Abstract:Large-scale policing data is vital for detecting inequity in police behavior and policing algorithms. However, one important type of policing data remains largely unavailable within the United States: aggregated police deployment data capturing which neighborhoods have the heaviest police presences. Here we show that disparities in police deployment levels can be quantified by detecting police vehicles in dashcam images of public street scenes. Using a dataset of 24,803,854 dashcam images from rideshare drivers in New York City, we find that police vehicles can be detected with high accuracy (average precision 0.82, AUC 0.99) and identify 233,596 images which contain police vehicles. There is substantial inequality across neighborhoods in police vehicle deployment levels. The neighborhood with the highest deployment levels has almost 20 times higher levels than the neighborhood with the lowest. Two strikingly different types of areas experience high police vehicle deployments - 1) dense, higher-income, commercial areas and 2) lower-income neighborhoods with higher proportions of Black and Hispanic residents. We discuss the implications of these disparities for policing equity and for algorithms trained on policing data.

What problem does this paper attempt to address?

The paper attempts to address the issue of the lack of police deployment data in the United States, particularly regarding the distribution of police forces in different communities. This data is crucial for detecting inequalities in police behavior and training policing algorithms. However, in the U.S., this data is typically not publicly available. Therefore, the authors propose a new method to quantify the presence of police cars in different communities by analyzing 24,803,854 dashcam images from New York City as a means to assess disparities in police deployment. Specifically, the goals of the paper include: 1. **Quantifying disparities in police deployment**: By detecting police cars in dashcam images, quantify the differences in police deployment levels between different communities. 2. **Revealing potential social inequalities**: The study finds that there is nearly a 20-fold difference between communities with the highest and lowest levels of police deployment. Communities with high police deployment fall into two categories: one with high population density, higher income, and frequent commercial activity; and another with lower income and higher proportions of Black and Latino residents. 3. **Discussing the impact on policing fairness and algorithmic fairness**: Explore the potential impacts of these disparities in police deployment on policing fairness and the fairness of algorithms trained on policing data (such as predictive policing algorithms). 4. **Proposing recommendations to improve data transparency**: Given the potential biases in current data, it is recommended that police departments release more reliable and direct aggregated police deployment data to enhance transparency and fairness. Through this method, the paper aims to fill the gap in police deployment data and provide a scientific basis for promoting fairness in policing and algorithmic fairness.

Detecting disparities in police deployments using dashcam data

Smartphone Data Reveal Neighborhood-Level Racial Disparities in Police Presence

A large-scale analysis of racial disparities in police stops across the United States

Do digital technologies reduce racially biased reporting? Evidence from NYPD administrative data

The racial composition of road users, traffic citations, and police stops

A Multi-Method Data Science Pipeline for Analyzing Police Service

Language from police body camera footage shows racial disparities in officer respect

A debiasing technique for place-based algorithmic patrol management

Surveilling Surveillance: Estimating the Prevalence of Surveillance Cameras with Street View Data

Social Media Representations of Law Enforcement within Four Diverse Chicago Neighborhoods

Causal inference and racial bias in policing: New estimands and the importance of mobility data

ESCORT: Fine-Grained Urban Crime Risk Inference Leveraging Heterogeneous Open Data

Filaments of crime: Informing policing via thresholded ridge estimation

Fine-Grained Car Detection for Visual Census Estimation

Towards Anomaly Detection in Dashcam Videos

Spatio-Temporal Analysis of Urban Crime Leveraging Multisource Crowdsensed Data

Dynamic road crime risk prediction with urban open data

On Racial Disparities in Recent Fatal Police Shootings

Evidence of What, for Whom? The Socially Contested Role of Algorithmic Bias in a Predictive Policing Tool

An algorithmic strategy for measuring police presence with GPS data

Leveraging body-worn camera footage to assess the effects of training on officer communication during traffic stops