Journal of Risk and Uncertainty

, Volume 46, Issue 1, pp 51–80

Deterring domestic violence: Do criminal sanctions reduce repeat offenses?

Authors

    • Department of EconomicsDuke University
  • Alyssa C. Platt
    • Duke Global Health & Department of Biostatics and BioinformaticsDuke University
  • Lindsey M. Chepke
    • Department of EconomicsDuke University
  • Claire E. Blevins
    • Clinical Psychology DepartmentVirginia Polytechnic Institute and State University
Article

DOI: 10.1007/s11166-012-9159-z

Cite this article as:
Sloan, F.A., Platt, A.C., Chepke, L.M. et al. J Risk Uncertain (2013) 46: 51. doi:10.1007/s11166-012-9159-z

Abstract

This study presents an empirical analysis of domestic violence case resolution in North Carolina for the years 2004 to 2010. The key hypothesis is that penalties at the level set for domestic violence crimes reduce recidivism (re-arrest on domestic violence charges or conviction in 2 years following an index arrest). We use state court data for all domestic violence-related arrests. Decisions to commit an act of domestic violence are based on a Bayesian process of updating subjective beliefs. Individuals have prior beliefs about penalties for domestic violence based on actual practice in their areas. An individual’s experience with an index arrest leads to belief updating. To address endogeneity of case outcomes, we use an instrumental variables strategy based on decisions of prosecutors and judges assigned to each index arrest in our sample. Contrary to our hypothesis, we find that penalities, at least as set at the current levels, do not deter future arrests and convictions.

Keywords

CrimeDomestic violenceDeterrenceSubjective beliefsProsecutorsJudges

JEL Classification

K14K36K42

Domestic violence (DV) is a major public health issue. Based on self-reported DV data from the National Violence Against Women Survey, 25% of women in the United States experience DV in their lifetimes, which equates to approximately 4.8 million incidents of DV against women annually (Tjaden and Thoennes 2000). Domestic violence accounts for an estimated 1,200 deaths and two million injuries among women each year (Black and Breiding 2008). In one study, the vast majority of DV arrests involved physical injury, happened in the home, in private, and by a closed fist (Schmidt and Steury 1989). Unlike other crimes, there are often several instances of abuse before a complaint is made to a law enforcement agency.

Consequences of DV are not limited to physical injury; the U.S. Centers for Disease Control estimates that the economic loss from DV is $5.8 billion annually, in part due to productivity declines and cost of medical care (Centers for Disease Control 2003). This cost is exclusive of non-economic loss such as for pain and suffering and loss of consortium. Aside from its immediate effects, there is also evidence that DV has long-term adverse effects on children of victims and/or batterers, including child maltreatment (Lee et al. 2004; Chang et al. 2008; Casanueva et al. 2009), delayed progress in school (Hurt et al. 2001), increased likelihood of inflicting emotional and/or physical harm on others including children (Baldry 2003), drug and alcohol abuse as adults (Felitti et al. 1998; Dube et al. 2002), and poor health outcomes as adults (Felitti et al. 1998; Dube et al. 2002).

While DV is not a new phenomenon, laws against it are fairly recent, having evolved in the last few decades of the 20th century. States and localities have implemented several policies aimed at reducing rates of DV. These policies include mandatory arrests, mandatory minimum sentences, limits to the discretion of the district attorney to drop DV cases, and implementing various types of specialty courts. Rates of prosecution for DV vary geographically within the U.S., but a large share of arrests do not result in prosecution (Klein 2004). One study reports DV prosecution rates as low as 10% of misdemeanor DV cases (Fagan 1996). Reasons for non-prosecution include lack of evidence to meet statutory burden of proof, refusal by the victim to participate in the prosecution, high workload, and/or lack of preparation on the part of prosecutors and judges (Robbins 1999; Klein 2004). A larger problem is that most cases of DV go unreported; estimates range from 33% to 72% (Rennison and Welchans 2000; Felson et al. 2002). Recent estimates of arrest rates conditional on police involvement typically range from 0.3 to 0.4.1 Thus, probabilities of arrest conditional on occurrence of DV are probably about 0.15 to 0.2.

