Judging prosecutors: Public support for prosecutorial discretion

Abstract

Prosecutors have immense discretion to determine which offenses to charge, which cases to take to trial, and which sentences to recommend. Yet, even though many of the prosecutors who exercise this discretion over important crimes must face the electorate to keep their jobs, we know little about how the use of this discretion affects prosecutors’ electoral fortunes. Drawing on two experiments embedded in a nationally representative survey, we demonstrate that the public is more supportive of prosecutors who issue lenient sentences for low-level crimes. The results have important implications for criminal justice reform inasmuch as they provide a linkage between progressive prosecutorial behavior and respondents’ vote intentions.

In recent years, voters in cities like Boston and Philadelphia elected district attorneys pledging to decline to prosecute a list of charges, refuse to seek cash bail, and divert individuals away from carceral sanctions. The electoral success of these candidates challenges an entrenched conventional wisdom that the American people are “tough on crime” and that candidates who boast their punitive bona fides are advantaged in the electoral arena (Krieger, 2011).

Drawing on two nationally representative survey experiments, we examine the electoral dynamics of lenient sentencing practices for prosecutors. Most local prosecutors in the United States have immense discretion about on-the-ground criminal justice policies and are also elected to office. Examining citizens’ responses to prosecutorial behavior provides a unique opportunity to test the extent to which progressive prosecutors receive some electoral benefit for leniency.

We reach two primary conclusions. First, contrary to popular wisdom—the American public is more (not less) supportive of prosecutors who issue lenient sentences for nonviolent offenses. Second, the size of this effect does not vary according to whether the prosecutor is a copartisan or outpartisan: voters are not willing to trade off their preferences regarding leniency to support an outpartisan prosecutor. The results have important implications for understanding criminal justice in modern America as a growing wave of reformers and “progressive prosecutors” promoting lenient policy are elected to office.

The dynamics of prosecutorial elections

Local prosecutors in 45 states are elected to office (Hessick and Morse, 2020). Yet, while the dynamics of state legislative races (e.g., Rogers, 2017), judicial races (e.g., Bonneau and Hall, 2009), and even sheriff’s contests (e.g., Thompson, 2020) have attracted the interest of scholars, prosecutorial elections have not received the level of scrutiny afforded to these other elections. This is surprising given the immense amount of discretion that prosecutors have about what charges to pursue, what crimes to prioritize, and what plea bargains to accept (Boyd et al., 2021). Despite previous observational studies on the structural qualities of prosecutorial elections—incumbency, competition, media coverage—few studies have addressed what motivates individual-level vote choice among the electorate (Hessick and Morse, 2020; Wright, 2009).

This omission has important consequences for our understanding of criminal justice, as local elections are often criticized for being low-salience contests that might inhibit democratic accountability through high rates of incumbent success and little meaningful electoral competition (Nelson, 2011). If, for example, partisan cues dominate the public’s calculus in these elections, then prosecutors in office may be empowered to implement their chosen priorities knowing that the cover provided by their partisan affiliation will help them achieve re-election. On the other hand, if voters respond to the severity of prosecutors’ sentencing decisions as they evaluate candidates, regardless of that official’s partisan identification, it would bode well for the level of democratic accountability in these elections. However, because individual-level studies of prosecutorial elections are scarce, evidence on this point is scant.

Looking at accountability from the other side of the coin, evidence suggests prosecutors behave as if they believe voters consider their behavior in office on election day: DAs in the run-up to elections (Okafor, 2021) and Republican DAs after election (Krumholz, 2018) incarcerate more people for longer, per capita. Nelson (2014) demonstrates that Colorado prosecutors adjusted their sentencing practices for low-level marijuana offenses after a failed statewide referendum provided them with constituency-level, issue-specific information about public preferences. However, these signals impart a weaker effect on prosecutors’ policy stances relative to other officials: Sances (2021) finds that California prosecutors are less likely than federal and state officeholders to match their constituents’ criminal justice beliefs as expressed through ballot propositions.

So, prosecutors appear to sense a threat of accountability from the public yet gain and retain office via often minimally competitive elections. We don’t know to what extent policy affects vote choice in prosecutorial elections nor how partisanship functions as a signal to voters in these contests. Testing the extent to which voters evaluate prosecutors on substantive measures like policy versus heuristics like partisanship offers insight into whether democratic accountability is at the heart of prosecutorial elections.

Electoral choice and prosecutorial discretion

If the public holds prosecutors accountable at the ballot box for their policies, then public opinion on criminal legal issues should relate to voter behavior in these elections. Elites’ pursuit of various punitive policies and rhetoric has long been associated with increases in public punitiveness: higher incarceration rates (Enns, 2014), greater emphasis on crime policy in campaigns (Holian, 2004), and the rapid diffusion of state punitive policies (Baumgartner et al., 2021). Yet public punitiveness has waned since its peak in the late 1990s (e.g., Enns, 2014), and prominent examples abound of the electoral success of progressive prosecutors.

