Abstract
We evaluate specialized domestic violence (DV) courts using administrative data on misdemeanor cases in Davidson County, TN from 2000–2006. We find that DV defendants arbitrarily assigned to specialized courts are more likely go to trial and less likely to be convicted than those in traditional court. Domestic violence victims in cases assigned to specialized courts are less likely to be involved in future DV incidents, but more likely to cooperate with police than those assigned to traditional courts. Defendants assigned to DV versus traditional courts are no more or less likely to be charged with a future crime.
I. Introduction
In 2012, more than 3,000 “specialized courts” were identified in the United States, representing 5 percent of all jurisdictions, with most created after 2001 (Strong, Rantala, and Kyckelhahn 2016). In contrast to traditional courts, specialized courts are dedicated to cases in which defendants have specific needs, such as mental health considerations, or are accused of specific types of crimes, such as drug offenses or domestic violence (DV).
There are multiple models of specialized courts, but all specialized courts share two features: (i) a judge specializing in a given case type and (ii) the specialized judge’s fixed courtroom personnel, typically probation and court officers (Strong, Rantala, and Kyckelhahn 2016). Existing research does not clearly show how specialized courts affect future criminal behavior or victim satisfaction relative to traditional courts. Indeed, in 2016, the Bureau of Justice Statistics noted that only 41 percent of specialized courts track the subsequent criminal behavior of individuals moving through them (Strong, Rantala, and Kyckelhahn 2016).
This study’s primary contribution is to estimate the causal effect of specialized, criminal DV courts’ central components on misdemeanor DV case outcomes and subsequent crime, in an established, rather than new, specialized court. We use novel data from the Tennessee General Sessions Court of Metropolitan Nashville and Davidson County’s specialized DV courts. The specialized DV courts consist of three actors that we collectively call a “judicial component”: a specialized judge, specialized probation officers, and specialized court officers. During our analysis period, 2000–2006, Davidson County operated two identically structured DV courts.
During this time, the two specialized DV courts, along with the other general courts, heard misdemeanor cases where the defendant had not posted bond (that is, bail) on a predetermined, rotating schedule. The court rotation process assigned misdemeanor defendants who could not post bond to specialized DV or traditional general courts by arrest date alone, which is uncorrelated with defendant or case characteristics that might predict conviction, incarceration, or recidivism. All DV cases, regardless of assignment to specialized or traditional courts, had access to additional resources for DV cases. These included specialized prosecutors and public defenders and a victim’s advocate. As such, our estimates should be considered to represent the causal impact of specialized judicial components only, and not the additional wrap-around services that are often available in specialized courts, on various DV outcomes.1
Comparing the DV case outcomes in specialized versus traditional courts, we find that DV cases are more likely to go to trial when assigned to courts with specialized judicial components. The increased use of trials leads to specialized judges exerting increased influence on outcomes, consistent with legal critiques of specialized judges issuing wide-ranging opinions in civil courts (for example, Rachlinski, Guthrie, and Wistrich 2006).
Our results furthermore show that defendants assigned to specialized DV courts are less likely to be convicted and incarcerated and have shorter expected sentences than those assigned to traditional courts. We do not observe that DV courts are less punitive than traditional courts in non-DV cases, which are also randomly assigned. Our finding suggests that the observed differences in DV case outcomes result from the way specialized judges, probation officers, and court officers react (or are expected by attorneys to react) to specialized cases, rather than a broader judge fixed effect. Taking judicial “taste for conviction” into account, DV judges are 10 percent less likely to convict in DV cases than in misdemeanors, on average. Furthermore, DV cases in DV court are 16 percent less likely to result in incarceration, and maximum sentences are 14 days shorter, than misdemeanors in general.
We next evaluate specialized courts’ effect on victims’ and offenders’ future behavior. We link the court data to incident and arrest records from Metropolitan Nashville Police Department, the largest police department in Davidson County, and estimate the impact of being assigned to a specialized court on multiple distinct measures of future criminal activity. In the subsample of cases with individual victim information in the court records, we find that victims whose cases are assigned to DV courts are 25 percent less likely to contact police about a future DV incident than those assigned to traditional courts.
We find the reduction is observed among victims whose cases are presented to specialized courts in a trial or to approve a plea bargain, not those whose cases are dismissed by prosecutors prior to involvement by the judicial component of the court. The finding suggests exposure to the specialized courts’ judicial component drives our outcomes of interest. Further, victims with previous cases handled by a DV court are up to 80 percent more likely to cooperate with police officers, when called, than those initially heard in traditional courts. Patterns of subsequent charging do not suggest that these results are driven by victims or witnesses allowing increased DV escalation before calling authorities.
We do not find evidence that specialized courts’ judicial component impacts recidivism rates, where recidivism is defined as the probability a defendant is charged with a misdemeanor or felony in the three years following their initial adjudication. We also fail to find evidence that defendants assigned to DV courts are, on net, any more or less likely to be arrested again than those assigned to general courts. Our data on victims and defendants collectively suggest that DV specialized courts can reduce costly convictions and incarceration while potentially reducing reoffending.
In Section II, we summarize the existing scholarship on judicial specialization and policy interventions intended to reduce DV. Section III provides an overview of how the Metropolitan Nashville and Davidson County Court handles misdemeanor cases. In Section IV, we explain our data, test our judicial component randomization, and outline our estimation strategy. We present our estimates of judicial specialization’s effect on case outcomes and future criminal charges and revictimization in Sections V and VI. Section VII examines the policy implications of our results.
II. Evidence on Judicial Specialization and DV Reduction
Our findings build on economic and legal scholarship on judicial specialization, as well as research on criminal justice interventions aimed at reducing DV’s social harms. As noted by Baum (2009), extant legal scholarship on judicial specialization primarily focuses on noncriminal federal courts making decisions about bankruptcy, taxes, or patents (for example, Miller and Curry 2009; Curry and Miller 2015; Baum 2011; Hansford 2011; Kesan and Ball 2011; Rachlinski, Guthrie, and Wistrich 2006; Howard 2005). This literature suggests that judges’ decision-making consistency increases with experience or expertise, and experienced judges may be less likely to be overturned on appeal than their inexperienced counterparts. Coviello, Ichino, and Persico (2019) empirically evaluate specialization’s impact in Italian labor dispute courts and find that judges with elevated experience in particular labor dispute types resolve the cases more quickly and with less likelihood of being overturned than those with less experience.
Other researchers have found evidence of specialization’s potential downsides. Rachlinski, Guthrie, and Wistrich (2006) and Miller and Curry (2009) find that politically polarized decision-making increases with judge experience. Baum (2011) argues that specialized or expert judges are more likely to issue broad decisions potentially influencing public policy than traditional court justices. In other words, specialization may increase the objective quality of judges’ decisions but inflate their confidence in their decision-making ability.