This study presents an empirical analysis of case resolution involving an arrest on a charge of DV in one U.S. state, North Carolina. Our key hypothesis is that recidivism is reduced by the imposition of penalties set at the level they are for DV-related crimes. Recidivism in our study is defined as a re-arrest on DV charges or, alternatively, conviction in 2 years following an index arrest for a DV charge. The index arrest is the first arrest for DV in 2007. We use state Administrative Office of the Courts (AOC) data for all arrests flagged by state courts as DV-related during 2004-February 2010. A look-back period of 3 years from the index offense date in 2007 is used to define variables for previous arrests and convictions for DV-related offenses. The data in the 2 years following the index arrest are used for analysis of recidivism.

Theoretically, decisions about whether or not to commit an act of DV are based on a Bayesian process of updating subjective beliefs. Individuals have prior beliefs about the consequences of being arrested for DV based on the actual penalties in their areas. An individual’s own experience with an index arrest leads to belief updating. To deal with endogeneity of case outcomes, we use an instrumental variables (IV) strategy based on decisions of prosecutors and judges assigned to each index arrest for DV in our sample. Empirical studies of deterrence face many challenges: (1) partial reporting of crimes; (2) reverse causality, for example, the deterrent, particularly in studies using aggregate data, may depend on the crime rate as well as the reverse; and (3) incapacitation may be indistinguishable from deterrence (Ehrlich 1981; Levitt and Miles 2007). The first is a challenge for our study. Even though we use data at the level of the individual arrest, heterogeneity in the offense and the offender remain a source of endogeneity. Incapacitation is not an issue given that jail terms are generally quite short relative to our follow-up period. Deterrence reflects the probability of arrest, fines and jail terms, statutory regulations, and judicial policies.

This study makes three major improvements over existing studies of the relationship between DV penalties and recidivism (Tolman and Weisz 1995; Thistlethwaite et al. 1998; Gross et al. 2000; Wooldredge and Thistlethwaite 2002; Ventura and Davis 2005). First, we model the Bayesian updating process of beliefs as reflected in changes in the probability of a repeat arrest for DV-related offense, distinguishing beliefs about sanctions for such offenses obtained from the experiences of others as reflected in area probabilities of sanctions and learning from one’s own experience with an arrest on a DV-related charge. Second, we use exogenous variation in the intensity of enforcement by using assignment of arrests to individual prosecutors and judges, thus avoiding endogeneity of sanctions that may arise due to omitted heterogeneity in characteristics in offenders and offenses. Third, we use a large sample of individual-level data allowing us to control for some important characteristics of offenders and offenses included in previous research to assess deterrence of criminal acts (Paternoster 2010; Apel 2012; Charles and Durlauf 2012). Deterrence is likely to be context specific, for example, varying to the extent that the laws are enforced. Increasing the maximum sentence for a crime is unlikely to lead to deterrence if the probability of arrest is low or once arrested, prosecutors decline to prosecute a high proportion of arrests.

Section 1 describes the Bayesian updating process of beliefs about the probability that sanctions will be imposed conditional on committing a criminal act and reviews prior literature on Bayesian updating as it applies to crime. Section 2 describes the data used in this study. Section 3 presents specifications of equations used to estimate parameters. Section 4 presents our findings. Section 5 discusses our findings in light of previous literature, strengths and weaknesses of our empirical analysis, and implications for public policy. Contrary to our hypothesis, we find that penalties, at least those set at the current levels, do not deter future arrests and convictions.

