Escalated police stops of Black men are linguistically and psychologically distinct in their earliest moments

Eugenia Ha Rim Rho, Maggie Harrington, Yuyang Zhong, Reid Pryzant, Nicholas P. Camp, Dan Jurafsky, Jennifer L. EberhardtProceedings of the National Academy of Sciences 2023 · DOI

Abstract

Across the United States, police chiefs, city officials, and community leaders alike have highlighted the need to de-escalate police encounters with the public. This concern about escalation extends from encounters involving use of force to routine car stops, where Black drivers are disproportionately pulled over. Yet, despite the calls for action, we know little about the trajectory of police stops or how escalation unfolds. In study 1, we use methods from computational linguistics to analyze police body-worn camera footage from 577 stops of Black drivers. We find that stops with escalated outcomes (those ending in arrest, handcuffing, or a search) diverge from stops without these outcomes in their earliest moments—even in the first 45 words spoken by the officer. In stops that result in escalation, officers are more likely to issue commands as their opening words to the driver and less likely to tell drivers the reason why they are being stopped. In study 2, we expose Black males to audio clips of the same stops and find differences in how escalated stops are perceived: Participants report more negative emotion, appraise officers more negatively, worry about force being used, and predict worse outcomes after hearing only the officer's initial words in escalated versus non-escalated stops. Our findings show that car stops that end in escalated outcomes sometimes begin in an escalated fashion, with adverse effects for Black male drivers and, in turn, police-community relations.

Introduction

While racial disparities in the frequency of police stops are well-documented, the internal dynamics of these interactions—the "black box" of the encounter—remain under-explored. Does escalation result from a gradual breakdown in communication, or is the tone set from the very first words? This research investigates the interactional trajectory of police stops involving Black men, seeking to identify linguistic patterns that predict escalated outcomes like arrest, search, or handcuffing.

Methodology

The researchers employed a two-stage approach to analyze the linguistic and psychological dimensions of police stops.

Study 1: Linguistic Analysis

The team analyzed transcripts from 577 body-worn camera (BWC) recordings of vehicle stops involving Black drivers in a racially diverse, medium-sized US city. The analysis focused on the "thin slice" of the interaction: the first 45 words spoken by the officer (approximately 27 seconds).

  • Manual Annotation: Research assistants coded "institutional dialog acts," such as greetings, providing a reason for the stop, issuing orders, and requesting documentation.
  • NLP Modeling: A DeBERTa V3 neural classifier was trained to predict stop outcomes based solely on these initial linguistic cues.

Study 2: Psychological Impact

To understand how these linguistic patterns affect civilians, 188 Black men listened to 100 audio clips of the officers' initial words. The driver's speech was masked to isolate the officer's impact. Participants then rated their emotional state, perceived the officer's demeanor, and predicted the stop's final outcome.

Key Insights: The Linguistic Signature of Escalation

The study reveals that escalation is often "front-loaded." Stops ending in arrest or search frequently begin with an escalated tone rather than reaching it through a slow decline in rapport.

Officers in escalated stops were found to be three times more likely to begin with a command and 2.5 times less likely to provide a reason for the stop. Non-escalated stops typically followed a "procedural justice" model: greeting, followed by a reason, then a request for documentation. In contrast, escalated stops often bypassed the greeting and reason, moving immediately to authoritative orders.

Results and Psychological Toll

The computational model (DeBERTa V3) achieved 70.83% accuracy in predicting escalated outcomes using only the first 45 words. Regression analysis confirmed that issuing an Order was a significant predictor of escalation (Odds Ratio = 1.29), while providing a Reason was a significant negative predictor (OR = 0.48).

Participants in Study 2 demonstrated a high sensitivity to these cues. Even without hearing the driver or seeing the scene, they accurately predicted negative outcomes based on the officer's tone and word choice.

Participants' confidence in predicting outcomes (handcuff, search, arrest) based on the first 45 words of a stop.

Furthermore, the linguistic style of escalated stops took a significant psychological toll. The presence of orders and the absence of reasons significantly increased participants' anxiety and worry about the potential use of force, regardless of the actual legal justification for the stop.

Percentage of trials where participants reported being worried about the use of force in escalated vs. non-escalated stops.

Case Study: The George Floyd Stop

The paper highlights the George Floyd stop as a tragic example of this pattern. In the first 27 seconds, the officer spoke 57 words—all of which were physical orders—despite Floyd offering apologies, explanations, and pleas. This exemplifies how an interaction can be fully escalated from its inception.

Conclusion

These findings suggest that the trajectory of a police stop is often established in its earliest moments. For Black men, the "linguistic signature" of an escalated stop—characterized by immediate commands and a lack of explanation—triggers immediate psychological distress and accurately signals a negative outcome.

While limited by its focus on a single city, this research highlights the potential for automated BWC analysis to serve as a tool for de-escalation training and monitoring police-community relations. By understanding these early interactional cues, departments can work toward more procedurally just encounters that prioritize communication and safety.

Summary updated

Abstract

Across the United States, police chiefs, city officials, and community leaders alike have highlighted the need to de-escalate police encounters with the public. This concern about escalation extends from encounters involving use of force to routine car stops, where Black drivers are disproportionately pulled over. Yet, despite the calls for action, we know little about the trajectory of police stops or how escalation unfolds. In study 1, we use methods from computational linguistics to analyze police body-worn camera footage from 577 stops of Black drivers. We find that stops with escalated outcomes (those ending in arrest, handcuffing, or a search) diverge from stops without these outcomes in their earliest moments-even in the first 45 words spoken by the officer. In stops that result in escalation, officers are more likely to issue commands as their opening words to the driver and less likely to tell drivers the reason why they are being stopped. In study 2, we expose Black males to audio clips of the same stops and find differences in how escalated stops are perceived: Participants report more negative emotion, appraise officers more negatively, worry about force being used, and predict worse outcomes after hearing only the officer's initial words in escalated versus non-escalated stops. Our findings show that car stops that end in escalated outcomes sometimes begin in an escalated fashion, with adverse effects for Black male drivers and, in turn, police-community relations.
BibTeX
@inproceedings{rho2023escalated,
  author = {Eugenia Ha Rim Rho and Maggie Harrington and Yuyang Zhong and Reid Pryzant and Nicholas P. Camp and Dan Jurafsky and Jennifer L. Eberhardt},
  title = {{Escalated police stops of Black men are linguistically and psychologically distinct in their earliest moments}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2023},
  doi = {10.1073/pnas.2216162120},
}
Figure