The nature of DV may lend itself to specialized policy responses (Dobash and Dobash 1992; Epstein 1999). Many DV-focused policy interventions involve hiring more women police officers, assuming they have expertise or experience that makes them better able to respond to DV victims than men. Some empirical evidence from the United States and Brazil (Perova and Reynolds 2017; Miller and Segal 2019) suggests the assumption may be true, but other efforts to create “women’s [police] stations” in India may have been less effective (Jassal 2020). Other criminal justice strategies to reduce DV involve reducing the need for victim cooperation, including “no drop” policies that require district attorneys to prosecute DV cases even if victims decline to cooperate (Aizer and Dal Bó 2009) and mandatory reporting laws requiring adults working with vulnerable individuals to contact police if they suspect family violence (Bullinger, Carr, and Packham 2021; Benson, Fitzpatrick, and Bondurant 2025). Notably, each of these interventions requires increased contact between offenders and the traditional criminal justice system.
A relatively small literature, primarily in criminology, has found mixed evidence on the effectiveness of specialized DV courts at reducing crime, particularly those following therapeutic jurisprudence models. An analysis of 20 peer-reviewed and non-peer-reviewed articles (Gutierrez, Blais, and Bourgon 2016) finds that DV court defendants are less likely on average to recidivate, but studies with more credible research designs (for example, using pre–post or matched comparison groups) find that introducing DV courts into existing court systems has small or null effects. Research on DV victims have been primarily qualitative, suggesting many—but not all—victims feel positively about the specialized court experience (Moore 2009).
We build on the policy intervention research by providing the first evaluation of a DV court that uses a criminal, rather than therapeutic, jurisprudence model. We also evaluate an established, rather than new, DV court with a specific and scalable structure and use an identification strategy closely mirroring random assignment to find credibly causal effects.
III. Institutional Setting
The Tennessee General Sessions Court of Metropolitan Nashville and Davidson County serves as the local judicial system’s entry point for those charged by the district attorney. It is a high-volume, limited jurisdiction court, meaning its judges can decide on civil, misdemeanor, traffic, environmental, and municipal ordinance violations. The judges can make only initial decisions in felony cases.2 During our analysis period, 87 percent of cases the court processed were misdemeanors. Court judges adjudicate misdemeanors by accepting plea agreements, presiding over bench trials (that is, hearing cases and making decisions without a jury), or dismissing cases prior to trial, which typically occurs when the district attorney’s office cannot produce evidence to substantiate warrant allegations. The court resolves most misdemeanors via a plea agreement negotiated by the district attorney and defendant and approved by a judge. The court employs 11 judges at any point in time (see Online Appendix Table A1 for basic demographic details of the judges in our sample), and each judge is elected to an eight-year term. For administrative purposes, each court’s judicial component is referred to by a numbered division, that is, Divisions 1–11.
In 1994, Nashville-area judges and victim’s advocates designated two of the 11 general sessions courts, Divisions 1 and 4, as DV courts.3 The new courts would hear all “DV-related cases,” or those where a police officer or court agent identified that the victim and offender shared a domestic relationship.4 The reform’s goal was for all DV cases to be heard by one of the two specialized courts. However, for budgetary and pre-trial detention reasons, only defendants released on bond had access to DV courts in 1994–2006. The reform further dictated that the DV courts would not require defendants to plead guilty, be limited to first-time offenders, or use nontraditional approaches like “restorative” or “therapeutic” justice. Rather, they would operate within the traditional criminal justice system but be directed by DV specialists.
A DV judge presides over each DV court. While no statute requires adherence to standards or judicial certification in DV issues, justices select into serving on the specialized courts; both DV judges in our sample, Gale B. Robinson and Gloria A. Dumas, specifically sought election to Division 1 and 4,5 and the justices are held in high esteem for their commitment to DV issues.6
Domestic violence court judges, like all the district’s general session judges, are assigned two probation officers. These officers receive training at conferences focused on managing DV defendant probation and understanding the dangers DV victims face.7 Defendants meet regularly with the probation officers, who file any probation violations, meaning these specialized officers have significant implications for accountability, recidivism, and victim support.
Domestic violence court judges, like all the district’s general sessions judges, are also assigned two court officers. The court officers’ primary function is to ensure courtroom safety and order. An observant court officer can help minimize witness intimidation or tampering, which can lead to victims or witnesses recanting testimony or leaving court.
All Davidson County DV cases are exposed to additional specialized resources not included as part of the courts’ judicial component. As in many jurisdictions, the district attorney’s office has lawyers and victim witness coordinators dedicated to DV cases. The public defender’s office also makes consistent assignments in DV cases. Community DV victim advocates, including professionals from shelters and legal aid and Batterer Intervention Program representatives, may be involved in any DV case. In addition to seeking justice for victims and defendants, these professionals provide DV courts continuous education and perspective.
Specialized courts’ judicial and nonjudicial components potentially affect the way DV cases proceed in two ways. First, DV judges may ultimately convict in DV cases at different rates than non-DV judges, either on average or in specific types of cases. The difference might be due to training they choose to receive, their expectation for their probation and court officers’ actions, their own experience, or their selection into the position. Second, the number and types of cases where a plea deal is reached may shift because of strategic attorney behavior in response to expectations about the courtroom experience and DV judges’ conviction tendencies. Domestic violence defendants may be willing to accept harsh plea conditions to avoid bench trials in front of high-conviction-rate judges. The legal literature suggests specialized judges may reject more plea deals than their peers because they prefer to have control over cases in their specialization area (Baum 2011). If a DV judge prefers to try DV cases, or a district attorney prefers to try DV cases in front of DV judges, the fraction of bench trials increases, mechanically lowering conviction rates relative to when more plea agreements are reached.
Domestic violence court judges do not have the authority to impose different sanctions than non-DV judges,8 but their experience and assistance from other specialized resources may help them select the criminal justice response best suited to reducing further victimization and offending in the circumstances of each offense. Moreover, DV defendants heard in DV courts may have different post-court experiences if specialized probation officers offer expertise not offered by standard officers.
Evaluating the specialized environment created by DV courts’ judicial component is complicated. Typically, the case assignment process creates a failure of common support; any case involving DV goes to a DV court, and only cases unrelated to DV are assigned to traditional courts. However, from 1994–2006 in Davidson County, cases were assigned to traditional versus DV courts based on their content only when the defendants had posted bond. We leverage the fact that defendants unable to post bond and held in jail are assigned to courts based solely on arrest date via the “jail docket.” As we will show, the assignment process meant that on average, jail docket defendants had a two in 11 chance of being assigned to a DV court’s judicial component and nine in 11 chance of being assigned to a non-DV court’s judicial component.