1 Bayesian updating of risk perceptions

1.1 Conceptual framework

Imposition of penalties may deter crime and arrest through several pathways. One channel is through affecting the probability of a penalty conditional on committing a crime. In this study, all persons were arrested and charged with a DV offense. Thus, relevant probabilities involve: (1) whether or not the person is arrested, conditional on committing a crime; (2) whether or not the person is prosecuted on at least one of the initial charges that led to the arrest—here the index arrest; (3) whether or not the person is convicted conditional on being arrested; and (4) whether or not the person is sentenced to jail conditional on having been convicted. As beliefs about the probabilities increase, the propensity of committing another offense presumably decreases. As documented below, there is little variation in jail terms and fines conditional on being convicted in our data on DV arrests.

We assume that the person contemplating committing a crime is a Bayesian updater. The person starts with a prior belief about the probabilities of various outcomes occurring conditional on choices he/she makes. These probabilities are updated as the person receives new information about the probabilities based on personal experiences.

Let pj be the person’s prior belief about outcome j, arrest, prosecution, conviction, and penalty and qj be the person’s subjective belief about the probability of outcome j after learning about outcomes from a particular incident, i.e., arrest (sj).2 The parameters γj and εj represent precision associated with the prior assessment of the probability and with information the individual obtains from personal experience, respectively. Prior beliefs may reflect experiences of other persons in the area; individuals may gain more information from personal experiences about how police and prosecutorial enforcement and judicial processes work in the person’s area. We employ a formulation used in many studies of individual risk perceptions (Viscusi and O’Connor 1984; Smith et al. 2001; Viscusi and Evans 2006; Lochner 2007).
$$ {q_j}=\frac{{\left( {{\gamma_j}{p_j}+{\varepsilon_j}{s_j}} \right)}}{{{\gamma_j}+{\varepsilon_j}}}=\left( {\gamma {\prime_j}{p_j}+\varepsilon {\prime_j}{s_j}} \right), $$
(1)
where
$$ \gamma {\prime_j}=\frac{{{\gamma_j}}}{{\left( {{\gamma_j}+{\varepsilon_j}} \right)}}\;\mathrm{and}\;\varepsilon \prime j=\frac{{{\varepsilon_j}}}{{\left( {{\gamma_j}+{\varepsilon_j}} \right)}}. $$

In our empirical analysis of the probability of recidivism, the pj are defined as prosecutorial or judicial district-specific mean probabilities of prosecution conditional on arrest (j = 1), conviction conditional on prosecution (j = 2), and sentences to jail for those who are convicted (j = 3). There are 42 prosecutorial and 40 judicial districts in NC, a state with a population of 9.5 million in 2010 (United States Census Bureau 2010). We assume that individuals’ prior subjective beliefs about penalties are a function of these district-specific probabilities, which individuals learn about from friends, relatives, colleagues, the media, and sources other than the individual’s own experience with arrests. Cet. par., high probabilities of adverse outcomes with pj, measured at the district level, are hypothesized to reduce recidivism. Personal experiences with the law should lead to updating of prior beliefs. In sum, by affecting subjective beliefs, higher values of the pj should reduce probabilities of a repeat arrest. Adverse personal experiences with a prior arrest as reflected in the sj should raise qj, making recidivism even less likely.

1.2 Prior studies of the Bayesian updating process in this context

There are few prior studies of Bayesian updating in the context of crime. In one such study, Lochner (2007) had direct retrospective measures of qj and sj from the 1997 cohort of the National Longitudinal Survey of Youth. He found that the qj, where j is the subjective probability of arrest, is lower among persons who engaged in criminal activity, as predicted by deterrence theory and belief updating. Yet the qj were only weakly related to county-measures of arrest-per-crime rates. Although the findings lend support to the notion that subjective beliefs are affected by actual events the person experiences, the study did not empirically evaluate the link between risk perceptions and actual arrest or crime rates. The survey data were based on respondent self-report. Thus, the precision as to timing of arrests and resolution of each arrest available from administrative arrest records was lacking.

