Abstract:  User-generated content (UGC) platforms increasingly enable content deletion to manage the online environment and user-posted content, which contributes to missing data issues in empirical research involving UGC data. Such missingness may be correlated with the unobserved data, resulting in the most challenging missingness type–Missing Not at Random (MNAR), which can lead to biased estimates in empirical analyses. Despite its prevalence, limited scholarly attention has been given to the missing data challenges arising from content deletion, and researchers lack theoretical guidance for diagnosing and handling such missingness. In this study, we conduct an in-depth investigation of the reasons for content deletion across various UGC platforms and identify variables susceptible to the MNAR issue in empirical studies. We also develop a theoretical framework for diagnosing missingness mechanisms and selecting appropriate correction methods. Most importantly, we propose theoretical formulas to correct biased estimation in two MNAR scenarios for scale-level studies. Extensive simulations and sensitivity tests demonstrate that our proposed methods effectively reduce the estimation bias under scale-level MNAR compared to benchmark approaches. Collectively, our work advances the missing data literature and provides practical implications for empirical researchers.

Keywords: Content deletion, content moderation, UGC platforms, missing data, missing not at random, estimation bias

Abstract: The online gaming industry increasingly incorporates virtual agents to enhance player experiences. Although prior literature has explored the provision of virtual agents in gaming, research on technological advancements remains limited. In this study, we investigate how introducing artificial intelligence (AI) powered agents as virtual opponents (versus rule-based opponents) influences human players’ engagement and performance. Leveraging a large-scale quasi-field experiment in a multiplayer online racing game, we employed difference-in-differences analyses with matching strategies. We show that the introduction of AI opponents can have “discouragement effects” on players, resulting in reduced player engagement and decreased performance. In addition, our mechanism exploration revealed that introducing AI opponents increases competition intensity and immersion in the game, and these two factors exhibit opposing influences on players’ subsequent behavior. Specifically, heightened competition hinders players’ further engagement and performance progression, whereas a more immersive experience encourages more gaming participation and better performance. Further, we found that the effects of AI opponents on player engagement and performance vary by player motivation and skill levels, such that competition-oriented and highly skilled players are more receptive to AI opponents. Moreover, our findings indicate that optimal engagement and performance outcomes occur when players compete against opponents with comparable and low competence levels, respectively. Lastly, we observed an inverted U-shaped relationship between the proportion of AI opponents and players’ engagement and performance. Our study contributes to the literature on human-AI interactions by offering novel empirical evidence on the impact of AI opponents on human players’ experiential and instrumental outcomes and disentangling the underlying mechanisms. This work also offers practical implications for game designers and policymakers regarding the design of AI-integrated competitive environments.

Keywords: Artificial intelligence opponents, online gaming, player engagement, player performance, self-efficacy theory, self-determination theory

Abstract: Online discussion platforms distribute content primarily through social and nonsocial newsfeed channels. Prior research has examined user engagement within individual channels, but less is known about their interactions. This study investigates the impact of reducing the visibility of a social channel (friends’ liked content) on the quantity and diversity of user engagement with other newsfeed channels. Drawing on the channel substitution and complementarity framework, we develop our hypotheses and test them using a newsfeed change that reduces the visibility of friends’ liked content. Our results show that this change leads to a general decline in both the quantity and diversity of user-engaged content. Across channels, users increase their engagement with other social channels (e.g., friends’ posts and trending topics), indicating a substitution effect among social channels. In contrast, they engage less with nonsocial content, suggesting a complementary relationship between social and nonsocial channels. Further, mechanism explorations indicate that the degree to which channels fulfill users’ social and informational needs influences whether they substitute or complement each other. Additionally, the decline in content diversity stems from reduced interaction with niche content liked by social connections. These findings offer key insights for content distribution strategies, newsfeed design, and efforts to mitigate echo chambers on digital platforms.

Keywords: Online discussion platforms, content distribution, newsfeed channels, user engagement, trending content, social networks

Abstract: With increasing volumes of participation in social media and online communities, content moderation has become an integral component of platform governance. Volunteer (human) moderators have thus far been the essential workforce for content moderation. Because volunteer-based content moderation faces challenges in achieving scalable, desirable, and sustainable moderation, many online platforms have recently started to adopt algorithm-based content moderation tools (bots). When bots are introduced into platform governance, it is unclear how volunteer moderators react in terms of their community-policing and -nurturing efforts. To understand the impacts of these increasingly popular bot moderators, we conduct an empirical study with data collected from 156 communities (subreddits) on Reddit. Based on a series of econometric analyses, we find that bots augment volunteer moderators by stimulating them to moderate a larger quantity of posts, and such effects are pronounced in larger communities. Specifically, volunteer moderators perform 20.9% more community policing, particularly over subjective rules. Moreover, in communities with larger sizes, volunteers also exert increased efforts in offering more explanations and suggestions after their community adopted bots. Notably, increases in activities are primarily driven by the increased need for nurturing efforts to accompany growth in subjective policing. Moreover, introducing bots to content moderation also improves the retention of volunteer moderators. Overall, we show that introducing algorithm-based content moderation into platform governance is beneficial for sustaining digital communities.

Keywords: Content moderation, human–machine collaboration, bot, volunteer moderators, platform governance

Abstract: We theorize peer awards’ effects on the volume and novelty of creative user-generated content (UGC) produced at online platform communities. We then test our hypotheses via a randomized field experiment on Reddit, wherein we randomly and anonymously assigned Reddit’s Gold Award to 905 users’ posts over a two-month period. We find that peer awards induced recipients to make longer, more frequent posts and that these effects were particularly pronounced among newer community members. Further, we show that recipients were causally influenced to engage in greater (lesser) exploitation (exploration) behavior, producing content that exhibited significantly greater textual similarity to their own past (awarded) content. However, because the effects were most pronounced among new community members, who also produce content that, in general, is systematically more novel than that of established members to begin with, this process yields a desirable outcome: larger volumes of generally novel UGC for the community.

Keywords: Peer awards, user-generated content, creativity, Reddit, text-mining, field experiment