Despite decreasing punitive public mood, we hypothesize that the public will still express greater support for punitive policymakers, relative to more lenient ones. Even today, research suggests that officials still believe punitiveness returns an electoral benefit. In highly contested electoral environments, Gunderson (2022) finds that Democratic governors outincarcerate Republican peers, and, at the local level, Taylor (2021) shows that elected state trial court judges outincarcerate trial court judges in states where they are not subject to democratic accountability. While the true size of this accountability effect may have decreased alongside punitive mood since the 1990s, we expect it is still directionally accurate. Thus:

H₁: The public is more supportive of punitive, rather than lenient, prosecutors.

Unlike governors and legislators, however, the public has little knowledge about prosecutors’ powers and behavior. In practice, data about prosecutors’ work and its effects—recidivism rates, misconduct, and charging decisions—are rarely readily available or apparent to the average voter. Rather, select cases are communicated through media and other courtroom observers (leading to “fire-alarm oversight” in which a handful of charging decisions attract the bulk of the public’s attention) (Pickett, 2019).

Given lower levels of public knowledge about prosecutors, voters are likely to depend on other heuristics to adjudicate prosecutorial candidates. Partisanship is the most fundamental heuristic Americans use to navigate politics (Bartels, 2000). The partisan label next to a prosecutor’s name on the ballot box gives voters important information about which candidate best aligns with their beliefs on crime policy, even when the voter lacks full information about the candidate or the office. Indeed, partisanship is particularly salient for voters in low-information races (Peterson, 2017).

Within a limited information environment, a single mismatch between the prosecutor’s sought sentence in a certain case and a respondent’s preference is unlikely to shake the strength of copartisan preference alone, so voters should be more likely to express electoral support for a copartisan, rather than an outpartisan, prosecutor who recommends a lenient sentence. Thus:

H₂: The public is more supportive of copartisan, rather than outpartisan, prosecutors.

H₃: The effect of a lenient sentence is mitigated for copartisan prosecutors.

Research design

We test our hypotheses via two experiments fielded on a module of the 2020 CES (Schaffner et al., 2021). We emphasize two features of our research design. First, the use of experiments has strengths and weaknesses. States’ sentencing laws vary, and an experiment allows for comparability unlikely to come at the expense of our estimands of interest (Brutger et al., 2022). However, although each respondent makes their decisions within identical information environments, our treatment effects may be overestimates, as not all respondents would give identical levels of attention to prosecutorial races and decision-making in the real world (Barabas and Jerit, 2010).

Second, we examine voters’ responses to common nonviolent offenses. Although our choices are more representative of a prosecutor’s caseload, violent and circumstantially salient offenses may also inspire public responses that impart different effects on electoral choice in prosecutorial elections, especially since the public is only likely to know about a few decisions made by a prosecutor over their term (Pickett, 2019). Examining how these effects generalize to violent crimes or complex information environments and how the public aggregates information across prosecutorial decisions are all worthy topics for future research. Our goal is to provide the first experimental tests of the effects of prosecutorial discretion in sentencing on individual-level vote choice.

Study 1: The direct effect of leniency

We use Study 1, fielded in the pre-election wave of the 2020 CES, to examine whether tough-on-crime prosecutors are rewarded by voters. The strength (and weakness) of this study is its simplicity, abstracting away partisanship from the decision-making environment to provide us with a clean test of our first hypothesis. ¹

Experimental design

We presented 1000 respondents with a vignette describing a fictitious cannabis possession case in a nearby state in which an individual pleads guilty to the charge. We select cannabis possession as the offense since it constitutes a meaningful share of the cases that come before a prosecutor’s desk: about 5% of all arrests (500,395) in 2019 were for cannabis possession, more than any other specific offense in data collected by the FBI (except driving while intoxicated, which we use in Study 2, and uncategorized assaults) (Federal Bureau of Investigation, 2019a, 2019b). All respondents read that, under state law, the 35-year-old man can be sentenced for up to 6 months in jail. ² Respondents are either treated with a prosecutor that recommends a 6 month sentence (harsh) or a 6 day sentence (lenient). ³ The judge accepts the recommended sentence.

As an example, here is a treatment from the pre-election experiment with a lenient prosecutor:

Suppose that the police in a nearby state arrested Matthew Anderson, a 35-year-old man, for the possession of marijuana. Anderson pled guilty to the crime. State law allowed Anderson to be sentenced to up to 6 months in jail. At the sentencing hearing, the local prosecutor recommended that Matthew Anderson serve 6 days in jail. The judge followed that recommendation. Anderson was sentenced to serve 6 days in jail.

