Relative Installed Base of Developers (Android-iOS, in millions) FIGURES 7a and 7b: Relative Installed Base of Apps (Android-iOS, in millions) 114 114 FIGURE 8a: Flow Charts of Matching Algorithm – The URL-matching Module 115 115 FIGURE 8b: Flow Charts of Matching Algorithm – The App Portfolio Matching Module 116 116 FIGURE 8c: Flow Charts of Matching Algorithm – Overview 117 117 APPENDICES 1a and 1b: Interaction Effect between Performance-Aspiration and Continuous Time (2008 July – 2015 October) 118 118 Robustness Checks of Interaction Effects between Performance-Aspiration Difference and Relative Installed Base Using Residual Inclusion Models APPENDICES 2a and 2b: Interaction with Relative Smartphone Shipments (Android-iOS, in billions) APPENDICES 3a and 3b: Interaction with Relative Installed Base of Developers (Android-iOS, in millions) APPENDICES 4a and 4b: Interaction with Relative Installed Base of Apps (Android-iOS, in millions) 119 CHAPTER 3 FOOLS RUSH IN? ENTRY INTO PLATFORM-BASED MARKET FOLLOWING ACQUISITIONS INTRODUCTION In 2014, Facebook acquired Whatsapp, the fast-growing social networking app, for $22 billion. 46 With this acquisition, nearly 450 million Whatsapp users were integrated into Facebook’s installed base. With the proliferation of companies building on network economies comes a new form of acquisitions: acquiring to enhance competitive advantage in customer bases. In fact, The Economist attempted to quantify such events in terms of how much an acquirer pays for each target user. For instance, in the Whatsapp acquisition, Facebook paid an equivalent of $49 for each user of the target company; Microsoft bought LinkedIn for $26.2 billion, or $60.5 per each of its 433 million users; and Activition Blizzard, one of the largest companies of console and computer games, acquired King.com, a major mobile game-developing company for $5.9 billion. 47 In essence, what these acquirers paid for were positive network effects started by the targets (Parker et al., 2016). Or, in other words, the ability to produce value for each user because of the target’s established user base. An intriguing question is: From the standpoint of potential entrants into corresponding markets, do such events present opportunities or potential threats? While such acquisitions may signal possible opportunities where positive network effects abound (thus making the market more attractive for potential entrants), the events could also reflect 46 The Economist, 2016 September 17, 2016. 47 The acquirer (Blizzard) lacked mobile-app users in its installed base, which mainly consisted of PC-game and console-game players. The target (King.com) was one of the largest mobile game companies with wide coverage of mobile game users. Many industry analysts speculate that, after the acquisition, the acquirer Blizzard might have the largest user base in the gaming industry, spanning multiple gaming platforms — PC, console, and mobile. As the company’s CEO put it succinctly, the acquisition “solidifies [its] position as the largest, most profitable standalone company in interactive entertainment.” 120 incumbents’ consolidation efforts (therefore posing significant competitive challenges for new entrants ex post). Hence, balancing the tradeoff between potential benefits and costs in making market entry decisions when responding to acquisitions is a critical issue and the focus of this study. As indicated through the examples, I situate the research in a nascent and fast-growing economic context: platform-based markets where network effects exist. Scholars of technological strategies have paid increasing attention to the importance of platforms and ecosystems in creating significant economic values (Adner & Kapoor, 2010; Eisenmann et al., 2006). An impressive amount of work has been done to understand platform- level competition (e.g., Boudreau, 2010; Eisenmann et al., 2011; Eisenmann et al., 2006), how platforms emerge and evolve (Gawer & Henderson, 2007), and how they attract and manage interdependencies with complementors (e.g., Cennamo & Santalo, 2013; Kapoor & Agarwal, 2017; Venkatraman & Lee, 2004). Indeed, both theorists and corporate practitioners agree that the success of a platform, to a large extent, relies on its complementors (or entities, either organization or individual, which provide products and/or services that augment what the platform delivers to users) (Shapiro & Varian, 1999). However, only few studies have recently started to study complementors’ strategic decisions in spite of the importance of such platform participants (e.g., Davis, Muzyrya, & Yin, 2014; Kapoor & Agarwal, 2017). Responding to this gap in the literature, my study puts complementors at the forefront and investigates one group of complementors’ category-entry behaviors in a platform-based market: app developers of Apple’s mobile app market (the App Store). The App Store is a typical example of a marketplace filled with information products in which (positive) network effects play an important role in determining a player’s competitive advantage. This is also a nascent and fast-growing market filled with uncertainties (Santos & 121 Eisenhardt, 2009). Under these market conditions, app developers (as platform complementors) need to respond swiftly to market opportunities. A key strategic move of current and potential app developers is to decide which product categories (or segments) to enter, and when. Moving too early means risking immature technologies and uncertain consumer demand (perhaps resulting in misidentifying opportunities), while moving too late creates the risk of missing opportunities (most likely encountering severe competition in a mature market). As the market shifts rapidly with fluctuating opportunity structures across different product categories, developers are likely to rely on external events as sources of market information in deciding when and which category(s) to enter. My study, thus, builds on the premise that mega events, such as acquisitions, will alter the market structure and hence influence developers’ category- entry decisions. As can be seen from the introductory examples of acquisitions, such deals have significant economic value and are often covered extensively in media reports. Consequently, they tend to garner a lot of attention from potential entrants. Accordingly to prior literature, acquisitions provide valuable information to market participants on the state of the market: the stage in the evolution cycle (Gort & Klepper, 1982), the extent of competition (e.g., Ashenfelter, Hosken, & Weinberg, 2015), and potential market opportunities and threats (e.g., Gaur, Malhotra, & Zhu, 2013). It is therefore logical to conclude that acquisitions will attract attention from potential entrants. The argument that acquisitions could influence market-entry decisions has been well documented (Caves & Porter, 1977; Stigler, 1950), especially from a theoretical standpoint. Empirically, however, whether acquisitions in a marketplace will attract or deter potential entries remains a puzzle. Essentially, the theoretical arguments on this topic suggest that acquisitions 122 send ambiguous signals towards potential entrants. On the one hand, theory suggests that acquisitions are incumbents’ consolidation attempts to build entry and mobility barriers for potential entrants (Caves & Porter, 1977; Klepper, 1997). In a platform market, the increased market power enjoyed by the merging companies is reflected in the consolidated installed base of users, which facilitates positive network effects in attracting further users and helps lock in existing users. Accordingly, one can deduce from this logic that intense acquisition activity in a marketplace would deter potential entrants. On the other hand, empirical evidence suggests that acquisitions send positive signals regarding the growth of the local market (Clougherty & Duso, 2009). This argument is especially relevant in a fast-growing platform market, such as the iOS mobile platform and its associated ecosystem. Unlike mature industries (e.g., auto manufacturing), the App Store is a dynamic marketplace that constantly spurs creativity and opens up opportunities for potential entrepreneurs. Acquisitions of startups by dominant companies are often viewed as the successful realization of entrepreneurial opportunities—as the opening examples suggests, such companies were acquired because of the network effects they started. This reasoning therefore offers an opposing prediction—intense acquisition activity in a marketplace will stimulate new entries. Given this ambiguity in the literature, the focus of the paper, therefore, is to resolve such uncertainty by carefully unpacking the way in which acquisitions influence developers’ category-entry decisions in the mobile app market. To further probe the underlying mechanism of network effects, I also examine how the signaling effects of acquisitions vary across product categories with strong-versus-weak network effects. Acquisitions are distributed across different categories of the platform market, therefore representing one system-level factor (Adner & Kapoor, 2010) that affects app developers’ entry 123 decisions by sending out signals. Naturally, one also needs to examine the side of signal recipients. That is, how app developers interpret such market signals, and, more interestingly, how they interpret the same signal differently. Building on the organizational experience literature of the behavioral theory of the firm (Argote & Epple, 1990; Eggers, 2012; Levinthal & Wu, 2010; March, 2010), I argue that, because app developers have heterogeneous experiences, their interpretations of acquisition intensity in a category will be different, leading to heterogeneous entry decisions when responding to the same market signals. The argument that experiences shape interpretation of events differs from the logic of learning from experiences, in which scholars have mainly focused on analyzing how firms build capabilities from experiences (Argote & Miron-Spektor, 2011). In contrast, the argument here reflects app developers’ incentives (in making category- entry decisions responding to acquisition signals) that are shaped by their prior experiences. This relies on an assumption of the platform market of this study (the App Store); i.e., that entry and mobility barriers to release products in a new category are marginal (e.g., Bresnahan et al., 2014; Parker et al., 2016). As a result, entry decisions are mainly determined by incentives rather than capabilities. I propose that different types of experiences will result in firms’ different ways of interpreting acquisition events, leading to heterogeneous incentives when making category-entry decisions. I therefore first examine how developers’ general experiences in the platform market shape their responses toward acquisition signals, then focus on the effects of two finer-grained types of experiences: one type differentiates whether prior experiences are proximate or distant (e.g., Levinthal, 1997; Nerkar and Roberts, 2004), the other differentiates whether prior experiences are broad or narrow in scope (e.g., Eggers, 2012). Each type of experience addresses a specific way in which prior experiences may shape app developers’ interpretations of 124 acquisition signals, and in turn, their incentives to enter the signaling category. A natural extension of this category-entry analysis is to study entrants’ post-entry performance—that is, to discern whether entrants who followed acquisition signals have made “wise” entry decisions, or whether they suffer in performance consequences as a result of rushing into seemingly attractive market segments. Theoretically, this follow-up analysis of post- entry performance addresses an important issue raised by prior literature. Prior conceptual work has suggested that post-entry performance is shaped by both market structures and firm attributes (e.g., Helfat & Lieberman, 2002), 48 but empirical evidence to substantiate this proposition is still limited (e.g., Franco, Sarkar, Agarwal, & Echambadi, 2009). 49 Consistent with this rationale, my research shows that entrants’ ex-post performance is influenced by both who they are (i.e., developer attributes in terms of heterogeneous experiences) and how they responded to acquisition signals (i.e., changes in market structures) when making initial entry decisions. To summarize, this study addresses the following tightly connected research questions: (1) Do app developers with heterogeneous experiences respond differently to market acquisition 48 This proposition mainly responds to the following argument that exclusively focuses on prior experiences’ direct effect on post-entry performance without accounting for the role of market structures. Many studies have argued and found that more experienced firms (mainly diversifying entrants) in general outperform less experienced ones (mostly entrepreneurial entrants), because the former possess pre-entry capabilities that enhance their post-entry adaptability. This thinking was empirically supported in different industry settings, such as U.S. manufacturing (Dunne, Roberts, and Samuelson, 1988), automobile (Carroll et al., 1996; Klepper, 2002), medical device (Adams, Fontana, and Malerba, 2016), and with different measures of post-entry performance, such as market share (Dunne et al., 1988), exit rate (Dunne et al., 1988), survival rate (Adams et al., 2016; Carroll et al., 1996; Klepper and Simons, 2000; Sleeper, 1998), financial performance (Chatterjee and Wernerfelt, 1991), sales (King and Tucci, 2002), and innovation (Klepper and Simons, 2000). 49 For instance, scholars suggest that young entrepreneurial firms expect to fare better in the early stages of industry lifecycle because their conditions (i.e., lack of a resource base but have radical innovative ideas) are more suitable for such market conditions, while diversifying firms may fare worse in such conditions due to rigidity in adapting to a fast-changing market (e.g., Helfat and Lieberman, 2002). Similarly, Franco et al. (2009) proposed that a key reason that the literature of first-mover advantage is inconclusive is that scholars did not incorporate both macro- level market conditions (e.g., who are first movers) and micro-level firm characteristics (e.g., their technological capabilities). By considering both aspects, Franco et al. (2009) found that being a first mover increases firms’ survival rate only when they possess high technological capabilities, but can decrease survival for those with low technological capabilities. In sum, the contingent arguments from these studies account for both (macro-level) market conditions and (micro-level) firm characteristics. 125 signals in making category-entry decisions? (2) When a developer enters the segment, how does acquisition intensity and the developer’s experiences at the time influence the subsequent likelihood of failure of its first product(s) to the category? The App Store of Apple’s iOS mobile platform is an ideal platform marketplace to study these two questions. First, the setting addresses a challenge that has limited prior studies of segment-entry analyses—that is, the platform marketplace is not subject to the unclear boundaries of product segments in most industry settings. This challenge was present in research on computer hardware and auto industries (e.g., Dobrev & Kim, 2006), wherein scholars devised ways, sometimes subjectively, to identify the boundaries of market segments. In this regard, the App Store as a research setting has unusual advantages. The platform owner, Apple, has clearly defined the segment boundaries in this market in terms of distinct app categories. The clearly defined app categories allow me to not only identify the distribution of acquisitions across categories and over time, but to also know the precise timing and categories that a developer chooses to enter (because mobile app developers typically have a portfolio of products that may fall into multiple product categories). In other words, the setting allows me to implement fine- grained analyses to detect the effects of acquisitions on category entries. Second, the App Store is a platform market in which network effects prevail (Hagiu, 2014; Parker et al., 2016; Shapiro & Varian, 1999). I propose that network effects are a major driving force of app developers’ category-entry decisions. Consequently, the theoretical insights generated from the study can be generalized to other types of platform markets where network effects exist (e.g., video-game platforms, movies and music distribution platforms, e-commerce platforms). The empirical context of the study, thus, sets the boundary of the theory—that is, network economies (Shapiro & Varian, 1999). 126 My empirical analysis is based on category-entry behaviors of more than 280,000 app developers from July 2008 to August 2015, covering nearly the entire time period that the iOS platform has been in existence and almost the entire population of iOS developers as of 2015. I study the likelihood that an app developer enters a product category in response to acquisition intensity in that category. In a developer-category-month analysis, I use a case-control design approach (e.g., Carnabuci, Operti, & Kovács, 2015; Rogan & Sorenson, 2014) and model the entry risk set as comprising all categories in which the focal app developer had not been active before the current entry event. I find that, in general, acquisition intensity has a significant and positive effect on an app developer’s entry probability. Moreover, developers with fewer general experiences, or whose experiences are distant or broad-scoped, are more likely to enter categories in response to acquisition events, while those with more general experiences, or whose experiences are proximate or narrow-scoped, are deterred. Finally, I estimate product failures at the post-entry period and find that products released immediately, following a category’s heyday of acquisitions, are less likely to fail in general. Products by certain developers (experienced, proximate, or narrow-scoped) are even less likely to fail. The paper contributes to the literature in several ways. First, it contributes to studies on technological strategies, and more specifically, on technological platforms and ecosystems. While most prior work in this field has been focused on competitive strategies at the platform- level (e.g., Boudreau, 2010; Caillaud & Jullien, 2003; Chao & Derdenger, 2013; Economides & Katsamakas, 2006; Eisenmann et al., 2011; Hagiu & Eisenmann, 2007), very few studies have given primacy to platform complementors (e.g., Clements & Ohashi, 2005; Corts & Lederman, 2009). Given that platform success is determined by the quantity and quality of complementors (e.g., Parker et al., 2016; Shapiro & Varian, 1999), this paper contributes to the platform 127 literature by studying how app developers (as platform complementors) make category-entry decisions in the iOS platform market by responding to acquisitions as well as to post-entry product performance. Based on the category-entry dynamics described in this study, we can draw implications regarding how the platform market will evolve resulting from developers’ entry behaviors. Furthermore, in the platform literature, scholars have treated a platform market as homogeneous (Boudreau, 2010; Caillaud & Jullien, 2003; Chao & Derdenger, 2013; Economides & Katsamakas, 2006; Eisenmann et al., 2011; Hagiu & Eisenmann, 2007). My research, however, relies on the assumption that a platform market can be segmented based on product attributes (Porter, 1985), thus creating segment heterogeneities. A specific source of segment heterogeneities that I will describe in greater detail is network effects—that is, certain product categories (e.g., iOS “social networking,” “games”) have stronger inherent network effects than others (e.g., iOS “utilities”). Examining fine-grained market segments instead of the platform in its entirety, and examining variation of network effects within a platform market, contributes to advancements in the platform literature. Such refined analyses reveal that the distribution of acquisitions across product categories and over time alters the market structure, increasing the attractiveness of certain categories for potential entrants while reducing that of others. Furthermore, this ecosystem-level effect is combined with complementor-level heterogeneities in explaining category-entry decisions and post-entry performance. Consequently, by examining complementors’ entry decisions and post- entry performance in a platform market, the paper helps to link the platform literature to a broader question that is of central interest to strategic management scholars: What explains heterogeneities in firms’ strategic choices and performance? Theoretically, I address this 128 question by integrating insights of the platform literature with two major theoretical perspectives—industrial organization (Caves & Porter, 1977; Porter, 1981)—and the literature on organizational experience from the behavioral theory of the firm (Cyert & March, 1963; March, 2010). The second contribution is that, by studying both stages of category entries and post-entry performance, the paper decomposes two effects of prior experiences. I argue that, while prior experiences shape a complementor’s incentive to enter a new segment following acquisition signals (because entry costs are negligible in the mobile app market) (Parker et al., 2016), such experiences may bestow on the complementor capabilities it might need to successfully navigate competitive challenges during the post-entry period (because unlike entry decisions, achieving high post-entry performance requires capabilities). The third contribution is to the industrial organization literature, and in particular to the inquiry regarding acquisitions’ effects on market entries. It is unclear, based on prior literature (e.g., Caves & Porter, 1977; Gaur et al., 2013; Klepper, 1997), whether acquisitions in a market will attract or deter subsequent market entries. This question becomes more interesting because acquisitions in a platform market have unique characteristics. First, acquisitions of successful startups (or products) send signals to the market that those who can start positive network effects will be rewarded. As a result, entrepreneurs tend to be attracted by the possibility of starting businesses that take advantage of network economies. However, acquisitions in a platform market also creates major players that could dominate a network economy, because enlarged installed bases of customers (a result of acquisitions) foreshadows their further exponential growth. Thus, monopoly power becomes possible. My study shows that, interestingly, both streams of thought apply. While acquisition signals attract inexperienced entrants who seek 129 opportunities, they also deter experienced (and potential major) players. The key is that market power accrued from acquisitions may create threats for other large incumbents, but not for niche- market companies. Consequently, both of the double-sided effects of acquisition signals on market entries can be explained by the same underlying driving force in a platform market: network effects. Finally, the study bridges the strategy and the entrepreneurship literatures. The entrepreneurship literature has mainly paid attention to the ways in which entrepreneurial firms actively pursue opportunities, but has largely ignored how they react to shifts in market structures. My findings suggest that acquisition distributions across market segments and over time function as a double-edged sword for entrepreneurs. On the one hand, they may hinder innovation from incumbents because these complementors tend to be deterred from entering segments where acquisitions occur intensely. On the other hand, acquisition events could spur entrepreneurship because inexperienced (and perhaps more entrepreneurial) complementors tend to rush into such segments. However, given that products by those who rush in tend to fail more frequently, one may conclude that the entrepreneurial waves created during acquisition heydays are high-cost experiments at the complementor level. Starting from the next section, I integrate insights from the literature on platforms (in particular, the role of network effects), industry organization, and organizational experiences to develop hypotheses of how acquisition activity in different categories of the iOS platform affects app developers’ category-entry decisions, how developers with heterogeneous experiences respond to acquisitions differently, and how entrants’ first products to a new category fare. 130 SEGMENT-ENTRY DECISIONS FOLLOWING ACQUISITIONS IN A PLATFORM- BASED MARKET Segment Entries in a Platform Market This study is one of the few that focuses on fine-grained segment-level entry decisions by platform complementors. It advances the platform and ecosystem literature in two ways. First, although entry into platform-based markets has been a topic of some pioneering work (e.g., Zhu & Iansiti, 2012), the scope has been limited to the platform-level. A platform can be defined as a technology (or system) that connects multiple groups of market participants, which could include platform complementors whose products and/or services augment the deliveries of the platform to customers. 50 In other words, platform complementors provide complementary products and services that can only be used by combining the platform’s offerings. Examples of platform complementors are game publishers of video-game platforms (e.g., Xbox) and app developers of mobile operating systems (e.g., iOS). Prior research has either examined market-entry decisions by platforms or complementors to a platform market (e.g., Zhu and Iansiti, 2012) without accounting for more fine-grained segment heterogeneities within a platform market. Therefore, this paper joins other recent work that has started to unpack segment heterogeneities within a platform and examine fine-grained, segment-level entry decisions (e.g., Doshi, 2014; Zhu & Liu, 2017). Second, the literature has largely focused on platform firms, while downplaying the importance of platform complementors. For instance, although scholars have examined platform- level strategies, such as openness versus closeness and pricing strategies (e.g., Bonaccorsi, Giannangeli, & Rossi, 2006; Boudreau, 2010; Economides & Katsamakas, 2006; Eisenmann et 50 Hardware providers are another group of market participants, but they are irrelevant in our context. One advantage of studying the iOS platform is that I have only one hardware provider, Apple. This enables me to eliminate the factors that affect app developers’ market entry decisions due to the competition in the hardware market. 131 al., 2011; Hagiu & Eisenmann, 2007), little research (e.g., Clements & Ohashi, 2005; Corts & Lederman, 2009) has been conducted on complementors’ strategic decisions. Shapiro and Varian (1999) succinctly highlight the importance of studying platform complementors. They argue that while “[t]raditional rules of competitive strategy focus on competitors, suppliers, and customers…. In the information economy, companies selling complementary components are equally important” (p.10). For scholars of technological strategies, complementary assets for a technology, akin to complementors to a platform, has been a central topic (e.g., Kapoor & Agarwal, 2017; Teece, 1986). My paper thus joins the few recent studies that focus on complementors (e.g., Kapoor & Agarwal, 2017), by specifically looking at how iOS developers (as complementors to iOS platform) choose categories to enter within the platform’s marketplace (the App Store). In the following discussion, I will therefore use the following terminologies interchangeably at each of the following three levels: (1) platform, iOS, and the App Store; (2) segments, and product categories; (3) complementors, app developers, entrants, and firms; and (4) customers and consumers. Segment-entry decisions in a platform market have unique characteristics compared to entry decisions in traditional marketplaces. A key attribute that differentiates a platform-based market from a traditional marketplace is network effects, according to which value to a potential platform participant is affected by the number of other participants, either on the same side (i.e., same-side network effects) or on a different side (i.e., cross-side network effects) (Eisenmann et al., 2006). In this current study, because I focus on one type of positive network effects—how the number of customers increases the utility of a complementor—I simplify the terminology by adopting what Shapiro and Varian (1999) referred to as “positive feedback.” While Shapiro and Varian’s (1999) discussion centered on positive feedback for 132 platforms, I argue that their logic also applies to the level of platform complementors. Complementors are motivated to build their own installed bases of customers, as the number of customers that a complementor could expect to acquire typically follows an exponential function. As a result, increasing the momentum in building the installed base of customers is critical for complementors, because, in a platform market with information products, the installed base is a critical source of competitive advantage (Shapiro & Varian, 1999; Sun & Tse, 2009). Even more important, once its installed base is established, a complementor can expect to lock in existing customers, because customers, either as individuals or a group, will experience significant switching costs when they adopt a different product. 51 In sum, if the installed base of customers provides a complementor with a competitive advantage in a platform market characterized by information products, then locking in customers based on their individual and collective switching costs provides the firm with a sustainable competitive advantage. To ensure competitive advantage, it is critical for complementors to be ahead of the curve on positive feedback. This means that complementors need to act swiftly in establishing a foothold in a market segment. Consequently, a complementor must discern whether a market segment is emerging and fast growing or is already consolidated and closed to further opportunities is critical. In other words, a complementor needs to choose the right entry timing. As a general rule, early entrants that capture market opportunities will be able to establish an installed base of users earlier than others, thus realizing the benefits of positive feedback (Shapiro & Varian, 1999). While such logic may bear a similarity to the first-mover advantages, the key question is not to discern the sequence of entries but the prudent time to enter a market segment. 51 For example, in the social-networking software segment, an individual customer’s switching costs arise from the investments that the customer makes in the product: personal information, network of friends, contacts, chat history, and so forth. Collective switching costs stem from the difficulty in shifting the entire network of connections or potential connections to a different software product. 133 Ambiguities of Acquisition Signals What, then, is a prudent time to enter a new market segment? I argue that, because an information-based market is characterized by fast growth and high uncertainties, companies need to rely on external events that convey information about a market to discern the situation. That is, firms rely on information of major external events to assess a segment’s current conditions. The logic that significant events will affect firm responses is not foreign to strategic management scholars. For instance, Figueiredo and Silverman (2007) described how the event of a dominant firm entering a market segment induces entries and exits by other companies. Diestre and Rajagopalan (2014) studied how chemical accidents affect related firms’ stock market performance. Acquisitions can signal to complementors whether an opportune moment of market entry exists. “Market dynamics provide information by giving managers and opportunity to learn about the market structure through the experience of others (Greve, 1996; Levitt & March, 1988)” (Greve, 1998: 65). Acquisition events, experienced by merging companies, provide valuable information to potential entrants on the state of the market: namely, the stage of the market in the evolution cycle (Gort & Klepper, 1982), the extent of competition (e.g., Ashenfelter et al., 2015), and the potential market opportunities and threats (e.g., Gaur et al., 2013). Hence, complementors are likely to rely on market signals such as those from acquisitions when making segment-entry decisions. 