A. Assigning Low-Income Defendants to Specialized or General Court
When police officers respond to any potential crime, they must complete an incident report containing a complaint number, incident characteristics, and victim information. If police officers have probable cause to believe a crime occurred in domestic incidents, Tennessee law “strongly encourages” them to arrest the offender. After an arrest is made, a night court commissioner sets the bond and bond conditions,9 and the case is subsequently transferred to the district attorney’s (DA’s) office. The DA can then either drop the case or pursue criminal charges in general sessions court.
Approximately 74 percent of people accused of any type of misdemeanor and 48 percent of people accused of misdemeanor DV are unable to make bond. Following Tennessee law, these defendants’ cases must be adjudicated by the general session court within ten days of their arrest, barring extenuating circumstances.10 To meet this statutorily mandated deadline, each judge must spend about four to five weeks annually hearing jailed defendants’ cases—what is known as “working the jail docket.”11 Per the rotating schedule, summarized in Figure 1, prior to 2006, a DV court’s judicial component—the judge and their courtroom working group—cycled through the jail docket twice every 11 weeks.12
Misdemeanor Case Assignment for Jailed Defendants in Davidson County, 2000–2006
Notes: Between 2000 and 2006, two DV judges and nine general judges served the Davidson County general sessions court. Two court officers and two probation officers are assigned to each judge, a set of five people called a “division.” Defendants who do not post bond are assigned to the division scheduled to adjudicate jailed cases that week according to a predetermined, rotating schedule. We do not examine cases where the defendant posts bond, since all DV cases and no non-DV cases are exposed to the entire DV court (DV division, prosecutors, public defenders, and victim’s advocates) when that occurs.
Online Appendix Table A2 shows how arrest days and times correspond to defendants’ first jail docket setting during our sample period; no judge, defendant history, or nature of offense is accounted for in determining this setting. Selection into a particular judge on the jail docket is possible only if a defendant aligns an offense, or police officers time an arrest, with a particular justice’s assignment. As we will show, we find no evidence that misdemeanor defendants post bond in response to the judge assigned to the jail docket at the time of their arrest.
B. Identifying the Impact of Court Specialization Using the Jail Docket
Case assignment to judges based solely on arrest date and the jail docket’s assignment rotation allows us to estimate the causal effect of specialized DV courts’ judicial component on case and defendant outcomes. We also examine non-DV jail docket cases arbitrarily heard by the DV court, allowing us to compare the two courts’ outcomes in DV and non-DV cases and to disentangle specialization’s impact from judge or court fixed effects.
Because all judges can assign the same punishments and all cases where the defendant cannot post bond receive nonjudicial DV services, our treatment mechanism is the collective specialized knowledge of the DV judge and their officers, rather than specific DV advocacy or defense strategies. Our findings may therefore be generalizable to other types of judicial specialization (that is, specific legal areas, such as veteran’s or drug court), where victim advocates or specialized legal representation may not be involved. Alternatively, if complementarities exist between DV courts’ judicial component and nonjudicial DV services, our results may overstate the impact of court specialization.13
IV. Data and Identification Strategy
A. Data
We use digitized data on all criminal charges filed in the Tennessee General Sessions Court of Metropolitan Nashville and Davidson County, up to 2018, taken from publicly available records posted on the county clerks’ web portal. Our analysis focuses on cases adjudicated between 2000 and 2006. When defendants did not post bail during the time period, their cases—and only their cases—were heard in Criminal Justice Center Room 132.
Our data begin in 2000, when the court system shifted to electronic warrants, making the information complete and sufficient for analysis. We do not analyze cases adjudicated after 2006, when the district restructured its jail docket to assign all DV cases exclusively to DV courts. While our analysis focuses on DV cases adjudicated in the court’s misdemeanor jail docket, we use the entire data set of court cases through 2018 to construct measures for defendant criminal histories and future interaction with the court system (that is, whether they are charged with a new crime before and/or after case disposition), including felonies and misdemeanors.
The court records include defendant information, as well as court appearance and case details. Defendant data include name, date of birth, race, and a unique, fingerprint-supported originating case agency (OCA) number generated upon jail booking. Appearance data include court dates, attorneys, courtroom assignment (allowing us to identify jail docket defendants), and judges (allowing us to identify when a DV judge and staff are assigned). The case details include convicted offense, disposition, disposition date, incarceration and/or probation length and location, court costs, restitution costs, and special/extra conditions. Using OCA numbers, we link court records to arrest and incident data obtained from the Metropolitan Nashville Police Department, the largest law enforcement agency in Davidson County.14 Incident data include victim names, ages, and genders,15 as well as the alleged offense and responding officer’s report on how the incident ended, that is, by arrest, offender death, determination of an unfounded crime, or victim’s refusal to cooperate.
We identify DV cases based on the charges filed in court. In our primary specification, our definition of a DV case is one that includes a charge for “domestic assault,”16 specifically a violation of Tennessee statute §39-13-111.17 Figure 2 presents the number of domestic assault cases heard in Nashville between 2000 and 2006, as well as the number heard by specialized judges. During our sample period, 12 judges ran the 11 judicial divisions. Consistent with jail docket cases being assigned to DV courts arbitrarily and defendants not differentially posting bond when assigned to DV courts, the two specialized courts heard about 18 percent of jail docket DV cases in our analytic sample.18
Total Number of Cases Heard
Notes: The figure presents the number of DV cases heard in metropolitan Nashville and Davidson County between 2000 and 2006, as well as the number of cases heard by DV courts. In our sample, a total of 12 judges ran 11 courts. Consistent with jail docket cases being assigned to DV courts arbitrarily, the two DV courts heard about 18 percent of jail docket DV cases on average in any given month.
B. Testing Random Assignment to Courts
We empirically investigate whether Davidson County jail docket defendants are assigned to courts arbitrarily. Following DiTella and Schargrodsky (2013), we estimate a regression of defendant attributes, such as race or age, on a DV court indicator, controlling for the variables on which randomization is conditioned (that is, year and week of the year fixed effects), as follows:
1In Equation 1, xijt represents a characteristic of defendant i, whose case is assigned to judge j at time t. (DV Court)j is an indicator taking the value of one if judge j, who hears a DV case at time t, is in a specialized DV court. The matrix λt contains year- and week-of-assignment fixed effects, and μijt is the case outcome’s unobserved component. We examine the following defendant characteristics: an indicator for being white, age, days from arrest to court, number of charges per case, an indicator for being previously charged with any crime, an indicator for being previously charged with DV, an indicator for being previously charged with simple non-DV assault (that is, violating Tennessee statute §39-13-101), and number of previous court appearances for any crime. Because of the small number of female defendants, we limit our sample to men.19
Table 1 presents the observable case and defendant characteristics from our sample across DV and non-DV (general) courts, including an indicator for whether the defendant is successfully linked to any Metropolitan Nashville Police Department arrest record with an OCA number. In Column 1, we present the unconditional mean of the non-DV court covariates. The second column contains the characteristics’ predicted value from an OLS regression on the DV court assignment indicator, controlling for time fixed effects. Most notably, the “days from arrest to court” variable is substantively identical and not statistically distinguishable across the two samples—if defendants were trying to delay or speed up their case to obtain a particular judge or court, we would expect that timing manipulation to appear here. The p-values reflect the coefficient’s statistical significance on the DV court assignment indicator (that is, π1 in Equation 1).