After presenting respondents with the vignette, we measure the respondent’s intent to re-elect the prosecutor in the upcoming election on a four-point scale. We also measure two alternative outcome variables—respondent’s approval of the prosecutor’s decision and perception of fairness—and combine all three outcome variables into a single scale for general incumbent support. Treatment effects for these three alternative measures are described in Supplemental Appendix C.

Results

Our first hypothesis suggested that respondents would judge a prosecutor more favorably if he recommended a harsher sentence. Figure 1 shows the average re-election support across treatments. This was not the case. Respondents who saw a lenient 6-day sentence reported significantly higher re-election support (p < 0.01) for the prosecutor. The difference in re-election support between the treatments is 0.56, or slightly more than one-half of a standard deviation in the outcome. This result alone suggests that the conventional wisdom of a public which desires harsh sentencing and holds officials accountable for being insufficiently punitive, all else equal, deserves additional scrutiny.

Figure 1.

Direct treatment effect, study 1. The plot shows the average value of re-election support (with 95% confidence intervals), by treatment. Higher values indicate stronger intent to re-elect the prosecutor in an upcoming election.

Study 2: Incorporating copartisanship

Although Study 1 points toward a public which desires lenient sentencing, we theorize that a more accurate representation of the information environment in which prosecutors are judged by the public must also include partisanship, given its utility as a heuristic for evaluating officials (particularly for low-information offices). In Study 2, we describe a case in which an individual is sentenced for driving while intoxicated (DWI) to test if the results from Study 1 hold in another context while identifying and varying the prosecutor’s partisanship.

Experimental design

In this experiment, we present the 675 respondents who identify as partisans and returned in the post-election wave with a vignette describing a DWI case in a nearby state in which an individual pleads guilty to the charge. We select DWI because, like cannabis possession, the charge constitutes a large share of prosecutions in many jurisdictions; the FBI estimates that over 10% of the country’s arrests in 2019 were for driving under the influence (Federal Bureau of Investigation, 2019b). We inform respondents the 35-year-old man can be sentenced for any time between 24 h and a year in jail under state law. ⁴ Respondents are either treated with a prosecutor that recommends a 1-year sentence (harsh) or a 1 day sentence (lenient). Further, we vary prosecutor partisanship, identifying the prosecutor as either a Republican or Democrat. Respondents who identify as partisan leaners are included in our analysis, while respondents who identify as true independents are excluded, such that all respondents see an outpartisan or copartisan prosecutor.

As an example, here is a vignette treatment with a lenient, Democratic prosecutor:

Suppose that the police in a nearby state arrested Colin Smith, a 35-year-old man, for driving while under the influence of alcohol. Smith pled guilty to the crime. Because it was Smith’s first offense, Smith could be sentenced to a period of time between 1 day to 1 year in jail. At the sentencing hearing, the local prosecutor, a Democrat, recommended that Smith serve 1 day in jail. The judge followed that recommendation. Smith was sentenced to serve 1 day in jail.

As in Study 1, the judge accepts the recommended sentence, and we present respondents with identical outcome questions (approval, fairness, and electoral support) as in Study 1. We present treatment effects on electoral support in Figure 2 and the rest in Supplemental Appendix C.

Figure 2.

Direct treatment effect, study 2. The plot shows the average value of re-election support (with 95% confidence intervals), by treatment. Higher values indicate stronger intent to re-elect the prosecutor in an upcoming election.

Results

Once again, the results explicitly contradict our first hypothesis. Averaging across the prosecutor’s party, respondents reported significantly higher re-election support (p < 0.01) for the lenient prosecutor with an average treatment effect of 0.27 (three-tenths of a standard deviation).

Second, we hypothesized that respondents would be more favorable toward a copartisan prosecutor than an outpartisan. This is the case. We find significant differences in re-election support (p < 0.01) between respondents who saw a copartisan prosecutor and respondents who saw an outpartisan prosecutor, averaging across the sentence length, with an average treatment effect of one-half a standard deviation. When accounting for all four possible treatment combinations, shown in Figure 2, we find significant differences in re-election support between respondents who saw a lenient sentence from a copartisan prosecutor and a harsh sentence from a copartisan prosecutor. However, we fail to see significant differences between the two leniency treatments from outpartisan prosecutors.

These results also bear out in regression analyses. In Table 1, we show four linear regression models with general incumbent support as the outcome variable. The first two bivariate models support the conclusions for Hypotheses 1 and 2. In Supplemental Appendix C, we present models that demonstrate that these conclusions hold over re-election support and our alternative outcome variables when we control for respondents’ political and demographic characteristics.

