Working Paper: CEPR ID: DP17353
Authors: Andrey Simonov; Raluca Ursu; Carolina Zheng
Abstract: We quantify the relative importance of beliefs-based suspense and surprise measures in the entertainment preferences of viewers of Twitch.tv, the largest online video game streaming platform. Using detailed viewership and game statistics data from broadcasts of tournaments of a popular video game, Counter-Strike: Global Offensive (CS:GO), we compute measures of suspense and surprise for a rational viewer. We then develop and estimate a stylized utility model that underlies viewers' decisions to both join and leave a game stream. Our method allows us to causally identify the direct effect of suspense and surprise on viewers' utilities, separating it from other sources of entertainment value (e.g. team skill) and from indirect/supply-side effects (e.g. word of mouth or advertising). We show that suspense enters a viewer's utility, but find little evidence of the effect of surprise. The magnitudes imply that a one standard deviation increase in round-level suspense decreases the probability of leaving a stream by 0.27 percentage points. We find no detectable effect of suspense and surprise on the decision to join a stream, ruling out indirect effects. Variation in suspense levels explains 9.2% of the observed range of the evolution of a stream's viewership. We use these estimates to evaluate counterfactual game and platform designs. We show that historical updates to CS:GO game rules have increased tournament viewership by 4.1%, that rules can be further modified to increase viewership, and that alternative platform designs that inform joining users of games' scores will additionally increase overall viewership by 1.3%. Together, these results illustrate the value of our method as a general tool that content producers and platforms can use to evaluate and design media products.
Keywords: media product design; platform design; entertainment preference; information
JEL Codes: C55; L15; L82; L83; L86; M31
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| suspense (Y60) | viewer retention (Y10) |
| surprise (Y60) | viewer behavior (Y10) |
| suspense (Y60) | viewership evolution (F62) |
| historical updates to CS:GO game rules (C70) | tournament viewership (Z21) |
| alternative platform designs (B50) | overall viewership (Z21) |