52 Yet, acquisitions that happen in a product category may lead to opposite interpretations 52 Many acquisitions are significant events in an industry that attracts great attention. For example, the acquisition of King.com by Blizzard (valued at $5.9 billion, source: http://investor.activision.com/releasedetail.cfm?releaseid=956435, accessed on May 26, 2016) and Facebook’s $22 billion acquisition of Whatsapp (the former a giant in social network services and the latter a rising star in the messaging category) were widely covered by the media. Such acquisitions are important market signals, because they often reflect prominent firms’ attempts to consolidate a market. 134 by app developers in the marketplace of iOS platform because such events can either be perceived as signals of growth and market opportunities or as signals of maturity and reduced market opportunities. 53 To understand whether acquisitions will attract or deter subsequent category entries, I invoke a simple, yet powerful, argument proposed by industrial organization scholars: developers, as potential entrants, will assess the potential benefits and costs when making category-entry decisions (Caves & Porter, 1977; Porter, 1985). Potential entry benefits. Potential benefits of entering a category following a period of intense acquisition activities have multiple, and tightly connected, components such as seeking market opportunities where network effects exist (i.e., “the next big thing”), replicating prior successes, the possibility of becoming the next acquisition target, and entering a market with weakened competition. First, an important incentive of app developers is to seek market opportunities where positive network effects exist (or, “the next big thing”). The opening examples suggest that the targets were rewarded by the acquisition market because of the positive network effects that they had started (Parker et al., 2016). Acquisitions, especially those completed by prominent firms, help to validate entrepreneurial ideas and send positive signals of the corresponding segment’s potential growth (e.g., Clougherty & Duso, 2009; Gaur et al., 2013). Once dominant firms enter a new market through acquisitions, the market expects to have further investments from such players, which could fuel the market’s next round of growth. 54 A second, 53 The reduced opportunity available to late entrants is recognized by app developers, as evident in the comments made by one of my interviewees who was clearly pessimistic about the recent (late 2015) entrants’ prospects in the social network segment. Referring to the saturated market opportunity in the social network segment, h/she observed that, “if you are going for the social network category now [late 2015], good luck [with laughter].” Related to this observation, another interviewee pointed out that, in the less-mature market of educational games, a major acquisition would signal opportunities, while similar events in the social network segment would signal reduced opportunities. 54 An example is Facebook’s acquisition of Oculus, the latter an emerging company in the field of virtual reality, and the former a dominant platform in social network businesses. Through this acquisition, Facebook signaled to the virtual reality community that the company planned to keep investing in this domain. It is reasonable to expect that such validation of the market will attract new entrants to the market segment. 135 and related point is that success breeds imitators. Prior studies on information-based products suggest that successful products, such as “killer apps” (Bresnahan et al., 2014), tend to attract flocks of imitators (Boudreau, 2015). Many types of market signals abound in the mobile app context that could suggest success. For instance, it could be top ranking and extensive coverage by media reports. Acquisition events are one type of market signals for other startups and potential entrepreneurs to realize the viability of certain technologies and ideas, thus making the decision to rush in. Third, acquisitions signal to potential entrants the possibilities of becoming the next targets, particularly when dominant market players acquire successful startups. In order to imitate and replicate prior success, entrepreneurs tend to enter segments frequented by acquisitions. 55 They expect to benefit from the increased financial returns from selling their own businesses and/or enhanced career opportunities at the (larger) acquiring firms. Additionally, realized acquisitions could increase bidding prices for similar future targets. 56 Finally, acquisitions could weaken competition in a market segment, therefore attracting potential new entrants. One of my interviewees 57 suggested that acquired targets often had a poorer performance at the post-acquisition stage because the acquirers, usually dominant market players, tended to interfere with the strategies of the former startups and reduced their flexibility, innovativeness, and product appeal. In extreme cases, empty spaces are created because the acquirer ex post withdrew the target’s products from the market. When such instances occur, opportunities are created for potential entrants as they expect decreased competition due to the acquisitions. Hence, I propose the following hypothesis: 55 As my field interviews suggest, many entrepreneurs consider being acquired a successful exit outcome and are hence inspired by successful historical acquisitions such as MySpace and YouTube. 56 Higher bidding prices, however, can increase entry costs for diversifying entrants, because they may want to enter the new category through acquisitions or engage in acquisition activities at the post-entry stage. 57 I made two field research trips to San Francisco to interview mobile app developers. In total, I conducted five field interviews. My interviewees include a former Facebook employee, the CEO of a mobile game company, the sales head of a mobile analytics company, a project manager of a mobile game company, and a cofounder of a mobile education-game company. 136 Hypothesis 2.1a: Acquisition intensity in a product category will be associated with an increased likelihood of an app developer entering the category. Potential entry costs. When Porter (1985) described the potential costs of entering a new segment, his discussion centered on the focal firm’s difficulty in organizing shared activities across multiple segments. In the discussion that follows, I examine potential costs in the form of competition and retaliation from established merging companies (Caves & Porter, 1977). The industrial organization literature proposes that acquisitions confer market power on the merging firms. After acquisitions, merging firms can enjoy greater bargaining power over suppliers and customers, and have fewer competitors (e.g., Haleblian et al., 2009; Kim & Singal, 1993). Incumbents use acquisitions as a defensive strategy to deter potential new entrants (Porter, 1985), because ex-post merging firms not only enjoy the benefits of scale economies, but they also make it difficult for potential new entrants to be profitable due to the increased minimum efficiency scale (Klepper, 1997). In an information-based industry, such as the App Store, acquisitions have unique characteristics. That is, such events may not only foster economies of scale on the supply side (i.e., larger firms can churn out products more efficiently), but they also provide the merging firms positive feedback on the demand side (i.e., increased installed bases of customers make their technologies more attractive to new customers). First, economies of scale, in a platform market characterized by information products, can arise from the fact that costs can be spread over numerous customers. Unlike traditional markets with physical products, information-based markets are characterized by high fixed costs but almost zero marginal costs of producing additional products (i.e., copies). Consequently, a firm can realize economies of scale by serving a large installed base of customers without incurring 137 significant marginal costs (Shapiro & Varian, 1999). To build an installed base of customers, many user acquisition techniques exist, including outsourcing to advertising companies. One of the most direct ways is to acquire a firm (and its products). Through an acquisition, the acquirer can retain the customer base of the target. Consequently, increased installed bases (Shapiro & Varian, 1999) as a result of acquisitions improve the merging firms’ competitive advantage, because they realize higher economies of scale. Such advantages of merging firms translate to entry costs for potential entrants. Second, because of increased installed bases due to acquisitions, merging firms also expect to benefit more from positive feedback on the demand side. A larger network of customers becomes more attractive to new customers. In a platform market, such as the App Store, an installed base of customers is a major source of competitive advantage (Shapiro & Varian, 1999; Sun & Tse, 2009), as the increases in the number of customers could grow exponentially once the installed base reaches a critical mass. The example of King.com acquiring Blizzard shows how a large firm solidifies its competitive advantage through acquiring the target and its installed base of customers. 58 If the abovementioned arguments suggest that an installed base could provide the firm with competitive advantage in a network economy, then locking in customers due to an enlarged installed base provides the merging firms with a sustainable competitive advantage. Information-based lock-in is more durable than equipment-based lock-in, a feature that characterizes the mobile app market, because “...equipment wears out, reducing switching costs, but specialized databases live on and grow, enhancing lock-in over time” 58 The acquirer (Blizzard) lacked mobile-app users in its installed base, which mainly consisted of PC-game and console-game players. The target (King.com) was one of the largest mobile game companies with wide coverage of mobile game users. Many industry analysts speculate that, after the acquisition, the acquirer Blizzard could have the largest user base in the gaming industry, spanning multiple gaming platforms—PC, console, and mobile. As the company’s CEO succinctly put it, the acquisition “solidifies [its] position as the largest, most profitable standalone company in interactive entertainment.” 138 (Shapiro & Varian, 1999: 115). Thus, building on the arguments of potential entry costs following acquisitions, I propose the following competing hypothesis: Hypothesis 2.1b: Acquisition intensity in a product category will be associated with a decreased likelihood of an app developer entering the category. Taken together, acquisitions in a product category could suggest both benefits and costs for a potential entrant. Given the opposing views, ex ante, one cannot determine whether, on average, the effect of acquisition intensity on a developer’s entry likelihood will be positive or negative. However, I argue that whether an app developer perceives more benefits or more costs depends on where it stands. Responding to the call to investigate heterogeneities of platform complementor (e.g., Gandal, Greenstein, & Salant, 1999), I theorize next on how heterogeneities among app developers influence their category-entry decisions in response to acquisitions, building on the organizational experiences literature (Argote & Epple, 1990; Eggers, 2012; Levinthal & Wu, 2010; March, 2010), and more broadly, the behavioral theory of the firm (Argote & Greve, 2007; Cyert & March, 1963; Gavetti, Greve, Levinthal, & Ocasio, 2012). HOW EXPERIENCES SHAPE RESPONSES TOWARDS ACQUISITION SIGNALS Experience is a teacher that guides a firm’s future conduct. In the vast literature on organizational experiences (e.g., Argote & Epple, 1990; Eggers, 2012; Levinthal & Wu, 2010; March, 2010; Wernerfelt, 1984), I stress one aspect that, to the best of my knowledge, previous empirical research has neglected: namely, that experiences shape a firm’s interpretation of events, and hence their decisions. This rests on the assumption that interpretations of historical events can be flexible, an argument that is vividly depicted by March (2010): “… just as the same face can easily produce quite different, even contradictory, portraits, the same observation of organizational experience can produce evaluations that differ profoundly” (p.111). Further, 139 “[s]ince the theory is predicated on the essential purposiveness of the firm, it assumes some attempt on the part of the organization to react to its environment through observation and interpretation” (Cyert and March, 1963; italics are added). Adopting this basic idea, I submit that developers’ idiosyncratic experiences will shape how they interpret, and thus react to, acquisitions differently. This is specifically reflected in how they evaluate the tradeoff between the potential benefits and costs of acquisitions when deciding whether to enter a corresponding market segment. I organize my discussion on developer-experience heterogeneity by following a framework that (1) first examines general platform experiences, and then distinguish in terms of (2) distant versus proximate experiences (e.g., Levinthal, 1997; Nerkar & Roberts, 2004), and (3) broad versus narrow experiences (e.g., Eggers, 2012; Porter, 1985). General Platform Experiences In the mobile app market of the iOS platform, developer experiences can be conceptualized in multiple ways. First, drawing from the traditional industrial-organization literature (e.g., Caves & Porter, 1977), the simplest way is to use a dichotomy to categorize potential entrants as either platform incumbents (those who have products on the platform but not yet in the focal product category), or platform entrants (those who have not yet adopted the platform). Second, app developers with many years of experience on a platform are more familiar with how the market has evolved over time and have more intimate knowledge about doing business on a specific platform (for instance, Apple’s App Store has different rules from Google’s Play Store). In other words, the length of time that a developer has been on a platform—or platform age—is another dimension to gauge whether the developer is experienced. Finally, the learning literature 140 suggests that one could differentiate firms by their product development experience (Argote & Epple, 1990): The more products that an app developers has released, the more efficient the developer is likely to be in repeating the task. To be parsimonious, instead of discussing each of the above distinctions separately, I refer to app developers as experienced versus inexperienced. However, the arguments advanced below are applicable to each type of general experience and the empirical tests will separately consider each of these different operationalizations of experience. I propose that experienced and inexperienced app developers will interpret acquisitions in a category differently when evaluating potential costs versus benefits of entries. First, they have different information access to acquisition events, and upon receiving such information, they interpret it differently. In general, when acquisitions take place, the information that inexperienced app developers have is thinner in content and less diverse, compared to the information that the experienced incumbents can access. This is because the inexperienced have weaker professional network connections (Uzzi, 1997) and they tend to rely on social information such as media reports. Yet, social information often contains systematic errors because information is often “sharpened” when moving from one cascade to the next (Gilovich, 1987). Media reports tend to exaggerate facts in order to provoke a strong emotion or trigger enthusiasm towards entrepreneurship. Business press 59 “probably acts primarily as a source of normative information, as extensive coverage of acquisitions is likely to carry the implicit message that acquisitions are the thing to do” (Haunschild & Beckman, 1998: 840). Learning from such media reports also means that inexperienced developers draw samples from a homogeneous group of examples—i.e., realized acquisitions of successful businesses. Such 59 Source: http://www.pbs.org/newshour/making-sense/unicorns-and-delusions-in-silicon-valleys-tech-bubble/, assessed on January 11, 2015. 141 “exposure to only good examples” does not provide contrasting examples for inexperienced app developers to make high-quality decisions (Haunschild & Beckman, 1998), while decision quality can be gauged by subsequent performance after entering a new product category. Therefore, inexperienced developers are likely to be dazzled and enter product categories that are frequented by acquisitions. Conversely, experienced app developers have access to richer content and more diverse acquisition information. In addition to access to social information, such as the business press (as the inexperienced developers do), more experienced developers also have alternative sources of information, such as their own experiences in the industry and their professional networks (Uzzi, 1997). Because of their access to multiple sources of information, experienced developers could draw on diverse examples when evaluating the potential benefits and costs of entering a category that has experienced frequent acquisitions. According to prior literature, information diversity improves decision quality (Beckman & Haunschild, 2002). For instance, through their industry networks, experienced developers could have deeper knowledge about a specific acquisition deal and may realize that the acquisition could be an attempt to nip an entrepreneurial idea in the bud before it becomes a serious threat for the acquirer, despite the rosy picture painted by media reports about the deal. The different nature of information received by experienced versus inexperienced app developers thus leads to different interpretations of acquisitions and conclusions. In general, upon receiving information of an acquisition, the inexperienced developer tends to interpret such a signal more positively, and is biased towards the aforementioned potential benefits (i.e., chase market opportunities where network effects abound, replicating prior successes, becoming the next acquisition target, and expecting weakened market competition). In contrast, the 142 experienced developer tends to react more negatively, seeing more of the potential costs (i.e., stronger ex-post competition from the merging firms due to improved economies of scales on the supply side, and positive feedback on the demand side). Moreover, the experienced app developer has more existing businesses at stake when it comes to competitive interactions with merging companies. The more experienced, the higher the stake, because such a developer fears that rushing into the product category of a merging firm will result in retaliation not only in the focal category, but also in other existing categories. The experienced developer thus tends to hold the view of “wait and see,” instead of rushing into the product category. 60 Second, from a learning perspective, experiential learning constrains a firm in making its future decisions (Baum, Li, & Usher, 2000) in response to external events. Given that an experienced app developer already has established positions in its existing product market segments, the developer is constrained in making decisions to enter new segments that respond to acquisition signals. The experienced developer fears that, if it enters new segments following such signals, it may spread itself too thin, and thus be unable to allocate sufficient attention to existing markets. Consequently, instead of investing in new segments, inertia encourages the experienced developer to continue to invest resources in existing segments, because this strategy ensures economies of scale. By doing so, the experienced developer can also avoid switching costs that arise from learning new domain-specific knowledge. 61 Moreover, routinization and inertia prevent the experienced from responding to both new 60 The following example illustrates the deterrence effect of acquisition signals on platform incumbents. When I interviewed a medium-sized mobile gaming company, the interviewee mentioned that the company’s employees and management discussed the implications of a major acquisition to understand how it changed the competitive landscape and its influence on the company’s product development strategies. The result of those deliberations was a more conservative market strategy moving forward. 61 Consistent with this argument, one of my interviewees noted that because they are a gaming app company, they would not imprudently enter a new domain, such as one that develops software providing data security to end users. Such an entry would put them at a competitive disadvantage relative to someone with 10 to 15 years of experience in that domain. 143 external opportunities and threats (Baum et al., 2000: 768-769). Actions of an experienced app developer are more likely to be shaped by internal forces of inertia rather than by external information from industry mega-events. Adopting the language of complementarity versus substitution effects, I argue that the effect of external acquisitions on an experienced developer’s segment-entry decision is partially substituted by its own experiences. This is because, based on the developer’s own industry experience, entry decisions could have been made earlier if the developer had thought that the acquired businesses were successful entrepreneurial opportunities. That is, the experienced developer has bypassed such opportunities earlier (for instance, when the eventually acquired entities had received their initial venture capital funding). As a result, when responding to acquisitions in making segment-entry decisions, experienced app developers are affected more by their internal inertia (King & Tucci, 2002; Miller & Chen, 1994) than by such external events. In contrast, inexperienced app developers are more nimble, less constrained by inertial forces (Kotha, Zheng, & George, 2011: 1014), and more likely to be influenced by external industry events. Consequently, they are more prone to respond to market information and make category-entry decisions. These arguments lead to the following hypothesis. Hypothesis 2.2: General platform experience will have a negative moderating effect on the relationship between acquisition intensity in a product category and the likelihood that the developer enters the category, such that acquisition intensity will be associated with increased entry probability by an inexperienced developer but a decreased entry likelihood by an experienced one. One caveat of the abovementioned argument is that the experienced developers are not necessarily making “unwise” decisions. The literature on experts versus novices suggests that, due to their abundant experiences, experts can make better decisions with limited information because they see patterns, instead of details, when making decisions (Simon, 1996). The decision of “holding back” when facing acquisition signals may indicate a better strategic decision, as 144 “[t]he essence of strategy is choosing what not to do” (Porter, 1996). Although this alternative argument leads to the same predicted direction of the hypothesized relationship, I will rely on the analysis of post-entry performance to draw normative implications. Proximate versus Distant Experiences The previous discussion of general platform experiences does not account for the fact that not all experiences are created equal (Eggers, 2012). In this and the following sections, I unpack general experiences along different dimensions. Experience proximity (versus distance) reflects the closeness of the focal developer’s experiences, and thus how salient the experiences are, in affecting the developer’s strategic decision making, such as segment-entry decisions. Proximity depends on whether it is about space or time. When product categories are viewed as spread out in a manner that resembles geographic regions, proximity essentially reflects how close a developer’s existing product categories are to the target category. Experience proximity, defined in this way, corresponds to the relatedness of experiences to the target category. Previous research has suggested that the greater the experience proximity, the more likely a firm is able to absorb new knowledge in neighboring knowledge domains (Cohen & Levinthal, 1990); thus, the better the firm can utilize its existing knowledge when searching in a new knowledge landscape (Levinthal, 1997), and the better the firm can leverage synergies between existing and new activities (Tanriverdi & Venkatraman, 2005). When measuring proximity in terms of time, we usually refer to distant experiences as those that were accumulated a long time ago. According to the learning literature, recent experiences (compared with older experiences) are more salient in affecting subsequent task performance. In essence, experience depreciates over time (Argote & Miron-Spektor, 2011). 145 Previous research has found that proximate experiences contribute positively to new product performance (Nerkar & Roberts, 2004) and R&D project performance (Macher & Boerner, 2006). Overall, both lines of research suggest that experience proximity enhances the focal firm’s ability in performing tasks. I extend this research by probing how experience proximity shapes a developer’s interpretation of acquisition intensity in a market segment. My argument continues with the cost-benefit analysis when developers make category-entry decisions, and the central theme when discussing experience proximity is the role of threat rigidity. Upon observing acquisitions, developers with proximal experiences will feel threatened by the approaching competition, and hence emphasize the costs of entering the new category and become more rigid. In contrast, developers with distant experiences will interpret such events less as threats and more as opportunities. First, experience proximity increases a developer’s access to information regarding acquisitions while, at the same time, shaping its interpretation of such information. Acquisitions in neighboring product categories are more informative than those that took place in distant categories, because the focal developer is more familiar with the competitive landscape and the business models. Developers that are in proximal domains have more direct knowledge about the state of the product category. When such events happen in neighboring categories, as opposed to distant product categories, they are more threatening to the focal developer’s competitive position. The merging companies’ intention to build a strong foothold in neighboring categories is also clearer, and they are more likely to be rivals with whom the focal developer has more contacts in multiple product categories. Consequently, the developer is more likely to forbear from entering corresponding product categories (e.g., Edwards, 1955; Gimeno, 1999). In contrast, 146 although developers with more distant experiences not only lack intimate knowledge about such events, they are also less threatened by the potential competition from the merging companies. Second, upon receiving acquisition signals, app developers with proximal experiences act in ways that reinforce their inertia. To address the concern that the approaching competition (due to acquisitions in neighboring segments) may lure its customers away in the near future, the developer will continue to invest in established routines and capabilities, expand the installed base of users through various approaches of user acquisition, and come up with technologies and services that could further lock in existing customers. All of these activities suggest that, upon receiving acquisition signals, firms with more proximate experiences will strengthen their commitments to existing businesses and be more cautious in making category-entry decisions. Moreover, because of their more intimate knowledge about the neighboring product categories, developers with proximate experiences could have made decisions to enter such categories earlier rather than waiting until acquisitions are realized, a sign of maturing entrepreneurial opportunities. That is, the effects of proximate experiences and external acquisition events may substitute for one another. In contrast, for app developers with distant experiences, information access is more limited in content and diversity, thus they are more prone to be influenced by positive pictures painted by business press. In addition, due to a lack of familiarity with such distant domains, the substitution effect from their prior experiences (as opposed to the influence from acquisitions) is weaker. To sum up, app developers with distant experiences are less subject to threat rigidity and therefore more flexible when making segment-entry decisions in response to external acquisition events. Although the above arguments are mainly based on experiences across market segments, 147 similar arguments can be made when experience proximity is measured in terms of time. That is, acquisitions are more salient for app developers who are currently active (i.e., with proximate experience), thereby more likely to be perceived as threatening and implying greater costs than benefits, than those who were once active (i.e., with distant experience), suggesting fewer costs than benefits. They are thus more deterred in entering corresponding market segments when observing acquisitions. Accordingly, Hypothesis 2.3: Experience proximity will have a negative moderating effect on the relationship between acquisition intensity in a product category and the likelihood that the developer enters the category, such that acquisition intensity will be associated with increased entry probability by a developer with distant experiences and decreased entry likelihood by a developer with proximate experiences. Narrow versus Broad Experiences Narrow experiences can be defined as, holding all else equal (e.g., product portfolio size, age on the platform), the extent to which a firm ’s products are created specifically to a narrow set of market niches, and, broad experiences in terms of the firm’s products to appeal to a wide range of niches (Eggers, 2012; Porter, 1985). Whether experiences are narrow or broad is therefore reflected in the firm’s current product portfolio. A key question in the experience-based learning literature is what antecedents predict a firm’s extent of experience breadth. In my current analysis, I invoke the argument that a firm’s current state of product portfolio is due to path dependency in its historical development. Embedded in the path dependency argument is the assumption that a firm has its own enduring nature that steers its strategic decision making, which has been defined in previous literature as the dominant logic of a firm’s diversification strategy (Prahalad & Bettis, 1986). Integrating this argument with Porter’s (1985) distinction between broad- versus narrow-scope companies, I 148 derive the following intuition. Companies that believe success comes from covering as many markets as possible eventually build a diversified product portfolio, while companies that perceive success comes from focus eventually build a more narrow product portfolio. In terms of their cost-benefit analysis, it can be argued that app developers with experience breadth will perceive more benefits (e.g., becoming “the next big thing,” replicating prior successes) than costs (i.e., stronger ex-post competition from the merging firms) from potential segment entries (Porter, 1985), while developers with narrow experience perceive more potential costs than benefits. Path dependency therefore creates broad- and narrow-scope developers’ different biases when responding to acquisition signals. Regarding information accessing, broad-scope app developers are exposed to more diverse (though not necessarily dense) information than narrow-scoped developers, because the former operate in multiple product