The effect of the election of prosecutors on criminal trials

Public Choice (2014) 161:141 156 DOI 10.1007/s11127-013-0144-0 The effect of the election of prosecutors on criminal trials Siddhartha Bandyopadhyay Bryan C. McCannon Received: 14 October 2012 / Accepted: 19 November 2013 / Published online: 29 November 2013 Springer Science+Business Media New York 2013 Abstract We examine whether elections of public prosecutors influence the mix of cases taken to trial versus plea bargained. A theoretical model is constructed wherein voters use outcomes of the criminal justice system as a signal of prosecutors quality, leading to a distortion in the mix of cases taken to trial. Using data from North Carolina we test whether reelection pressures lead to (a) an increase in the number and proportion of convictions from jury trials and (b) a decrease in the average sanction obtained in both jury trials and pleas. Our empirical findings are consistent with our theoretical predictions. Keywords Crime Election Plea bargaining Prosecutor Trials 1 Introduction Local prosecutors in the United States exercise enormous discretion in how they handle criminal cases. There are 2344 local US prosecutor offices, which collectively handle 2.3 million felony cases each year (Perry 2006). This number represents approximately 95 % of all criminal prosecutions (Simmons 2004). A common feature of almost all of these local offices is that the chief prosecutor is elected by popular vote. 1 Given prosecutors importance in the functioning of the criminal justice system, one would like to understand the impact elections have on how they exercise their discretion. However, with a few exceptions this has not been formally analyzed. We fill the void by asking: do elections of prosecutors influence the way they handle cases? In particular, do such elections affect which cases they take to trial and which cases are plea bargained? We show how jury trials are used by 1 The three states that do not elect public prosecutors are Alaska, Connecticut, and New Jersey (Perry 2006). S. Bandyopadhyay Department of Economics, University of Birmingham, Birmingham B152TT, UK B.C. McCannon (B) Saint Bonaventure University, Saint Bonaventure, NY 14778, USA e-mail: bmccannon@sbu.edu

142 Public Choice (2014) 161:141 156 prosecutors to signal their quality to voters. We analyze a model in which there is asymmetric information about prosecutor quality. Prosecutors signal by distorting the mix of cases they take to trial versus those plea bargained. We use a detailed dataset from North Carolina to assess whether empirical evidence exists of the distortions predicted by our theoretical model. The model we use is an adaptation of Bandyopadhyay and McCannon (2014). That theoretical model assumes that higher quality prosecutors are more likely than their colleagues to be successful in the courtroom and can thus secure stiffer penalties. As quality is not directly observable, voters can use information on sentences obtained. Incumbent prosecutors respond to the information asymmetry by taking more cases to trial. In pooling equilibria, lower quality prosecutors aggressively take cases to trial to mimic the high-quality type, while in separating equilibria higher quality prosecutors increase the number to differentiate themselves. In both situations the model predicts that when reelection pressures are strong: (i) the number of cases obtained from jury trials rises, (ii) the number of convictions from trial increases and the number of guilty pleas decreases, so that the proportion of convictions that are obtained from jury trials increases, and (iii) the average sentence obtained in both trials and plea bargaining goes down, as marginal cases with weaker evidence are prosecuted. We test these hypotheses using a district-level, panel dataset from North Carolina. The data cover all 43 prosecutorial districts between 1997 and 2009. Data on electoral contests over this period are also collected. Our theoretical results receive empirical support. We find that the number of convictions obtained in jury trials (relative to the total number of convictions) increases in the year before an incumbent runs for reelection. In the year of reelection, if the incumbent faces a challenger either in the primary or the general election, we observe a further increase in the relative number of convictions obtained from jury trials. We estimate that the reelection incentives result in a 9.7 % increase in the proportion of cases taken to trial, while the presence of a challenger results in an additional 14.7 % increase. If a district does not have a contested election in any year within the sample, fewer convictions arise from trials. Another finding, as predicted by the theory, is that the yearly average maximum sanction obtained falls. Early analysis of prosecutors focused on the allocation of the office s budget and the role of plea bargaining (Landes 1971). Boylan (2005) analyzes data on chief federal prosecutors and their subsequent careers. He shows that prosecutors maximize sentence length, which supports our choice of evaluation metric used in the theory presented here. Boylan and Long (2005) show that prosecutors use trial experience as a career advancement strategy. Glaesar et al. (2000) find evidence that career concerns affect the decisions of US Attorneys and have led to an expansion in the federalization of drug crimes. The focus in this paper is on state-level prosecutors who face a different retention process. A work closely related to ours is that of Rasmusen et al. (2009), who investigate the decision about whether to prosecute a case or to dismiss it. They consider an environment wherein a prosecutor is interested in allocating the office s budget and effort over convictions, improving conviction rates, and fulfilling personal goals. The authors use a cross-sectional dataset to assess the impact of an increase in the office s budget on case outcomes, finding that larger budgets raise both conviction rates and the number of convictions. They do not consider the effect of the signaling problem arising from reelection pressure and do not analyze the decision about whether to engage in plea bargaining or to proceed to trial, as we do here. Our findings complement theirs since they show that elected prosecutors have relatively high conviction rates, while we show that the greater success of elected prosecutors coincides with an expanded use of the courtroom. Another paper closely connected to

