UC San Diego

This dissertation studies viewer engagement on livestreaming platforms, with a focus on Twitch.tv. Across three chapters, it investigates how the real-time interactive nature of livestreaming shapes the relationship between streamers and their audiences, and how that relationship in turn shapes viewer responses to commercial and creative interventions.Chapter 1 develops and validates a measurement framework for emotional attachment in livestreaming. Because emotional attachment cannot be directly observed from platform behavior data, the chapter combines a survey of 146 active Twitch viewers with a one-month snapshot of real-world Twitch broadcast data on the same streamers. The survey-based analysis establishes that a streamer's conversational engagement cultivates viewer attachment, and that attachment in turn governs viewers' tolerance for sponsored content. Linking the survey responses to the real-world broadcast data then shows that a streamer's proportion of broadcasts in Twitch's "Just Chatting'' content category serves as a reliable proxy for the construct, closing a gap that platform research typically leaves open.Chapter 2 applies this validated measurement to study how sponsored content affects concurrent viewership. Using twelve months of session data on 24,080 Twitch streamers and a matched-sample event study with 6,548 treated-control session pairs, the chapter shows that switching from organic to sponsored content reduces concurrent viewership by approximately 13.85%, but that this effect is significantly attenuated for streamers who sustain conversational engagement. A within-streamer decomposition identifies emotional attachment as a moderator of sponsorship avoidance specific to real-time interactive media.Chapter 3 examines streamer-to-streamer collaborations, exploiting Twitch's launch of the @-tag feature on 17 February 2023 as a platform-wide natural experiment. Drawing on more than two years of Twitch broadcast data, a segmented interrupted time-series analysis shows that the launch produced substantial expansion along every collaboration margin while leaving broader engagement metrics largely unaffected. A streamer-level stacked event study around 7,220 collaboration events documents an on-day spike in follower acquisition of 5.4% to 8.7%, with a roughly monotonic asymmetry across the partner-status differential consistent with brand-alliance theory: lower-status partners benefit disproportionately. Together, these chapters develop a unified empirical and theoretical framework for understanding viewer behavior in livestreaming.