categories. Broad-scope developers are therefore more sensitive towards information from the marketplace, such as acquisition events. Therefore, compared with narrow-scoped developers, they are more responsive toward such signals when making category-entry decisions. Taken together, due to their different information accesses and biases towards acquisitions, broad- and narrow-scoped developers will respond differently in making category- entry decisions. When observing acquisitions in a new product category, developers’ inherent strategic orientations are triggered. The signals sent by such events reinforce broad-scope developers’ bias towards potential benefits of entering the corresponding new segment. The contrary is true for developers with focused product portfolios. Thus, diversified developers tend to rush in, while focused ones are likely to withhold in response to acquisition signals. Hypothesis 2.4: Experience breadth will have a positive moderating effect on the relationship between acquisition intensity in a product category and the 149 likelihood that the developer enters the category, such that acquisition intensity will be associated with decreased entry probability by a narrow-scope developer but increased entry likelihood by a broad-scope developer. CATEGORY HETEROGENEITIES IN NETWORK EFFECTS Building on the argument from previous research that segment heterogeneities exist within an industry (e.g., Figueiredo & Silverman, 2007; Jacobides & Tae, 2015), I submit that, within a platform market, product categories vary in terms of network effects. That is, products of certain categories could inherently better leverage (positive) network effects than those in other categories. In the mobile app market of the iOS platform, one example is the “social networking” category. As illustrated in the opening example of Facebook acquiring Whatsapp, the value of social apps for potential consumers increases as the number of users using such products increases. Another example is the “games” category (e.g., Miric, 2017). During the era of internet games, connectivity among game players becomes increasingly valued, to the extent that even single-player game subcategories start to stress the importance of social connections. For instance, even though the popular game Candy Crush is essentially a single-player game, the social connectivity aspect is reflected by its function of requesting a “life” from a friend who also plays the game. Because of the network effects started by this popular game and a consequential large installed base of mobile users, the parent company King.com was acquired by a major gaming company, Blizzard. Products in other categories (e.g., iOS “utilities”), however, are inherently less capable of leveraging network effects. Hence, I propose that acquisition signals from categories with strong versus weak network effects will lead to different responses from potential entrants. Building on a hypothesis developed in a previous section (Hypothesis 2)—that inexperienced developers are attracted, while the experienced are deterred, by acquisition intensity when making entry decisions—I 150 further propose that such opposing responses toward acquisition signals will be amplified in categories with stronger network effects. That is, in categories with stronger network effects, the inexperienced will be even more likely to rush in, while the experienced will be further deterred. On the one hand, potential costs for experienced developers to enter a category with stronger network effects are increased. In the aftermath of acquisition heydays, merging firms are likely to become major market players. Because a market space can only support a certain number of dominant market players, it becomes more difficult for additional major players, presumably experienced developers with already-installed bases of customers, to establish a foothold (e.g., Hannan & Freeman, 1977). As the growth magnitudes of merging firms are greater in categories with stronger network effects, it can be expected that the deterring effect of acquisitions on experienced developers will be stronger in a strong network-effect category than in a weak network-effect category. On the other hand, potential benefits for inexperienced developers are strengthened in categories with stronger network effects. Although acquisitions create major market players, thus making a category less attractive for experienced developers, opportunities remain for niche market players, who, as presumably inexperienced developers have not yet built meaningful installed bases of customers. Admittedly, if we adopt the assumption that a market space is fixed in terms of available resources (i.e., customer base in a platform market), then we can make the following logical conclusion. When acquisitions occur in a category with stronger network effects, the growth magnitudes of merging firms are greater, so they will occupy a larger portion of the resource space. Consequently, according to this logic, the space of niche markets left for inexperienced developers is smaller and therefore they should be deterred from entering. This conclusion, however, needs to be modified if we relax the assumption that the market space is 151 fixed. I propose that in a growing market, such as the mobile app industry, the market space of a product category is not fixed but expanding (e.g., customer base). As the previous arguments on potential benefits of acquisitions suggest, acquisitions in a category may also contribute to the expansion of market space by opening up new opportunities. Such potential benefits from acquisitions should be stronger in categories with stronger network effects. Interestingly, the expanded market space due to acquisitions may sustain niche market players but not major ones. Consequently, I submit that the facilitating effect of acquisitions in attracting entry from inexperienced developers will be stronger in a strong network-effect category than in a weak network-effect category. Taking the two sides of the arguments together, I propose: Hypothesis 2.5: In a category with stronger network effects (e.g., social networking, games), different responses toward acquisition intensity from inexperienced and experienced developers will be amplified when they make entry decisions. That is, (1) the positive effect of acquisition intensity on the entry likelihood by inexperienced developers and (2) the negative effect of acquisition intensity on the entry likelihood by experienced developers will both be strengthened. POST-ENTRY PERFORMANCE OF FIRST PRODUCTS One way to draw normative implications of segment-entry decisions is to examine the performance of entrants’ first products to a new category. I define first products in the mobile app setting as apps that are launched by a developer at the time when it enters a new category, such as the first month. As the previous category-entry analyses have foreshadowed, it is unclear, ex ante, whether acquisitions ultimately offer openings of market opportunities or suggest closures of opportunities. Accordingly, I address both sides of the argument in the following section that probes how acquisition intensity at the time when a developer enters a category shapes the subsequent likelihood of failure of its first product(s). 152 Acquisitions as Signals of Market Opportunities One side of the argument is that acquisitions open opportunities for subsequent entrants, whose first products to a new category will therefore perform well. First, while acquisitions in a product category may ultimately create dominant market players that occupy most of the resource space, opportunities for new entrants still exist in niche markets. Entrants that target the niche markets when introducing their first products will be able to enjoy a certain market share without an attack by the dominant market players. Second, acquisitions are in waves. Realized successful acquisitions suggest that opportunities exist in nearby space of opportunities. Entrants that join a new category to capture market opportunities where network effects exist may expect to have decent products for subsequent buyers. For instance, products as imitators of “killer apps” could also be successful, sometimes becoming targets for potential acquirers who have missed initial opportunities of seizing top-performing startups. Hence, according to the logic of acquisitions opening market opportunities, one would expect that first products launched during acquisition heydays will be less likely to fail, leading to the following hypothesis. Hypothesis 2.6a: Acquisition intensity in a category at the time when a developer enters the category will be negatively associated with the likelihood of failure of the developer’s first product(s) to the category. Acquisitions as Signals of Market Competition Contrary to the previous logic of acquisitions opening further market opportunities, the opposing argument is that acquisitions are consolidation attempts, therefore posing threats of ex-post competition for potential entrants. Several literatures have reached this same conclusion. First, product categories that have recently experienced acquisition heydays are likely to be associated with severe competition, characterized by higher market concentration (e.g., Dafny, Duggan, & 153 Ramanarayanan, 2012), increased price competition (e.g., Ashenfelter et al., 2015; Focarelli & Panetta, 2003), greater market power of the merging firms (e.g., Kim & Singal, 1993; Prager, 1992), and their more competitive products (e.g., Lubatkin, Schulze, Mainkar, & Cotterill, 2001). According to the industrial organization literature, competitive intensity of a market segment will influence product performance in that segment. It follows that an entrant’s first products following acquisition heydays tend to encounter more severe competition, and therefore are more likely to fail. Second, one could argue that acquisitions could enable merging firms to occupy most of, if not all, the resource space (i.e., customer base). If positive network effects are so strong that the merging firms could cover almost the entire available customer base, no niche markets may be left for potential entrants. Under such circumstances, one could argue that entry products to a new category after the category has experienced intense acquisition activities will be more likely to fail. In sum, this line of argumentation that associates acquisitions with heightened market competition suggests a different prediction: Hypothesis 2.6b: Acquisition intensity in a product category at the time when a developer enters the category will be positively associated with the likelihood of failure of the developer’s first product(s) to the category. Moderating Effects of Experiences The opposing effects of acquisition intensity on the performance of first products can be succinctly captured using the classic strategy literature distinction of (acquisitions in a category as either) opportunities (H5a) or threats (H5b). Naturally, one would expect that developers would vary in their abilities to capture opportunities and mitigate threats. According to the literature, a good proxy of firm capabilities is experiences. I therefore adopt a capability view to unpack performance heterogeneities of first products by entrants with different experiences. A major difference between the analyses of post-entry performance and the previous 154 category-entry decisions is the different roles played by organizational experiences in each of the two stages. In the category-entry analyses, I argued that experiences will shape how developers interpret acquisition signals, building on the assumption that, in the mobile app market, it takes marginal resources to enter a new category (Parker et al., 2016). However, in the post-entry performance stage, I argue that the role of experiences become sources of developers’ capabilities to capture market opportunities and address competitive threats. This is because, although entries can be relatively cost-free in this market, performing well requires capabilities. Inherent in each of the three types of experiences are different capabilities. General platform experience. General experiences reflect generic capabilities on the platform, which enable a developer to capture opportunities and address competition. Such generic capabilities accrue when a developer operates in the platform market. Some examples include familiarity with the platform’s rules (e.g., how app ranking works), specific value- creation strategies that are consistent with such rules (e.g., how to increase positions in app ranking), user demographics, and user-acquisition tactics in a particular platform market. Such familiarity with how to run businesses on a platform is generic in the sense that it can be applied across different product categories of the platform. Based on my field interviews, for instance, a type of generic capability in the mobile app industry that has become increasingly important is business intelligence and live operations, or, the ability to constantly update content and provide ongoing user support. The quote below from a project manager from a major mobile gaming company illustrates this point: Developers started to emphasize live operations. That means developers now need to make sure [to] have content for a long time, for instance, several years. An example would be Clash of Clans, which has existed for three to four years. Such games provide live content and support for such content. The games that were developed within three to six months usually do not have a good core game to support a particularly long lifecycle. I argue that app developers with greater general platform experiences are more likely to 155 possess such capabilities, which in turn enable them to better capture opportunities (or address competition) that are associated with market conditions of intense acquisition activities. On the one hand, if opportunities arise with acquisition intensity in a new category (e.g., ideas for the “next big thing,” market space for niche players), generic capabilities on the platform then enable a developer to be more accurate in identifying opportunities where network effects might exist, as well as rapidly develop and deliver higher quality products that appeal to customer demands. On the other hand, if acquisition intensity is associated with severe competition in the newly entered category, generic capabilities allow the experienced developers to withstand ex-post competition from merging firms, either by claiming certain parts of the market space or diverting resources from existing businesses to support direct competition from merging firms. When a market is relatively mature, experiencing low growth rate, and thus highly competitive, lack of experience becomes a liability; however, established firms are better positioned to carry out defensive strategies to protect existing territories and carefully expand businesses. In contrast, developers without much experience on the platform are less capable of dealing with ex-post competition and capturing opportunities that are opened up by acquisition signals. If further market opportunities follow acquisition signals, inexperienced entrants are less capable of capturing such opportunities. If intense competition is associated with acquisition signals, such competition will amplify the limited resources of inexperienced new entrants, putting them at a competitive disadvantage. To summarize, the value of general platform experiences will be better reflected when market opportunities or threats are present (such as in times of intense acquisition activities). Accordingly, Hypothesis 2.7: There will be a negative interaction effect between a developer’s general platform experience and acquisition intensity at the time of its entry in influencing the likelihood of failure of its first products. 156 Distant versus proximate experience. Possessing proximate experiences means that the developer has the capability to capture opportunities and deal with competition in adjacent markets (or in a recent time of the market). Knowledge about local domains takes time to develop and is difficult to replicate by firms that operate in distant domains. From my field interviews, for instance, app developers of certain categories (e.g., gaming) suggest that they have developed their domain-specific skills and knowledge for more than a decade. They suspect that (capable) developers in other areas have also taken a long time to develop their knowledge assets. Therefore, they conclude, it would be imprudent for them to enter (distant) markets to compete with companies that have specialized in that area for a long time. As a result, between entrants with proximate versus distant experiences I argue the former will have better product performance when entering a category during its acquisition heydays. First, regarding the aspect of acquisitions signaling market opportunities, developers with proximate experiences are better equipped with abilities to recognize and capture opportunities in neighboring domains (Cohen and Levinthal, 1990). Second, regarding the argument that acquisitions may signal severe ex-post competition, previous research has suggested that related competencies reduce firm failure when entering a new market (e.g., Carroll et al., 1996; Dosi & Teece, 1998). I further submit that the advantages of related competencies (e.g., Davis & Thomas, 1993) are realized when the market is highly competitive, because under such market conditions product success (e.g., Lee, Venkatraman, Tanriverdi, & Iyer, 2010; Nerkar & Roberts, 2004) and survival (e.g., Dencker, Gruber, & Shah, 2009; Franco et al., 2009) are more challenging. Therefore, I expect that, among entrants to a segment following acquisition heydays, the more proximate experiences they possess, the better their first products are likely to fare. That is, 157 Hypothesis 2.8: There will be a negative interaction effect between a developer’s proximate experience and acquisition intensity at the time of its entry in influencing the likelihood of failure of its first products. Narrow versus broad experiences. What differentiates broad scope- versus narrow-scope developers is that the former possess more generic capabilities (that can be applied across market segments) while the latter have domain-specific capabilities (that may be associated with deeper knowledge of the specialized market domains). I argue that, in general, when entering a category following acquisition signals, broad- scope developers tend to have worse first-product performance than narrow-scope developers. First, if acquisition intensity signaled market competition, then scope diseconomies—the situation in which the costs of coordinating activities across segments may outweigh the benefits from synergies (e.g., Hashai, 2015; Macher & Boerner, 2006; Porter, 1985)—associated with broad-scope developers will be amplified. This is because competitive intensity pushes the productivity frontier further to the extent that specialization, rather than diversification, will be rewarded by the market (e.g., Teodoridis, 2016). Broad-scope developers without deep capabilities in any product category are less capable of withstanding the pressures of intensified competition. Second, if acquisitions signaled market opportunities, narrow-scoped developers would tend to have stronger capabilities to capture the opportunities. Per the logic of path dependency, when narrow-scope developers enter new product categories we would expect that these are adjacent categories (because such developers have the tendency to maintain a narrow scope). As a result, based on the logic of the previous hypothesis (H7), domain-specific capabilities of narrow-scope developers will have the skills to better capture opportunities in an adjacent-product category. In sum, among entrants who enter a category in the aftermath of intense acquisition 158 activities, broad-scope developers are likely to witness worse entry-product performance (than narrow-scope developers) because they are less capable of addressing ex-post competition and capturing market opportunities. These arguments lead to the final research hypothesis. Hypothesis 2.9: There will be a positive interaction effect between a developer’s broad-scope experience and acquisition intensity at the time of its entry in influencing the likelihood of failure of its first products. METHODOLOGY Setting and Data The setting to test my hypotheses is the mobile app market for Apple’s mobile platform iOS— the App Store. I specifically focus on one group of complementors of the platform: iOS-app developers. The setting satisfies several critical assumptions of the theory and thus is ideally suited for testing the predictions. First, the mobile-app market is a typical platform-based market with information products in which positive feedback exists. I have argued that positive feedback is a critical motivation for app developers to act swiftly in making segment-entry decisions, and/or to acquire other app companies (and hence, their installed base of users) to strengthen competitive advantages. Second, the iOS platform and its associated App Store have clearly defined market segments, or app categories. Because mobile app developers typically have a portfolio of products, which may fall into multiple product categories, I am able to precisely identify the timing and categories that a developer chooses to enter. To test the hypotheses, I combined two large groups of data. The first data set consists of all mobile-app developers and their app products. The data were compiled through a laborious process of scrapping multiple web sources, designing and constructing data formats to process raw data, importing to the database management software MySQL, and combining data from different sources. I obtained the product-category information of 1,199,021 unique apps in the 159 U.S. market from Apple’s iTunes website as of October, 2015, and release dates of 1,437,777 apps from a private analytics company. Release-date information consisted of an app’s initial releases and subsequent version releases. I cross-checked and cleaned the data to ensure that the app’s release date was the date of its earliest version. After dropping apps with missing values of release dates or category information, the final sample consists of 1,137,298 unique apps, which is quite comprehensive, and approximates the population of both developers and apps. 62 The second data set consists of acquisition events that occurred in the iOS platform mobile app market. I combined two sources of data and hand-coded acquisition events. Because acquisitions in this industry typically occur with small startups, and because databases such as SDC Platinum may not cover such cases exhaustively, I mainly relied on two comprehensive acquisition lists: Crunchbase, and a list provided by a proprietary data company. For instance, from the Crunchbase acquisition list, I first identified the events whose associated acquirers and/or targets have mobile businesses by searching keywords in the companies’ category lists. 63 The keyword search narrowed the sample down to more than 5,000 acquisitions. I then worked with three trained research assistants to read case-by-case in order to identify acquisitions that had occurred in the iOS mobile market. I coded the initial several hundred acquisition events, then verified and finalized all the remaining events identified by the research assistants. To ensure accuracy, many identified acquisition events went through multiple iterations during the coding process. We cross-checked event-associated companies’ websites, iOS developer pages, news reports, and the Internet Archive (internetarchive.org), to match event companies to iOS 62 The entire population of iOS apps, according to Apple’s World Wide Developer Conference in June 2015, is 1.5 million. Since many apps were withdrawn from the App Store, the accessible population of developers and apps (of the U.S. market) can be best reflected in our data extracted from the iTunes website, which consists of 1,199,021 ID’s. Thus, my final sample represents most of the population (nearly 95%). 63 These include 58 keywords in the “company_category_list” and “acquirer_category_list” fields of the Crunchbase data. A few example keywords include: iOS, iPad, iPhone, iPod Touch, App Discovery, App Store, Apps, Mobile, Mobile Analytics, Mobile Security, Mobile Shopping, and so forth. 160 developers and corresponding product categories. An event might be recorded for multiple categories if the acquirer and/or target spanned multiple app categories. That is, multiple market segments can be affected by a single event (e.g., Kim & Singal, 1993). In total, I identified 1,009 acquisitions that occurred in this market between July 2008 and November 2015 (Figure 9) and across different product categories (Figure 10), with a total of 1,059 unique category-months that received event treatment (note that some acquisitions span multiple categories). Finally, I expanded the acquisition data by obtaining supplementary information (such as acquirer size in terms of number of employees and assets) on the same deals from SDC Platinum. ----------------Figures 9 and 10 about here---------------- Firm Category-level Analyses Predicting Category Entry I tested Hypotheses 1–4 that predicted entry mainly at the developer-category level. The data structure I used is a developer-category-month panel, with each observation reflecting !"#"$%&"' ( ′* decision to enter +,-".%'/ 0 at 1%2- ℎ 4 . The iOS platform includes 23 developer- level categories (e.g., Books, Entertainment) as of early 2015. I used conditional logit models (Carnabuci et al., 2015; McFadden, 1973) to estimate how acquisition intensity and its interaction effects with the hypothesized moderators affect the probability that an app developer enters a category. The model is equivalent to a fixed-effects logit model in which the interest is in within-group effects—that is, why segment entries are made for certain developer-category-months conditional on existing alternatives available in the group. I thus followed the logic of case-control designs (e.g., Carnabuci et al., 2015; Rogan & Sorenson, 2014) to identify the set of realized entry events and control cases. For the realized- case group, I extracted all developer-category-months where an entry occurred. There are 161 702,942 developer-category-month observations in the group of realized cases, which were carried out by 281,771 unique iOS developers. To build control cases, I followed prior research and applied two approaches (e.g., Carnabuci et al., 2015). The first approach I used included all the other categories that the focal developer could have entered in the month but did not (i.e., all !"#"$%&"'−+,-".%'/ 0 −1%2- ℎ observations where an entry did not occur for the developer in the month), resulting in 8,839,252 developer-category-month observations in the control-case group. Because of missing values on some variables (e.g., size of acquirer), a certain number of observations were dropped in regression analyses, for instance, resulting in a total of 7,889,259 developer-category-month observations in conditional logit models (i.e., 83% of the total number of realized- and control-case observations). The second approach I used was a matched sample after applying coarsened exact matching. 64 The main model that is used to conduct segment-entry analysis is in the following form: 6'%7 82-'/ (04 = 1|< 0 4=> ? ,A (4 ,B 0 = C(< 0 4=> ? ∗F+A (4 +B 0 ), where 82-'/ (04 is a dummy variable that takes a value of 1 when !"#"$%&"' ( enters +,-".%'/ 0 at 1%2- ℎ 4 , and 0 otherwise. < 0(4=>) ? is the vector of time-variant category characteristics, which includes Acquisition Intensity, measured as the cumulative number of acquisitions that occurred in the preceding three months in the category (or IJKBL*L-L%2 0(4=>) M >NO ), 65 B 0 is category fixed effects, and P and F are regression coefficients. In conditional logit models, I defined the group as a developer-month. Thus, developer-month fixed effects (A (4 ) are automatically accounted for. 64 Results of analyses based on the CEM-matched sample are presented in Appendix 6. This alternative approach yielded results largely consistent with the reported main results. 65 In the supplementary analyses, I tested robustness of findings by using lagged count measures of acquisitions (rather than the cumulative measure). I lagged by different periods, ranging from one month to six months. The rationale to lag the explanatory variable by six months as the maximum is based on my observations and field interviews. Quick response is important for app developers to catch up with the market. An interviewee from a gaming-app company suggested that game developers with fast turn-around could release products in three months, but for others focused on building a strong core game, the process could take six months. 162 All variation in the dependent variable is explained by time-variant category characteristics, and their interactions. To test Hypotheses 2–4, I examined the interactions between cumulative acquisitions and the following variables related to developer experiences. General platform experience. General platform experiences are reflected by multiple measures. First, I categorized developers that enter a category into two groups: (1) platform entrants, who launched their first product(s) in the platform in the month, and (2) platform incumbents, who were already in the platform but did not yet have a product in the category- month. I coded a dummy variable, platform incumbent, as 0 to indicate if the firm is a platform entrant in the month, and 1 when it becomes an incumbent. Second, platform age was measured as number of months since a developer launched its first product, scaled to years. Finally, following the learning literature (e.g., Argote & Epple, 1990), I measured a developer’s product- development experience as the number of apps it had developed for the platform. Proximate versus distant experiences. Experience proximity can be reflected in both space and time dimensions. With respect to the space dimension, I measured proximate experience in terms of closeness of a developer’s existing product categories to the target category, or category relatedness. The variable consists of two parts: the first is the closeness between two categories: + (0 /+ ( , where + (0 is the count number of developers that are simultaneously (1) in an existing category of the focal developer (+,-".%'/ ( ) and (2) in the target category (+,-".%'/ 0 ). The second part accounts for the developer’s number of products in its existing categories (R ( ). Thus, the developer's category relatedness to the target category (+,-".%'/ 0 ) is measured as R ( ×(+ (0 /+ ( ) ( , where L covers all existing categories of the developer. 66 66 The category-relatedness measure is adapted from Silverman’s technological-relatedness measure (1999). 163 With respect to the time dimension, I built on previous research (e.g., Macher & Boerner, 2006) and measured time-depreciated experience in the form of δ ( UV ( 8<6 4=( , where 8<6 4=( is the number of products released in time -−L, where L traces backwards to the initial month (2008 July). δ is the depreciation factor that took a value of 70%. 67 Narrow versus broad experiences. A developer’s product scope is measured in two ways. The first is a simple count of a developer’s product categories (Cottrell & Nault, 2004). This approach, however, does not account for product dispersion across categories. The second measure, diversification index, is an adaptation of the Herfindahl index and is in the form of 1− W ( M , where W ( is the developer’s percentage of products in category L (Teodoridis, 2016). The higher the value, the more diversified the developer’s product portfolio. Control variables. Because the explanatory power of variations in the dependent variable resides in time-variant category attributes, I controlled for another type of event that occurs in a category-month: the number of initial public offerings (IPOs). To account for how density of developers in a segment affects acquisition and competitive entries, I controlled for segment size in terms of number of developers. 