Public Choice (2014) 161:141 156 143 our theoretical model is that of Gordon and Huber (2002), who consider the moral hazard problem arising when voters use observable signals of performance to monitor prosecutorial effort. They do not consider adverse selection issues. A qualitative analysis of the related matter of the retention of local prosecutors is Wright (2009), who presents stylized facts concerning media coverage of prosecutor elections. The model of this paper has similarities to incumbent-challenger models (Besley 2006), which consider the agency issues involved in elections. Such models have been used to analyze a host of situations, including political business cycles (Rogoff and Siebert 1988; Rogoff 1990), the continuation of inefficient policy (Majumdar and Mukand 2004), and the persistence of conflict (Bandyopadhyay and Oak 2010). Our model predicts an excessive use of trials, in line with most results in the literature. The growing literature on how career concerns affect the behavior of public officials borrows the central insight that politicians manipulate outcomes before elections. Leaver (2009) and Shotts and Wiseman (2010) both consider environments of asymmetric information on the quality of regulators. Hanssen (1999, 2000), Shepherd (2009), Berdejo and Yuchtman (2010), and Lim (2013) all provide empirical verification of the hypothesis that the independence and retention concerns of judges affect their decisionmaking. While not about prosecutorial behavior, these papers demonstrate that retention motives affect outcomes of the justice system. 2 Model We now develop a model of prosecutorial decisionmaking that will allow us to investigate the incentives created by elections. Consider a two-period model. In the first period there is a single prosecutor of unknown quality who is to decide how to handle cases brought before her. She takes one of two quality types q {H,L}, referred to as high and low respectively. She is high quality with probability γ (0, 1). Let the parameter θ beasummary measure of the strength of evidence she has against the defendant in any given case. The parameter may also capture information the prosecutor has concerning the skill of the defendant s attorney. Assume that θ [0,θ M ] where θ M <, with this being known to the defendant as well. Observing θ for a particular case, the prosecutor may either take the case to trial or engage in plea bargaining. 2 Assume that a large number of cases come up in the first period, which we can think of as a term in office. Denote s as the sanction received if the prosecutor is successful in the courtroom. The sanction is known and is exogenously set by, for example, sentencing guidelines. 3 Alternatively, if there are judicial discretion, uncertain parole outcomes, or appeals, s is best thought of as the expected sanction. We also take account of the probability that the prosecutor is successful at trial, a probability which depends on the quality of the prosecutor and the evidence. A high-quality prosecutor wins at trial with probability p H (θ), while if she is low quality she wins with probability p L (θ). Assume that 1 >p H (θ) > p L (θ) > 0 θ, p q (0) = 0, and dpq 0 q. Finally, if she takes the case to trial and obtains a conviction, a dθ cost c>0 is borne by the prosecutor and the defendant. 4 For simplicity we assume that c does not depend on the strength of the evidence or the prosecutor s ability. 2 We assume that the choices of whether to file charges and which charges to file have already been made. 3 The analysis is unaffected if one assumes that s = s(θ),where dθ ds 0. 4 Assuming different costs of trials for the two parties does not make any substantive change in the analysis.

144 Public Choice (2014) 161:141 156 With regard to plea bargaining, let g(θ) denote the agreed plea bargain sentence. We ignore the possibility that the plea outcome depends on the quality of the prosecutor. As the prosecutor s type is assumed to be the private information to the defendant, we assume the outcome under plea bargaining cannot be conditioned on type. One could alternatively argue, however, that even though the skill of the incumbent is not known to the voting public, defense attorneys may, in view of their repeated interaction, gain this information. In that case, g would be a function of the prosecutor s type as well as of θ. Extending the environment to allow for type-dependent plea bargains simply adjusts the bounds of the set of separating and pooling equilibrium. Being cumbersome, the analysis is not presented here, but is available from the authors upon request. How might the function g be determined? One way is to assume that a prosecutor makes the best offer that the defendant will accept. Since trials are lengthy and the final sanction of a successful trial comes with a delay, compared with the outcome of plea bargain, which is settled more quickly, the defendant will discount the future disutility of the cost of trial. Thus, he or she will agree to a sanction no greater than δps +c where p = γp H +(1 γ)p L. Alternatively, with optimism or self-serving bias (as is often assumed in the plea bargaining literature; see Burke 2007 and Farmer and Pecorino 2002 for a discussion), a defendant may believe that he will receive a sanction kps where k<1, and so even with a take it or leave it offer will not accept a sanction greater than kps + c. Consequently dpq dθ s>dg dθ 0and g(θ) > 0 θ. Such results are consistent with the usual assumption that in a plea bargain the sentence received is less than after a trial conviction. Finally, assume that if the prosecutor is indifferent between the two options then she chooses to plea bargain. Let w(g(θ)) be the welfare generated from a case that results in g(θ),andw(p q (θ)s) c be the welfare generated from a case that goes to trial. Assume that w is a strictly increasing function. Let W(q)be the welfare generated over the prosecutor s entire first term if she is of quality q. To link the two concepts, assume that θ for each case is a random variable. Across cases, values of θ are independent and identically distributed. The likelihood of a particular value is determined by the distribution function F :[0,θ M ] [0, 1]. Assume that a large number of cases arise during the prosecutor s term and that the number of cases disposed of does not depend on the manner in which they are handled. The expected welfare from a case then equals the average welfare generated from each case over the course of the term. Given that this is so, assume that first-period welfare equals the expected welfare from a randomly selected case. Thus, if a prosecutor chooses to take every case to trial where θ θ and plea out those with θ< θ, then first-period welfare is θ W(q)= w ( g(θ) ) θm [ ( df(θ)+ w pq (θ)s ) c ] df (θ). (1) θ=0 θ=θ The assumption of welfare being monotonic in sanctions is based on the presumption that the sanction has been optimally chosen by society. This does not necessarily mean that society benefits from having every defendant punished as harshly as possible, but rather that given the option to impose the harsher sanction deemed appropriate by the judge (or chosen by the sentencing board or legislature) or to accept a plea offer, welfare is greater if the prosecutor achieves the stiffer of the two penalties. One may also think of welfare as the expected welfare allowing for the possibility of wrongful convictions. 5 The cost of trial appears as a negative term because with greater expenditures on a trial, the resources available for disposing of other cases shrink. Finally, let V(q ) be second-period welfare if the prosecutor is of quality q,where dv dq > 0. Consequently, total welfare is W(q)+ V(q ). 5 See McCannon (2013) for an analysis of the effect of prosecutor elections on mistakes and appeal[s].

