Debbie Amewu

In many jurisdictions, defendants charged with certain offenses are presumed ineligible for bail based on the charge alone, a categorical rule that treats the charge as a proxy for risk regardless of a defendant's individual circumstances. This paper studies Virginia's 2021 repeal of its Presumptive Denial of Bail statute, which previously made defendants charged with specified violent and serious offenses presumptively ineligible for bail on the basis of charge alone. Using administrative court records from the Virginia court system and a difference-in-differences design comparing defendants with affected and unaffected charges before and after the reform, I estimate the effects of removing charge-based bail presumptions. The repeal increased pretrial release among affected defendants, accompanied by a shift away from monetary bail toward other forms of release. Consistent with the literature on the costs of detention, this expansion in releases coincided with more lenient case dispositions—incarceration and guilty pleas both declined, though the decline in guilty pleas did not translate into a statistically significant drop in conviction rates. The reform did not increase failures to appear or short-run rearrest, but longer-run rearrest rose modestly. These results provide new evidence on bail reform for defendants charged with violent and serious offenses, a population largely absent from a literature focused primarily on lower-level offenses.

How disruptive are leadership changes in local law enforcement? While a large body of literature studies the performance effects of electoral turnover among local officials, little is known about county sheriffs. Sheriff elections generate abrupt leadership changes in agencies with substantial authority, raising questions about whether turnover disrupts operations or whether institutional continuity mitigates such effects. This paper investigates whether electoral turnover in law enforcement impacts arrest rates. Using hand-collected data on sheriff elections merged with arrest records, I employ a difference-in-differences design to examine how arrest rates respond, first, to an incumbent sheriff's electoral defeat and, second, to the subsequent transition to new leadership. I find that election losses lead to modest declines in arrests for lower-level offenses, while no significant effects are observed for arrest rates of severe offenses. In contrast, the transition to a new sheriff shows little impact on arrest activity. Analysis of deputy employment data suggests that staffing declines may drive the decrease in arrests following election losses.

Owing to the explosion in incarceration rates, coupled with high crime and overcrowded prisons in America, much research has been conducted to analyse the causes of such phenomena. This paper contributes to such research by conducting a policy analysis of the effectiveness of the Justice Reinvestment Initiative, a reformative legislation instituted to reform the justice system and tackle the issue. Since its inception, no empirical analysis has been conducted in the literature to ascertain its impact on crime rates. This paper employs the Difference-inDifferences estimator to review the impact the policy has on crime rates. The results show that the policy's implementation led to significant increases in property and total crime rates, contrary to popular belief in its successes.

Pretrial risk assessment tools play an increasingly important role in informing judicial decisions about bail and pretrial release. Among the most widely adopted instruments is the Public Safety Assessment (PSA), which predicts the likelihood that a defendant will fail to appear in court or be rearrested while awaiting trial. Despite its widespread use, relatively little evidence exists on how the PSA's predictive performance compares with modern machine learning methods when applied to large administrative datasets. This study evaluates whether machine learning algorithms can improve upon the predictive accuracy of the PSA for forecasting failure to appear among pretrial defendants. Using a large dataset of court cases from the Virginia Pretrial Data Project, I will develop and compare several predictive models, including logistic regression, decision trees, and random forests. The performance of these models will be assessed using standard classification metrics and compared directly with the PSA's failure-to-appear score to determine whether more flexible algorithms provide meaningful improvements in predictive performance. By examining the strengths and limitations of both traditional statistical methods and machine learning approaches, this study aims to contribute to the growing literature on evidence-based pretrial decision-making and the use of predictive analytics in the criminal justice system.