68 To further tease out effects of market competition, I also controlled for concentration ratio (measured as percentage of apps released by top-10 developers). Another time-variant category control is acquirer size, measured as either total assets or number of employees, 69 to partially account for acquisition heterogeneities. I added a control variable of market growth to address the concern that entrants might be attracted due to market hotness. I used a flow variable—number of apps released in a category- 67 I conducted robustness checks by setting the depreciation factor at alternative values (i.e., 80% and 60%). The results, reported in Appendix 7, are substantively the same. 68 Alternatively, I also tested with models controlling for category size as number of apps in a category-month. The results are presented in Appendix 5. 69 Appendix 6 presents the results when acquirer size was measured as number of employees. 164 month—as the building block to construct the market growth measure. The purpose was to reflect a product category’s growth speed (or market hotness). I build on prior literature’s (Dess & Beard, 1984) theoretical argument to create the measure, and the procedure can be summarized in the following two steps. In the first step, I regress the flow variable (number of released apps in a category-month) on dummy variables of years in a random-coefficient maximum likelihood model. The model is in the form of X = P MOOU Y +P MOOZ Y [ MOOZ +⋯+ P MO]^ Y [ MO]^ +_, where each coefficient P Y has its own mean and standard deviation and can vary across categories. [ MOOZ through [ MO]^ are indicator variables of years, and the year 2008 was set as the baseline (as reflected by the intercept). In the second step, I obtained the estimated coefficients for each category; for instance, the “Games” category has a set of predicted regression coefficients: P MOOU abcde through P MO]^ abcde . I then adopted the logic suggested by prior studies (Dess & Beard, 1984) to use the predicted coefficients to divide the average value of the flow variable (monthly app releases) during the entire period of 2008–2015 (i.e., P MOOUfgh=MO]^ib> abcde for the “Games” category). Thus, the (yearly) market growth measure is in the form of a predicted regression coefficient divided by a corresponding average value. For instance, the market growth for the games category in 2010 is j klml nopqr j kllstuvwklmxyoz nopqr . Finally, in the conditional logit models, the developer-month fixed effects are accounted for through group specification. The developer-month fixed effects thus absorb both month fixed effects (because there was no within-group variation for months) and developer fixed effects. Additionally, category fixed effects were added to tease out time-invariant variations across categories. Results. Table 11 provides descriptive statistics of the key variables used in the analysis. 165 The table shows that the maximum number of acquisitions in 3 months (or Acquisition Intensity) is 30; there are more category entries from platform entrants (accounting for 68% of entries) than from incumbents (32%); the maximum number of IPOs in a category-month is 3. ----------------Table 11 about here---------------- Results of conditional logit models are presented in Tables 12, in which the dependent variable is the Entry dummy variable. It is important to note that, due to the developer-month fixed effects, experience variables that are developer-specific were absorbed. ---------------- Table 12 about here---------------- Hypotheses 2.1a (H2.1a) and 2.1b (H2.1b) predict opposing effects of acquisition intensity on entry probability. The coefficient of acquisitions is positive and statistically significant (p < .01) in Model 1, thus supporting H1a. With respect to effect sizes, because conditional logit models do not have intercepts, and because fixed effects are omitted in the estimation, interpretation of effect magnitudes will be based on odds (of entry versus non-entry probability) and odds ratio (of how the odds change given one unit change of an independent variable). All else equal, increasing Acquisition Intensity by 1, the odds change by a factor of 1.002 (or " O.OOM ), or, the odds of entry increases by 0.2%. Increasing Acquisition Intensity by 30 (i.e., maximum in the data) will result in a change in odds by 1.062 (or " O.OOM∗|O ), or the odds of entry increases by 6.2%. Hypothesis 2.2 predicts that general platform experience will negatively moderate the effect of acquisition intensity on entry probability. In Model 2, we can see that the interaction between Platform Incumbent and Acquisitions is negative (p < 0.001). Model 3 shows that the interaction effect between Platform Age and Acquisitions is also negative (p < 0.001). Similarly, Product-development Experience also has a negative interaction effect with Acquisitions (p < 166 0.001) (Model 4). Thus, all three measures of general platform experience have negative moderating effects on the link between acquisition intensity and the likelihood of segment entry, providing strong evidence in support of Hypothesis 2.2. In terms of magnitudes of the effects, let us refer to odds ratio as the case when Acquisition Intensity is increased by 5 (the mean) (which equals to }~~e Äd> fYÅg(e(4(}> Ç>4d>e(4ÉN^ }~~e Äd> fYÅg(e(4(}> Ç>4d>e(4ÉNO ). Then, the odds ratio for a platform entrant (Platform Incumbent = 0) is 1.025 (i.e., " O.OO^∗^ ), or an increase in the odds of entry by 2.5%; however, the odds ratio for a platform incumbent (Platform Incumbent = 1) is 0.970 (i.e., " O.OO^∗^ ∗" =O.O]]∗^ ), or a reduction in the odds of entry by 3%. The odds ratio for a developer with 0 years of experience is 1.015, while for a developer with 2 years of experience it is 0.975. Similarly, the odds ratio for a developer with 0 apps is 1.010, while for a developer with 10 existing apps the corresponding ratio is 0.961. To show the interaction effects graphically, I applied a simulation approach for non- linear logit models (Zelner, 2009). Although logit model specifications using the simulation approach are not the same as those in the conditional logit models (in that the large number of developer-month fixed-effect dummy variables exceeds the limit of allowed variables by Stata, and therefore not included), the magnitudes of coefficients are qualitatively similar. Table 13 presents the results of logit models based on which interaction graphs were generated. Figures 11–13 depict the interaction effects between acquisitions and different measures of general platform experience. In the figures, the vertical axis shows the probability of entering a category, and the horizontal axis depicts the cumulative number of acquisitions in the past three months. Experienced versus less experienced developers are distinguished by solid versus dashed lines. Areas surrounding the lines are 95% confidence intervals. The figures indicate that, while the probability of entering a category increases as acquisition intensity increases for the 167 inexperienced (including platform entrants, those with lower platform age, and those with less product-development experiences), the entry probability decreases as acquisition intensity increases for the experienced (including platform incumbents, those with higher platform age, and those with more product-development experiences). This finding is confirmed by tests of the slope for inexperienced (platform incumbent = 0, platform age = 0, and product experience = 0) and the slope for the experienced (platform incumbent = 1, platform age = 2, and product experience = 10), 70 based on conditional logit models and treating the dependent variable as log odds (i.e., linear regression line). ----------------Figures 11, 12, and 13 about here---------------- In H2.3, I predicted that proximate experience would negatively moderate the effect of acquisition intensity on the likelihood of category entry. In support of this hypothesis, Models 5 and 6 of Table 12 show that the interactions between acquisitions and (1) time-depreciated experiences, and (2) category relatedness are significantly negative (p < 0.001). In terms of effect sizes, the odds ratio for a developer with 0 time-depreciated experience is 1.010 (i.e., " O.OOM∗^ ), while that of a developer with more experience (time-depreciated experience = 2) is 0.980 (i.e., " O.OOM∗^ ∗" =O.OO|∗^∗M ), a difference of 3% for the two groups. The odds ratio for a developer with low category relatedness (= 0) is 1.015, while that for a developer with high category relatedness (= 3) is 0.985. These effects are also shown in figures generated via simulation approaches (Figures 14 and 15). Slope tests based on conditional logit model results suggest that the slope for developers with proximate experiences (time-depreciated experiences = 0, category relatedness = 0.27) is positive, while the slope for developers with distant 70 Results of slope tests are: when platform incumbent = 1 (chi-squared = 91.77, p < 0.001); when platform age = 2 (chi-squared = 52.48, p < 0.001); when product experience = 10 (chi-squared = 65.71, p < 0.001). 168 experiences (time-depreciated experiences = 2.1, category relatedness = 2.23) 71 is negative. Thus, acquisition intensity reduces the entry probability for those with more proximate experiences (in both time and space) but increases the likelihood of entry by those with more distant experiences. ----------------Figures 14 and 15 about here---------------- H2.4 predicts that a developer’s experience breadth will positively moderate the effect of acquisitions on entry likelihood (i.e., a positive interaction effect). The hypothesis is supported for the diversification index of product scope (Model 8) but not number of product categories (Model 7), which, contrary to predictions, show a negative interaction effect. Therefore, the hypothesis is partially supported. To have a sense of the effect size (based Model 8), the odds ratio for a developer with a narrow-scope product portfolio (diversification index = 0) is 0.966 (i.e., " =O.OOV∗^ ) while that for a broad-scope developer (diversification index = 0.5) is 1.051 (i.e., " =O.OOV∗^ ∗" O.O|Ñ∗^∗O.^ ). What is meaningful is the difference across the two groups—an 8.6% difference in odds ratio when responding to acquisitions. Interaction effects based on results of Table 13 are depicted in Figures 16 and 17. For instance, from Figure 17, we can see that a more diversified developer is more likely to enter a category than a focused one (i.e., the main effect, irrespective of the influence of acquisitions), and that the entry probability of such a broad-scope developer is positively associated with acquisition intensity. ----------------Figures 16 and 17 about here---------------- To examine whether the three types of experiences distinctively explain the variance in entry decisions, I included one measure for each experience type in Models 9 and 10. As can be seen from the models with all three types of experiences included, the effect directions are 71 Results of slope tests are: when time-depreciated experiences = 2.1 (chi-squared = 65.38, p < 0.001), and when category relatedness = 2.23 (chi-squared = 31.75, p < 0.001). 169 consistent and effect sizes are stable, thus lending further support to my theoretical predictions. Intending to probe heterogeneities of network effects across product categories, Hypothesis 2.5 predicts that the opposing reactions by experienced versus inexperienced developers towards acquisitions will be amplified if such signals are from strong-network-effect categories than if from weak-network-effect categories. To test the hypothesis, I build on the model of interaction effects between acquisition intensity and platform incumbent (Model 2) and add a third conditioning dummy variable, which indicates whether the focal category has stronger network effects. Amongst the 23 primary categories of iOS platform, I select two categories that, according to research and business press, represent the strongest network effects—games (e.g., Miric, 2017) and social networking—to code dummy variables, comparing them with the rest of the categories (with weaker network effects). In Models 11 and 12, I add the interaction effects with the dummy variable “games.” Models 13 and 14 enter interactions with “social networking” dummy. Models 15 and 16 enter interaction terms with a dummy variable that combines the two categories: “game-and-social- network.” Because the three sets of models suggest the same pattern of results, I interpret results in Models 15 and 16 in detail. In Model 15, in which acquisition intensity and its interaction term with the “game-and- social-network” dummy are entered, we can see that both the main effect of acquisition intensity and its interaction with the “game-and-social-network” dummy variable are positive. This suggests that acquisitions are positively associated with entry likelihood in general, and that the positive effect is stronger in “games” and “social networking” categories. To test H2.5, I focus on the results of Model 16. First, in categories with weaker network effects (i.e., when the “game-and-social-network” dummy = 0), the association between acquisition intensity and entry 170 likelihood for a platform entrant is 0.004. Second, in categories with weaker network effects, the association between acquisition intensity and entry likelihood for a platform incumbent is -0.007. Third, in categories with stronger network effects (i.e., when the “game-and-social-network” dummy = 1), the association between acquisition intensity and entry likelihood for a platform entrant is 0.022. Finally, in categories with stronger network effects, the association between acquisition intensity and entry likelihood for a platform incumbent is -0.023. Comparing the four coefficients as described in the previous four points, we can see that acquisition intensity’s opposing effects on platform entrants versus platform incumbents are amplified in “games” and “social networking” categories (i.e., categories with stronger network effects). The empirical evidence thus supports Hypothesis 5. Predicting Post-Entry App Failure An important metric of product performance, in the App Store, is whether a developer keeps updating an app to newer versions. Version updates could be either dealing with problems of a product (e.g., fixing bugs) or introducing novel features, especially in times when the market has discontinuous changes (such as operating system major updates) (Kapoor & Agarwal, 2017). An app that has not been updated for a relatively long period, such as one year, is most likely to have been abandoned by the developer and forgotten by consumers. Industry practitioners label such products “zombie apps,” or what I refer to as failed products. In fact, to be considered for the award of best apps of the year on certain platforms (e.g., Android), one criterion is that the app has been updated in the past 12 months. Thus, I adopt a dichotomous metric to differentiate failed first products to others entry products—that is, whether it has been updated for at least a year since its last version update. 171 To examine a developer’s post-entry product performance, I coded the dependent variable as a dichotomous variable that indicates whether a first product by a category entrant is failed. A first product is an app that the developer launches in the first month it enters a category. In cases when the developer releases multiple products in the same month, all first products were included in analyses. Thus, the dependent variable, App Failure, is operationalized as a dummy variable that equals to 1 if the app has not been updated for at least 12 months 72 from its most recent version update to the month when I started data collection of app version history (i.e., January 2015), and 0 otherwise. Data structure and analysis. The level of analysis for post-entry performance is changed to the more fine-grained app level—a category-app, that is, an app that is released in a category as a developer’s first product. Notice that this is different from an app as a unit of analysis, as an app can be associated with multiple categories. The sample, therefore, consists of all category- apps that are first products to a newly entered category. The data structure for post-entry performance analyses, therefore, is a pooled cross-section of all category-apps as first products to newly entered categories. Based on the category-app data structure, I merged the variables that take the same values as used to predict entry decisions. Because a time lag exists between releasing a first product when entering the category and eventual app failure, the analyses are akin to testing category-entry conditions’ imprinting effects on a first product. I also merged the lambda variable of the inverse Mill’s ratio that is generated from a first-stage conditional logit model (Model 10 in Table 12) estimating entry probability. This helps to address the concern over potential selection in that I could only observe first products of developers who decided to enter a category (as opposed to those who decided not to). Descriptive statistics of the variables 72 I conducted a robustness check with alternative cutoffs that define an app failure (i.e., 6, 18, or 24 months). The results are qualitatively the same when defined as 6 and 24 months, but are less similar when defined as 18 months. 172 are presented in Table 14. ---------------- Table 14 about here ---------------- I used logit models to estimate the likelihood of app failure. At the product-category level, I controlled for the same set of category-specific variables (Category Size, Market Concentration Ratio, Market Growth, Average Acquirer Size), as well as category fixed effects. At the developer level, developer fixed effects were controlled to wipe out the effects of developer- specific time-invariant unobservables. 73 I entered inverse Mill’s ratio in all the models to address the concern that only products of developers who decided to enter a category (versus other unrealized potential entrants) are observed. To have an initial understanding of the dependent variable, I tabulated app failure across the explanatory variables in Table 15. The mean values of app failure in the table suggest that the failure likelihood is lower when acquisition intensity is low, when developer experience is high, or when narrow in scope. To have a sense of whether the descriptive statistics lend support to the moderating effect hypotheses, we could look at the second-order difference in each of the sections that tabulate an experience variable and acquisition intensity. For instance, the first 73 It is important to note that because of adding developer fixed effects, the sample is further restricted to category- apps (1) whose corresponding developers have multiple products, and (2) within a developer there are variations in the dependent variable—that is, the developer has both failed and non-failed first products. A major advantage of the developer fixed-effects specification is that if developer-specific invariant unobservables are correlated with both explanatory covariates and the dependent variable, then such confounding effects are accounted for when estimating regression coefficients. A disadvantage, however, is that findings are not generalizable to category-apps (and developers) that are not included in the analyses. Generalizability of findings, however, can be realized in models with developer random effects, which are reported in Appendix 8. Although results of the full-sample developer- random-effect models (Appendix 8) and results of the selective-sample, developer-fixed-effects models (Table 16) are similar, there are some dissimilarities regarding whether hypotheses are supported. To briefly summarize findings of the developer random-effects models: First, H2.5a, which predicts a positive effect of acquisition intensity on the likelihood of app failure, is supported by results in Model 1 of Appendix 8; second, H2.6, which states a negative moderating effect by general platform experience, is supported by results in Model 2 (platform incumbent) and Model 3 (platform age), but not by Model 4 (product-development experience); third, H2.7, which predicts a negative moderating effect by proximate experience, is not supported by results in both Models 5 (time- depreciated experience) and Model 6 (category relatedness); and finally, H2.8, which predicts a positive moderating effect of broad-scope experience, is supported by results in both Models 8 (number of product categories) and Model 9 (diversification index). 173 order difference for platform entrants when acquisition intensity moves from low to high is 0.02 (or 0.74 - 0.72), and for platform incumbents is -0.20 (or 0.35 - 0.55). The second-order difference, reflecting the interaction effect, is -0.22 (or -0.20 - 0.02), suggesting that first products by platform incumbents when acquisition intensity is high are even less likely to fail. For formal tests of the post-entry performance hypotheses, I next provide and interpret results of regression analyses. ---------------- Table 15 about here ---------------- Results. Model 1 in Table 16 includes the main effect of acquisition intensity and category-level control variables. From the results of category control variables in Model 1, we can see that a category-app is more likely to fail if it was launched in categories that are small (Category Size), that have low market concentration (Market Concentration Ratio), that has low growth rate (Market Growth), and where acquirers in acquisitions deals are smaller (Average Acquirer Size in Total Assets). In the sampled 154,100 category-apps (as first products to newly entered categories), 85,125 category-apps (or, 55%) eventually fail (based on the assumption that an app is failed if it has not been updated for at least 12 months). 74 ---------------- Table 16 about here ---------------- Hypotheses 2.6a and 2.6b are competing hypotheses, with the former predicting that first products introduced in a category with intense acquisition activities are less likely to fail (i.e., the market opportunity argument), while the latter predicting a greater likelihood of failure for such products (i.e., the market competition argument). As can be seen from Model 1, the main effect of acquisition intensity on the likelihood of app failure is negative and significant (P = 74 The percentage of failed apps, as identified in this study, is smaller than those reported by app analytic companies. If a more stringent criteria is used (for instance, an app needs to be ranked on top-ranking lists to be considered alive), the percentage of failed apps can be higher (e.g., more than 80%) according to some industry reports (e.g., https://techcrunch.com/2015/01/30/zombie-apps-on-the-rise-83-of-apps-not-on-top-lists-up-from-74-last-year/, accessed on 4-16-2017). 174 −0.038,& <0.001). Thus, the finding supports H6a, or the market opportunity argument. Models 2–4 test Hypothesis 2.7, which predicts that general platform experience will have a negative interaction effect with acquisition intensity in influencing app-failure likelihood. As can be seen, the interaction terms of all three experience measures—platform incumbent (P = −0.093,& <0.001 ), platform age ( P = −0.030,& <0.001 ), and product-development experience (P = −0.002,& <0.001)—with acquisition intensity have a negative and significant effect on the likelihood of app failure. These findings suggest that, among first products that are launched during a category’s acquisition heydays, products created by experienced developers are even less likely to fail than those by inexperienced developers. Hypothesis 2.7 is therefore supported. Similarly, Hypothesis 2.8 predicts a negative interaction effect between proximate experiences and acquisition intensity. As can be seen in Model 5 and Model 6, the interaction terms between acquisition intensity and the two aspects of proximate experiences—time- depreciated experience (P = −0.006,& <0.001), and category relatedness (P = −0.003,& < 0.001)—are all significantly negative. These findings suggest that, among products that are launched during a category’s acquisition heydays, those by developers with more proximate experiences are even less likely to fail, thus supporting H2.8. Finally, in Hypothesis 2.9 I propose that broad-scope experience will have a positive interaction effect with acquisition intensity in influencing first products’ likelihood of failure. Models 7 and 8 test the hypothesis. Hypothesis 2.9 is supported in the model using one measure of broad-scope experience (diversification index in Model 8; P = 0.037,& <0.01), but is not supported in the model using the other measure (number of product categories in Model 7; P = −0.002, not statistically significant). Given that diversification index is a more fine-grained 175 measure that accounts for the proportion of products in the focal developer’s categories, the empirical evidence therefore provides support to the hypothesis (H2.9). To mitigate the concern that these three types of experiences may not be distinct due to high correlations between experience variables, I provide a more fully specified model (Model 9) in Table 16, which includes one measure from each of the three types of experiences—platform incumbent, category relatedness, and diversification index. As can be seen from Model 9, although magnitudes of coefficients and significance levels drop for interaction effects with category relatedness (P = −0.001, now not statistically significant) and diversification index (P = 0.021,& <0.1), directions of all the interaction terms remain the same. In addition, the interaction term between acquisition intensity and platform incumbent (P = −0.091,& <0.001) shows strong support for the hypothesis. Hence, I conclude that Hypotheses 2.6–2.9, which are the theoretical predictions regarding experiences’ moderating effects, are generally supported by the empirical evidence. DISCUSSION AND CONCLUSION This paper was motivated by my desire to understand how market structure of a platform-based market and complementor heterogeneities jointly determine category-entry decisions and post- entry product performance. While impressive work has been done on how platforms compete with one another (e.g., Boudreau, 2010; Eisenmann et al., 2011; Eisenmann et al., 2006; Zhu & Iansiti, 2012), and on how platform policies influence platform participants’ decisions to join (e.g., Boudreau, 2010; Majumdar & Venkataraman, 1998; Venkatraman & Lee, 2004), less is known about the strategic decisions of complementors. Overall, the theory on platform-based competition in the strategy field, particularly in relation to how complementors compete, has 176 received little development. In order to address this gap in the literature, I drew upon insights from the platform literature along with two major theories in strategic management—industrial organization (Porter, 1981; Porter, 1985) and organizational experiences from the behavioral theory of the firm (e.g., Argote & Epple, 1990; Eggers, 2012; Levinthal & Wu, 2010; March, 2010; Wernerfelt, 1984)—to examine the following research questions. First, how do complementors with heterogeneous experiences respond to acquisition signals from a market segment when making segment-entry decisions, and how do such responses differ in categories with strong versus weak network effects? Second, how do acquisition intensity and developers’ experiences, specifically at the time when a developer enters a segment, influence the subsequent likelihood of failure of the developer’s first product(s)? A key insight that my paper provides is that both external market shifts and internal firm- specific heterogeneities shape entry decisions, consistent with the core theoretical arguments of two major schools of thought in the strategy field—industrial organization and the behavioral theory of the firm. The finding that acquisition intensity in a segment could either deter or encourage subsequent entries highlights the explanatory power of market structure shifts. The conclusion that complementors with heterogeneous prior experiences react to acquisitions in different ways indicates that the two major perspectives in the field actually complement one another in providing a more complete explanation for heterogeneities in firms’ strategic choices (Porter, 1981). My study reveals that, while app developers with certain types of experiences (general, proximate, and narrow-scope) are deterred from entering a segment following acquisitions, developers with the opposite experiences (less general, distant, and broad-scope) tend to rush into segments in which acquisitions abound (e.g., approximately 70% of category- entries were by platform entrants). At the post-entry stage, products launched during acquisition 177 heydays witness higher performance than usual, and those by developers with more general, proximate, and narrow-scope experiences benefit even more. I discuss specific findings next. First, acquisition intensity in a market segment increases the subsequent likelihood of a developer entering the segment. This finding responds to a longstanding debate on whether acquisitions facilitate or deter competition (e.g., Clougherty & Duso, 2009; Gaur et al., 2013; Kim & Singal, 1993; Prager, 1992; Stigler, 1950). One explanation for this finding is that, when app developers assess the potential costs and benefits of entering a new segment upon receiving acquisition signals, in general they perceive that benefits outweigh costs. In a platform-based market characterized by information products, major sources of entry costs are potential competition and retaliation from merging firms. Potential benefits accrue from the possibilities of becoming “the next big thing,” the next acquisition target, and/or entering a validated market with weakened competition. Overall, the empirical findings in support of Hypothesis 1a appear to be aligned with “the bright side” of acquisitions, consistent with prior studies that suggest acquisitions could provide validation of a market segment (e.g., Clougherty and Duso, 2009; Gaur et al., 2013). Second, the findings of the interaction effects on segment-entry probability suggest that not all developers respond to acquisition signals in the same manner. I argued that a developer’s perception of potential costs and benefits of entering a segment frequented by acquisition events are shaped by the developer’s experiences. The findings support the prediction that general platform experiences deter developers from entering categories where acquisitions abound, because they trade potential benefits (e.g., replicating prior successes to capture opportunities where network effects are strong) in favor of costs (e.g., potential competition from merging companies). Interestingly, the findings suggest that the less experienced developers seem to put 178 greater value on potential benefits and are thus more likely to enter a segment in which acquisitions occur frequently. Such opposing effects for experienced and inexperienced are further manifested in the findings on the moderating effect of proximate experiences. I found that developers with more proximate experiences are more deterred—while those with less proximate experiences are encouraged—by acquisition signals when making entry decisions. Finally, I also found that acquisition intensity is associated with an increased likelihood of entry for broad- scope developers, but a decreased entry probability for focused developers. Taken together, it appears that, when deciding whether to enter a segment following acquisitions, some developers are motivated by what they may gain, while others are constrained by what they may lose. Because the theory proposes that the common driving force for such opposing effects are positive network effects in the platform market, I further probed the role of network effects by leveraging advantages of the empirical setting. In the App Store, I identified two categories with the strongest network effects: “games” and “social networking.” I found that the aforementioned opposing reactions toward acquisitions by experienced versus inexperienced developers are enhanced if acquisition signals are from these two categories. When investigating the post-entry likelihoods of failure of entrants’ first products, I found empirical support for the theoretical prediction that, in general, acquisition intensity decreases first-product failure. The finding confirms the argument that acquisitions in a product category may open new market opportunities for potential entrants. This finding is robust even after accounting for endogeneities of entry decisions in analyses. Furthermore, I found that products during acquisition heydays by developers with certain experiences (more platform experience, proximate, and narrow-scoped) perform even better, suggesting that capabilities behind the experiences enable developers to better capture opportunities and address competition. 