Public Choice (2014) 161:141 156 145 2.1 First best We now describe the first-best outcome in this environment as it is a useful benchmark. Taking a case to trial is best if w(p q (θ)s) c>w(g(θ)). Trial costs are assumed not to be too great: c<w(p L (θ M )s) w(g(θ M )). Otherwise, even with very good evidence the expected welfare of going to trial is insufficient to motivate that action. Given the way we have modeled plea bargaining, dpq dθ s>dg dθ and p q(0)s = 0 <c, it follows that there exists a threshold value of θ, denoted θ q, where welfare is equal between the two options. 6 The threshold value depends on the prosecutor s quality, since the probability of conviction depends on her abilities. Consequently, θ L > θ H > 0. If θ> θ q,thenw(q) is improved if the case proceeds to trial. The evidence is so substantial that the expected sanction is severe enough to make the trial preferable. If θ< θ q, then the best outcome is for the cases to be resolved by plea bargaining. The expected sanction is small and, given the high cost of trial, negotiating and accepting a plea generates a better outcome. Note also that the first-best outcome requires that a prosecutor of high quality be retained since V(H)>V(L). Also, if the prosecutor is low quality, then a new one should be selected, since EV > V (L). 2.2 The model with asymmetric information Consider the principal-agent problem that arises when the principal (in this case the voting public or, with ideological heterogeneity, the median voter) wants to maximize total welfare, but does not know the quality of the prosecutor or of the evidence in each particular case. With perfect information only high-quality prosecutors are retained. However, with a bonus for being retained, the prosecutor s payoff does not coincide with the median voter s. Thus, both types of prosecutor may try to signal that they are of the high type. We are interested in understanding how such signals distort outcomes in the criminal justice system. Suppose that the benefit received by the prosecutor from a particular case is either u(p q (θ)s) c or u(g(θ)). Her total expected utility, U q, then aggregates the expected benefits derived in each case. Assume that the expected utility of a randomly selected case equals the average utility generated over the prosecutor s term. Also, let b>0 denote a retention bonus, which the prosecutor receives if and only if she is reelected. One may think of the bonus as future wages earned, but it could also represent the future gains for an altruistic prosecutor who, if retained, gains utility from prosecuting more cases. 7 Thus, if a prosecutor takes a case to trial if and only if θ θ q,then θq U q = u ( g(θ) ) θm [ ( df(θ)+ u pq (θ)s ) c ] df(θ)+ b (2) θ=0 θ=θ q if retained and U q b if not retained. To simplify the analysis we assume, absent compensation, that the preferences of the prosecutor are related directly to welfare, u(p ) = αw(p) for α>0. Thus, the environment may be best thought of as addressing the decision making of the chief prosecutor who is directly accountable to the voting public. 6 With the assumption that p q (θ)s and g(θ) are continuous functions and c is bounded from above, the intermediate value theorem guarantees that these thresholds (and all thresholds derived throughout the analysis) exist. 7 This setup is similar to the incumbent-challenger models of political agency, which consider the reelection motives of political leaders (Besley 2006). It is possible to make b type-dependent as well, i.e., a better prosecutor could also enjoy higher utility in period 2 because society s welfare is higher. As it does not change the primary predictions of the model, such complications are not presented.