Testing Random Assignment to DV Court for DV and All Other Cases
The sample contains approximately 1.5 times more Black than white defendants, and about 85 percent of the 1,045 DV defendants have previous court appearances. More than 22 percent of individuals in the sample have been charged with DV before their current case, and the average defendant age is about 35 years.20 Two out of 18 p-values associated with the defendants’ differences across specialized and nonspecialized judges are smaller than 10 percent, and the means of the characteristics of the litigants and cases seen by judges are substantively similar. We conclude that defendant and case characteristics are not correlated with judge specialization, consistent with our institutional knowledge of jail docket case assignment.
C. Estimation Strategy
Given arbitrary assignment of jail docket cases, we use basic multivariate regression to evaluate whether DV cases assigned to DV courts are adjudicated differently than they would be in traditional courts.21 We first analyze the district attorney’s initial decisions on dismissal and plea bargaining, then turn to four main categories of adjudication outcomes: (i) any conviction, (ii) number of convictions, (iii) incarceration, and (iv) maximum sentence length. Finally, we examine postcase outcomes using incident, arrest, and court data. Our initial specification takes the form:
2where Yijt is the outcome of interest for defendant i facing the judge in court j (either specialized or not) in year t. (DV Court)j is an indicator variable for whether the judge and court j staff are specialized. λt are year- and week-of-case fixed effects, and uijt is an unobserved error term, which we adjust for arbitrary correlation in outcomes within defendants and weeks.22 Under our identifying assumptions of arbitrary assignment, 
provides an unbiased estimate of the effect of a DV offender being assigned to a specialized court relative to a nonspecialized court. To confirm the conditionally independent case assignment to judges and increase precision, we show results including time varying case (Xit) and judge (Jjt) controls, including defendant race, age, age-squared, whether they have previously been charged with a crime, and number of previous charges, as well as judge gender and time since appointment.
The β1 estimate reflects two things: (i) the average difference among all cases assigned to specialized courts (that is, a court fixed effect) and (ii) the average difference in outcomes for specialized versus nonspecialized cases in specialized courts. The underlying arbitrary case assignment mechanism allows us to differentiate between the two impacts because the judge assigned to work the jail docket hears all cases each week, rather than simply DV cases. We therefore estimate the following difference-in-differences (DID) model:
3Here, we essentially estimate a court fixed effect α1 using a comparison group of misdemeanor cases unlikely to involve DV—primarily drug offenses, theft, and fraud—and allow for an average difference in outcomes across DV and non-DV case types c.23 Notably, and important for the ultimate interpretation of these results, in order to ensure that DV victims were not denied services due to their alleged attacker’s financial constraints, all nonjudicial DV court services (for example, Batter Intervention Program representatives, specialized public defenders, and prosecutors with DV expertise) were provided on the jail docket. Therefore, by extension, all jail docket cases—even those unrelated to DV—received the full slate of nonjudicial DV court services. Therefore, the estimated value of δdd is the impact of judicial specialization on the specialized cases.
Finally, we will also show that the inclusion of judge, defendant, and case characteristics, including the judge’s gender and tenure, the total number of charges, the month of the case, the race and age of the defendant, and the defendant’s previously number of DV and non-DV charges, does not substantively affect our estimates of either Equations 2 or 3, consistent with court assignment being arbitrary, and not subject to selection.
V. Results
A. How Attorneys Make Initial Decisions in Specialized Courts
A case sent to the DA’s office has several potential outcomes. The most likely in our sample is dismissal or retirement prior to trial (44 percent of all DV cases and 23 percent of all—DV and non-DV—misdemeanors). Either outcome can occur when the state is unable to prosecute. For example, the victim may not appear in court or recant previous statements, or the prosecutor may determine the facts of the case do not constitute a crime. Judges cannot force a case to trial when the prosecutor drops all charges.
The assigned judge adjudicates the remaining, nondismissed cases, unless they are pled. Plea bargains must be agreed upon by prosecutors and defense attorneys, who know the plea must be accepted by the presiding judge. The attorneys’ decision whether to reach a plea agreement is therefore conducted “in the shadow of a trial” and might differ based on court assignment.
Domestic violence trials may be more likely if prosecutors seek harsher verdicts when assigned to DV courts—conditions they know could be subsequently rejected by the defense. At the same time, legal critiques of specialization suggest judges might issue broad and far-reaching decisions in their area of expertise, meaning DV judges would be more likely to reject plea deals and personally resolve DV cases. It is also possible that non-DV judges might prefer to avoid hearing DV cases because they fail to understand the underlying dynamics.
Table 2 presents our estimates of specialization’s impact on initial case processing. We estimate a statistically insignificant 2.7–3.8 percentage point increase in the probability that all charges are dismissed in DV cases assigned to a DV judge versus a non-DV judge, depending on additional control variables. The result is not consistent with district attorneys strategically pursuing DV charges in objectively weak cases when assigned to DV courts. We find an eight to ten percentage point reduction (16–20 percent of the sample mean) in the probability that a DV case assigned to a DV court versus a non-DV court is fully resolved via plea bargain. Finally, we see a five to six percentage point increase in the probability that a DV case will be heard in a bench trial when assigned to a DV judge relative to a DV case assigned to a traditional judge, essentially doubling the probability of a DV trial in which the judge engages in direct decision-making.
How DV Courts Process DV Cases
Our difference-in-differences models, shown in Table 2, Panel B, imply DV cases are, on average, twice as likely as other types of cases to be dismissed. When assigned to DV instead of non-DV courts, these cases are 6.8 percentage points (about 10 percent) less likely to end in plea bargains. About half the of the five to six percentage point increased propensity to go to trial observed in Panel A represents a court fixed effect; DV judges are about three percentage points more likely to hear a DV case than any other misdemeanor case involving a jailed defendant relative to nonspecialized judges. While smaller in absolute terms than Panel A and imprecisely estimated, the result corresponds with a 100 percent increase in the probability of a bench trial relative to cases on the misdemeanor jail docket.
