Impact Analysis of the Chesa Boudin Administration

TL;DR

This study empirically evaluates the impact of Chesa Boudin's tenure as San Francisco district attorney, revealing significant reductions in prosecutions and convictions, and increased case diversions, using regression discontinuity design.

Contribution

It provides the first empirical analysis of Boudin's policies, employing regression discontinuity and machine learning to assess their effects on prosecution and crime levels.

Findings

36% reduction in prosecutions for all crimes

58% increase in successful case diversions

Potential for machine learning to improve estimation precision

Abstract

Claims of soft-handed prosecutorial policies and increases in crime were precipitating factors in the removal of Chesa Boudin as district attorney of the city and county of San Francisco. However, little research has been conducted to empirically investigate the veracity of these indictments on the former district attorney. Using regression discontinuity design (RDD), I find that the Boudin administration led to a 36\% and 21\% reduction in monthly prosecutions and convictions respectively for all crimes. Moreover, his tenure increased monthly successful case diversions by 58\%. When only looking at violent crimes during this period, the SFDA's office saw a 36\% decrease, 7\% decrease, and 47\% increase in monthly prosecutions, convictions, and successful case diversions respectively. Although, the decrease in monthly convictions was not statistically significant for the violent crimes subset. Additionally, I did identify a potentially causal relationship between lower numbers of prosecutions and higher levels of criminal activity, however, such findings did not meet the standard for statistical significance. Finally, I conclude that using machine learning algorithms, such as neural networks and K-nearest neighbors, in place of ordinary least squares regression for the estimation of the reduced form equation possibly may decrease the size of the standard errors of the parameters in the structural equation. However, future research needs to be conducted in this space to corroborate these initially promising findings.