146 Public Choice (2014) 161:141 156 We model the situation as a signaling game. Prosecutors signal their quality by taking a certain number of cases to trial. Voters observe the outcomes, namely the sentences obtained, in trial, and then decide whether to retain the prosecutor. If the prosecutor is not retained, a randomly chosen challenger takes the incumbent s place. We will analyze the perfect Bayesian equilibria of this game. 3 Prosecutor evaluation The question becomes how the prosecutor s behavior responds to the way in which she believes voters will evaluate her. The voters, who do not have access to information about her true type, may use a number of evaluation criteria. One possibility is that prosecutors can signal to voters their quality by getting heavy sentences at trial. Thus, we analyze a particular metric of quality, namely aggregate sentence lengths obtained. Evidence has been presented that this is one of the primary measurements used to assess incumbent prosecutors in media coverage of elections (Wright 2009). Boylan (2005) provides evidence suggesting that this is the quantity that prosecutors try to maximize. Accordingly, suppose that voters assess the quality of the prosecutor by the lengths of the sentences she obtains. Sanctions may be achieved either through plea bargaining or by courtroom victories. Hence, the aggregate sanction imposed over the course of her term can be used to assess her quality. As described above, for each trial an independent draw of θ is taken from the distribution function F. The prosecutor selects a threshold value of θ, denoted θ q, for her term. If θ> θ q the case is taken to trial; if θ< θ q a plea bargain results. This prosecutor s choice arises because the difference between the expected sanction from trial and the plea bargain grows as the quality and quantity of evidence improves. The threshold she chooses may depend on her quality. The expected sanction generated is θq θm S q = g(θ)df(θ) + θ=0 p q (θ)s df (θ). θ=θ q (3) Since it is assumed that a large number of cases are decided in a term, the expected sanction of a randomly selected case represents the actual aggregate sentence when the number of cases is normalized to unity. The voters observe the aggregate sanction achieved and decide whether to retain or replace the incumbent. If the aggregate sentence length exceeds a threshold, then the prosecutor is retained. If the threshold is not achieved, then we assume that with probability z the prosecutor is replaced. This allows for uncertainty in the process and can be a summary measure of the likelihood of a suitable replacement being found. For example, z may represent the probability that the opposing party organizes a campaign to replace the incumbent. Thus, the parameter z serves as a measure of the competitiveness of the electoral environment. 3.1 Separating equilibria Consider, first, separating equilibria. In such outcomes voters keep the prosecutor if S matches or exceeds S e, in the belief that a prosecutor whose aggregate sanction is greater than or equal to S e is type H and those below are type L. Which values of S e are voters able to obtain in a separating equilibrium?

Public Choice (2014) 161:141 156 147 As a useful point of reference, consider the expected aggregate sanction that arises in the first-best outcome. When θ H = θ H θ q θm S q = g(θ)df(θ) + p q (θ)s df (θ). (4) θ=0 θ= θ q Since a prosecutor of high quality wins at trial with a higher probability and takes more cases to trial, it is straightforward to verify that S H > S L. Thus, sentence length is a reasonable metric to use to distinguish between the types. If the benefit to being retained is relatively small, then a low-quality prosecutor is uninterested in increasing the number of cases she takes to trial so as to mimic a high-quality prosecutor. This self separation occurs if the probability of being replaced is small. If the probability of a low-quality prosecutor being replaced is small, then she will select S L = S L and take her chances on Election Day. Therefore, there exists a threshold z defined by z 1 b θ L where if z<z, then self separation occurs. θ H [ u ( g(θ) ) u ( pl (θ)s ) c ] df (θ), (5) Lemma 1 If z<z, then the unique equilibrium has S L = S L and S H = S H. Therefore, consider the environment where z z so that both types are willing to achieve S H to be retained, since the benefit to being retained is sufficiently high and the probability of being replaced is not too low. It follows that in the separating equilibrium S e must be greater than S H. One also finds, since for a fixed b the prosecutor s utility is proportional to welfare, that a high-quality prosecutor is interested, if she remains in her job, in achieving the shortest sentence length that secures her retention, or rather, S = S e. Thus, in a separating equilibrium a threshold θ H = θ e H,below θ H, is set so that S H = S e. Consequently, an excessively aggressive aggregate sentence length is obtained by trying in court some defendants who in the first-best outcome would have received a plea offer. For this aggressive policy to yield an equilibrium she must be willing to push for these harsher sanctions. If she is not willing, then the best choice for her is S = S H. Therefore, selecting θ q = θ e H is preferable if θ e q 0 u ( g(θ) ) df(θ)+ θm θ e q [ u ( pq (θ)s ) c ] df(θ)+ b θ q u ( g(θ) ) θm [ ( df(θ)+ u pq (θ)s ) c ] df(θ)+ (1 z)b, (6) 0 θ q where q = H. There exists a threshold level of θ, denoted φ q,where(6) holds with equality. Hence, if θ e H φ H, then the high-quality prosecutor is willing to take enough cases to court to remain in her position. The cutoff value φ q corresponds to a level of S denoted by S φ q (> S q ). Now consider the incentives of the prosecutor of low quality. If she does not attempt to increase S to S e, then she will not be retained. Since her preferences are proportional to welfare it follows that if she chooses not to achieve S e she selects S = S L by adopting θ L = θ L. Alternatively, she may choose to misrepresent her type by taking more cases to

