How Do Prosecutors’ Beliefs About Violent Re-Offense Shape Their Charging and Sentencing Recommendations?

This study asks whether prosecutors’ incorrect beliefs about who is likely to commit violent crimes affect their charging and sentencing recommendations. The authors link survey data from 171 North Carolina prosecutors to more than 600,000 felony cases from 1995–2019. They find that prosecutors systematically underestimate how much violent re-offending declines with age and overestimate how strongly criminal history predicts future violence—by about ninefold. These mistaken beliefs meaningfully shape decisions: prosecutors who think a group is riskier are about 1.3 percentage points more likely to recommend incarceration. More accurate prosecutors simultaneously reduce violent re-arrests by 6.4 percent and lower incarceration by 4.2 percent.

The Policy Scientist's Perspective

This article addresses a broadly significant policy question: whether inaccurate beliefs about violent re-offense shape prosecutorial decision-making, thereby influencing both incarceration patterns and public safety. Its importance extends beyond criminal justice, as it speaks to how expert judgment, misinformation, and institutional practice interact in high-stakes environments. The dataset used in the study is unusually rich, linking survey-elicited beliefs to more than two decades of administrative case records, enhancing credibility. The quasi-random allocation of cases provides a credible causal framework. This study builds on influential work on prediction error and decision-making but extends the literature by directly connecting belief distortions to real prosecutorial outcomes, making it a notable and timely contribution.

Full Citation and Link to Article

Harrington, E., Murdock, W. III, & Shaffer, H. (forthcoming). Prediction Errors, Incarceration, & Violent Crime: Evidence from Linking Prosecutor Surveys to Court Records. American Economic Journal: Economic Policy. Retrieved from https://www.aeaweb.org/articles?from=f&id=10.1257%2Fpol.20230812

Central Research Question

The article examines whether prosecutors’ inaccurate beliefs about the determinants of violent re-offending—particularly defendant age and criminal history—systematically influence their charging and sentencing recommendations, and whether these belief distortions ultimately affect both incarceration rates and future violent crime. The authors focus on two linked questions: first, whether prosecutors’ subjective predictions about group-level risk diverge from empirical risk patterns; and second, whether those divergences shape real case outcomes in a way that suggests meaningful policy consequences. In addition, the study evaluates whether prosecutors with more accurate beliefs are more effective at reducing violent re-arrest while simultaneously avoiding unnecessary incarceration. 

Previous Literature

This study contributes to several established research traditions on forecasting error, discretion, and criminal-justice decision-making. Work in psychology and behavioral economics, beginning with Kahneman and Tversky, highlights how individuals often rely on heuristics when confronting uncertainty. More recent economic studies show that decision-makers frequently develop systematic misperceptions about risk, and these misperceptions can distort choices in contexts such as lending, medical diagnosis, and judicial bail decisions. Central to this literature is the question of whether observed disparities in outcomes reflect informational errors, taste-based preferences, or structural constraints.

Within criminal justice, prior empirical work has pursued two main strategies. One line of research uses quasi-random judge assignment to isolate judicial preferences and estimate causal effects on incarceration, recidivism, and labor-market outcomes. A second line explores the predictive validity of criminal-risk assessments and sentencing guidelines. However, few studies have been able to directly measure the beliefs of decision-makers and compare those beliefs to actual risk patterns. A major challenge is that beliefs and preferences are usually not separately identifiable using observational data. The authors build on recent advances showing that belief elicitation, when linked to rich administrative data, can overcome this identification problem by permitting direct comparison between subjective expectations and objective empirical patterns. This article is one of the first to apply that insight to prosecutors—a group that exercises substantial influence over charging, plea bargaining, and ultimately incarceration. 

Data

The study links two unique datasets: (1) an original survey of prosecutors in North Carolina, and (2) twenty-five years of administrative felony court records. The survey was administered in cooperation with the North Carolina Conference of District Attorneys. Sixteen of the forty-three offices agreed to participate, yielding 186 surveyed prosecutors, 171 of whom completed the belief-elicitation questions. More than ninety percent could be matched to administrative data using a combination of fuzzy name-matching and office records.

The administrative dataset covers all felony cases handled in North Carolina Superior Courts between 1995 and 2019. These records include information on charges, criminal histories, demographic characteristics of defendants, sentencing outcomes, and subsequent re-arrests for violent crime within five years of case resolution. Prosecutors handled an average of 682 cases over the period, giving the linked dataset more than 600,000 analyzable cases. Violent re-arrest is defined consistently across both the survey instrument and the administrative records, covering offenses such as robbery, assault, homicide, serious sex offenses, kidnapping, burglary, and arson.