B. Specialized Judges’ Decision-Making
Table 3 presents our estimates of specialization’s effect on case outcomes based on Equation 2. As in Table 2, columns labeled 1 contain only time effects, and columns labeled 2 add controls for defendant characteristics (race and age) and criminal history (whether previously charged and number of previous charges) and judge characteristics (gender and experience, that is, years of judicial service). In Panel A, we examine DV assault cases only. Domestic violence court judges are six to eight percentage points (11.6–13.3 percent) less likely to convict in these cases than non-DV court judges. On average, DV case defendants heard in specialized courts are convicted of 13–20 percent fewer charges. Specialized judges are three to five percentage points (7–10 percent) less likely to incarcerate DV defendants relative to non-DV judges, although the effect is not precisely estimated. We do not find strong evidence that maximum sentence lengths differ between specialized and nonspecialized judges, although the expected maximum sentence is roughly ten days shorter for specialized courts due to their reduced convictions. Figure 3 plots the assigned sentences for charges in our sample conditional on conviction. The distribution is similar across courtroom types, suggesting that DV judges decline to convict at the same rate in all types of cases, rather than in only, for example, the least serious or weakest cases.
DV Judge Decisions in DV Cases
Distribution of Incarceration Time
Notes: The figure shows the distribution of observed sentences for charges in our sample, conditional on conviction. Domestic violence courts frequently assign short sentences, but the overall distribution is similar across courtroom types.
We find that DV defendants are more likely to be acquitted by DV courts than non-DV courts, an unexpected result given the specialized courts’ victim-centered focus. One explanation is that DV judges are less punitive than their peers; however, even allowing for differences in how DV judges adjudicate non-DV cases (Panel B), they remain less likely to convict DV offenders relative to non-DV offenders. We find strong statistical evidence that DV defendants assigned to DV courts can expect less punitive sentences than they would in traditional courts. For DV defendants, DV courts offer a 10 percent reduction in conviction probability relative to misdemeanors on average, a slightly larger but not statistically distinguishable effect than that shown in Panel A. We find qualitatively similar and more precisely estimated reductions in number of convicted charges, a 16 percent reduction in incarcerations for DV cases (relative to the overall incarceration rate) heard in specialized courts versus general courts, and a two-week shorter expected sentence. In other words, our DID results suggest that DV judges are specifically less likely to convict in DV cases relative to their decision-making in non-DV cases.
1. Low-discretion DV cases
Our findings are not clearly consistent with the historical motivation for creating specialized courts—namely, to increase successful prosecution rate of DV crimes. We therefore examine other court outcomes and case subsamples to help understand our Table 3 estimates.
Violations of an order of protection (VOP) are a type of contempt of court charge unique to DV cases. In VOP cases, judicial discretion plays a diminished role in both fact finding and sentencing. Similar to DUI cases, which require evidence of a specific blood alcohol content, VOP case defendants either have or have not contacted the individual in violation of the order; context and intent do not affect legal culpability as it might in a DV assault case. The penalty for each VOP violation is statutorily defined as ten days to serve, day-for-day; in other words, the sentence cannot be shortened for “good behavior” or other stipulations.
We therefore next focus on jail docket cases including only VOP charges, where judges only decide how many times a VOP occurs in each case.24 As shown in Table 4, Panels A and B, we find that specialized and nonspecialized judges make statistically identical decisions in low-discretion VOP cases. Violations of an order of protection cases may be more likely to end in convictions and carceral outcomes in DV courts than non-DV courts, but as in domestic assault cases, differences in judicial taste appear to drive the outcomes, with specialized judges being more likely to incarcerate. Taken as a whole, the result suggests judicial discretion may a role in DV judges making pronounced choices in their area of specialty; when that discretion is limited, court outcomes are similar across all judges, and point estimates may suggest that specialized judges are more punitive in low-discretion decisions.
DV Judge Decisions in Violations of Order of Protection (VOP) Cases
2. Judge-specific fixed effects
Our estimate of court specialization’s impact on our outcomes of interest essentially represents the difference in judge fixed effects across the general and specialized groups. In Figure 4, we present corollary estimates of Table 3, Panel B, Column 2, where for each model we either (i) compare one judge to all others or (ii) exclude a judge from the sample. In addition to the two specialized DV judges, two additional judges, Moreland and Holt, presided over an order of protection (OP) docket during our analysis period. The OP docket contains all civil proceedings in which a petitioner (victim) asks the court to prohibit a respondent (aggressor) from contacting them. While Moreland and Holt were not formally designated as DV judges, by having the OP docket as part of their responsibilities, Moreland and Holt may have gained expertise on DV cases that their peers did not possess.25
Conviction Rates
Notes: Panel A shows conviction rates in DV courts by estimating 12 models, each identifying a separate fixed effect for each judge relative to all others. We indicate the estimated fixed effect with a circle and 95 percent confidence interval with a dashed line. For each court, we estimate a difference-in-differences model (δdd) from the following equation: Convictionicjt = α0 + α1(DV Court)j + α2(DV Case)c + δdd (DV Court × DV Case)cj + λt + uicjt. Panel B plots the estimated coefficient on the effects of DV courts when one non-DV judge is excluded. Each reported coefficient δdd is estimated using Equation 3. Percentage changes from baseline are in parentheses.
Relative to other judges and relative to non-DV misdemeanor cases, the two judges in specialized DV courts are less likely to convict in DV cases, except when compared to Moreland, who is 7.7 percent less likely to convict in DV cases than his peers. One interpretation of this result is that experience in a specific type of case, rather than any training per se, may contribute to what makes specialization different. When we sequentially exclude one judge from the sample, our estimates range from a seven to 11 percentage point reduction in conviction probability, and the values are always statistically distinguishable from zero.
3. Broad definition of DV cases
As a final robustness test, we broaden our DV definition to include other charges commonly associated with domestic crime. Based on case facts not observable to researchers (like the relationship between the defendant and victim), such charges might be routed to DV court when the defendant posts bond. Misdemeanor assault (§39-13-101) is particularly likely to include an unobserved-to-the-researcher DV element, as Tennessee did not create its specific law against domestic assault (§39-13-111) until 2000. Figure 5 shows the frequency with which general (§39-13-101) and domestic assault charges are filed in court (§39-13-111), revealing evidence that the two charges may be used to describe the same criminal act and used interchangeably. The practice declines over time as the legal change is normalized in the criminal justice system.26
Total DV and Assault Cases Filed in General Sessions Court
Notes: Cases aggregated by month.
As shown in Table 5, we find qualitatively similar estimates of court specialization’s impact on the new sample, although the results are somewhat attenuated and less precise. “Possibly DV” cases heard in DV courts are less likely to result in convictions, particularly relative to other misdemeanors heard in the same court. Domestic violence judges are also more likely to impose incarceration sentences in DV (§39-13-111) and assault (§39-13-101) cases, but the result appears to be a judge fixed effect, rather than a decision specific to DV or possibly DV cases.