Table 1.

Regression models, study 2.

	Re-election support
	(1)	(2)	(3)
Lenient sentence (1 Day)	0.35*** (0.07)		0.24* (0.10)
Copartisan prosecutor		0.39*** (0.07)	0.29** (0.09)
Copartisan prosecutor × lenient sentence			0.17 (0.14)
Constant	2.29*** (0.05)	2.26*** (0.05)	2.15*** (0.06)
N	675	675	675
Log likelihood	−886.82	−883.36	−871.23
AIC	1777.64	1770.73	1750.46

*p < 0.05; **p < 0.01; ***p < 0.001.

Table 1: Linear Regression Results, Study 2. The table presents the results of a linear regression with re-election support as the outcome variable. Higher values of the outcome indicate stronger intent to re-elect the prosecutor in an upcoming election. * indicates p < 0.05.

To test our third hypothesis, Model 3 includes an interaction between the two treatments to determine whether or not the effect of a lenient sentence varies according to the respondent’s copartisanship. Contrary to our third hypothesis, we find no evidence to support the idea that respondents treat copartisans and outpartisans differently according to their sentencing decisions.

Discussion

From our two studies, we make three major conclusions. First, contrary to popular wisdom—the American public is more (not less) supportive of prosecutors who issue lenient sentences for nonviolent crimes. Across all our outcome variables—including vote intention—and two distinct nonviolent and low-level charges, we find that respondents are more supportive of prosecutors who issued a sentence at the low end of the statutory range than one who sought a sentence at the high end of that range. Second (and unsurprisingly), people are more likely to support and to indicate a vote intention for a copartisan (rather than an outpartisan) prosecutor. Third, however, we found no evidence that copartisan prosecutors who supported lenient sentences were evaluated more favorably than outpartisan prosecutors who asked for the same sentence.

This is a finding that deserves additional scrutiny. That partisanship dominates decision-making in this experimental setting—where information about prosecutorial behavior was easily available to all respondents and there was no countervailing information (as would happen in a campaign) to contextualize the decision—raises serious concerns for the efficacy of these sorts of elections. Future work should probe the extent to which other considerations might mitigate (or exaggerate) the effects of partisanship on these sorts of elections to illuminate voters’ calculi in these elections. Were it the case that partisanship swamped the effects of all performance-related considerations, then reformers may need to weigh the costs and benefits of nonpartisan elections to these sorts of offices.

Our study has its limits. First, 46% of pre-election and 55% of post-election respondents express some degree of intent to re-elect the prosecutor in our experiments, reflecting a weaker preference for incumbents than observational studies on prosecutor elections have previously identified (Hessick and Morse, 2020; Wright, 2009). Thus, although we identify substantive effects of lenient sentencing on re-election intent, they may be attenuated by stronger incumbent advantages in the low-information environments characteristic of many prosecutor elections (unlike the uniform information environment in our experiment). Second, our experiments use two different, specific charges. Yet prosecutors’ caseloads are substantially more diverse than simply DWIs and cannabis possession, and other types of cases (especially violent crimes) are salient to the public. Salience and resulting media coverage structure the information upon which criminal legal officials are held accountable, as the public zeroes in on a few cases or decisions, particularly instances of underpunishment, when evaluating an official electorally (Gordon and Huber, 2002; Pickett, 2019). In other words, the structure of the experimental information environment is distinct from the story voters face while in the voting booth, complicated by high caseloads, media coverage, and variation in offense salience.

Further analyses of prosecutorial behavior and public opinion should test our conclusions with different offenses and greater variation in salience and sentence lengths to develop a fuller picture of the limits of progressive prosecution. Alternative research designs that manipulate prosecutors’ complex profiles (e.g., conviction rates, declination policies, prosecutorial misconduct history) can better identify how voters evaluate district attorneys on a broader set of merits and discretionary policies than a sentencing recommendation in a singular case, as we test here. Further, we would not be surprised to learn that the effects here are unique to nonviolent offenses: citizens may be willing to support prosecutors who offer lenient sentences to individuals who do not pose a violent threat but recoil from prosecutors who act with leniency toward violent offenders. This is a vital topic for follow-up research.

On the whole, our results emphasize just how much fertile ground there is for scholars who seek to understand the dynamics of criminal justice reform in the United States. Prosecutors have an extraordinary amount of political power but have received far too little scrutiny from political scientists. This should change.

Supplemental Material

Supplemental Material - Judging prosecutors: Public support for prosecutorial discretion

Supplemental Material for Judging prosecutors: Public support for prosecutorial discretion by Michael J. Nelson and Taran Samarth in Research & Politics