TikTok comedy during COVID-19: a multimodal study of platform data, audience engagement, and creative performance

This thesis investigates how multimodal features, platform infrastructure, and creative strategy jointly shape audience engagement in TikTok comedy content produced during the COVID-19 period (2020 to 2022). Drawing on a shared corpus of 741 videos from 14 high-profile creators active in the United Kingdom and the United States, three interrelated empirical studies address this aim from complementary analytical angles: evaluating the data infrastructure through which platform content is accessed, modelling how visual, audio, linguistic, and platform features jointly predict audience engagement, and examining how a specific creative strategy independently predicts engagement above and beyond correlated production characteristics. The three studies form a progressive analytical sequence in which each paper establishes the methodological and conceptual foundation for the next. The first paper subjects TikTok’s Research API to the first systematic comparative evaluation of its kind, contrasting its outputs against a commercial third-party analytics service and manual data collection across a broader corpus of 6,373 videos. The API offers genuine strengths, including scalability and access to 22 variables, eight of which are unavailable through alternative methods. The limitations are substantial: transcription coverage reaches approximately ten per cent of content, with systematic demographic skews favouring American male creators, and engagement metrics are inconsistent across data sources. The paper advocates multi-method auditing protocols as a necessary condition of validity for platform-based social science research. The second paper examines how visual composition, audio context, linguistic markers, and platform affordances jointly predict three engagement outcomes in TikTok comedy videos, using Generalised Additive Mixed Models with creator-specific random intercepts. Likes, comments, and shares respond to qualitatively distinct multimodal configurations rather than a single underlying engagement factor. Likes function as sensory verdicts, sensitive to visual composition and prosodic expressiveness. Comments reflect cognitive verdicts, driven by video duration, title sentiment, and linguistic complexity. Shares embody social verdicts, shaped by relationship-focused content, transcript sentiment, and lexical accessibility. Video description length is the sole universal predictor across all three models. Creator gender consistently moderates multimodal effects throughout. The third paper investigates role-play, a comedic strategy in which creators explicitly embody characters distinct from their everyday personas, accounting for 57.4 per cent of the corpus. Role-play videos display a consistent and identifiable production signature, characterised by significantly higher rates of accessory use, distinct costuming, and relationship-focused narratives. Role-play independently predicts approximately 21 per cent higher like counts and 42 to 48 per cent higher share counts after controlling for the full production feature set, yet shows no significant independent effect on comment counts. Male creators derive substantially greater sharing gains from same-gender role-play than female creators, a pattern consistent with accounts of gender-incongruent performance in algorithmically mediated contexts. Across the three papers, two findings recur with sufficient consistency to constitute thesis-level contributions. Engagement on TikTok is irreducibly multidimensional: likes, comments, and shares are driven by distinct feature configurations and cannot be collapsed into a single outcome measure without substantial loss of analytical precision. Creator gender is a structural moderating variable that shapes not only baseline engagement levels but the mechanisms through which multimodal features and creative strategies translate into audience response. Together, the three studies advance multimodal communication theory, extend performance and gender performativity frameworks into short-form digital contexts, and contribute to an emerging methodological literature on the conditions of validity for platform-based social science.