The authors emphasize two strengths of this combined dataset: the ability to link subjective beliefs to objective case outcomes and the presence of quasi-random case assignment within offices. The rotation systems used by many North Carolina offices appear to generate meaningful balance on observable characteristics. This balance supports the credibility of causal estimation strategies that rely on cross-prosecutor variation in beliefs. 

Methods

The authors use a multi-stage empirical strategy. First, they elicit prosecutors’ beliefs about violent re-arrest rates across five age groups and five criminal-record categories. Rather than rely on absolute probability judgments—which prosecutors struggled to provide in pilots—the survey asks prosecutors to make relative comparisons across groups by adjusting sliders that map perceived risk.

Second, they compare these subjective beliefs to empirical re-arrest patterns derived from the administrative data. Because the empirical risk profiles may be affected by selection—since only released defendants can re-offend—they use an “identification-at-infinity” approach borrowed from prior work to confirm that age-crime profiles remain stable even when selection is addressed.

Third, to estimate the influence of beliefs on decisions, the authors use within-prosecutor regressions. They examine whether a prosecutor who views a particular age or criminal-record group as riskier is more likely to recommend incarceration for defendants in that group relative to how other prosecutors treat similar defendants. These specifications include prosecutor fixed effects, group fixed effects interacted with office, and controls that mirror North Carolina’s sentencing-guideline inputs, such as offense severity and criminal-record points.

Finally, the authors investigate whether prosecutors with more accurate beliefs produce better aggregate outcomes. Accuracy is defined as the mean squared error between each prosecutor’s perceived and empirical risk profiles. They then estimate the relationship between belief accuracy and two key outcomes: violent re-arrest among defendants in prosecutors’ cases and overall incarceration rates. The identifying assumption is conditional quasi-random assignment of cases to prosecutors. They supplement this analysis by estimating the “mechanical” effect of incarceration on violent re-arrest using two independent designs—a split-sample instrumental-variables method based on across-prosecutor variation, and a sentencing-discontinuity design aligned with prior literature. 

Findings/Size Effects

The study reports three major findings.

First, prosecutors exhibit systematic prediction errors. They underestimate how sharply violent re-arrest declines with age. Empirical patterns show substantial “aging out” of violent crime, but prosecutors believe the decline is only about half as steep. They also overestimate the predictive power of criminal history—by a factor of roughly nine. Although they correctly perceive the direction of these relationships, they misjudge magnitudes.

Second, these misperceptions meaningfully influence case outcomes. Prosecutors who believe a group is riskier than the average prosecutor is tend to impose higher incarceration-recommending sentences on that group. A one-standard-deviation increase in perceived risk is associated with a roughly 1.8 percent rise (0.36 percentage points) in the probability of recommending incarceration for defendants in that group. Put differently, a prosecutor who perceives a group as twice as likely to re-offend is about 1.3 percentage points more likely to recommend incarceration for that group, controlling for all guideline-based factors and fixed characteristics.

Third, accuracy matters for public-safety outcomes. Prosecutors whose beliefs are one standard deviation more accurate reduce violent re-arrest in their caseloads by 0.57 percentage points, or 6.4 percent relative to the average. At the same time, they also reduce incarceration by 0.86 percentage points, or 4.2 percent. These dual gains suggest that accurate prosecutors more effectively identify high-risk defendants for incapacitation while avoiding unnecessary incarceration of lower-risk individuals. The authors estimate that the difference between highly accurate and highly inaccurate prosecutors corresponds to the impact of roughly twelve additional years of prosecutorial experience.

The authors also note that these effects appear similar for Black and non-Black defendants, suggesting that improved accuracy need not exacerbate racial disparities. 

Conclusion

The article demonstrates that prosecutors’ beliefs about risk diverge meaningfully from empirical patterns and that these incorrect beliefs distort real sentencing recommendations. Because prosecutors occupy a central position in criminal case processing and exercise wide discretion, their belief errors contribute to systematic variation in both incarceration rates and future violent crime. The authors show that increased accuracy generates simultaneous reductions in incarceration and violent re-arrest, suggesting that informational interventions—such as training, feedback mechanisms, or analytic tools—may hold promise for improving decision quality. More broadly, the paper highlights the role of subjective beliefs in institutional decision-making and extends the literature by showing how directly measured belief errors translate into measurable impacts on public-safety outcomes.