179 Theoretical Contribution My study has theoretical implications for multiple streams of research. First and foremost, it contributes to the literature on platform-based competition (e.g., Boudreau, 2010; Eisenmann et al., 2011; Eisenmann et al., 2006; Zhu & Iansiti, 2012) by developing a theoretical framework to examine complementors’ strategic choices. This addresses an essential gap in the platform literature, which has predominantly focused on platforms (Boudreau, 2010; Caillaud & Jullien, 2003; Chao & Derdenger, 2013; Economides & Katsamakas, 2006; Eisenmann et al., 2011; Hagiu & Eisenmann, 2007). The key insight of my study is that complementors make the strategic choice of segment entry based on both external market shifts and internal experiences. My empirical findings are that the less experienced and broad-scope developers are more likely to enter a product category following a period of intense acquisition activity in the category, while the more experienced and narrow-scoped developers appear to decide with more caution. Such opposing reactions towards acquisition signals suggest the functioning of the underlying mechanism of the platform market: (positive) network effects. On the one hand, acquisitions signal potential market opportunities where network effects exist, thus attracting inexperienced and broad-scoped developers to rush in to capture such opportunities. On the other hand, acquisitions create major market players who enjoy stronger network effects due to enlarged installed bases of customers, thus making the market segment less attractive for potential big players who tend to be more experienced and specialized. Such dual roles played by the underlying network-effect mechanisms are further manifested in the findings on category heterogeneities. I found that the aforementioned opposing reactions toward acquisitions by experienced versus less experienced developers are enhanced if acquisition signals are from categories with stronger network effects (e.g., iOS “social networking” and “games”). This 180 finding lends further support to the functioning of network effects in this platform market. Second, the effects of acquisitions on segment entries contribute to the industrial organization literature, specifically to the impact of acquisitions on market competition, in two ways. First, the literature on acquisition and market competition has mainly focused on two (opposing) mechanisms—market power, namely that merging firms accrue greater power such as greater bargaining power towards suppliers, and greater pricing power towards consumers (Fee & Thomas, 2004; Haleblian et al., 2009; Kim & Singal, 1993), and productivity efficiency, meaning that merged entities can have productivity efficiency gains as a result of the combination of complementary resources (Ashenfelter et al., 2015; Fee & Thomas, 2004; Yu, Umashankar, & Rao, 2016). Most prior research in this stream, however, has focused on the effects of acquisitions on merging firms. Only a few studies have recently started to examine the effects of acquisitions on rival firms, with the general finding that acquisitions have positive spillovers on rival firms’ stock market returns (e.g., Clougherty & Duso, 2009; Gaur et al., 2013). My paper follows this line of inquiry by adopting the perspective of rival firms—how potential competitors react to acquisition signals. It further extends this stream of research by studying market entry as an alternative dependent variable. Second, my study sheds useful additional light on a longstanding debate in the literature regarding whether acquisitions have positive or negative impacts on market competition (e.g., Clougherty & Duso, 2009; Gaur et al., 2013; Kim & Singal, 1993; Prager, 1992). I found that acquisitions generally encourage subsequent segment entries, and that entry products during a segment’s acquisition heydays are less likely to fail. The findings are therefore consistent with the argument that acquisitions generally send positive signals about a market segment. The empirical evidence on post-entry product performance further suggests that acquisitions open novel market opportunities, for instance, the possibilities 181 of leveraging network effects in a platform market. Building on my study, future research could directly investigate the effects of acquisitions on specific competitive behaviors from rivals such as pricing, R&D investments, and product releases. Third, an important contribution of this study lies in the insights it provides to a central question in the strategic management literature: What explains heterogeneities in firms’ strategic choices? In the past several decades, two broad theoretical perspectives—the behavioral theory of the firm (Argote & Greve, 2007; Cyert & March, 1963; Gavetti et al., 2012; March, 2010), and industrial organization (Caves & Porter, 1977; Porter, 1981; Porter, 1985)—have helped scholars probe this question. In this research I integrate insights from the platform literature with these two perspectives, relying on the assumption that one side alone cannot explain the heterogeneities in firm strategies and performance. As an example, let us consider the dichotomous difference between diversifying (de alio) and entrepreneurial (de novo) entrants. It would be a mistake to conclude that more experienced firms would always outperform less experienced ones, because such an argument overlooks the importance of market conditions. Similarly, it would also be imprudent to conclude that the entrants in early stage of market evolution will have an advantage over entrants of later stages, because, as prior research suggests, this argument misses the important role of firm heterogeneity (Franco et al., 2009). Hence, when we examine the role of firm heterogeneities to explain firms’ strategic choices, we also need to account for shifts in market conditions. My study shows that each factor is a necessary, but not sufficient, condition. By focusing on acquisition surges in different market segments of the iOS platform, my paper probes one source of shifts in market conditions, namely, how acquisitions alter the perceived opportunity structure of the market. By combining market shifts with heterogeneities of firm experiences, I examined which firms are more or less likely to respond to 182 acquisitions in making segment-entry decisions. Fourth, my study also contributes to the literature on organizational experiences (e.g., Argote and Epple, 1990; Eggers, 2012; Levinthal and Wu, 2010; Wang and Rajagopalan, 2015). One unresolved tension in this literature is that prior experiences can be both constraining (in that they can be sources of inertia) (e.g., Christensen, 1997) and enabling (in that they provide capabilities to create and capture value) (e.g., Wang & Rajagopalan, 2015). Previous research has adopted one of these two perspectives in examining the effects of experiences. For instance, one argument is that diversifying firms tend to outperform entrepreneurial entrants because of their pre-entry capabilities. However, this argument does not enable us to unpack the real effect of experiences. That is, we cannot parse whether determinants of a post-entry performance is because only firms with higher capabilities choose to enter to begin with, or because such firms leverage their superior capabilities after entry to outperform their rivals. Empirically, most prior research has either studied the direct effect of pre-entry experiences on market entry decisions or the direct effect of experiences on post-entry performance. By separating the two stages of segment-entry decision and post-entry performance, I was able to isolate the constraining effect of experiences in the first-stage analysis (where I proposed that inertia results in conservativeness in experienced developers’ segment-entry decisions), while at the same time demonstrate the enabling effect of experiences at the second-stage (i.e., capabilities inherited from pre-entry experiences enhance the survival of experienced entrants’ first products). The two stages, considered together, enabled me to unpack how experiences lead to experienced firms’ lower incentives in making segment-entry decisions following acquisition signals, but their better capabilities in navigating the competitive landscape at the post-entry stage. Thus, my paper provides useful new insights that may help us reconcile the conflicting expectations on whether 183 prior experiences are constraining or enabling. What is more meaningful and interesting is that, according to the empirical findings, such constraining and enabling effects of experiences are more obvious in product categories with stronger network effects. Finally, my study builds an important bridge between the strategy literature and the entrepreneurship literature. The entrepreneurship literature has mainly paid attention to how entrepreneurial firms make decisions to pursue opportunities, while largely ignoring how they react to current market conditions. In this research, I look at how new entrants’ decisions are conditioned by existing firms and opportunity structures in the market. The findings may suggest that acquisitions could potentially reduce innovation by incumbents, but acquisitions could also encourage entrepreneurship by bringing in new entrants into a platform-based market. Furthermore, the findings suggest a potential challenge for entrepreneurial firms. Namely, on the one hand, market shifts may pressurize them to enter market segments, and some of this pressure can come from investors who push them to seek new opportunities. On the other hand, such firms may seriously lack the experience (and the capabilities that such experience bestows on them) to address post-entry competition. Thus, entrepreneurial firms have to learn to balance the tension between pursuing potentially lucrative opportunities and dealing with limited experience that makes them vulnerable in a competitive market. Limitations and Future Research In closing, I acknowledge several limitations of my study and offer directions for future research. First, the sample of apps and corresponding developers that was used for the analysis only consists of those that survived until the year of data collection (2015). Therefore, I did not observe the developers who withdrew all products from the App Store. Within the current 184 sample of developers, I also did not observe their withdrawn products. If the withdrawn products are the ones that determine initial category entries, then for the affected cases the category-entry months were postponed to the first surviving products that were released within corresponding categories. This survival bias would lead to a concern over the estimated effect of acquisition intensity on segment-entry probability, if the missing data are systematic (e.g., if all of them are concentrated in category-months following low acquisition intensity). As long as the missing product information is random, which is likely to be the case because data from acquisition event and apps are collected from multiple different sources, the estimated effects should be unbiased. Second, although most findings in the paper have strong generalizability (given the approximate population data on iOS developers in 2015), the empirical evidence on experiences’ moderating effects on the relation between acquisition intensity and post-entry app failure should be interpreted with caution. This is because the developer fixed-effects model specification constrains the sample to developers with multiple first products, some of which failed and others that survived. As a result, the conclusions drawn from fixed-effect models can only be applied to this group of app developers. Nevertheless, results of developer random-effect models that use the full sample of first products are provided in Appendix 8, where, as can be seen, the main effect of acquisition intensity on the likelihood of app failure is consistent across different model specifications as well across different dependent variables (app failure and overall app rating). 75 Third, while my study only examined firm heterogeneity in terms of product-market experiences (i.e., from an output perspective), future research could also study heterogeneity that originates from technological experience (i.e., from an input perspective). For instance, technologies included in an app can be reflected in its permissions, which relate to the app’s abilities to interact with user devices such as accessing to Wifi and camera, but, as the name 75 Results when the dependent variable is app rating are presented in Appendix 9. 185 suggests, permissions require user authorization. Such technologies can be grouped, and then used to categorize apps and developers. The key is to focus on how firm attributes interact with market shifts in affecting developers’ strategic choices. Fourth, in this study I only looked at how one type of event (i.e., acquisitions) leads to market shifts. Future research could examine other market shifts triggered by other mega events (e.g., funding events). It is also possible that different types of industry mega events can interact in influencing developer behaviors. For instance, acquisitions may be followed by investors herding to fund similar businesses, which in turn attracts potential entrants. Admittedly, a weakness of the current study is that I did not control for funding events in my analyses because coding such data is a challenge given its abundance in the industry. This limitation, however, suggests meaningful future research. Furthermore, I invoked network effects as the mechanisms that acquisition intensity in a product category influences potential entrants’ entry decisions. However, my operationalization of the construct (acquisition intensity) could not entirely capture the cross-side network effects (from the side of customers to the side of app developers). To fully capture the invoked network effects, I first need to separate acquisitions that involve acquiring customer bases and those that do not (because many acquisitions are for technologies). I then need to learn about customer bases involved in each acquisition deal. Such information is difficult to obtain comprehensively. Therefore, missing such information in the current data and analysis becomes a limitation, but at the same time, provides a future research opportunity. A related concern is that other theoretical frameworks might also be able to explain the phenomena that I examine in this research, the industry evolution literature (e.g., Klepper, 1997) being a prominent example. Following the framework of industry evolution, one could argue that whether acquisitions positively or 186 negatively affect subsequent market entries depend on whether such events occur in the growth, mature, or decline stages of a market evolution. The theory that I develop in this research integrates insights from this literature while at the same adding novel mechanisms that are unique in platform-based markets (e.g., network effects). My theory also advances the industry evolution literature by addressing some interesting and unresolved issues in the literature, such as within-industry segment heterogeneities and ambiguities of acquisitions signals. Hence, by relying on network effects as the foundation of theoretical development, my research extends, rather than alienates, the industry evolution literature. Finally, although in this study I intend to unpack incentive versus capability aspects of prior experiences, due to the nature of archive-data research I was unable to specifically detail the mechanisms underlying the moderating effects of experiences. One argument could be that the more experienced a developer, the more options the developer has in addition to entering the category where acquisition signals abound. Building on this logic, one might argue that this is perhaps why experienced developers are more deterred from entering a category with acquisition signals. Empirically, this concern could be partially addressed by the fact that, in the conditional logit model specification, within a developer-month set, the experienced has fewer options, because, in general, the experienced will have more existing product categories. Furthermore, in pooled logit models estimating entry probability, I found that experience variables have positive main effects on the entry likelihood, supporting the capability argument regarding entry decisions. The fact that interaction effects with acquisition intensity are negative provides further evidence that what I am capturing empirically reflects incentives instead of capabilities. However, admittedly, I am not able to capture all unobservables, or mechanisms, that are associated with experience variables, and this is a limitation of the current study. 187 Conclusion To conclude, in this study I draw upon and integrate insights from industrial organization and the behavioral theory of the firm to test hypotheses that advance our understanding of complementors’ behaviors in a platform market where products are information-based. I hope the study demonstrates the value of simultaneously considering the role of both external market shifts and internal firm heterogeneities in explaining market entries. 188 TABLE 11: Descriptive Statistics of Variables Predicting Segment Entries Mean S.D. Min Max 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 Entry 0.07 0.26 0 1 1.00 2 Acquisition Intensity 5.00 5.06 0 30 0.12 1.00 3 Platform Incumbent 0.32 0.47 0 1 -0.01 -0.02 1.00 4 Platform Age.L3 0.35 0.83 0 6.92 -0.01 0.01 0.62 1.00 5 Product-development Experience.L3 1.36 8.91 0 2218 0.00 -0.01 0.22 0.28 1.00 6 Time-depreciated Experience.L3 0.29 2.43 0 666.68 0.00 -0.01 0.18 0.11 0.71 1.00 7 Category Relatedness.L3 0.27 1.96 0 1002.67 0.02 0.04 0.20 0.24 0.78 0.60 1.00 8 Number of Product Categories 1.88 0.73 1 20 0.07 -0.04 0.27 0.13 0.18 0.16 0.16 1.00 9 Diversification Index 0.22 0.13 0 0.69 0.08 -0.04 0.09 0.01 0.01 0.02 0.01 0.80 1.00 10 Category Size (Developers in 1000s) 28.31 32.74 0 177.94 0.14 0.50 -0.05 0.02 -0.01 -0.02 0.04 -0.08 -0.08 1.00 11 Category Size (Apps in 1000s) 41.28 51.09 0 317.97 0.14 0.44 -0.05 0.01 -0.01 -0.02 0.04 -0.07 -0.08 0.99 1.00 12 Concentration Ratio (Top 10 Developers) 0.06 0.07 0.01 1 -0.07 -0.31 0.01 -0.03 0.00 0.01 -0.03 0.05 0.05 -0.32 -0.29 1.00 13 Category-level Market Growth (yearly) 1.73 0.95 -2.79 4.70 -0.09 -0.12 0.02 0.08 0.02 0.00 -0.03 -0.05 -0.07 -0.04 -0.05 0.09 1.00 14 IPOs (3-month cumulated) 0.06 0.28 0 3 0.06 0.17 -0.01 -0.01 -0.01 0.00 0.02 0.00 0.01 0.16 0.14 -0.08 -0.13 1.00 15 Average Acquirer Size (# Employees in 10 thousands) 3.46 5.15 0 112.34 0.04 0.08 -0.01 0.00 0.00 0.00 0.01 -0.02 -0.01 0.14 0.14 -0.12 -0.10 0.05 1.00 16 Average Acquirer Size (Total Assets in 10 billions) 3.77 4.24 0 26.66 0.03 0.19 -0.01 0.03 0.00 -0.01 0.01 -0.04 -0.05 0.29 0.27 -0.21 -0.03 0.09 0.65 1.00 Note: L3 denotes that the variable was lagged by 3 months. 189 TABLE 12: Developer-Category-Month Analyses: Conditional Logit Models Predicting Category-Entry Likelihood Hypothesis DV: Entry = 1/0 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) H2.1a / Acquisition Intensity 0.002** 0.005*** 0.003*** 0.002*** 0.002*** 0.003*** 0.009*** -0.007*** 0.010*** -0.003*** H2.1b (0.000) (0.001) (0.001) (0.000) (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) H2.2 Platform Incumbent X Acquisitions -0.011*** -0.009*** -0.010*** General (0.001) (0.001) (0.001) Platform Platform Age X Acquisitions -0.004*** Experience (0.000) Product-development Experience X Acquisitions -0.001*** (0.000) H2.3 Time-depreciated Experience X Acquisitions -0.003*** Proximate (0.000) v.s. Category Relatedness 0.040*** 0.035*** 0.036*** Distant (0.002) (0.002) (0.002) Experience Category Relatedness X Acquisitions -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) H2.4 Number of Product Categories X Acquisitions -0.004*** -0.002*** Narrow v.s. (0.000) (0.000) Broad Diversification Index X Acquisitions 0.034*** 0.036*** Experience (0.002) (0.002) Category Size (Developers in 1000s) -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Mkt Concentration Ratio (Top 10 Developers) -0.455*** -0.450*** -0.448*** -0.451*** -0.454*** -0.456*** -0.457*** -0.449*** -0.454*** -0.446*** (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) Category-level Market Growth (yearly) 0.589*** 0.589*** 0.589*** 0.589*** 0.589*** 0.589*** 0.589*** 0.589*** 0.588*** 0.589*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) IPOs (3-month Cumulative) -0.041*** -0.040*** -0.040*** -0.040*** -0.041*** -0.040*** -0.041*** -0.041*** -0.040*** -0.040*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Average Acquirer Size (Total Assets in 10 billions) 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Category Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-month Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-category-months 7889259 7889259 7889259 7889259 7889259 7889259 7889259 7889259 7889259 7889259 Developer-months 415937 415937 415937 415937 415937 415937 415937 415937 415937 415937 Log Likelihood -1521124 -1520962 -1521045 -1520943 -1520933 -1520900 -1521060 -1521004 -1520758 -1520649 Notes: (1) Standard errors are in parentheses. (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (3) The group of the fixed-effect Logit models was set at developer-month level. (4) Effects of firm attributes (or main effects of developer-experience variables) were absorbed by developer-month fixed effects. (5) Time effects (i.e. month fixed effects) are also absorbed by developer-month fixed effects. (6) The following explanatory variables were lagged by 3 months (i.e., before the focal month) to mitigate concerns over reverse causality: Platform Age, Product-development Experience, Time-depreciated Experience, and Category Relatedness. 190 TABLE 12: Conditional Logit Models Predicting Likelihood of Entering a Category (Continued) (11) (12) (13) (14) (15) (16) Acquisition Intensity 0.001** 0.005*** 0.001** 0.004*** 0.001* 0.004*** (0.000) (0.001) (0.000) (0.001) (0.000) (0.001) Platform Incumbent X Acquisitions -0.012*** -0.010*** -0.011*** (0.001) (0.001) (0.001) Games(0/1) X Acquisitions 0.005* 0.018*** (0.002) (0.002) Games(0/1) X Platform Incumbent(0/1) X Acquisitions -0.054*** (0.002) Social Networking(0/1) X Acquisitions 0.011*** 0.016*** (0.002) (0.002) Social Networking(0/1) X Platform Incumbent(0/1) X Acquisitions -0.015*** (0.002) Game-and-Social-Network(0/1) X Acquisitions 0.009*** 0.018*** (0.001) (0.002) Game-and-Social-Network(0/1) X Platform Incumbent(0/1) X Acquisitions -0.034*** (0.001) Category Size (Developers in 1000s) -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** -0.003*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Mkt Concentration Ratio (Top 10 Developers) -0.444*** -0.426*** -0.455*** -0.447*** -0.436*** -0.416*** (0.056) (0.056) (0.056) (0.056) (0.056) (0.056) Category-level Market Growth (yearly) 0.586*** 0.587*** 0.585*** 0.583*** 0.581*** 0.579*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) IPOs (3-month Cumulative) -0.039*** -0.038*** -0.041*** -0.041*** -0.038*** -0.038*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Average Acquirer Size (# Employees in 10 thousands) 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** 0.004*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Category Fixed-effects Yes Yes Yes Yes Yes Yes Developer-month Fixed-effects Yes Yes Yes Yes Yes Yes Developer-category-months 7889259 7889259 7889259 7889259 7889259 7889259 Developer-months 415937 415937 415937 415937 415937 415937 Log Likelihood -1521121 -1520523 -1521109 -1520901 -1521107 -1520571 191 TABLE 13: Pooled Logit Models (to Generate Interaction Graphs via Simulation) DV: Category Entry (1/0) (1) (2) (3) (4) (5) (6) (7) (8) H2.1a / H2.1b Acquisition Intensity 0.00816*** 0.00694*** 0.00589*** 0.00601*** 0.00573*** 0.0115*** 0.00678*** 0.00553*** (15.05) (13.02) (11.17) (11.31) (10.84) (18.72) (10.02) (7.48) H2.2 Platform Incumbent 0.0645*** -0.555*** (15.40) (-94.27) Platform Incumbent X Acquisitions -0.00891*** -0.0118*** (-18.46) (-18.26) Platform Age 0.0493*** (20.02) Platform Age X Acquisitions -0.00384*** (-13.93) Product-development Experience 0.00371*** (4.82) Product-development Experience X Acquisitions -0.000300*** (-4.84) H2.3 Category Relatedness 0.0183*** 0.0167*** (5.41) (5.82) Category Relatedness X Acquisitions -0.00110*** -0.000860*** (-5.03) (-4.62) Time-depreciated Experience 0.00786* (2.32) Time-depreciated Experience X Acquisitions -0.000912** (-2.72) H2.4 # of Product Categories 0.115*** (117.97) # of Product Categories X Acquisitions -0.00211*** (-18.09) Diversification Index 1.653*** 2.801*** (136.87) (160.95) Diversification Index X Acquisitions -0.00323* 0.0136*** (-2.36) (6.84) Category Size (Developers in 1000s) -0.780*** -0.749*** -0.761*** -0.757*** -0.766*** -0.807*** -0.765*** -0.716*** (-13.20) (-12.70) (-12.90) (-12.83) (-12.98) (-13.51) (-12.74) (-11.97) Concentration Ratio (Top 10 Developers) 0.677*** 0.689*** 0.682*** 0.680*** 0.679*** 0.689*** 0.690*** 0.704*** (12.92) (13.17) (13.02) (12.98) (12.95) (13.02) (13.04) (13.36) IPOs (3-month cumulated) -0.0769*** -0.0766*** -0.0769*** -0.0768*** -0.0769*** -0.0776*** -0.0778*** -0.0773*** (-16.92) (-16.85) (-16.91) (-16.88) (-16.91) (-17.00) (-16.96) (-16.79) Average Acquirer Size (Total Assets, in billions) 0.000987*** 0.000981*** 0.000985*** 0.000984*** 0.000984*** 0.000996*** 0.000991*** 0.000982*** (21.19) (21.07) (21.14) (21.12) (21.13) (21.29) (21.15) (20.89) Intercept -3.911*** -3.901*** -3.897*** -3.894*** -3.896*** -4.189*** -4.402*** -4.632*** (-62.90) (-62.63) (-62.57) (-62.54) (-62.55) (-69.15) (-74.33) (-78.97) Developer-month-categories 8216469 8216469 8216469 8216469 8216469 8216469 8216469 8216469 Month Fixed Effects Y Y Y Y Y Y Y Y Category Fixed Effects Y Y Y Y Y Y Y Y * p<0.05, ** p<0.01, *** p<0.001 192 TABLE 14: Descriptive Statistics of Variables Predicting Post-Entry App Failures Mean S.D. Min Max 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 1. App Failure (1/0) 0.46 0.5 0 1 1.00 2. Acquisition Intensity 7.29 6 0 30 -0.32 1.00 3. Platform Incumbent 0.31 0.46 0 1 0.07 -0.03 1.00 4. Platform Age.L3 0.3 0.74 0 6.33 -0.04 0.03 0.61 1.00 5. Product-development Experience.L3 1.46 8.78 0 1707 0.01 -0.02 0.25 0.31 1.00 6. Time-depreciated Experience.L3 0.34 2.9 0 666.68 0.02 -0.03 0.18 0.12 0.69 1.00 7. Category Relatedness.L3 0.46 2.71 0 347.1 0.02 0.01 0.26 0.29 0.83 0.56 1.00 8. Diversification Index 0.26 0.12 0 0.69 0.08 -0.05 0.06 -0.04 0.00 0.01 0.00 1.00 9. # of Product Categories 2.17 1.04 1 17 0.09 -0.07 0.23 0.07 0.17 0.13 0.15 0.72 1.00 10. Average Acquirer Size (# Employees in 10 thousands) 4.08 7.48 0 112.34 0.02 -0.04 -0.01 -0.02 -0.01 0.00 0.00 0.00 0.00 1.00 11. Average Acquirer Size (Total Assets in 10 billions) 3.93 3.73 0 26.66 -0.24 0.11 0.00 0.04 0.01 0.00 0.01 -0.05 -0.03 0.50 1.00 12. Category Size (Apps in 1000s) 53.69 53.19 0 244.43 -0.33 0.45 -0.03 0.03 -0.02 -0.03 0.03 -0.12 -0.09 0.02 0.26 1.00 13. Category Size (Developers in 1000s) 36.8 34.22 0 144.6 -0.35 0.53 -0.04 0.03 -0.03 -0.03 0.03 -0.12 -0.09 0.00 0.26 0.99 1.00 14. Concentration Ratio (Top 10 Developers) 0.05 0.05 0.01 1 0.14 -0.30 -0.01 -0.06 0.00 0.02 -0.02 0.09 0.09 0.02 -0.09 -0.34 -0.37 1.00 15. Category-level Market Growth (yearly) 1.27 0.68 -2.79 4.7 -0.48 0.24 0.04 0.13 0.04 0.01 -0.01 -0.10 -0.06 -0.08 0.22 0.26 0.28 -0.17 1.00 16. Inverse Mill's Ratio (lambda) 0.14 0.2 0 5.36 0.08 -0.46 0.06 0.03 0.01 0.01 -0.04 0.05 0.07 -0.10 -0.19 -0.47 -0.51 0.56 0.23 1.00 17. App Overall Rating 3.77 1.02 0 5 -0.14 0.15 -0.08 0.00 -0.04 -0.03 -0.04 -0.08 -0.13 -0.07 0.06 0.25 0.26 -0.22 0.26 -0.11 1.00 Notes: (1) The data structure is cross-sectional category-apps. (2) All variables take the same values as that were used to estimate entry decisions (i.e., at the time of entry), except for the following ones: App Failure, Inverse Mill’s Ratio, and App Overall Rating. 193 TABLE 15: Tabulate App Failure across Acquisition Intensity and Experience Variables Acquisition Intensity Low High Total Platform Incumbent (0/1) Platform Entrant 0.72 0.74 0.73 Platform Incumbent 0.55 0.35 0.44 Platform Age Low 0.68 0.61 0.65 High 0.42 0.22 0.30 Product-development Exp. Low 0.68 0.61 0.65 High 0.50 0.29 0.39 Time-depreciated Exp. Low 0.65 0.53 0.59 High 0.55 0.36 0.45 Category Relatedness Low 0.66 0.62 0.64 High 0.55 0.33 0.42 Number of Product Categories Low 0.62 0.47 0.55 High 0.64 0.50 0.58 Diversification Index Low 0.57 0.38 0.47 High 0.64 0.53 0.59 Total 0.62 0.48 0.55 Sample size 77,922 76,178 154,100 Notes: (1) Sample includes category-apps from developer fixed-effects models. (2) The following are the cut-off points to differentiate low versus high values of a variable: Acquisition Intensity (6, or 50 percentile), Platform Age (1, between 75 and 90 percentiles), Product-development Exp. (1, or 75 percentile), Time-depreciated Exp. (0.34, between 75 and 90 percentiles), Category Relatedness (0.46, between 75 and 90 percentiles), Number of Product Categories (2, between 50 and 75 percentile), and Diversification Index (0.26, between 50 and 75 percentiles). The cut-off values were chosen based on the criteria of ease for interpretation, as well as sufficient non-zero observations on each side of the cut-offs. 194 TABLE 16: App-level Developer-Fixed-Effects Logit Models Estimating Likelihood of App Failure (1) (2) (3) (4) (5) (6) (7) (8) (9) Acquisition Intensity (value at time of entry) -0.038*** 0.029*** -0.001 -0.030*** -0.033*** -0.034*** -0.034*** -0.048*** 0.024*** (0.002) (0.003) (0.003) (0.002) (0.002) (0.002) (0.004) (0.004) (0.005) Platform Incumbent (at time of entry) 0.394*** 0.383*** (0.026) (0.026) Platform Incumbent X Acquisitions -0.093*** -0.091*** (0.003) (0.003) Platform Age.L3 (at time of entry) -0.419*** (0.022) Platform Age X Acquisitions -0.030*** (0.002) Product-development Experience.L3 (at time of entry) -0.005*** (0.001) Product-development Experience X Acquisitions -0.002*** (0.000) Time-depreciated Experience.L3 (at time of entry) 0.000 (0.004) Time-depreciated Experience X Acquisitions -0.006*** (0.001) Category Relatedness.L3 (at time of entry) 0.007 -0.007+ (0.005) (0.004) Category Relatedness X Acquisitions -0.003*** -0.001 (0.000) (0.000) Number of Product Categories (at time of entry) 0.017 (0.011) Number of Product Categories X Acquisitions -0.002 (0.001) Diversification Index (at time of entry) -0.170 -0.017 (0.105) (0.106) Diversification Index X Acquisitions 0.037** 0.021+ (0.011) (0.012) Category Size (Apps in 1000s) (at time of entry) -0.019*** -0.019*** -0.010*** -0.017*** -0.019*** -0.019*** -0.019*** -0.019*** -0.019*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Mkt Concentration Ratio (Top 10) (at time of entry) -3.335*** -3.217*** -6.128*** -3.465*** -3.336*** -3.433*** -3.338*** -3.354*** -3.313*** (0.231) (0.233) (0.255) (0.232) (0.231) (0.232) (0.232) (0.232) (0.233) Category-level Market Growth (at time of entry) -1.477*** -1.182*** -0.880*** -1.489*** -1.472*** -1.440*** -1.475*** -1.460*** -1.171*** (0.051) (0.053) (0.053) (0.052) (0.052) (0.052) (0.051) (0.052) (0.054) Average Acquirer Size (Assets in 10 billions) (at time of entry) -0.057*** -0.054*** -0.052*** -0.056*** -0.057*** -0.057*** -0.057*** -0.057*** -0.054*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Inverse Mill's Ratio (lambda) 1.474*** 2.390*** 0.435* 0.952*** 1.374*** 1.555*** 1.490*** 1.536*** 2.350*** (0.202) (0.206) (0.202) (0.210) (0.206) (0.208) (0.203) (0.203) (0.212) Developer Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Category Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Category-apps 154100 154100 154100 154100 154100 154100 154100 154100 154100 Developers 28386 28386 28386 28386 28386 28386 28386 28386 28386 Log Likelihood -46078 -45553 -45355 -45902 -46000 -46027 -46077 -46072 -45535 Pseudo R-squared .26 .26 .27 .26 .26 .26 .26 .26 .26 Notes: (1) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (2) Standard errors are in parentheses. (3) I define an app failure as when the number of months of an app not being updated, since its most recent version update until the right censoring month (January 2015), is equal to or greater than 12. 