Signal The Effect of Perceived Race and Gender on Police Language Use: Experimental Evidence from VR Simulations

Summary

Sandra C. Sandoval, Navita Goyal, Rashawn Ray, Long Doan, Rachel Rudinger, and Hal Daumé published a paper on arXiv (cs.CY) on August 5, 2026, examining how deferentially police officers speak to virtual characters depicted as Black adult males in virtual reality simulations, using a causal inference framework where assignment of the Black man character to an officer and simulation is the treatment variable. The study found that most officers speak less deferentially to Black man characters, except for White, biracial, and multiracial female officers, with the effect especially pronounced when the VR character was known to be a suspect. Across a full conversation in a typical VR scene, these marginal treatment effects can result in notable changes in deference of tone, two to several points on a 0-10 scale, beyond the initial effect of perceiving a Black male character. The researchers warn this can contribute to conversation breakdowns that potentially result in violence or danger to both the public and police. The paper also explored large language models for treatment effect estimation, recommending mixed effects models with inverse propensity treatment weighting using LLM-generated text features, while noting LLM fine-tuning for prediction models requires further development.