DV Judge Decision-Making in Assault Cases
4. Who is convicted in DV court?
Our finding that DV defendants assigned to DV courts are less likely to be convicted on average does not imply that all DV defendants are equally less likely to be convicted in DV court. In Table 6, we characterize defendants on the jail docket convicted and not convicted by DV and general judges. As in Table 1, we present unconditional mean values for defendants assigned to traditional courts and predicted means for defendants assigned to DV courts, accounting for time (year and week of the year) fixed effects. Cases involving multiple charges are more likely to end in conviction in both courts, and defendants with previous assault records are more likely to be convicted, although the effect is only marginally significant in DV courts. Young DV defendants appear more likely to be assigned to traditional courts than specialized courts—recall that we found that on average DV defendants in DV court were roughly 1.5 years older and had one additional previous conviction. Defendants convicted by DV judges are almost 2.5 years older on average than those convicted by traditional judges and have appeared in court two additional times. In contrast, defendants not convicted in either court appear to be more similar in terms of demographics and criminal history.
Who Is Convicted in DV Court
Overall, this shift in average values implies that DV judges appear to be less willing to convict defendants with less previous criminal justice contact, perhaps entirely due to their age. In Figure 6, we confirm the result by reestimating our baseline model for defendants in four age categories. A reduced conviction rate for the youngest defendants assigned to DV versus general courts drives the overall lower conviction rate in specialized courts.
DV Court’s Estimated Effect by Defendant Age
Notes: Figure 6 plots the estimated coefficient on the effects of DV courts for defendants of different ages. Each reported coefficient δdd is estimated using Equation 3. Percentage changes from baseline are in parentheses.
C. Specialized Courts and Future Domestic Violence
1. Domestic violence victims’ contact with police
If DV judges are less likely than general judges to convict young DV defendants with short criminal histories than general judges, what are the consequences for subsequent recidivism? We measure recidivism in multiple ways. First, we use a victim-centered definition. Are DV victims assigned to DV courts differentially likely to contact police in the future relative to those assigned to general courts?
In 58 percent of our observed DV cases, victim names are included. We compare the names in our court data with complainant information in all incidents recorded by the Metropolitan Nashville Police Department from 2000–2016. In non-DV cases, which in our sample are primarily drug offenses, the victim is generally listed as the “State of Tennessee,” meaning we cannot estimate Equation 3 for the outcomes, as it requires comparing revictimization between DV and non-DV cases.27
Table 7 shows our results using the victim-centered recidivism approach. Though we cannot distinguish between a judge fixed effect and specialization’s impact, we find that DV victims whose cases are assigned to DV judges are ten to 12 percentage points (18–22 percent) less likely to contact Nashville police, for any reason, in the three years following initial adjudication relative to DV victims assigned to general court. We find the specialized court leads to a similar 11–12 percentage point reduction, just over 20 percent of the 46 percent mean, in the probability of victims filing a police report for an assault within three years.
DV Courts’ Effect on Revictimization: DV Cases in DV Court versus DV Cases in General Court
The observed reduction could be due to increased victim safety or victims having decreased trust in courts, potentially due to their original assailant not being convicted. The results could also be affected by differential mobility, which we will address with our offender-based definition of recidivism (Section V.C.2). We find no strong evidence for substantive changes in whether DV victims are later involved in cases where an arrest is made (that is, when authorities believe a crime has occurred), although our point estimates are negative and correspond with a 5–10 percent reduction. We also analyze the likelihood that police say a victim refuses to cooperate in subsequent incident reports. Victims previously assigned to DV courts are 4.7–5.6 percentage points (69–82 percent) less likely to be recorded as refusing to cooperate with police at subsequent crime scenes. Estimating the probability of victim cooperation conditional on being involved in a future incident does not substantively change our results. This finding is important for two reasons. First, successful prosecution of DV crimes often relies on victim cooperation, and second, it signals that, although overall conviction rates are decreased, DV judges may increase future victim participation with the criminal justice system.
2. Domestic violence offenders’ criminal justice system contact
We next define recidivism based on whether defendants have future arrests by the Metropolitan Nashville Police Department or felony or misdemeanor charges filed against them in the three years after initial adjudication. Figure 3 shows that misdemeanor offenses are generally associated with sentences no more than three months long, meaning the scope for differential incapacitation affecting our recidivism results over three years is relatively small.
Table 8 presents our estimates of court specialization’s impact on recidivism defined as an offender either being arrested by Nashville police (Columns 1, 3, and 5) or appearing in general sessions court (Columns 2, 4, 6, and 7) within three years for a DV assault charge (felony or misdemeanor), any assault charge (felony or misdemeanor), any criminal charge (felony or misdemeanor), or any DV-related felony charge. We present results only from models including defendant and case controls.28
DV Courts’ Impact on Future Criminal Justice Contact
In Panel A, we find a negative, statistically imprecise relationship between whether a DV offender is assigned to a DV court and various rearrest measures for DV assault or assault generally.29 We also find a negative, statistically imprecise reduction in the probability of future DV-related felony charges, which is not clearly consistent with acquitted defendants escalating criminal behavior. Whether using arrests or court charges, the point estimates correspond to a roughly 10 percent reduction in criminal justice contact, but we cannot reject large reductions or small increases in the defendant-focused recidivism measure.
In DV cases where we observe the victim’s name, we can also test whether a DV defendant assigned to DV court is differentially likely to be charged with a future violent crime against the same victim, which we observe after 16 percent of all DV cases over three years, and after 22 percent of all DV cases over ten years. The point estimates of this effect are highly imprecise, and consistent with up to a 40 percent (35 percent) increase in same-victim recidivism over three (ten) years and a 14 percent (13 percent) reduction in same-victim recidivism over three (ten) years.30
When we compare DV defendants heard in DV or general courts to misdemeanants assigned to DV or general courts in a DID framework (Table 8, Panel B), we find similarly imprecise estimates, although we observe that defendants charged with DV offenses are more likely than those charged with non-DV offenses to be rearrested for assault. In contrast with our victim-centered results, the arrest and court data do not reveal strong evidence about specialization’s impact on the likelihood that a defendant is charged with a future crime.31 The similarity of our results across these two measures generally rules out the that the differential mobility of treated offenders—specifically to places outside of Nashville—is biasing our arrest results.
3. Appearing before a judge
Does judicial action actually drive the differences in recidivism outcomes we observe? Cases dismissed by prosecutors never appear before a judge, but we find little evidence that prosecutors are differentially likely to drop all charges in DV cases assigned to DV judges. If DV court exposure impacts our outcomes of interest, we would expect the differences in victim-oriented recidivism to be largest for victims encountering DV judges. In Table 9, we compare our victim-centered outcomes between cases assigned to general and DV courts, distinguishing between cases never appearing before a judge and those appearing before a judge only to accept a plea. We report only unconditional means due to the small sample size; these findings should be interpreted as suggestive only.