An earlier study by Pogarsky and Piquero (2003) sought to explain two patterns reported in prior studies. One is that persons who were punished were more, not less, likely to offend in the future. The second is that subjective probabilities of punishment conditional on having committing a crime are lower for those who have received punishment than those who have not. The authors distinguished between a “selection effect” and a “resetting effect.” The former is simply that recidivists have persistent beliefs that the probability of punishment is lower. Alternatively, the resetting hypothesis is based on the gambler’s fallacy. This fallacy is that previous failures, e.g., losing a bet, indicate to the gambler an increased probability of future success, i.e., winning the bet on subsequent attempts. The fallacy is in thinking that the events are negatively correlated, when in fact they are independent. Data for Pogarsky and Piquero’s empirical analysis came from a survey of students. The analysis revealed mixed results on the importance of the selection versus the resetting hypothesis. Although the questions dealt with driving while intoxicated, a salient issue to many students, generalizability to adult offenses such as domestic violence and offenses with identifiable victims is questionable. For many crimes, e.g., drug possession or driving while intoxicated, there is no identifiable victim in the vast majority of cases. Or if there is a victim, the victim is unknown to the offender prior to the time of the crime.

Subsequently published empirical applications to crime do not find support for the pattern of updating implied by the gambler’s fallacy, but rather find that arrests lead to an increased subjective belief about the probability of arrest (Pogarsky et al. 2004; Matsueda et al. 2006; Anwar and Loughran 2011). Matsueda et al. (2006) infer from their results that youth who commit crimes initially overestimate the probability of an arrest but after eluding arrest, adjust their subjective beliefs downward. However, Pogarsky et al. (2005), using different data from Pogarsky et al. (2004), report that having been arrested between interviews is unrelated to perceptual change although the coefficients are consistently positive. All of these studies used survey data on youths.

There is also empirical support for the notion of persistence in beliefs about adverse consequences of committing crimes, with one type persistently attaching lower probabilities to punishment conditional on committing a criminal act (Nagin and Paternoster 1993; Loughran et al. 2009). None of this evidence pertains directly to DV; rather the evidence pertains to youths and young adults in “victimless” crimes often committed for financial gain.

While empirical support for the gambler’s fallacy is lacking and heterogeneity among offenders in subjective beliefs plausibly explains some of the prior results, we cannot rule out the probability that Bayesian updating in the context of DV leads persons sanctioned by having been arrested or convicted of this type of crime to reduce their subjective probabilities of being sanctioned. A possible reason is that instead of ceasing to victimize one’s family member, the arrestee may think that if he or she is tougher on the victim, the latter will be less likely to report the offense to public agencies in the future. But this belief is disproven when the victim subsequently reports the offense and there is a repeat arrest. Another possibility is that the arrestee may learn that while the police may make an arrest, the charge is not pursued in the courts in the majority of cases.

2 Data

We obtained access to a unique database for purposes of this study. North Carolina’s Administrative Office of the Courts maintains a database containing information on criminal arrests and case disposition at the charge/individual arrestee level called the Automated Criminal Infractions System (ACIS). ACIS includes each criminal charge organized by the day that the charge was made and lists the NC General Statute Code and offense descriptions. The data are from 2004 through early 2010.

There is no crime of DV in the North Carolina Code; rather criminal infractions that have a DV component have been flagged by individual courts since 2004 when there is evidence of DV. All charges associated with the arrest are also flagged, whether or not they constitute DV. Most of the 100 NC counties flag for DV charges; the non-reporting counties are typically less populous counties.

These data have several advantages. First, they allow analysis at the level of the individual arrestee as opposed to an aggregate at the county or state level, as in many studies of criminal behavior. The data contain identifying information, including the person’s name, birth date, gender, and last four digits of the person’s Social Security number (SSN).

Second, with created unique identifiers, ACIS data permit tracking of offenses and associated outcomes over time. Information on how the defendant arrived in court (for example, citation, warrant, criminal summons), the type of legal representation the defendant had (e.g., court-appointed, public defender, waived, privately retained), and method of disposition of charge, including verdict and sentences (e.g., jail terms, fines, community service hours). The method of disposition includes trial or dismissal. Third, for a category such as DV, the population of offenses is large. North Carolina classifies offenses in broad categories: misdemeanors; felonies; and infractions. Within categories, offenses are classified by severity. Fourth, most states do not have a central standardized system for maintaining court records; data must be obtained from individual courts.

