Concetti Chiave
Developing a data-driven method to evaluate racial bias in law enforcement systems using a multi-stage causal framework.
Sintesi
The content introduces a causal framework to assess racial bias in law enforcement systems. It addresses limitations in prior works by incorporating criminality and multiple stages of interactions. Three scenarios are identified to determine the primary source of bias based on race and criminality. Empirical studies using police-civilian interaction data from NYC and New Orleans reveal instances of counter-intuitive phenomena related to racial bias.
-
Introduction
- Renewed interest in societal decision-making based on race.
- Importance of evaluating racial disparities in law enforcement for reforms.
-
Causal Framework
- Multi-stage process involving race, reporting, stops, and law enforcement actions.
- Notion of racial parity among criminals and innocents.
-
Scenarios
- Scenario 1: Airport security checks show bias against minority innocents.
- Scenario 2: AI-empowered policing reveals bias against minority criminals.
- Scenario 3: Police-civilian interactions highlight various sources of observational disparity.
-
Empirics
- Utilization of publicly available datasets for NYC and New Orleans.
- Test statistics to verify racial disparity and identify primary sources based on scenarios.
-
Findings
- Distribution of precinct-level disparities indicating bias against minorities.
- Predominant observation bias against minorities compared to majorities in NYC and vice versa in New Orleans.
-
Conclusion
- Primary source of bias identified as policing actions against minorities in NYC and public reporting biases against majorities in New Orleans.
Statistiche
"In NYC, there are observational biases against minority races (Black/Hispanic) compared to the majority (White)."
"In New Orleans we find the exact opposite - an observational bias against the majority race."
"Through careful processing and stitching such data, we construct the multi-stage dataset of various incidences."
Citazioni
"We do not intend to advocate for the implementation of AI technologies in law enforcement."
"Understanding whether there is evidence of bias either in law enforcement or reporting or both given the observed data."