195 FIGURE 9: Number of Acquisitions in the iOS Mobile App Platform Over Time FIGURE 10: Developers and Acquisitions across iOS Categories 0 10 20 30 2008m1 2009m1 2010m1 2011m1 2012m1 2013m1 2014m1 2015m1 2016m1 0 50 100 150 200 250 300 350 0 10000 20000 30000 40000 50000 60000 70000 80000 iOS Developers Acquisitions 196 FIGURE 11: Interaction Effect between Acquisition Intensity and Platform Incumbent on the Likelihood of Segment Entry FIGURE 12: Interaction Effect between Acquisition Intensity and Platform Age on the Likelihood of Segment Entry FIGURE 13: Interaction Effect between Acquisition Intensity and Product- development Experience on the Likelihood of Segment Entry .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Platform_Incumbent takes value of 0 Predicted probability when Platform_Incumbent takes value of 1 .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Platform_Age takes value of 0 Predicted probability when Platform_Age takes value of 2 .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Product_Develop_Exp takes value of 0 Predicted probability when Product_Develop_Exp takes value of 10 197 FIGURE 14: Interaction Effect between Acquisition Intensity and Time- depreciated Experiences on the Likelihood of Segment Entry FIGURE 15: Interaction Effect between Acquisition Intensity and Category Relatedness on the Likelihood of Segment Entry FIGURE 16: Interaction Effect between Acquisition Intensity and Number of Categories on the Likelihood of Segment Entry FIGURE 17: Interaction Effect between Acquisition Intensity and Diversification Index on the Likelihood of Segment Entry .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Time_Depreciated_Exp takes value of 0 Predicted probability when Time_Depreciated_Exp takes value of 2.1 .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Category_Relatedness takes value of .2822366356935377 Predicted probability when Category_Relatedness takes value of 2.327018536538529 .045 .05 .055 .06 .065 .07 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Number_Categories takes value of 1 Predicted probability when Number_Categories takes value of 4 0 .05 .1 .15 .2 Pr(Entry=1) 0 10 20 30 Acquisition Itensity Predicted probability when Diversification_Index takes value of 0 Predicted probability when Diversification_Index takes value of .6 198 APPENDIX 5: Developer-Category-Month Conditional Logit Models Predicting Category-Entry Likelihood (with alternative controls) Hypothesis DV: Entry = 1/0 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) H2.1a / Acquisition Intensity 0.001* 0.004*** 0.003*** 0.002*** 0.002*** 0.002*** 0.008*** -0.008*** 0.009*** -0.004*** H2.1b (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) H2.2 Platform Incumbent X Acquisitions -0.011*** -0.009*** -0.010*** General (0.001) (0.001) (0.001) Platform Platform Age X Acquisitions -0.004*** Experience (0.000) Product-development Experience X Acquisitions -0.001*** (0.000) H2.3 Time-depreciated Experience X Acquisitions -0.003*** Proximate (0.000) v.s. Category Relatedness 0.044*** 0.040*** 0.040*** Distant (0.002) (0.002) (0.002) Experience Category Relatedness X Acquisitions -0.003*** -0.002*** -0.002*** (0.000) (0.000) (0.000) H2.4 Number of Product Categories X Acquisitions -0.004*** -0.002*** Narrow v.s. (0.000) (0.000) Broad Diversification Index X Acquisitions 0.036*** 0.038*** Experience (0.002) (0.002) Category Size (Apps in 1000s) -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Mkt Concentration Ratio (Top 10 Developers) -0.550*** -0.545*** -0.541*** -0.545*** -0.549*** -0.550*** -0.553*** -0.542*** -0.550*** -0.538*** (0.057) (0.057) (0.057) (0.057) (0.057) (0.056) (0.057) (0.056) (0.057) (0.056) Category-level Market Growth (yearly) 0.588*** 0.588*** 0.589*** 0.589*** 0.588*** 0.588*** 0.588*** 0.589*** 0.588*** 0.589*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) IPOs (3-month Cumulative) -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** -0.041*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Average Acquirer Size (Total Assets in 10 billions) 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Category Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-month Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-category-months 7829570 7829570 7829570 7829570 7829570 7829570 7829570 7829570 7829570 7829570 Developer-months 415366 415366 415366 415366 415366 415366 415366 415366 415366 415366 Log Likelihood -1512504 -1512340 -1512420 -1512326 -1512318 -1512246 -1512446 -1512373 -1512106 -1511982 Notes: (1) Standard errors are in parentheses. (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (3) The group of the fixed-effect Logit models was set at developer-month level. (4) Effects of firm attributes (or main effects of developer-experience variables) were absorbed by developer-month fixed effects. (5) Time effects (i.e. month fixed effects) are also absorbed by developer-month fixed effects. (6) The following explanatory variables were lagged by 3 months (i.e., before the focal month): Platform Age, Product-development Experience, Time-depreciated Experience, and Category Relatedness. 199 APPENDIX 6: Conditional Logit Models Estimating Category-Entry Likelihood (with Coarsened-matched Sample) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) H2.1a / Acquisition Intensity 0.003* 0.011*** 0.006*** 0.005*** 0.005*** 0.005*** 0.034*** 0.009*** 0.031*** 0.015*** H2.1b (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.002) (0.002) H2.2 Platform Incumbent X Acquisitions -0.026*** -0.017*** -0.020*** (0.001) (0.002) (0.002) Platform Age X Acquisitions -0.008*** (0.001) Product-development Experience X Acquisitions -0.002*** (0.000) H2.3 Time-depreciated Experience X Acquisitions -0.007*** (0.001) Category Relatedness 0.269*** 0.235*** 0.242*** (0.010) (0.010) (0.010) Category Relatedness X Acquisitions -0.007*** -0.003*** -0.004*** (0.000) (0.001) (0.001) H2.4 Number of Product Categories X Acquisitions -0.017*** -0.012*** (0.001) (0.001) Diversification Index X Acquisitions -0.031*** -0.022*** (0.005) (0.005) Category Size (Developers in 1000s) -0.030*** -0.029*** -0.030*** -0.029*** -0.029*** -0.030*** -0.029*** -0.029*** -0.029*** -0.029*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Mkt Concentration Ratio (Top 10 Developers) -1.795*** -1.779*** -1.793*** -1.789*** -1.789*** -1.805*** -1.775*** -1.786*** -1.787*** -1.791*** (0.112) (0.112) (0.112) (0.112) (0.112) (0.112) (0.112) (0.112) (0.112) (0.112) Category-level Market Growth (yearly) 0.609*** 0.608*** 0.610*** 0.610*** 0.609*** 0.612*** 0.605*** 0.607*** 0.609*** 0.610*** (0.020) (0.020) (0.020) (0.020) (0.020) (0.021) (0.020) (0.020) (0.021) (0.021) IPOs (3-month Cumulative) -0.053*** -0.054*** -0.053*** -0.054*** -0.053*** -0.054*** -0.053*** -0.053*** -0.054*** -0.054*** (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) Average Acquirer Size (# Employees in 10 thousands) -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** -0.005*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Category Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-month Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Developer-category-months 636066 636066 636066 636066 636066 636066 636066 636066 636066 636066 Developer-months 222193 222193 222193 222193 222193 222193 222193 222193 222193 222193 Log Likelihood -202934 -202761 -202880 -202834 -202830 -202464 -202782 -202914 -202303 -202360 Notes: (1) Standard errors are in parentheses. (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (3) CEM matches a realized-case (entered) category to similar control-case categories. (4) The group of the fixed-effect Logit models was set at developer-month level. (5) Effects of firm attributes (or main effects of developer-experience variables) were absorbed by developer-month fixed effects. (6) Time effects (i.e. month fixed effects) are also absorbed by developer-month fixed effects. 200 APPENDIX 7: Conditional Logit Models Estimating Category-Entry Likelihood (with Alternative Ways of Coding Explanatory Variables) (1) (2) (3) (4) (5) (6) (7) (8) Acquisition Intensity (1-month cumulative) 0.004*** (0.001) Acquisition Intensity (2-month cumulative) 0.003*** (0.001) Acquisition Intensity (3-month cumulative) 0.001* 0.002*** 0.002*** (0.000) (0.000) (0.000) Acquisition Intensity (4-month cumulative) 0.001* (0.000) Acquisition Intensity (5-month cumulative) 0.000 (0.000) Acquisition Intensity (6-month cumulative) -0.000 (0.000) Acquisitions X Time-depreciated Experience (depreciation factor = 60%) -0.005*** (0.000) Acquisitions X Time-depreciated Experience (depreciation factor = 80%) -0.002*** (0.000) Category Size (Apps in 1000s) -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Mkt Concentration Ratio (Top 10 Developers) -0.554*** -0.560*** -0.550*** -0.545*** -0.538*** -0.546*** -0.550*** -0.549*** (0.056) (0.056) (0.057) (0.057) (0.057) (0.058) (0.057) (0.057) Category-level Market Growth (yearly) 0.588*** 0.587*** 0.588*** 0.589*** 0.590*** 0.594*** 0.588*** 0.588*** (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) IPOs (3-month Cumulative) -0.041*** -0.041*** -0.041*** -0.041*** -0.040*** -0.040*** -0.041*** -0.041*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Average Acquirer Size (Total Assets in 10 billions) 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** 0.003*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Category Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Developer-month Fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes Developer-category-months 7829794 7829794 7829570 7827086 7823744 7821009 7829570 7829570 Developer-months 415441 415441 415366 415171 414928 414713 415366 415366 Log Likelihood -1512570 -1512569 -1512504 -1511973 -1511265 -1510653 -1512330 -1512307 APPENDIX 8: App-Level Developer Random-Effects Logit Models Predicting Likelihood of App Failure (1) (2) (3) (4) (5) (6) (7) (8) (9) Acquisition Intensity (value at time of entry) -0.082*** -0.076*** -0.080*** -0.082*** -0.082*** -0.082*** -0.094*** -0.095*** -0.089*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003) (0.004) Platform Incumbent (at time of entry) 0.772*** 0.793*** (0.022) (0.023) Platform Incumbent X Acquisitions -0.020*** -0.023*** (0.002) (0.002) Platform Age.L3 (at time of entry) 0.420*** (0.015) Platform Age X Acquisitions -0.013*** (0.001) Product-development Experience.L3 (at time of entry) -0.002* (0.001) Product-development Experience X Acquisitions 0.000*** (0.000) Time-depreciated Experience.L3 (at time of entry) -0.010*** (0.003) Time-depreciated Experience X Acquisitions 0.002*** (0.000) Category Relatedness.L3 (at time of entry) 0.017*** -0.014*** (0.004) (0.004) Category Relatedness X Acquisitions 0.000 0.002*** (0.000) (0.000) Number of Product Categories (at time of entry) 0.071*** (0.011) Number of Product Categories X Acquisitions 0.006*** (0.001) Diversification Index (at time of entry) -0.525*** -0.559*** (0.096) (0.096) Diversification Index X Acquisitions 0.051*** 0.046*** (0.010) (0.010) Category-level Market Growth (at time of entry) -5.536*** -5.744*** -5.782*** -5.543*** -5.541*** -5.551*** -5.515*** -5.538*** -5.756*** (0.029) (0.030) (0.031) (0.029) (0.029) (0.029) (0.029) (0.029) (0.030) Inverse Mill's Ratio (lambda) -5.875*** -5.909*** -5.525*** -5.903*** -5.910*** -5.815*** -5.764*** -5.854*** -5.950*** (0.123) (0.124) (0.125) (0.125) (0.124) (0.124) (0.123) (0.123) (0.125) Category Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Category-apps 593228 593228 593228 593228 593228 593228 593228 593228 593228 Developers 218875 218875 218875 218875 218875 218875 218875 218875 218875 Log Likelihood -229495 -228559 -228984 -229486 -229486 -229457 -229409 -229478 -228529 Notes: (1) Standard errors are in parentheses. (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (3) Sampled includes apps released on the first month a developer enters a category. (4) All models include developer random-effects. (5) Several category-level control variables (Category Size, Market Concentration, and Average Acquirer Size) were dropped, because including the full set of controls leads to the maximum likelihood estimation not converge. 202 APPENDIX 9: App-level OLS Models Estimating App Rating (1) (2) (3) (4) (5) (6) (7) (8) (9) Acquisition Intensity (value at time of entry) 0.015*** 0.013*** 0.001+ 0.013*** 0.014*** 0.013*** 0.017*** 0.018*** 0.015*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Platform Incumbent (at time of entry) 0.117*** 0.109*** (0.006) (0.006) Platform Incumbent X Acquisitions -0.003*** -0.003*** (0.001) (0.001) Platform Age.L3 (at time of entry) 0.235*** (0.004) Platform Age X Acquisitions -0.004*** (0.000) Product-development Experience.L3 (at time of entry) 0.003*** (0.000) Product-development Experience X Acquisitions 0.000*** (0.000) Time-depreciated Experience.L3 (at time of entry) 0.000 (0.000) Time-depreciated Experience X Acquisitions 0.000** (0.000) Category Relatedness.L3 (at time of entry) 0.011*** 0.008*** (0.001) (0.001) Category Relatedness X Acquisitions -0.000 0.000 (0.000) (0.000) Number of Product Categories (at time of entry) -0.006* (0.003) Number of Product Categories X Acquisitions -0.001** (0.000) Diversification Index (at time of entry) -0.249*** -0.215*** (0.033) (0.033) Diversification Index X Acquisitions -0.012** -0.010* (0.004) (0.004) Inverse Mill's Ratio (lambda) -0.614*** -0.501*** 0.031 -0.529*** -0.607*** -0.548*** -0.622*** -0.603*** -0.457*** (0.021) (0.022) (0.023) (0.022) (0.022) (0.022) (0.021) (0.021) (0.022) Developer Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Category Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Category-apps 285145 285145 285145 285145 285145 285145 285145 285145 285145 Developers 119618 119618 119618 119618 119618 119618 119618 119618 119618 R-squared .012 .016 .039 .014 .012 .014 .012 .013 .018 R-squared within .012 .016 .039 .014 .012 .014 .012 .013 .018 Notes: (1) Standard errors are in parentheses. (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. (3) Sample includes apps that are released in the first month of a developer entering a category. 203 CHAPTER 4 DISCUSSION AND CONCLUSION Summary of Main Findings The main findings of my dissertation essays can be summarized in terms of the key concepts underpinning my conceptual framework, namely market structures from the industrial organization literature and firm heterogeneities from the behavioral theory of the firm and capability-based view. I argue that characteristics of the market structure of the platform-based market—the marketplaces of iOS and Android platforms—shape app developers’ market entry decisions, either across platforms or within a single platform market. In the first essay, the market structure is determined by two competing platforms with different strategic emphases— one quality-driven (iOS), the other quantity-driven (Android). I used installed base differences of software and hardware complementary products between the quantity- and quality-driven platforms to reflect an important aspect of the market. I find that installed base differences have an inverted-U-shaped effect on a complementor’s likelihood of moving to the target platform for both directions of cross-platform mobility. This finding supports my theoretical prediction for the direction of mobility from iOS to Android, but not for the other direction (for which I predicted a U-shaped relation). Consequently, it seems that the same mechanisms apply for both directions of mobility—that is, as installed base differences increase initially, market growth and legitimation of the target platform encourages developers adopting the platform, and as installed base differences further increase, market saturation and de-legitimization reduce app developers’ incentives to move. The market structure aspect in the second essay, in which I study within-platform category entries, is captured by the frequency of acquisition events across product categories and 204 changes in such frequency over time. The finding is that, on average, acquisition intensity in a product category is positively associated with the probability that an app developer enters the category. This finding suggests that, when evaluating potential benefits versus costs of entering a category that is frequented by acquisitions, developers generally weigh benefits more. In the post-entry performance analysis, the results indeed support the rationale that developers benefit from entering categories during acquisition heydays. To examine the post-entry performance effects, I adopted a research design that enabled me to examine imprinting effects and found that acquisition intensity of the category at the time of a developer launching its first products tends to be negatively associated with the first products’ failure probability. Thus, the empirical findings of both essays align well with the industrial organization literature framework that states market structures shape firm conduct. To further understand heterogeneous market-entry behaviors by app developers, I integrated industrial organization arguments with more internally oriented theoretical perspectives, namely the behavioral theory of the firm and capability-based view. In the first essay on cross-platform mobility, app developer heterogeneity was captured by aspiration- performance differences, which enabled me to identify high versus low performers after taking into account both social comparison with similar others and historical self-comparisons. The findings of proportional hazard models suggest that both above and below aspirations increase the hazards of a developer moving to the competing platform. Below-aspiration developers are more likely to abandon the home platform after mobility and consist of the group of movers that I refer to as platform switchers. Above-aspiration developers, in contrast, are less likely to abandon the home platform, forming the group of movers that I refer to as multihomers. These findings, taken together, indicate that cross-platform mobility decisions and platform- 205 abandonment decisions by different groups of app developers are motivated by different incentives, and support my theoretical arguments. That is, above-aspiration developers move and then multihome to pursue greater positive network effects and the possibilities of winner-take-all in their own niche markets, while below-aspiration developers move and then abandon the home platform because they are competitively crowded out and thus engage in a problemistic search of alternatives. This rationale is further supported by my findings of post-mobility performance analyses, in which I found that, compared with multihomers, platform switchers fare worse both in terms of performance on the home platform and on the target platform. I captured firm heterogeneities in the second essay by building on the organizational experiences literature. I systematically categorized developers along three dimensions of prior experiences—general platform experience, experience proximity, and experience scope. My findings suggest that developers with heterogeneous prior experience react differently toward acquisition signals in a product category, such that acquisition intensity of a category is positively associated with entry probability by developers who have fewer general platform experiences and whose experiences are more distant and broader in scope. In contrast, acquisition intensity tends to be negatively related to the likelihood of entry by developers who are more experienced and whose experiences are more proximate and narrower in scope. Heterogeneities across developers with different prior experiences are also reflected in the performance of the first products they introduce when they enter a category. In the post-entry performance analyses, my findings support the argument that experiences bestow upon developers the capabilities to capture opportunities and address market competition. First products introduced in a category during its acquisition heydays are less likely to fail if released by developers with more platform experience and whose experiences are more proximate and 206 narrower in scope. These findings, along with those of the first essay, echo a fundamental assertion of the strategic management literature—strategic decisions (e.g., market entries) and performance are heterogeneous across firms. Perhaps the most intriguing and meaningful set of findings of this dissertation is the interactions between environmental and developer-specific factors, which are based on my theoretical integration of the external view (e.g., industrial organization) and more internal perspectives (e.g., behavioral theory of the firm). In the first essay on cross-platform mobility, I interacted aspiration variables with installed base differences (between quantity- and quality- driven platforms). The findings suggest that, in addition to a statistically significant inverted-U relation between installed base differences and the likelihood of moving to the competing platform, the effect is moderated by a performance-aspiration difference. The moderating effects can be summarized as follows: high performers tend to react earlier, compared with low performers, during market growth when making mobility decisions and tend to desist from making such decisions later during market maturation. In the second essay on category entries following acquisitions, such interactions between environmental and developer-specific attributes are reflected, as discussed earlier, in the heterogeneous responses of developers toward acquisition signals as well as post-entry performance heterogeneities. In sum, the findings of the two essays reinforce a fundamental belief in the strategy literature, namely, that external environments and internal firm heterogeneities jointly shape firm behaviors and performance. While most of the empirical evidence used to validate this belief has focused on well-established (and mainly bricks and mortar) industry environments, this dissertation demonstrates the validity of this assertion in an emerging, fast-growing market— platforms—characterized by network effects that are unique to information-based products. 207 Contributions The essays in this dissertation are among the first empirical examinations to exclusively focus on the active roles of platform complementors in shaping one of the largest and fast-growing platform markets—the iOS and Android mobile operating systems, which account for the majority of the global mobile device market share. From a theoretical perspective, the dissertation contributes to the following literatures: platform-based competition, industrial organization, organizational capabilities, and behavioral theory of the firm. Platform-based competition. This work builds on and extends the rising literature of platforms in the field of strategic management. It contributes to the platform literature in several ways. First, this dissertation shifts prior literature’s almost exclusive focus on platform-level strategies to the fundamental source of competitive advantages of a platform—complementors— and investigates ways that complementors’ market-entry behaviors shape platforms’ competitive advantages and long-term evolution. Positioned in this way, this dissertation joins the most recent stream of research in the platform literature that studies the relationship between platform owners and complementors (e.g., Edelman, 2014; Eisenmann et al., 2011; Zhu & Liu, 2017). A related second contribution examines the issue of competition between two dominant technological platforms, herein iOS and Android. My findings suggest that, during the battle for dominance between the two mobile platforms, the Android platform has been accruing competitive advantages by benefiting from both inflows and outflows of platform migrators between the two platforms. Because cross-platform mobility is a dynamic process, studying this phenomenon sheds light on the topic of long-term evolution of the two competing platforms, leading to the dissertation’s third contribution to the platform literature—whether competing platforms will converge or diverge. 208 Third, I conceptualized the market structure of the two competing platforms in terms of underlying differences in their strategic emphases, one quality-driven (iOS) and the other quantity-driven (Android). These different strategic orientations are, however, not necessarily stable attributes over time. Convergence would be the situation under which the two platforms adjust their strategic orientation and become increasingly similar to each other: that is, the quality-driven platform starts to stress quantity (of complementors and complementary products), while the quantity-driven platform starts to improve quality. 76 Although in reality the two platforms would not be exactly the same (at least in the short term, because Android is an open system while iOS is closed), theoretically if the two platforms converged in strategic orientations—mixing quality and quantity-driven orientations, for instance—one would eventually become the winner in the standards war. Divergence in strategic orientation, in contrast, might ensure the coexistence of both platforms, perhaps splitting the market share relatively equally (Hossain, Minor, & Morgan, 2011). However, findings of the dissertation indicate that perhaps platform convergence is the trend and that Android is gaining more competitive advantages, at least in terms of volume if not profitability. Finally, my research suggests that, within a single platform market, industry mega-events such as acquisitions shape the evolution of its market segments. The second essay on segment entries has implications for the ways in which acquisition events shape segment evolution. Market segments that are frequented by acquisitions tend to attract more products from inexperienced (and thus entrepreneurial) developers, thus stimulating entrepreneurship. But 76 The most recent industry development, in my view, has shown evidence of convergence in strategic orientations of the two platforms. For instance, at the 2017 WWDC, Apple announced that it is doing its best to reduce hurdles that developers must overcome in releasing products for the App Store by specifically reducing the timelines of reviewing and accepting app submissions. This reflects the company’s changes in strategic orientation toward stressing quantity (of complementary products). Conversely, Google launched its own mobile phone on October 20, 2016. This gesture was interpreted by industry experts and platform scholars as an evidence of Google’s attempt to tighten control over Android. The goal is to improve user experiences and quality of the platform. 209 experienced incumbents tend to avoid such segments, thereby reducing innovation and investments by incumbents. In this way, within a platform market (e.g., the App Store), we expect to observe that segment heterogeneities will be amplified as a result of industry mega- events. Industrial organization. My research contributes to the industrial organization literature in the field of strategic management in the following ways. First, a critical theoretical link suggested by prior research is between acquisition events and subsequent market entries (e.g., Caves and Porter, 1977; Stigler, 1950; Klepper, 1997). This relationship, however, has only been argued theoretically and has not been sufficiently tested empirically. Furthermore, prior empirical work has not provided a clear answer to the question of whether acquisitions encourage or deter entries. My essay on within-platform segment entries contributes to the literature by clarifying this linkage. Meanwhile, my research identifies conditions under which acquisitions attract versus deter entries, thus providing a boundary condition for this theoretical relationship—prior experiences. More broadly, such a research design and finding echo the view that entry and mobility barriers have heterogeneous implications for different potential entrants (Caves and Porter, 1977). Second, a critique of the industrial organization arguments is that the perspective only focuses on inter-industry differences and is unable to explain intra-industry variances. Recent developments in the literature—in particular, recent advancements in the industry lifecycle literature—start to unpack segment heterogeneities within a single industry (e.g., Jacobides and Tae, 2015). My empirical essay on app developer’s category-entry decisions following acquisitions joins this line of inquiry and thereby contributes to the development of this literature. Key insights generated from the second essay regarding this issue include: product categories 210 differences in terms of strengths of network effects; categories’ variance in terms of lifecycle growth stages as signaled by mega-events such as acquisitions; and, the way entry behaviors by platform participants shape evolution of product categories, thus further enhancing category heterogeneities. Firm capabilities and behavioral theory of the firm. Finally, the dissertation contributes to the internal views of firm capabilities and the behavioral theory of the firm. In the vast literature of organizational experiences, an unresolved issue hinges on whether prior experiences have enabling or constraining effects. Enabling effects are mainly invoked by scholars who focus on how experiences provide companies with the capability to engage in strategic endeavors, such as manufacturing, contracting, strategic alliances, and market entries. Constraining effects, in contrast, are often conceptualized in terms of routinization and inertia. Empirically, these two aspects are often convoluted and scholars have struggled to tease them apart. The second essay of this dissertation helps to at least partially reconcile these contradictory views by separately examining the roles of prior experiences in affecting market entries and post-entry performance. The key insight is that the constraining aspect of prior experiences comes into play when experiences inhibit developers from entering a category following acquisition signals; the enabling aspect of experiences is reflected by the fact that prior experiences grant developers the capability to deal with post-entry competition. The second contribution is to the behavioral theory of the firm, in which the mechanisms are unclear regarding ways in which above aspiration also drives organizations to seek changes. Integrating the theory with insights from the platform literature, I proposed that the pursuit of greater network effects and the possibilities of a winner-take-all objective is a strong motivation behind above-aspiration developers. In this sense, my work contributes to the performance feedback component of the behavioral theory of the firm 211 by offering a theoretical element that has been missing in the literature. Practical Implications This dissertation has significant practical implications, because it not only provides descriptive evidence, but also offers normative lessons. The descriptive aspect is reflected by the first-stage analyses in both essays—that is, how app developers make market entry decisions either across platforms or within a single platform market. From the standpoint of platform owners, the pattern of complementors’ behaviors is that, although both high and low performers tend to move to the competing platform, platforms mainly lose developers who were initially lower performers. As a result, it appears that platform owners do not need to worry much about cross-platform mobility. Moreover, market segment changes within a platform are apparently shaped by segment entries by platform complementors, whose decisions are motivated by industry events such as acquisitions. In order to manage the direction of segment evolution of a platform market, platform owners need to monitor and perhaps control the amount and types of industry events, not necessarily restricted to acquisitions, but also to other major events that could send signals to potential entrants, such as funding events, IPOs, and opening parts of the platform that only affect certain market segments. The more intriguing aspect is the normative implications implied by the analyses of post- entry performance. With respect to this, the