A disadvantage of the data is lack of information on more than a limited number of personal characteristics, e.g., educational attainment, employment, and household income. Further, we have no information on heterogeneity of types of offenders in the domain of subjective beliefs about the chances of being punished for their offenses.

We organize the data into a file for index offenses. An index offense is the first flagged offense occurring on or after January 1, 2007 and up to December 31, 2007. The offenses are organized by offense date, with all offenses occurring on the same date recorded as part of the index offense. We use North Carolina structured sentencing policies to categorize offenses in order of severity (North Carolina Sentencing and Policy Advisory Committee 2009). Three categories of punishment are assigned for offenses in the classification system: (1) active punishment (NC Gen. Stat. 15A-1340.11(1)), requiring that the offender be sentenced to jail or prison; (2) intermediate punishment (NC Gen. Stat. 15A-1340.11(6)), requiring a sentence of supervised probation with at least one of the following conditions: special probation, residential program, house arrest with electronic monitoring, intensive supervision, day reporting center, or drug treatment court program; and (3) community punishment (NC Gen. Stat. 15A-1340.11(2)) consisting of any authorized condition of probation except for those defined as intermediate punishments, outpatient drug and alcohol treatment, community service, referral to mental health or substance abuse services, restitution, or fines. For felonies, we reduce the number of classifications to four categories; in decreasing order of severity: 3—class G, H, I offenses: up to a year of active, intermediate or community punishment for minimum prior record level; 2—class E, F offenses: from 1 to 2 years of active or intermediate punishment for minimum prior record level; 1—class A, B1, B2, C, D offenses: 4+ years of active punishment for minimum prior record level; and a fourth category of unclassified offenses. Misdemeanors and traffic offenses are separated into three categories in decreasing order of severity: 2—class A1 and 1 offenses: upper range of community, intermediate, or active punishment from 45 to 60 days; 1—class 2 and 3 offenses: upper range of community punishment from 10 to 30 days; and a third category of unclassified offenses.

3 Specification

3.1 Overview

The major econometric problem in estimating deterrent effects with micro data is that the penalty the individual obtains, sj, may be correlated with unobserved personal characteristics such as the propensity to commit crimes. To the extent this is so, we may observe that people who have adverse experiences with criminal law enforcement are more, not less likely, to engage in repeat offenses just because they have particular attributes unobserved by the researcher. To deal with this issue, we specify a two-equation model. The main equation is for the probability of re-arrest or conviction for a DV-related offense in a two-year period following the index arrest for DV which occurred in 2007.

The main equation is for recidivism:
$$ {r}={\theta_0}+{\theta_1}{{\mathrm{q}}_j}+{\theta_2}X+{\theta_3}Y+{\varepsilon}, $$
(2)
where r is the probability of a re-arrest or conviction during the two-year follow-up period, and X and Y are characteristics of the index arrest, other individual and area characteristics, respectively. Since all sample persons were arrested for the index offense, arrests in the follow-up are rearrests. The conviction during follow-up may or may not be a re-conviction for DV. For example, a person could have had an index arrest but this arrest was not prosecuted. Subsequently, the person could have been arrested and convicted for a DV offense.
The subjective belief about incurring a penalty for DV is a function of sj and pj. Thus (2) can be rewritten as
$$ {r}={\theta_0}+{\theta_{11 }}{s_j}+{\theta_{12 }}{p_j}+{\theta_2}X+{\theta_3}Y+{\varepsilon}. $$
(2')
The first-stage equation is
$$ {s_j}={\phi_0}+{\phi_1}{p_j}+{\phi_2}IV+{\phi_3}X+{\phi_4}Y+{v_j}. $$
(3)

IVs are included in (3) but excluded from (2').