148 Public Choice (2014) 161:141 156 court and plea bargaining less. Since she is less successful in court, even more cases must be tried than if she were more skilled. Thus, to obtain S e, her threshold, θ L, must be set at a value less than θh e. As before, denote the value of θ that generates S L = S e by θl e.forse to be supported as a separating equilibrium, (6) must fail for θ q = θq e when q = L. Thus, a low-quality prosecutor is unwilling to mimic the high-quality prosecutor if θl e <φ L,or rather if S e >S φ L. Combining these two results, if S e > S H a separating equilibrium exists so long as S e (S φ L,Sφ H ].Sincep H (θ) > p L (θ) θ if a low-quality prosecutor is willing to achieve S e,so too is a high-quality one. Therefore, S φ H >Sφ L and the interval is non-empty. Proposition 1 There exists a range of aggregate sanctions, (S φ L,Sφ H ], in which separating equilibria exist. As a consequence of Proposition 1, the incentives of a prosecutor encourage one of high quality to engage in an excessive number of trials. This increase in tried cases must be so large that a low-quality prosecutor would be unwilling to plea bargain so few cases. Interestingly, in these equilibria it is the low-quality prosecutor who selects the welfaremaximizing number of trials. 3.2 Pooling equilibria In a pooling equilibrium both types achieve an aggregate sentence length S e and the voters believe that a prosecutor who achieves S e or above is type H with probability γ and type L with probability 1 γ. What outcomes can arise in such an equilibrium? First, S e S H since if this were not true a high-quality prosecutor would prefer to deviate to S H. Second, for any outcome to be a pooling equilibrium each agent must be unwilling to deviate. Since the outcome results in a greater aggregate sanction than what the prosecutor prefers, regardless of her type, only deviations that will lower aggregate sentences need be considered. If the voters believe that any such deviations are from low-quality prosecutors, then they will choose not to retain one who makes the choice. Thus, a pooling equilibrium requires that deviating to S q, the most preferred deviation, results in a lower utility than achieving S e and being retained. Again, this requires that (6) hold when θq e is the cutoff value of θ that if used obtains S = S e. Hence, a prosecutor of quality q is unwilling to pool with the other type if θq e <φ q. As a result, S e can be supported as a pooling equilibrium only if S e Sq φ q. Finally, since p H (θ) > p L (θ) θ, any aggregate sentence length that a low-quality prosecutor is willing to achieve to be retained, a high-quality one is also willing to obtain, so as to remain in the position. Thus, S φ H >Sφ L. Additionally, since it is assumed that b b (so that self separation is not an outcome), it must be that S φ L > S H. As a result, the interval [ S H,S φ L ] is non-empty and describes the set of pooling equilibria. Alternatively, if S e > S H, then it may be reasonable for voters to believe that a deviation to S H comes from a high-quality prosecutor. With these beliefs no pooling equilibrium other than S e = S H exists. As is typical in signaling models, the size of the set of pooling equilibria depends on beliefs regarding the likelihood of deviations to non-equilibrium outcomes. Proposition 2 If it is believed that a deviation to a shorter aggregate sentence length comes from a low-quality prosecutor, then there exists a range of aggregate sanctions, [ S H,S φ L ],

Public Choice (2014) 161:141 156 149 in which pooling equilibria exist. If such deviations are believed to have to come from a high-quality prosecutor, then the unique pooling equilibrium is S H. As a consequence of Proposition 2, a prosecutor of low quality is encouraged to engage in an excessive number of trials in order to raise the lengths of her obtained sentences. Interestingly, if the voter believes that shorter aggregate sentence lengths come from a low-quality prosecutor, then there exist equilibria in which both types of prosecutors engage in insufficient plea bargaining. Alternatively, with more reasonable beliefs, high-quality prosecutors engage in the first-best number of prosecutions. 3.3 Testable predictions Our theoretical model leads to several predictions that can be used as a test of the theory. First, consider cases that go to trial. As shown in Lemma 1, if reelection pressures are weak, then the first-best outcome is selected by the prosecutor, regardless of her type. Propositions 1 and 2 illustrate that in all separating and pooling equilibria the number of cases taken to trial is larger than what would be selected in the first-best outcome. Thus, Result 1 Propositions 1 and 2 establish that if reelection pressures are strong, i.e., z>z, there is an increase in the number of cases arising from trials and a decrease in the number of cases plea bargained. As an immediate consequence, the number of convictions adjusts: Result 2 If reelection pressures are strong, then there is an increase in the number of convictions arising from trials increases and a decrease in the number arising from plea bargaining. Consequently, the proportion of total convictions coming from trial increases. As the number of trials increases, the total number of jury trial convictions must also increase, while the total number of plea bargains decreases. Thus, the ratio of the two must also adjust. This distortion reduces the average sentence obtained in a year: Result 3 An immediate consequence of Eq. (6) is that if reelection pressures are strong, then the average sentence obtained in jury trials declines, as does the average number of guilty pleas. More marginal cases, with lower expected sanctions, are taken to trial when the prosecutor expands her use of the courtroom. This reduces the average sentences obtained in trial even though the aggregate sentences obtained increases. 4 Data We now examine whether the effects predicted by the theory arise in practice. Data on crime convictions and prosecutor elections in North Carolina are collected. While there are 100 counties in the state, there are only 43 prosecutorial districts. More heavily populated counties comprise entire districts. Less-populated counties are grouped together into a single prosecutorial district. Each district has one chief public prosecutor (known as the district attorney) who is elected by voters to serve a four-year term. Each district attorney has a staff