Post-Case Outcomes by Court Assignment and Prosecutorial Dismissal
Table 9 shows that the observed differences in outcomes are largest for victims whose cases are prosecuted, meaning the victim and offender interact with the DV court’s judicial component. Given that DV crimes often take place in private settings with few nonfamilial witnesses, most of the cases can only be prosecuted with the victim present. Appearing in court may matter for the victims regardless of case outcome because victims may hear their judge instruct defendants about their past or future behavior and may be addressed themselves.
In prosecuted cases, again a potential proxy for victims interacting with the judge, 54 percent of victims assigned to non-DV courts contact the police in the future, regardless of the extent to which the case goes before a judge. When assigned to a DV court, 57 percent of victims contact the police again after their cases are dismissed versus 35 percent of victims whose cases appeared in front of the DV judge. Domestic violence victims assigned to DV courts are less than half as likely to be involved in a future incident where the police make an arrest. Unconditionally, DV victims interacting with DV courts are one percentage point (3 percent versus 2 percent) more likely to contact and cooperate with police in the future than DV victims assigned to general court. Conditional on a reported incident occurring, DV victims whose previous cases involved a DV judges are two percentage points (33 percent) more likely to cooperate with police than DV victims who were assigned to DV court, but whose previous cases were dismissed prior to any court appearances. We observe no difference in outcomes when DV cases are assigned to general courts, implying that whether a victim or defendant actually sees a non-DV judge is a poor predictor of post-case behavior.
The observed patterns appear to be unique to our victim-centered outcomes. We find that both DV and non-DV courts have, substantively and statistically, the same offender-centered recidivism measures regardless of judge interaction, despite DV judges being less likely to convict or incarcerate their defendants.
VI. Conclusion
Specialized criminal courts are increasingly popular among local governments to address crimes involving victims or offenders with needs believed to render traditional adjudication and incarceration ineffective. Our analysis of court, arrest, and incident data in Davidson County and Nashville, TN suggests that such specialization may reduce both crime and punishment. More specifically, our observed potential reduction in revictimization and failure to find evidence of changes in recorded recidivism suggest that DV courts may reduce type 1 errors in misdemeanor courts (by declining to convict “nonguilty” defendants who will not commit future offenses) without increasing type 2 errors (the rate at which offenders are not punished). The finding aligns with recent evidence on the benefits of declining to prosecute first-time misdemeanants in Suffolk County, Massachusetts (Agan, Doleac, and Harvey 2023) and Harris County, Texas (Mueller-Smith and Schnepel 2021). We find some suggestive evidence that repeat experience with a particular type of case may be important for judicial decision-making and that limiting the scope for judicial decision-making reduces the impact of specialization.
Of course, one possibility is that the reduced use of conviction and incarceration in DV cases may reduce the deterrent effect of the criminal justice system. While this is not the focus of our study, we can shed some suggestive light on this issue by examining how arrests and charges for DV changed after 2006, when the DV jail docket expanded. After this date, the probability that a DV case appeared in DV court went from two in 11 to one. As we show in Online Appendix Figure A4, there is little evidence that the associated reduction in expected formal punishment was associated with an increase in the number of DV arrests or misdemeanor charges per 1,000 Davidson County residents.
Future work examining why victims contact police should explore whether our observed DV court impacts on future behavior are truly the result of increased victim cooperation with police and prosecutors, as our data suggest. This quantitative result could be consistent with DV courts’ goal of improving victim outcomes; however, it could also suggest a reluctance to contact police in marginal domestic incidents. While we do not observe evidence suggesting victims assigned to DV court wait to contact police until violence escalates, the misaligned outcome does warrant additional research.
Acknowledgments
The authors thank Esther Arenas-Arroyo, Deborah Epstein, J.J. Prescott, CarlyWill Sloan, Jane Stoever, Jennifer Doleac, and two anonymous reviewers for helpful comments and suggestions. They appreciate feedback from the General Sessions Court of Metropolitan Nashville and Davidson County and participants at the 2021 Chicago/LSE Conference on the Economics of Crime, the European Society for Population Economics 2021 Conference 2020 NBER Summer Institute, the 2020 Southern Economics Association Annual Meetings, the 2020 APPAM Fall Research Conferences, and UCI’s Initiative to End Family Violence Research Colloquium. The authors also thank Meghan Ballard for excellent research assistance and Negar Ghorbani and Arian Akhavan Niaki for extensive data collection assistance. They acknowledge financial support from the UCI Initiative to End Family Violence. All opinions and errors are those of the author. This paper uses restricted data obtained via MOU from the Nashville Police Department. Identifying information in this restricted data is necessary to access publicly available data from the General Sessions Court of Metropolitan Nashville and Davidson County. A zero-observation data set and all code necessary to replicate the results in the paper are available on the Harvard Dataverse (https://doi.org/10.7910/DVN/ODVENA).
Footnotes
↵1. Strong, Rantala, and Kyckelhahn (2016) estimate that as of 2012, there were 170 therapeutic DV Problem Solving Courts in the United States. There are 3,244 counties or county equivalents in the US. If each county represents one court system (a conservative estimate), then only 5 percent of US courts handle DV cases in a therapeutic system. Moreover, even in systems with a therapeutic DV system, only certain DV cases meet the selection criteria for inclusion. By our estimates, therefore, 95 percent of US courts either have no DV court specialization or have the ability to implement one within their criminal court system.
↵2. The judges determine if there is probable cause that a defendant committed a felony.
↵3. For the DV docket’s history, visit Nashville.gov.
↵4. The “DV-related” designation can be given by any court agent but is typically applied by an arresting police officer. Some criminal charges in Tennessee are unique to DV. The state maintains a separate statute for DV (§39-13-111) and non-DV (§39-13-101) assault. Although the statutes’ penalties and legal requirements are identical, the separate charges allow researchers to identify at least a subset of DV assaults. Other misdemeanors not classified as assault, such as vandalism or forcibly entering a former romantic partner’s home, do not have separate DV and non-DV statutes. In such cases, including when non-DV assault (§39-13-101) is reported, the police officer executing the warrant must indicate if the victim and defendant share a domestic relationship. For example, an assault occurring between a dating couple would be formally charged as “assault” (§39-13-101) but routed and staffed as a DV case. The information is used to assign cases to DV courts but not available to researchers.
↵5. Two additional judges, Aaron Holt and Casey Moreland, were not elected DV judges but heard requests for protection orders. In our main empirical specification, we treat Holt and Moreland as general sessions judges, but we explore how our estimates change if we examine each judge’s decision-making separately.
↵6. For example, Robinson has received the Nashville Coalition Against Domestic Violence’s Award of Excellence and Tennessee Task Force Against Domestic Violence’s Judge Wheatcraft Award.
↵7. Davidson County domestic violence probation officers attend training at the local level conducted by the YWCA Domestic Violence Shelter, district attorney’s office, and local police department’s domestic violence division. They attend national training and conferences like the American Probation and Parole Association Conference and National College of District Attorney’s National Conference, according to an email correspondence with Davidson County Probation Officer Maria Schaffner September 2021.