second-stage findings on performance due to cross- platform mobility modify the initial understanding that platform mobility might not hurt a platform’s competitive advantages because higher performers tend to multihome while lower performers switch platforms. The findings of the first essay, in aggregate, suggest that the quality-driven platform (iOS) is losing competitive advantages towards the quantity-driven 212 platform (Android), largely because migrators’ performances are influenced by their platform mobility decisions on both home and target platforms. One major reason is that multihoming is costly, and as such, not all developers are able to afford it. The other reason is learning. The findings indicate that the Android platform benefits from platform migrations, because when iOS developers move to Android, and when Android developers port on iOS, they learn from the more demanding environment of the iOS platform. Learning in turn feeds their product performance on the Android platform. This finding echoes arguments from the broader literature on competitive advantages of nations (Porter, 1990), according to which, when doing business in an environment with more demanding customers (e.g., American companies selling products in Japan), such firms learn to improve their product quality not only for the market of the target country (Japan) but also for the home country (the U.S.). Consequently, asymmetric learning occurs between the two competing platforms, with one able to absorb more knowledge from the other side. For platform owners, being on the losing side of such asymmetric learning could forecast a not-very-promising future in the battle for the dominant technological standard. How to manage and mitigate this general trend thus becomes a critical issue for platform owners. Limitations and Directions for Future Research The following limitations of this dissertation provide avenues for future research. The first concerns data. Although the comprehensiveness of the proprietary data on apps and developers used in this dissertation is noteworthy, there remains much latitude for refinements. I mentioned in the second essay that a concern is survival bias—that is, only those developers and apps that survived until my data collection are accounted for. In future research, if scholars could obtain data of deceased app products and corresponding developers, such data could be used as other, 213 perhaps even more robust, indicators of market exits and product failures. Relatedly, meaningful research questions can be derived based on theoretical perspectives such as industry evolution and population ecology. Another data limitation concerns the coding of release dates. Release dates of iOS apps are based on data from one app analytics company. Given that the company collaborated with Apple via the latter’s Enterprise Partner Feed, the data are expected to be comprehensive and reliable. As suggested in the first essay, release dates of Android apps are proxied by the earliest version date collected by another analytics company. This limitation will be more challenging to overcome in future research. Assuming the availability of comprehensive data on Android apps’ version histories since their initial releases, more fine-grained research questions, such as the following, can be asked: How do developers allocate resources in updating products on both platforms? Whether and why do they shift resources in updating one side to updating the other side? The final data limitation is that I have only aggregated hardware data. For future research, with more detailed hardware data, scholars could effectively integrate with information on apps and developers and examine research questions to tap into cross-side network effects. For instance, how do available technologies on the hardware side (e.g., force touch, finger print recognition) facilitate innovation and diffusion of certain applications? Reversely, how do software advancements (e.g., artificial intelligence) stimulate the growth and innovation in hardware devices (e.g., healthcare equipment, home applications, auto-driving technologies)? The second set of limitations concerns the potential to develop more market-structure characteristics that are specific to platform-based markets. As suggested in my empirical essays, there are several potential directions to explore. One direction is to use alternative industry mega-events to reflect the conditions of a platform market. Candidate events include IPOs and 214 funding events, which can be further broken down in terms of entrepreneurial stages— bootstrapping, seed funding, angel investments for earlier entrepreneurial attempts; venture capital rounds and institutional investments for more mature entrepreneurial opportunities. Moreover, future research could adopt an ecosystem perspective in examining the structural features of a platform market (Iansiti & Levien, 2004a; Iansiti & Levien, 2004b). Some recent developments in the literature are along this direction, such as the work by (Argarwal & Kapoor, 2017) who proposed a structural variable of ecosystem connectedness. The final set of suggestions for future research is related to alternative means of capturing firm heterogeneities. One example that I gave in the second essay on segment entries is to leverage data on technologies within apps (e.g., permissions) to categorize developers. For this line of inquiry, scholars could build on innovation adoption literature by examining questions, such as (a) Where do new technologies for applications originate? (b) What kind of developers are more likely to be early adopters of available technologies, and how do they shape the further adoption of such technologies and evolution of the competitive landscape? (c) Whether and why do platforms intentionally stimulate certain types of technologies while inhibiting the development of others? And, (d) how do such interventions by platform owners affect the long- term development of the platforms and relative standing of platform complementors? Concluding Remarks I embarked on an ambitious research agenda in this dissertation. I believe that the topics of the two empirical essays—cross-platform mobility and within-platform segment entries—have profound implications for other platform-based markets as well. These topics not only shed light on important business phenomena that are of concern to both practitioners and academics, such 215 as whether a technological platform could eventually win a standards war, but they also build on and extend existing theoretical perspectives of strategic management to incorporate a novel form of business transaction—platforms as intermediaries. The theoretical development and empirical examination in this dissertation provide validity to a fundamental belief of strategic management scholars—when studying firm behaviors and performance consequences, both market structures and firm heterogeneities matter. Standing on the shoulders of giants, this dissertation invokes the unique attributes of platform markets and network economies to complement and advance received wisdom. 216 REFERENCES Adams, P., Fontana, R., & Malerba, F. 2016. User-industry spinouts: Downstream industry knowledge as a source of new firm entry and survival. Organization Science, 27(1): 18- 35. Adner, R. & Kapoor, R. 2010. Value creation in innovation ecosystems: How the structure of technological interdependence affects firm performance in new technology generations. Strategic Management Journal, 31(3): 306-333. Argarwal, S. & Kapoor, R. 2017. Two faces of value creation in business ecosystems: Leveraging complementarities and managing interdependencies. University of Pennsylvania, Philadelphia, PA. Argote, L. & Epple, D. 1990. Learning curves in manufacturing. Science, 247(4945): 920-924. Argote, L. & Greve, H. R. 2007. A behavioral theory of the firm—40 Years and counting: Introduction and impact. Organization Science, 18(3): 337-349. Argote, L. & Miron-Spektor, E. 2011. Organizational learning: From experience to knowledge. Organization Science, 22(5): 1123-1137. Armstrong, M. 2006. Competition in two-sided markets. RAND Journal of Economics, 37(3): 668-691. Ashenfelter, O. C., Hosken, D. S., & Weinberg, M. C. 2015. Efficiencies brewed: Pricing and consolidation in the US beer industry. RAND Journal of Economics, 46(2): 328-361. Barney, J. 1991. Firm resources and sustained competitive advantage. Journal of Management, 17(1): 99-120. 217 Baum, J. A. C., Li, S. X., & Usher, J. M. 2000. Making the next move: How experiential and vicarious learning shape the locations of chains' acquisitions. Administrative Science Quarterly, 45(4): 766-801. Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y.-T. 2005. Dancing with strangers: Aspiration performance and the search for underwriting syndicate partners. Administrative Science Quarterly, 50(4): 536-575. Beckman, C. M. & Haunschild, P. R. 2002. Network learning: The effects of partners' heterogeneity of experience on corporate acquisitions. Administrative Science Quarterly, 47(1): 92-124. Bonaccorsi, A., Giannangeli, S., & Rossi, C. 2006. Entry strategies under competing standards: Hybrid business models in the open source software industry. Management Science, 52(7): 1085-1098. Boudreau, K. 2010. Open platform strategies and innovation: Granting access vs. devolving control. Management Science, 56(10): 1849-1872. Boudreau, K. 2015. Competition, innovation, and experimentation on the Apple AppStore: Developer 'types', incentive levels and project choices. Boudreau, K. J. 2012. Let a thousand flowers bloom? An early look at large numbers of software app developers and patterns of innovation. Organization Science, 23(5): 1409-1427. Bresnahan, T., Orsini, J., & Yin, P.-L. 2014. Platform choice by mobile app developers. National Bureau of Economic Research, Cambridge, MA. Brown, S. L. & Eisenhardt, K. M. 1997. The art of continuous change: Linking complexity theory and time-paced evolution in relentlessly shifting organizations. Administrative Science Quarterly, 42(1): 1-34. 218 Caillaud, B. & Jullien, B. 2003. Chicken & egg: Competition among intermediation service providers. RAND Journal of Economics, 34(2): 309-328. Campbell, B. A., Ganco, M., Franco, A. M., & Agarwal, R. 2012. Who leaves, where to, and why worry? Employee mobility, entrepreneurship and effects on source firm performance. Strategic Management Journal, 33(1): 65-87. Carare, O. 2012. The impact of bestseller rank on demand: Evidence from the app market. International Economic Review, 53(3): 717-742. Carnabuci, G., Operti, E., & Kovács, B. 2015. The categorical imperative and structural reproduction: Dynamics of technological entry in the semiconductor industry. Organization Science, 26(6): 1734-1751. Carroll, G. R., Bigelow, L. S., Seidel, M.-D. L., & Tsai, L. B. 1996. The fates of de novo and de alio producers in the American automobile industry 1885-1981. Strategic Management Journal, 17(S1): 117-137. Casadesus-Masanell, R. & Zhu, F. 2013. Business model innovation and competitive imitation: The case of sponsor-based business models. Strategic Management Journal, 34(4): 464- 482. Caves, R. E. & Porter, M. E. 1977. From entry barriers to mobility barriers: Conjectural decisions and contrived deterrence to new competition. Quarterly Journal of Economics, 91(2): 241-261. Caves, R. E. 1992. American Industry: Structure, Conduct, Performance (7th ed.). Englewood Cliffs, New Jersey 07632: Prentice-Hall Inc. Cennamo, C. & Santalo, J. 2013. Platform competition: Strategic trade-offs in platform markets. Strategic Management Journal, 34(11): 1331-1350. 219 Chao, Y. & Derdenger, T. 2013. Mixed bundling in two-sided markets in the presence of installed base effects. Management Science, 59(8): 1904-1926. Chatterjee, S. & Wernerfelt, B. 1991. The link between resources and type of diversification: Theory and evidence. Strategic Management Journal, 12(1): 33-48. Christensen, C. M. 1997. The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail. Cambridge: Harvard Business School Press. Clements, M. T. & Ohashi, H. 2005. Indirect network effects and the product cycle: Video games in the U.S., 1994-2002. Journal of Industrial Economics, 53(4): 515-542. Clougherty, J. A. & Duso, T. 2009. The impact of horizontal mergers on rivals: Gains to being left outside a merger. Journal of Management Studies, 46(8): 1365-1395. Cohen, W. M. & Levinthal, D. A. 1990. Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35(1): 128-152. Corts, K. S. & Lederman, M. 2009. Software exclusivity and the scope of indirect network effects in the U.S. home video game market. International Journal of Industrial Organization, 27(2): 121-136. Cottrell, T. & Nault, B. R. 2004. Product variety and firm survival in the microcomputer software industry. Strategic Management Journal, 25(10): 1005-1025. Cyert, R. M. & March, J. G. 1963. A Behavior Theory of the Firm. Dafny, L., Duggan, M., & Ramanarayanan, S. 2012. Paying a premium on your premium? Consolidation in the U.S. health insurance industry. American Economic Review, 102(2): 1161-1185. 220 Davis, J. P., Muzyrya, Y., & Yin, P. L. 2013. Innovation strategies in entrepreneurial firms: Market maturity and killer apps in the iPhone ecosystem, West Coast Research Symposium. Seattle, WA. Davis, J. P., Muzyrya, Y., & Yin, P.-L. 2014. Experimentation strategies and entrepreneurial innovation: Inherited market differences in the iPhone ecosystem. In INSEAD (Ed.). Davis, R. & Thomas, L. G. 1993. Direct estimation of synergy: A new approach to the diversity- performance debate. Management Science, 39(11): 1334-1346. Dencker, J. C., Gruber, M., & Shah, S. K. 2009. Pre-entry knowledge, learning, and the survival of new firms. Organization Science, 20(3): 516-537. Dess, G. G. & Beard, D. W. 1984. Dimensions of organizational task environments. Administrative Science Quarterly, 29(1): 52-73. Dierickx, I. & Cool, K. 1989. Asset stock accumulation and sustainability of competitive advantage. Management Science, 35(12): 1504-1511. Diestre, L. & Rajagopalan, N. 2014. Toward an input-based perspective on categorization: Investor reactions to chemical accidents. Academy of Management Journal, 57(4): 1130-1153. Dobrev, S. D. & Kim, T.-Y. 2006. Positioning among organizations in a population: Moves between market segments and the evolution of industry structure. Administrative Science Quarterly, 51(2): 230-261. Doshi, A. 2014. The impact of high performance outliers on two-sided platforms: Evidence from crowdfunding: working paper. Harvard Business School Working Paper. Dosi, G. & Teece, D. 1998. Organizational competencies and the boundaries of the firm, Markets and Organization. Springer Berlin Heidelberg. 221 Dranove, D. & Gandal, N. 2003. The DVD-vs.-DIVX standard war: Empirical evidence of network effects and preannouncement effects. Journal of Economics & Management Strategy, 12(3): 363-386. Dunne, T., Roberts, M. J., & Samuelson, L. 1988. Patterns of firm entry and exit in U.S. manufacturing industries. RAND Journal of Economics, 19(4): 495-515. Economides, N. & Katsamakas, E. 2006. Two-sided competition of proprietary vs. open source technology platforms and the implications for the software industry. Management Science, 52(7): 1057-1071. Edelman, B. 2014. Mastering the intermediaries. Harvard business review, 92(6): 86-92, 138. Edwards, C. D. 1955. Conglomerate bigness as a source of power, Business Concentration and Price Policy: 329-358: Princeton University Press. Eggers, J. P. 2012. All experience is not created equal: Learning, adapting, and focusing in product portfolio management. Strategic Management Journal, 33(3): 315-335. Eggers, J. P. & Kaul, A. 2017. Motivation and ability? A behavioral perspective on the pursuit of radical invention in multi-technology incumbents. Academy of Management Journal. Eisenmann, T., Parker, G., & Van Alstyne, M. W. 2006. Strategies for two-sided markets. Harvard business review, 84(10): 92-101. Eisenmann, T., Parker, G., & Van Alstyne, M. 2011. Platform envelopment. Strategic Management Journal, 32(12): 1270-1285. Eisenmann, T. R., Parker, G., & Van Alstyne, M. 2008. Opening platforms: How, when and why? In A. Gawer (Ed.), Platforms, Markets and Innovation: 131-162: Edward Elgar Publishing. 222 Evans, D. S. 2009. How catalysts ignite: The economics of platform-based start-ups. In A. Gawer (Ed.), Platforms, Markets and Innovation. Cheltenham, UK • Northampton, MA, USA: Edward Elgar Publishing Limited. Fee, C. E. & Thomas, S. 2004. Sources of gains in horizontal mergers: Evidence from customer, supplier, and rival firms. Journal of Financial Economics, 74(3): 423-460. Figueiredo, J. M. d. & Silverman, B. S. 2007. Churn, baby, churn: Strategic dynamics among dominant and fringe firms in a segmented industry. Management Science, 53(4): 632- 650. Focarelli, D. & Panetta, F. 2003. Are mergers beneficial to consumers? Evidence from the market for bank deposits. American Economic Review, 93(4): 1152-1172. Franco, A. M., Sarkar, M., Agarwal, R., & Echambadi, R. 2009. Swift and smart: The moderating effects of technological capabilities on the market pioneering–firm survival relationship. Management Science, 55(11): 1842-1860. Gambardella, A., Ganco, M., & Honoré, F. 2015. Using what you know: Patented knowledge in incumbent firms and employee entrepreneurship. Organization Science, 26(2): 456-474. Gandal, N., Greenstein, S., & Salant, D. 1999. Adoptions and orphans in the early microcomputer market. The Journal of Industrial Economics, 47(1): 87-105. Garud, R. & Kumaraswamy, A. 1993. Changing competitive dynamics in network industries: An exploration of sun microsystems' open systems strategy. Strategic Management Journal, 14(5): 351-369. Gaur, A. S., Malhotra, S., & Zhu, P. 2013. Acquisition announcements and stock market valuations of acquiring firms' rivals: A test of the growth probability hypothesis in China. Strategic Management Journal, 34(2): 215-232. 223 Gavetti, G., Greve, H. R., Levinthal, D. A., & Ocasio, W. 2012. The behavioral theory of the firm: Assessment and prospects. Academy of Management Annals, 6(1): 1-40. Gawer, A. & Henderson, R. 2007. Platform owner entry and innovation in complementary markets: Evidence from Intel. Journal of Economics & Management Strategy, 16(1): 1- 34. Gawer, A. 2014. Bridging differing perspectives on technological platforms: Toward an integrative framework. Research Policy, 43(7): 1239-1249. Gilovich, T. 1987. Secondhand information and social judgment. Journal of Experimental Social Psychology, 23(1): 59-74. Gimeno, J. 1999. Reciprocal threats in multimarket rivalry: Staking out ‘spheres of influence’ in the U.S. airline industry. Strategic Management Journal, 20(2): 101-128. Gort, M. & Klepper, S. 1982. Time paths in the diffusion of product innovations. The Economic Journal, 92(367): 630-653. Greve, H. R. 1996. Patterns of competition: The diffusion of a market position in radio broadcasting. Administrative Science Quarterly, 41(1): 29-60. Greve, H. R. 1998. Performance, aspirations, and risky organizational change. Administrative Science Quarterly, 43(1): 58-86. Greve, H. R. 2003. A behavioral theory of R&D expenditures and innovations: Evidence from shipbuilding. Academy of Management Journal, 46(6): 685-702. Gupta, S., Jain, D. C., & Sawhney, M. S. 1999. Modeling the evolution of markets with indirect network externalities: An application to digital television. Marketing Science, 18(3): 396-416. 224 Hagiu, A. & Eisenmann, T. 2007. A staged solution to the catch-22. Harvard Business Review, 85(11): 25-26. Hagiu, A. 2011. Quantity vs. quality: Exclusion by platforms with network effects. Harvard Business School, Boston, MA. Hagiu, A. 2014. Strategic decisions for multisided platforms. MIT Sloan Management Review, 55(2). Hagiu, A. & Wright, J. 2015. Marketplace or reseller? Management Science, 61(1): 184-203. Haleblian, J., Devers, C. E., McNamara, G., Carpenter, M. A., & Davison, R. B. 2009. Taking stock of what we know about mergers and acquisitions: A review and research agenda. Journal of Management, 35(3): 469-502. Hallen, B. L., Davis, J. P., & Yin, P.-L. 2017. Network isolates: Entrepreneurial boostrapping and the social disconnection of new organizations in the mobile app ecosystem. University of Washington. Hannan, M. T. & Freeman, J. 1977. The population ecology of organizations. American Journal of Sociology, 82(5): 929-964. Hashai, N. 2015. Within-industry diversification and firm performance—an S-shaped hypothesis. Strategic Management Journal, 36(9): 1378-1400. Haunschild, P. R. & Beckman, C. M. 1998. When do interlocks matter?: Alternate sources of information and interlock influence. Administrative Science Quarterly, 43(4): 815-844. Helfat, C. E. & Lieberman, M. B. 2002. The birth of capabilities: Market entry and the importance of pre-history. Industrial and Corporate Change, 11(4): 725-760. Hossain, T., Minor, D., & Morgan, J. 2011. Competing matchmakers: An experimental analysis. Management Science, 57(11): 1913-1925. 225 Hsiao, C. 2014. Analysis of Panel Data: Cambridge University Press. Hsieh, K.-Y. & Vermeulen, F. 2013. The structure of competition: How competition between one's rivals influences imitative market entry. Organization Science. Huang, P., Ceccagnoli, M., Forman, C., & Wu, D. J. 2013. Appropriability mechanisms and the platform partnership decision: Evidence from enterprise software. Management Science, 59(1): 102-121. Iansiti, M. & Levien, R. 2004a. Keystones and dominators: Framing operating and technology strategy in a business ecosystem. Harvard Business School, Boston, MA. Iansiti, M. & Levien, R. 2004b. Strategy as ecology. Harvard business review, 82(3): 68-81. Isaacson, W. 2011. Steve Jobs. New York: Simon & Schuster. Jacobides, M. G. & Tae, C. J. 2015. Kingpins, bottlenecks, and value dynamics along a sector. Organization Science, 26(3): 889-907. Kacperczyk, A., Beckman, C. M., & Moliterno, T. P. 2015. Disentangling risk and change: Internal and external social comparison in the mutual fund industry. Administrative Science Quarterly, 60(2): 228-262. Kapoor, R. 2013. Collaborating with complementors: What do firms do?: Emerald Group Publishing Limited. Kapoor, R. & Furr, N. R. 2014. Complementarities and competition: Unpacking the drivers of entrants' technology choices in the solar photovoltaic industry. Strategic Management Journal, 36(3): 416-436. Kapoor, R. & Agarwal, S. 2017. Sustaining superior performance in business ecosystems: Evidence from application software developers in the iOS and Android smartphone ecosystems. Organization Science, forthcoming. 226 Kim, E. H. & Singal, V. 1993. Mergers and market power: Evidence from the airline industry. American Economic Review, 83(3): 549-569. King, A. A. & Tucci, C. L. 2002. Incumbent entry into new market niches: The role of experience and managerial choice in the creation of dynamic capabilities. Management Science, 48(2): 171-186. Klepper, S. 1997. Industry life cycles. Industrial and Corporate Change, 6(1): 145-182. Klepper, S. & Simons, K. L. 2000. Dominance by birthright: Entry of prior radio producers and competitive ramifications in the U.S. television receiver industry. Strategic Management Journal, 21(10-11): 997-1016. Klepper, S. 2002. The capabilities of new firms and the evolution of the U.S. automobile industry. Ind Corp Change, 11(4): 645-666. Kotha, R., Zheng, Y., & George, G. 2011. Entry into new niches: the effects of firm age and the expansion of technological capabilities on innovative output and impact. Strategic Management Journal, 32(9): 1011-1024. Lee, C.-H., Venkatraman, N., Tanriverdi, H., & Iyer, B. 2010. Complementarity-based hypercompetition in the software industry: Theory and empirical test, 1990–2002. Strategic Management Journal, 31(13): 1431-1456. Lee, J., Song, J., & Yang, J.-S. 2016. Network structure effects on incumbency advantage. Strategic Management Journal, 37(8): 1632-1648. Levinthal, D. A. 1997. Adaptation on rugged landscapes. Management Science, 43(7): 934-950. Levinthal, D. A. & Wu, B. 2010. Opportunity costs and non-scale free capabilities: Profit maximization, corporate scope, and profit margins. Strategic Management Journal, 31(7): 780-801. 227 Levitt, B. & March, J. G. 1988. Organizational learning. Annual Review of Sociology, 14: 319- 340. Liu, Y. 2014. Mobile app platform choice: An application of strategic games on big data. University of California, Berkeley, Berkeley, CA. Loch, C. H. & Huberman, B. A. 1999. A punctuated-equilibrium model of technology diffusion. Management Science, 45(2): 160-177. Lounsbury, M. & Leblebici, H. 2004. The origins of strategic practice: Product diversification in the American mutual fund industry. Strategic Organization, 2(1): 65-90. Lubatkin, M., Schulze, W. S., Mainkar, A., & Cotterill, R. W. 2001. Ecological investigation of firm effects in horizontal mergers. Strategic Management Journal, 22(4): 335-357. Macher, J. T. & Boerner, C. S. 2006. Experience and scale and scope economies: Trade-offs and performance in development. Strategic Management Journal, 27(9): 845-865. Majumdar, S. K. & Venkataraman, S. 1998. Network effects and the adoption of new technology: Evidence from the U.S. telecommunications industry. Strategic Management Journal, 19(11): 1045-1062. March, J. G. 2010. The Ambiguities of Experience: Cornell University Press. McFadden, D. 1973. Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Frontiers in econometrics: 105-142. New York: Academic Press. Miller, D. & Chen, M.-J. 1994. Sources and consequences of competitive inertia: A study of the U.S. airline industry. Administrative Science Quarterly, 39(1): 1-23. Miric, M. 2016. How do non-monetary motives influence the rate and direction of innovative activity? A study of competition and innovation in digital markets. University of Southern California, Los Angeles, CA. 228 Miric, M. 2017. Building network effects via business model design: A study of game center and freemium models on the App Store. University of Southern California, Los Angeles, CA. Nair, H., Chintagunta, P., & Dubé, J.-P. 2004. Empirical analysis of indirect network effects in the market for personal digital assistants. Quantitative Marketing and Economics, 2(1): 23-58. Navis, C. & Glynn, M. A. 2010. How new market categories emerge: Temporal dynamics of legitimacy, identity, and entrepreneurship in satellite radio, 1990–2005. Administrative Science Quarterly, 55(3): 439-471. Nelson, R. R. & Winter, S. G. 1982. An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge, MA. Nerkar, A. & Roberts, P. W. 2004. Technological and product-market experience and the success of new product introductions in the pharmaceutical industry. Strategic Management Journal, 25(8-9): 779-799. Parker, G., Van Alstyne, M. W., & Choudary, S. P. 2016. Platform Revolution (First ed.). New York: W. W. Norton & Company, Inc. Pontikes, E. G. 2012. Fitting in or starting new? Invention, leniency, and category emergence in the software industry, 1990–2002. University of Chicago, Chicago, IL. Porter, M. E. 1981. The contributions of industrial organization to strategic management. Academy of Management Review, 6(4): 609-620. Porter, M. E. 1985. Competitive advantage: Creating and sustaining superior performance: Collier Macmillan Publishers, New York, N.Y. Porter, M. E. 1990. The Competitive Advantage of Nations. Harvard Business Review, 68(2): 73-93. 229 Porter, M. E. 1996. What is strategy? Harvard Business Review, 74(6): 61-78. Prager, R. A. 1992. The effects of horizontal mergers on competition: The case of the northern securities company. RAND Journal of Economics, 23(1): 123-133. Prahalad, C. K. & Bettis, R. A. 1986. The dominant logic: A new linkage between diversity and performance. Strategic Management Journal, 7(6): 485-501. Rao, H., Davis, G. F., & Ward, A. 2000. Embeddedness, social identity and mobility: Why firms leave the NASDAQ and join the new york stock exchange. Administrative Science Quarterly, 45(2): 268-292. Rochet, J.-C. & Tirole, J. 2006. Two-sided markets: A progress report. RAND Journal of Economics, 37(3): 645-667. Rogan, M. & Sorenson, O. 2014. Picking a (poor) partner: A relational perspective on acquisitions. Administrative Science Quarterly, 59(2): 301-329. Santos, F. M. & Eisenhardt, K. M. 2009. Constructing markets and shaping boundaries: Entrepreneurial power in nascent fields. Academy of Management Journal, 52(4): 643- 671. Schilling, M. A. 2002. Technology success and failure in winner-take-all markets: The impact of learning orientation, timing, and network externalities. Academy of Management Journal, 45(2): 387-398. Schilling, M. A. 2009. Protecting or diffusing a technology platform: Tradeoffs in appropriability, network externalities, and architectural control. In A. Gawer (Ed.), Platforms, Markets and Innovation: 192-218. Cheltenham, UK: Edward Elgar Publishing Limited 230 Shapiro, C. & Varian, H. R. 1999. Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press, Cambridge, MA. Silverman, B. S. 1999. Technological resources and the direction of corporate diversification: Toward an integration of the resource-based view and transaction cost economics. Management Science, 45(8): 1109-1124. Simon, H. A. 1996. The Sciences of the Artificial. MIT Press, Cambridge, MA. Sleeper, S. 1998. The role of firm capabilities in the evolution of the laser industry: The making of a high-tech market. Unpublished Doctoral Dissertation, Carnegie Mellon University, Pittsburgh, PA. Stigler, G. J. 1950. Monopoly and oligopoly by merger. American Economic Review, 40(2): 23- 34. Suchman, M. C. 1995. Managing legitimacy: Strategic and institutional approaches. Academy of Management Review, 20(3): 571-610. Sun, M. & Tse, E. 2009. The resource-based view of competitive advantage in two-sided markets. Journal of Management Studies, 46(1): 45-64. Tanriverdi, H. & Venkatraman, N. 2005. Knowledge relatedness and the performance of multibusiness firms. Strategic Management Journal, 26(2): 97-119. Tanriverdı ̇ , H. n. & Lee, C.-H. 2008. Within-industry diversification and firm performance in the presence of network externalities: Evidence from the software industry. Academy of Management Journal, 51(2): 381-397. Teece, D. J. 1986. Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy. Research Policy, 15(6): 285-305. 231 Tellis, G. J., Yin, E., & Niraj, R. 2009. Does quality win? Network effects versus quality in high- tech markets. Journal of Marketing Research, 46(2): 135-149. Teodoridis, F. 2016. Understanding team knowledge production: Generalists and specialists. University of Southern California, Los Angeles, CA. Terza, J. V., Basu, A., & Rathouz, P. J. 2008. Two-stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3): 531-543. Uzzi, B. 1997. Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42(1): 35-67. Venkatraman, N. & Lee, C.-H. 2004. Preferential linkage and network evolution: A conceptual model and empirical test in the U.S. video game sector. Academy of Management Journal, 47(6): 876-892. Wang, Y. & Rajagopalan, N. 2015. Alliance capabilities: Review and research agenda. Journal of Management, 41(1): 236-260. Wernerfelt, B. 1984. A resource-based view of the firm. Strategic Management Journal, 5(2): 171-180. Winter, S. G. 2003. Understanding dynamic capabilities. Strategic Management Journal, 24(10): 991-995. Xu, X., Venkatesh, V., Tam, K. Y., & Hong, S.-J. 2010. Model of migration and use of platforms: Role of hierarchy, current generation, and complementarities in consumer settings. Management Science, 56(8): 1304-1323. 232 Ye, G., Priem, R. L., & Alshwer, A. A. 2012. Achieving demand-side synergy from strategic diversification: How combining mundane assets can leverage consumer utilities. Organization Science, 23(1): 207-224. Yu, Y., Umashankar, N., & Rao, V. R. 2016. Choosing the right target: Relative preferences for resource similarity and complementarity in acquisition choice. Strategic Management Journal, 37(8): 1808-1825. Zelner, B. A. 2009. Using simulation to interpret results from logit, probit, and other nonlinear models. Strategic Management Journal, 30(12): 1335-1348. Zhao, E. Y., Ishihara, M., & Jennings, P. D. 2013. Adaptive emulation to trend setter categories: Following in the U.S. video game industry, 1995–2012, West Coast Research Symposium. Seattle, WA. Zhu, F. & Iansiti, M. 2012. Entry into platform-based markets. Strategic Management Journal, 33(1): 88-106. Zhu, F. & Liu, Q. 2017. Competing with complementors: An empirical look at Amazon.com. Harvard Business School, Boston, MA.