150 Public Choice (2014) 161:141 156 of assistant district attorneys. The number of assistant district attorneys varies across the districts. Data for felony prosecutions are collected from the North Carolina Sentencing and Policy Advisory Commission. Each year the Commission publishes the Structured Sentencing Statistical Report for Felonies and Misdemeanors. 8 The Report provides data about convictions obtained for the fiscal year running from June 1 to July 30 of the following year. Data are collected for the 1997 1998 fiscal year to the 2008 2009 fiscal year. Only felony convictions are considered here. In each year a variety of information is available. First, in each district the total number of convictions obtained on the basis of guilty pleas and jury trials is reported. Guilty pleas are obtained primarily through plea bargaining. The variable jury measures the proportion of total convictions obtained in jury trials in a district for a year. The primary testable prediction of the theoretical model is that when reelection pressures are strong prosecutors take more cases to trial and plea bargain less so that more convictions arise from the courtroom relative to plea bargains. Thus, the hypothesis is that jury is distorted by elections. Measuring jury convictions as a proportion of total convictions controls for scale effects from districts differing in the total number of crimes committed. We also have information regarding the type of punishment obtained. The type is determined by the most serious sentence an offender received. Each sanction falls into one of three types: community, intermediate, and active punishments. Community punishments include community service, outpatient drug or alcohol treatment, and fines. Examples of intermediate punishments in North Carolina include house arrest with electronic monitoring, residence in a substance abuse facility, or drug treatment court. Active sanctions are those that result in incarceration. North Carolina uses sentencing guidelines to determine minimum and maximum sanctions. District attorneys complete a worksheet to calculate a score, which measures the number and severity of past offenses. Given the offense the individual is convicted of and his/her prior record score, the guidelines provide for a range of acceptable sanctions. Judges may, though, choose to deviate from these bounds. North Carolina is unique in that the sentencing guidelines also lay out a range of acceptable sanctions for both aggravated and mitigated punishments. 9 If a mitigated or aggravated sanction is selected, the judge is required to provide his/her reason. Additionally, defendants may appeal rulings on these plus and minus factors. The data also contain information about the duration of the active punishments. For each district for each year the average maximum sanction received by those convicted is reported. 10 All of these sanction variables serve as measurements of the particular distribution of cases and crimes that arise in a district over the course of the year. The seriousness of the offenses committed in a year affects the decision to go to trial, independent of reelection motivations. Socioeconomic variables are created. Population data are collected from the North Carolina Office of State Budget and Management. 11 Annual county-level population estimates 8 www.nccourts.org/courts/crs/councils/spac/publication/statistical/annual. 9 In fact, the National Center for State Courts ranks North Carolina as having the most mandatory of sentencing guidelines in the country. 10 The size of the gap between the minimum and maximum sanction is fixed by the sentencing guidelines. Thus, only the maximum is considered here. 11 www.osbm.state.nc.us/ncosbm/facts_and_figures/socioeconomic_data/population_estimates/county_ estimates.shtm provides the data and a description of the estimation procedures used.

Public Choice (2014) 161:141 156 151 are provided. Hence, the population density and the proportion of a district s population that is between the ages of 16 and 24 are recorded. The Office also reports data on the number of males and the number of whites in each county in each year. Finally, district-level labor market data are collected from the Employment Securities Commission of North Carolina. The unemployment rate is used as a control for economic opportunities and the opportunity cost of crime. 12 In four cases, the composition of the district changed during our sample period. While in no circumstance was a county split between multiple districts, in four cases a prosecutorial district that contained multiple districts was split into two districts. In the fiscal years 1997 1998 to 2005 2006 there were 39 districts. In 2005 2006 two districts were split so that then there were 41 in the state. Again in 2008 2009 two districts divided. Thus, for the 2008 2009 fiscal year there were 43 prosecutorial districts. Consequently, our sample contains 476 observations. Finally, election data are collected from the North Carolina s State Board of Elections. 13 We create three dummy variables to capture election outcomes. The dummy variable CI equals one if in a district in that year an incumbent runs for reelection and has a challenger (either in the primary or the general election). 14 CI equals zero if there is no election, if the incumbent is not running for reelection, or if the incumbent running for reelection is unchallenged. While DAs serve four-year terms, elections are staggered so that some occur every biennium. Second, reelect equals one if it is observed in the year before an incumbent runs for reelection. Reelect equals zero if it is not the year before an election or if it is the year before an election but the incumbent does not run for reelection the following year. Third, the dummy variable never equals one if the district has not had either a contested primary election or a contested general election in any election cycle between 1998 and 2008. A value of zero for never occurs if during an election in any year of the sample period the voters in the district had a choice between candidates. Competitive prosecutorial elections are somewhat common in North Carolina. In 22.4% of the elections there was a contest in the general election, while in 24.8 % of the elections there was a contest in a primary election. As a result, the mean value of reelect is 0.2668, CI is 0.0259, and never is 0.2668. These three variables are used to measure reelection pressures and serve as proxies for z in the theoretical model. It is posited that in the year before a reelection campaign an incumbent, if she does adjust her behavior to the election cycle, will respond more in that year than in the previous two years. In the year of the election the presence of a challenger allows for the possibility of not being retained and, thus, compared to years without an election or without challengers, the theory predicts an increase in the use of the courtroom relative to the use of plea bargaining. Therefore, reelect captures the motivation to remain in office, while CI measures the pressure applied directly by a challenger. We assume that voters focus more heavily on the prosecutor s performance in the year prior to and in the year of the election. This is a common assumption in the literature on political business 12 Labor market data are obtained from www.nces.com. The labor force participation rate is statistically insignificant in the results and, hence, is omitted. Furthermore, minor issues arising from the merging of the data, such as aggregating county-level data into district-level observations and public financing issues are discussed in detail in an online appendix (http://faculty.sbu.edu/bmccannon/supplement.htm). 13 www.sboe.state.nc.us. 14 The conviction data cover fiscal years beginning on July 1. Since the general election is in November of the calendar year, the election is recorded in our data as being in the same year if the general election falls within the fiscal year. This implies, though, that the primaries occur in the prior year within the dataset.