↵8. The maximum sentence for a Class A misdemeanor is 11 months and 29 days; judges can apply sentences of varying lengths below the maximum. In DV cases, jail time can be commuted if a defendant completes an in-jail Batterers Intervention Program. Judges can also choose between various conditions, including: (i) supervised probation, (ii) stay of execution (SOE), (iii) court ordered DV treatment, (iv) stay-away and no-contact orders, (v) community service, (vi) weapons forfeit orders, and (vii) victim restitution. Probation and SOE orders cannot exceed maximum jail time. Should probation or SOE orders be revoked, defendants serve their time at a predetermined percentage (30 percent, 40 percent, etc.).
↵9. The commissioner also decides whether to issue a warrant after a warrant-less arrest. The warrant contains the offender’s demographic information (for example, name, gender, date of birth, race).
↵10. Cases can be set beyond the tenth day at the defense’s request or due to circumstances like a pending mental health evaluation. If the tenth day is on a weekend, the next court date is typically the following Monday.
↵11. General Sessions judges work other dockets when they are not scheduled for the misdemeanor jail docket. On weeks DV judges are not assigned to the misdemeanor docket, they may hear cases on the felony jail docket or DV bond docket. Non-DV judges rotate through the felony jail, non-DV bond, traffic, civil, mental health, and forfeiture dockets. All judges have weeks with no scheduled cases.
↵12. The non-judicial DV services were available in all jail docket cases; our estimates therefore do not reflect the impact of specialized defense, prosecution, or victims’ services on case outcomes.
↵13. While judges may develop expertise from presiding over certain types of cases during their career, we are unaware of any jurisdiction where policy randomly compels judges to preside over specialized courts against their will. Our results would not generalize to such a situation.
↵14. According to data reported to the State of Tennessee (https://www.tn.gov/tbi/crime-issues/crime-issues/crime-statistics.html, accessed January 28, 2026), the Nashville Police Department makes approximately 70 percent of all arrests in Davidson County.
↵15. Most of the incidents are drug offenses, with the victim listed as the “State of Tennessee.” Victim information is included in most incidents allegedly involving domestic violence (§39-13-111).
↵16. Prior to 2000, domestic assault was charged as “assault.” As shown in Table 6, our main results are robust to defining “DV cases” as all misdemeanor assault charges. All our results are robust to varying definitions of DV, as well as analysis at the charge, rather than case, level. Additional tables and figures are available on request.
↵17. The statute defines domestic assault as a misdemeanor assault where the adult or minor victim is related to the defendant as a (i) current or former spouse, (ii) current or former cohabiter, (iii) current or former sexual partner, (iv) relative by blood or adoption, (v) current or former relation by marriage, or (vi) child of someone whose relationship with the defendant is described previously.
↵18. Judges technically decide on charges, rather than cases. However, charge bargaining during adjudication means case-level conviction and sentencing rates are strongly correlated within cases. Recidivism is also best examined at the case level. Still, estimating our results at the charge level and allowing for arbitrary outcome correlations within defendants (and cases) yields identical conclusions.
↵19. Including the 178 female defendants does not substantively change our results.
↵20. Online Appendix Figure A3 shows the full distribution of defendant age in specialized and nonspecialized courts. The 1.5 year mean age difference is driven by a small cluster of young defendants heard in general courts, creating a more pronounced bimodal age distribution relative to DV courts. Otherwise, the age distributions are nearly identical.
↵21. We treat each case, which may involve one or more charges, as inclusive of all the defendant’s charges. We also run our analyses at the charge level, and our results are robust to the unit of analysis.
↵22. Note that the random assignment of cases to judges, confirmed for observable characteristics by our balance tests, means we find no correlation in case or defendant characteristics within judges. While there is surely correlation in judicial decision-making within judges, the absence of correlation in regressors within a randomly assigned judge indicates that not clustering at the judge level does not introduce bias in our estimated standard errors (Cameron and Miller 2015). Our estimates of robust standard errors are qualitatively identical to standard errors clustered at the judge level. With only 12 judges in our sample, estimated standard errors clustered at the judge level are unlikely to have converged to the correct standard errors (Cameron, Gelbach, and Miller 2008).
↵23. We exclude cases involving assault, stalking, harassment, vandalism, cruelty to animals, and VOP charges, all commonly associated with DV and assigned to DV courts even without formal domestic assault (§39-13-111).
↵24. A single VOP warrant may allege multiple violations. A warrant with one violation would result in ten days in jail, a warrant with two violations would result in 20 days in jail, and so forth. We do not have the text of the warrant available and, therefore, cannot know how many violations occur on each warrant in our data.
↵25. Nashville and Davidson County assigned the two additional GS divisions to hear OP petitions due to urgency and volume. When petitioners seek OPs, the court grants a temporary order that could preclude respondents from living in their residence and contacting their children. The order in some cases could prevent a respondent from possessing a firearm. In 2006, DV divisions began hearing all OP dockets, misdemeanor DV criminal charges, and initial probable cause hearings/evaluations for felony DV criminal charges.
↵26. The gradual increase in DV assaults implies that new DV courts may function differently than established courts, complicating the external validity of identification strategies based on the DV courts’ introduction.
↵27. Our main sample includes DV cases in the jail docket comprising 1445 observations. Following restrictions reduce the number of observations in the revictimization sample to 836 observations: (i) information on the victim’s name is included in incident reports (84.4 percent of previous sample), (ii) victim is female (85.4 percent of previous sample), (iii) the absolute value of age difference between victim and offender is smaller than 15 years (86.5 percent of previous sample), and (iv) keep the first court appearance for victims (92.8 percent of previous sample). Revictimization sample as a percentage of main sample is 58 percent. Note that the case outcome for offenders based on this smaller sample is quantitatively identical to our main sample.
↵28. Defendant and case controls include defendant race, age, age-squared, whether they have previously been charged with a crime, and number of previous charges, as well as judge gender and time since appointment. Results were robust to inclusion of controls, and results without controls are available upon request.
↵29. The probability that any defendant is rearrested by Nashville police (16–20 percent) is roughly the same as that of a victim being involved in a future incident where an arrest occurs (18 percent). The probability of the defendant appearing in Davidson County Court after an arrest by any agency increases. The finding could be due to missing ID numbers, offenders being rearrested by other Davidson County law enforcement agencies, or both.
↵30. These results are available on request.
↵31. We examine the recidivism differences among DV offenders not convicted in a bench trial but still appearing before a judge, either in an acquittal or plea bargain negotiation. The sample size (20 defendants acquitted in a bench trial, 44 pleading guilty on all charges) is too small for formal regression analysis but suggests defendants appearing before a specialized judge may be less likely to recidivate.
- Received March 2022.
- Accepted January 2024.
This open access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: https://jhr.uwpress.org.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.