Research

Abstract: Thumbnails, reduced-size preview images or clips, have emerged as a pivotal visual cue that helps consumers navigate through video selection while "previewing" for what to expect in the video. We study how thumbnails, relative to video content, affect viewers’ video behavior (e.g., views, watchtime, preference match, and engagement). We propose a video mining procedure that decomposes high-dimensional video data into interpretable features (image content, affective emotions and aesthetics), leveraging computer vision, deep learning, text mining and advanced large language models. Motivated by behavioral theories such as expectation-disconfirmation theory and Loewenstein's theory of curiosity, we postulate that thumbnails may affect video reactions through the roles of expectation-based reference points (which shape viewers’ expectations for content prior to watching a video) or informational reference points (which build anticipation for the upcoming video content). These two roles can create opposing effects on consumers’ video viewing behavior. Based on these theories,  we construct measures to evaluate the thumbnail relative to the video content to assess the degree to which the thumbnail is representative of the video. Using both secondary data from YouTube and a novel video streaming platform called “CTube” that we build to exogenously randomize thumbnails across videos, we find that aesthetically pleasing thumbnails lead to overall positive outcomes across measures (e.g., views and watchtime). On the other hand, content disconfirmation between the thumbnail and the video leads to opposing effect, generating more views but lower post-video engagement (e.g., likes and comments) for a video. However, for ongoing engagement such as watchtime, content disconfirmation leads to higher watchtime, suggesting that the role of thumbnails as generating curiosity for what may come next in the video dominates. We contrast a series of alternative stylized thumbnail candidates with predicted model-based optimal frames and platform-recommended frames. Based on the results, we then characterize  “optimal” thumbnails and design criteria for different video outcomes to guide creators and platforms in thumbnail selection and creation. 

Abstract: We build a Bayesian learning model to investigate how thumbnails, previews of video content, affect consumers’ reactions to videos in terms of video choice and watchtime via two roles: thumbnails as expectation-based reference points that shape viewers’ expectations of the video content before they start watching a video, and thumbnails as informational reference points that build viewers’ anticipation for a video. We model consumers' decisions to click on a video and continue watching the video as based on their priors (the thumbnail) and updated beliefs of the video content (the video's frames, characterized as multi-dimensional and correlated video topic proportions). We create a novel video streaming platform called "CTube" (a simplified version of YouTube) to conduct a novel field-framed online experiment by randomizing thumbnails viewers see to start a video. Leveraging the high-frequency clickstream data tracked by our video platform, we estimate the bayesian learning model by restoring from clickstream data individuals’ exposed video thumbnails, video choice and watchtime decisions. Our results suggest that  viewers overall prefer watching videos longer when there is a higher disconfirmation between their initial content beliefs based on the thumbnail and updated beliefs based on the observed video scenes (signals). In addition, viewers prefer less content disconfirmation before observing the thumbnail, highlighting that the role of disconfirmation may change before and after viewers observe the moment highlighted by the thumbnail. Based on the model's estimates, we then run a series of counterfactual analyses to propose optimal thumbnails and compare them with current practices of thumbnail recommendation to guide creators and platforms in thumbnail selection. 

Abstract: The necessity of content regulation on digital platforms, particularly concerning misinformation and harmful content, has sparked a growing debate. While many platforms have increasingly relied on self-regulation to address these issues, the effectiveness of such practices remains uncertain, as platforms may prioritize profits over consumer protection and thus have misaligned incentives with regulators. We investigate the effectiveness and market outcomes of content self-regulation by studying Twitch’s ban on online gambling livestreams in October 2022, using a novel high-frequency panel dataset covering the top 6,000 Twitch streamers. We leverage video analysis on historical video clips, high-frequency stream titles, and in-stream chat analysis to identify banned content and streamers affected by the policy. To tackle key identification challenges, we use three causal estimators: two-way fixed effects DiD, Synthetic DiD, and the doubly-robust estimator of group-time average treatment effects, and employ network analysis to construct valid treated and control groups. On the supply side, we find that the policy caused a reduction in weekly gambling streams by 63.2% for streamers whose content was banned and 12.2% for streamers whose content was not banned. However, the policy also decreased non-gambling streams, resulting in an unwanted reduction in total content production and content diversity on the platform. In particular, the more popular streamers experienced a higher content reduction, driven by two underlying mechanisms: lower reliance on gambling content and concerns for reputation. On the demand side, we find that the policy reduced total viewership and low-tier subscriptions among affected streamers but did not reduce revenue from their loyal viewers. We discuss the implications of Twitch’s policy ban and the broader practices of content self-regulation on platforms in general.

Abstract: We study content collaboration in the creator economy, in which competing creators mutually agree to collaborate on joint content and negotiate on content production and revenue sharing. Using a game theory model with creators competing for consumers on a Hotelling line, we show that collaboration allows creators to use the jointly-produced content to moderate competition, while using their individual content to expand into new audiences. This increases content diversity but also leads to increased monetizability of content. In general, collaboration among creators has an effect of increasing the profits of creators while reducing consumer surplus. When creators create content with heterogeneous entertainment values, the creator producing content of lower entertainment value has an incentive to free ride on the collaborative content. This free riding may increase surplus for consumers (who without collaboration would watch content of low entertainment value), thereby improving creators' profits as well as consumer surplus. Our results provide guidance to content creators, to platforms designing tools to facilitate collaborations, and to policy makers.

Abstract: Online video platforms rely on creators to supply various types of content to viewers, potentially leading to production inefficiencies in the marketplace. We define and separate out two mechanisms behind such inefficiencies: creators’ profit motives and direct preferences for producing content. “Creators-as- firms” care about profits and may over-enter into broadcasting popular content with potential high returns. “Creators-as-consumers” choose content they like to produce (e.g. which games they play) that may not align with what viewers want to watch. These two mechanisms have drastically different policy prescriptions for how platforms can reduce inefficiencies and increase profitability. We separate out profit-driven and preference-driven motives behind creators’ broadcasting decisions on Twitch.tv, the largest online video game streaming platform. By monitoring livestreams of pre-selected 30,000 streamers (based on a 1-month pilot study) and their viewers every 15 minutes for nearly a year, we construct a novel high-frequency dataset covering both streamers’ broadcasting decisions and individual-level records of viewers’ consumption decisions. Based on separate network analysis on streamers and viewers’ game decisions, we find evidence of production inefficiencies in the market (e.g, extraneous content on the platform that few viewers watch, and misallocation of production decisions of streamers even for content that is watched). Using detailed individual-level viewership data, we estimate heterogeneous viewers’ preferences for content and streamers. Using high frequency data on streamers’ production decisions, coupled them with demand estimates, we then estimate weights on streamers’ profit-driven and preference-driven incentives, documenting substantial heterogeneity in streamers’ tastes. We use the estimates to quantify the size of market inefficiencies and evaluate counterfactual policy prescriptions for welfare improvement.