152 Public Choice (2014) 161:141 156 cycles (Rogoff 1990; Rogoff and Siebert 1988). Additionally, we predict that the set of equilibria involve outcomes where there is less of use of the courts when z, the probability of being removed from office by the voters if the incumbent underperforms, is small. Further, we posit that if a district has not experienced a single contested election over the entire sample period in either primary or general elections, then the incumbent prosecutors of these districts face lower zs than prosecutors in districts with electoral contests. Hence, never captures environments where electoral concerns are minimal. 5 Results When an incumbent runs for reelection, the data indicate that there is an increase in the number of trial convictions per year of 7.2 % and an increase in jury trial convictions relative to total convictions per year of 18.3 %. When she faces a challenger, additional increases of 13.0 % and 23.7 %, respectively, arise. An incumbent who is immune from such pressures prosecutes less (11.6 % and 23.8 %). A more rigorous econometric test is needed, though, to verify that such effects actually do occur. The variables reelect and CI measure the existence of a reelection campaign and the presence of a challenger, respectively. Jointly, these identify the potential distortions caused by retention motivations. CI is obviously collinear with never since for the former to equal one it is required that the latter equals zero. Hence, separate specifications are estimated to identify the impact of the election variables on the use of jury trials. Note also that since the bulk of the elections occur in the 1998 2002 2006 cycle, CI and reelect are highly correlated with the year fixed effects. Finally, the variable never is a district-level control and, hence, never and district fixed effects cannot be entered simultaneously in a regression specification. Consequently, the econometric models to be estimated are jury dt = α 0 + α 1 CI dt + α 2 reelect dt + α 3 S dt + α 4 X dt + α 5 D d + ɛ jury dt = β 0 + β 1 never d + β 2 S dt + β 3 X dt + β 4 Y t + ɛ dt, (7) where S is the sanction variables, X comprises the socioeconomic controls, and D and Y are district and year fixed effects, respectively. Theory predicts that α 1 > 0andα 2 > 0, along with β 1 < 0. Table 1 presents results of tests of the corresponding null hypotheses. The coefficients of CI and reelect are positive and statistically significant. If an incumbent is going to run for reelection and is interested in increasing her perceived toughness, then she increases the number of convictions that come from jury trials in the prior year. This is in order to gain attention in the news, appear hawkish to voters, obtain greater sanctions, deter challengers in the general election, and to be successful in the primary. If in the year of reelection a challenger enters, then she increases the number of convictions obtained from jury trials relative to the number of convictions obtained from guilty pleas. This represents at the mean a 9.7 % increase in the year before the reelection and an additional 14.7 % increase if in the reelection year a challenger enters the race. Put another way, the average number of jury convictions in a district in a year in North Carolina is 17.66 and the average number of guilty pleas is 700.92. Thus, there are 1.80 more convictions from jury trials in the year before the reelection and 2.73 more convictions in the year of the contested race, for a total of 4.53 more jury trial convictions. The coefficient of never is negative and statistically significant. Thus, a district that has not had an election contest between 1997 and 2009 experiences fewer jury trials. This corresponds to a 10.8 % decrease in the number of jury trial convictions.

Public Choice (2014) 161:141 156 153 Table 1 Effect of the election of prosecutors (dep. var. = jury, N = 476) I II CI 0.0038** (0.0018) reelect 0.0025** (0.0011) never 0.0028** (0.0012) Population density 2.0 10 6 (1.6 10 5 ) 1.0 10 5 *** (0.0000) Unemployment rate 0.0211 (0.0295) 0.2466*** (0.0778) Average maximum sanction 0.0004*** (0.0001) 0.0005*** (0.0001) % of convictions with an 0.0341** (0.0159) 0.0048 (0.0136) aggravated sentence % of convictions with a mitigated 0.0159** (0.0067) 0.0070 (0.0074) sentence % of convictions with an 0.0254** (0.0104) 0.0334*** (0.0101) intermediate sanction % of convictions with a community 0.0243* (0.0142) 0.0178 (0.0131) punishment % of population that is male 1.1489*** (0.3932) 0.3050*** (0.0593) % of population that is white 0.0148 (0.1103) 0.0181*** (0.0053) % of population between 16 and 24 0.1047 (0.1291) 0.0079 (0.0273) Year controls No Yes District controls Yes No adj R 2 0.588 0.370 * 10% level; ** 5% level; *** 1% level. Heteroscedasticity-robust standard errors are reported The sanction control variables are, for the most part, statistically significant and have the predicted signs. The socio-economic variables help explain the use of jury trials to obtain convictions as well. Robustness checks including alternative specifications and standard errors, along with other election-related variables, are conducted and available in an online supplement. One important concern is the potential endogeneity problem of the choice of a challenger to enter the race. Suppose a challenger runs because the incumbent is more aggressively prosecuting cases (resulting in more acquittals or a greater use of public funds, for example). The results presented in Table 1, then, may arise not from reelection concerns, but rather from the choices made by the challenger. To address this concern we adopt an IV approach as a robustness check. To capture election pressure we need a variable correlated with CI but uncorrelated with jury. Specifically, an instrument is needed that measures the competitiveness of the political offices in an area. We consider judicial contests at the court of first instance for felony cases (known as the Superior Court) as an instrument. The state is divided into superior-court districts, which (roughly) coincide with prosecutorial districts. These judicial districts are organized into eight judicial divisions. Superior Court judges rotate between districts within their divisions. A term in office is eight years long and elections for those judges are staggered within the judicial division. In the year in which an incumbent prosecutor of a district is running for reelection, judge equals the proportion of Superior Court judicial seats that are contested within the division that contains the district. In years in districts without reelection campaigns for the DA position, judge equals zero. Correlation between prosecutor and judicial contests could arise because of public dissatisfaction with elected officials, or because they are both more likely to occur in a relatively politically