Abstract (if available)

Abstract Platforms have emerged as an important form of economy in the digital revolution age. Although existing research on platform-based competition has mainly focused on platform-level strategies, little empirical research exists on the strategic behaviors of platform complementors. In this dissertation, I investigate market entries by software application developers, acting as complementors of two dominant mobile platforms (iOS and Android), across and within the competing platforms, and the performance consequences of their strategic choices. I propose that developers’ market-entry decisions and post-entry performances are determined by both market structures and developer heterogeneities. ❧ The first empirical essay in the dissertation examines how developers’ cross-platform mobility shapes platform-level competition between iOS and Android. I argue that the market structure of the two competing platforms with different strategic emphases—one quality-driven (iOS), one quantity-driven (Android)—and app developers’ performance feedback jointly explain variance in developers’ mobility patterns across platforms. Although both high performers and low achievers tend to move to the competing platform, they are motivated by different incentives. High performers move across platforms in order to pursue greater network effects and the possibilities of “winner-take-all” in their niche markets

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Empirical essays on relationships between alliance experience and firm capability development

PDF

Essays on online advertising markets

PDF

Efficient policies and mechanisms for online platforms

PDF

Housing market crisis and underwater homeowners

PDF

Theoretical and data-driven analysis of two-sided markets and strategic behavior in modern economic systems

PDF

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

Asset Metadata

Creator Wang, Yongzhi (author)

Core Title Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Publication Date 07/18/2021

Defense Date 05/17/2017

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Android,app developers,behavioral theory of the firm,big data,capabilities,category entry,ecosystems,experiences,industrial organization,iOS,market entry,market structure,mobile apps,mobile platforms,network effects,OAI-PMH Harvest,performance feedback,platform migration,platform mobility,platforms,resources

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Rajagopalan, Nandini (committee chair), Hsiao, Cheng (committee member), Teodoridis, Florenta (committee member), Wu, Brian (committee member), Yue, Lori Qingyuan (committee member)

Creator Email wang.2877@osu.edu,yongzhiw@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-407593

Unique identifier UC11265799

Identifier etd-WangYongzh-5568.pdf (filename),usctheses-c40-407593 (legacy record id)

Legacy Identifier etd-WangYongzh-5568.pdf

Dmrecord 407593

Document Type Dissertation

Rights Wang, Yongzhi

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA