Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Uncovering the Effect of Toxicity on Player Engagement and its Propagation in Competitive Online Video Games

Jacob Morrier, Amine Mahmassani, R. Michael Alvarez

TL;DR

The paper tackles the problem of estimating the causal impact of exposure to toxic language on player engagement and the propagation of toxicity in competitive online games. It employs a structural model and a leave-one-out instrumental variable to achieve a 2SLS estimate that isolates exogenous variation in exposure, addressing endogeneity and peer-effects. The study reveals significant, context-dependent effects: exposure increases time to next match and boosts the probability of using similar language, with heterogeneous magnitudes across opponents vs teammates, same vs different party, and match outcomes, along with pronounced propagation among teammates in the same party. Practically, the results offer actionable guidance for publishers on where to allocate toxicity-mitigation resources, balancing goals of reducing engagement harm and limiting toxicity spread, especially under loss conditions.

Abstract

This article seeks to provide accurate estimates of the causal effect of exposure to toxic language on player engagement and the proliferation of toxic language. To this end, we analyze proprietary data from the first-person action video game Call of Duty: Modern Warfare III, published by Activision. To overcome causal identification problems, we implement an instrumental variables estimation strategy. Our findings confirm that exposure to toxic language significantly affects player engagement and the probability that players use similar language. Accordingly, video game publishers have a vested interest in addressing toxic language. Further, we demonstrate that this effect varies significantly depending on whether toxic language originates from opponents or teammates, whether it originates from teammates in the same party or a different party, and the match's outcome. This has meaningful implications regarding how resources for addressing toxicity should be allocated.

Uncovering the Effect of Toxicity on Player Engagement and its Propagation in Competitive Online Video Games

TL;DR

The paper tackles the problem of estimating the causal impact of exposure to toxic language on player engagement and the propagation of toxicity in competitive online games. It employs a structural model and a leave-one-out instrumental variable to achieve a 2SLS estimate that isolates exogenous variation in exposure, addressing endogeneity and peer-effects. The study reveals significant, context-dependent effects: exposure increases time to next match and boosts the probability of using similar language, with heterogeneous magnitudes across opponents vs teammates, same vs different party, and match outcomes, along with pronounced propagation among teammates in the same party. Practically, the results offer actionable guidance for publishers on where to allocate toxicity-mitigation resources, balancing goals of reducing engagement harm and limiting toxicity spread, especially under loss conditions.

Abstract

This article seeks to provide accurate estimates of the causal effect of exposure to toxic language on player engagement and the proliferation of toxic language. To this end, we analyze proprietary data from the first-person action video game Call of Duty: Modern Warfare III, published by Activision. To overcome causal identification problems, we implement an instrumental variables estimation strategy. Our findings confirm that exposure to toxic language significantly affects player engagement and the probability that players use similar language. Accordingly, video game publishers have a vested interest in addressing toxic language. Further, we demonstrate that this effect varies significantly depending on whether toxic language originates from opponents or teammates, whether it originates from teammates in the same party or a different party, and the match's outcome. This has meaningful implications regarding how resources for addressing toxicity should be allocated.
Paper Structure (12 sections, 4 equations, 6 figures, 2 tables)

This paper contains 12 sections, 4 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset Description
  4. Model Specification
  5. Causal Identification Problems
  6. Identification Strategy
  7. Estimation
  8. Results
  9. Probability of Exposure to Toxic Language
  10. Effect of Toxic Language from Opponents and Teammates
  11. Effect of Toxic Language from Different-Party and Same-Party Teammates
  12. Discussion

Figures (6)

  • Figure 1: Daily Evolution of the Share of Unavailable Exposure Data
  • Figure 2: Daily Evolution of the Probability of Exposure to Toxic Language
  • Figure 3: Probability of Exposure to Toxic Language from March 4 to April 12, 2024
  • Figure 4: Probability of Exposure to Toxic Language
  • Figure 5: Effect of Exposure to Toxic Language from Opponents and Teammates
  • ...and 1 more figures