154 Public Choice (2014) 161:141 156 Table 2 Two-stage least squares results (dep. var. = jury; N = 476) * 10% level; ** 5% level; *** 1% level I II CI 0.0121* (0.0069) 0.0119* (0.0070) reelect 0.0028** (0.0014) 0.0028** (0.0014) never 0.0019 (0.0015) Controls: socio-economic Yes Yes sanction Yes Yes adj R 2 0.318 0.320 Table 3 Effect of prosecutor elections on sentencing (N = 476) dep. var. % of convictions with community punishment % of convictions with community punishment average maximum sanction CI 0.0213** (0.0098) 2.640* (1.3575) reeleect 0.03383*** (0.0070) 1.4316* (0.7805) never 0.0248*** (0.0054) District controls No No Yes Year controls No Yes No adj R 2 0.129 0.290 0.439 * 10% level; ** 5% level; *** 1% level. Heteroscedasticity-robust standard errors are reported active, contested area. Our variable satisfies the criteria of an instrument, i.e., it is highly correlated with prosecutor contests, but not with jury. Table 2 presents the two-stage least squares estimates. The significantly positive coefficients of CI and reelect show that our main results are confirmed when entry is instrumented by judicial contests. Another prediction of the theoretical model is that the average sentence length a prosecutor achieves in jury trials in a year should tend to decline when reelection pressures are relatively strong. We propose two ways to identify this effect. First, if the prosecutor is trying to appear tough and achieve longer incarcerations, then a reduction in the percentage of convictions that result in community punishments as the most serious sanction should occur. Since community punishments are expected to arise more frequently from plea bargaining than from trial cases, a decrease in the percentage of convictions resulting in community punishments is evidence that prosecutors are pursuing relatively more severe sanctions. Second, if relatively more cases are being taken to court, and fewer are being plea bargained, then the marginal cases tried in court are those with the weakest evidence. Such cases will result in lower expected sanctions than those tried regardless of the retention motivation, ceteris paribus. Thus, our model predicts that the average sanction obtained should decrease with reelection pressures. Table 3 presents the corresponding results. 15 15 The first two columns include all socio-economic controls. The third and fourth columns include the sentencing guideline controls along with population density. The specification with never as the independent variable and the average maximum sanction as the dependent variable is omitted because of never s statistical insignificance.

Public Choice (2014) 161:141 156 155 The predictions of the theoretical model are supported by these results. If a district has an incumbent running for reelection, and a challenger has entered the race, the number of convictions that result in community punishments falls by 10.5 %. Similarly, the change in the average maximum sanction declines by 6.2 % in the year of a contested election, as would be predicted if prosecutors increase the number of cases they take to trial by including those likely to receive lower than average sanctions. Never having had a contested election increases the proportion of convictions that have community punishments, while it does not have a statistically significant effect on the sanctions for active punishments (jail sentence). Interestingly, in the year before a reelection campaign, there is a greater reliance on community punishments, and there is an increase in the average maximum active sanction. This could be explained, for example, by a forwardlooking incumbent dispensing with weaker cases in the year prior to the reelection, by encouraging pleas to community punishments so that in the reelection year resources can be devoted to securing tougher sentences through trials (as would happen in Rasmusen et al. 2009 framework). Thus, it seems that the contested incumbent influences which cases go to trial. Given the strict sentencing guidelines used in North Carolina, the result implies that prosecutors must expand the scope of cases prosecuted at trial and are unable to push for harsher sentences. 6 Conclusion In the United States, prosecutors exercise a significant amount of discretion. Understanding how direct elections affect prosecutors choices is crucial to developing the best mechanisms for implementing the preferences of the majority voting. There is a serious void in the analysis of the decisions of prosecutors. In this paper we analyze prosecutors decisions to take cases to trial or plea bargain. We investigate how the desire to win reelection affects this decision. We develop a theoretical model of asymmetric information to explore how trial outcomes can be used as a signal to the voting public. The theory predicts that when reelection pressures are strong, prosecutors increase the number of cases taken to trial and plea bargain less. The empirical evidence is consistent with the theory. One might be concerned with the role of the judges, since their electoral cycles are correlated with prosecutor electoral cycles. However, prosecutors (and not judges) make the decision to take cases to jury trials. Thus, one would not expect the increase in the number of jury trials prior to the prosecutorial election to be a result of judges reelection pressures. One can argue that there is an indirect effect if a judge, running for reelection, levied enhanced punishments. That may cause prosecutors to revise their expectations about the expected sanction if they took the case to trial and, consequently, induce prosecutors to avoid trials and accept plea bargaining yielding tougher sentences. Our model suggests that is precisely what will happen. Plea bargain outcomes adjust as defense attorneys anticipate this behavior and accept tougher sentences in plea bargains. Consequently, the incentive to take cases to trials should not change. Further, our empirical analysis shows that the judge variable is uncorrelated with the proportion of convictions that arise from trials. This suggests that the results are not influenced by judicial retention concerns. The results strongly suggest that election pressures affect decisions. Whether this is suboptimal depends on whether alternative explanations can be ruled out. For instance, retention concerns may induce additional effort. If prosecutors are motivated to work harder, the average sanction achieved in trial presumably would increase, as a result of the additional resources expended on preparing cases. Hence, the result that the average maximum sanction