We conduct separate analysis of arrests and convictions during the 2 years following the index arrest on three samples. The first is for all persons with an index arrest with s1 a binary variable set to 1 if the person was prosecuted for the index offense and p1 for the proportion of DV arrests in the prosecutorial district that were prosecuted during 2007. The majority of DV offenders are identified not by patrol and policing, but rather by reports from the victim (Rennison and Welchans 2000). After a suspect is arrested, a prosecutor generally decides whether or not to charge the person with specific crimes. Prosecutors may drop cases by: (1) declining to prosecute; (2) filing charges but subsequently dropping them before trial or requesting that the court dismiss them.

The second analysis is for all persons prosecuted for the index offense with s2 set to 1 if the person was convicted for the index offense, and p2 is the proportion of DV prosecutions in the country’s judicial district that resulted in a conviction in 2007. The third analysis is for all persons convicted of the index offense with s3 equal to 1 if the person was sentenced to jail for the index offense and p3 is the proportion of convictions on DV charges in the county’s judicial district that resulted in a jail sentence. We use the same strategy in our analysis for DV conviction during follow-up.

3.2 Characteristics of the index arrest

Characteristics of the index arrest (part of X) are for: (1) the most serious index charge; (2) legal representation and victim witness for the index arrest; (3) and the most serious DV charge or conviction in the 3 years before the date of the index arrest. To the extent that more serious offenses imply that the individual is more violent and/or crime-prone, the recidivism probability should increase with crime severity. However, more serious DV crimes occur less often. For this reason, persons arrested or convicted of a more serious crime should be less likely to repeat it. In the analysis of re-arrest, we use the most serious index charge. For the conviction analysis, we use the most serious offense of which the person was convicted.

ACIS reports the type of attorney assigned for defense and prosecution. Defense attorneys are categorized as: court-appointed; public defender; privately retained; and waived (defendant represents self, “pro se”). We assign binary variables for each of these categories with privately retained as the omitted reference group. Defendants who retain a private attorney are likely to be relatively affluent. Meanwhile, persons represented by a public defender or court-appointed private attorney must be indigent according to criteria specified by North Carolina statute, and there is no choice of government-supplied attorney except perhaps under very unusual circumstances (e.g., conflict of interest) when the government’s choice can be rejected by the defendant and replaced by another attorney. Private attorneys may exert greater effort in representing the defendant’s interests. For this reason, we expect that clients of private attorneys are less likely to incur penalties than are others. No direct measure of household income is available from ACIS.

ACIS provides a list of witness types for each case: complainant; defense; inactive; law enforcement; complainant non-witness; state; and victim. We include a binary variable to indicate that a certain individual’s case included a victim witness. We expect that the defendants are less likely to be acquitted or have their charges dismissed when there is a victim witness.

We use data from 3 years prior to the index offense date in 2007 as a look-back period to document prior DV offenses. Prior studies indicate that persons with prior criminal records should be more likely to recidivate following the index offense (Kingsnorth 2006; Hilton et al. 2007; e.g., Ahlin et al. 2011). The vast majority of persons with DV crimes had no prior DV-related crimes in the 3 years before the index offense (79.1%).

3.3 Other individual characteristics

We include explanatory variables for gender, age at the date of the index offense, and race/ethnicity—black, Hispanic ethnicity, other race, and white (omitted). Age is coded into several mutually exclusive categories: under 21 (omitted), 21–25, 26–34, 35–44, 45–54, 55–64, and 65+. The median age of sample persons at the time of the index offense is between 26 and 34. Sample persons are all age 18 or older at the time of arrest.

3.4 Area-specific factors

We include the arrest rate in the county in 2007 as an explanatory variable. Cet. par., one expects higher levels of deterrence in counties with higher arrest rates. However, the arrest rate also reflects the area prevalence of DV, and a higher propensity to arrest means that more DV cases end up in court. In analysis not shown, we found that arrest rates were higher in counties with higher rates of hospitalization for women admitted for treatment for DV-type injuries. Also, arrest rates were higher in rural and low-income counties. We also include explanatory variables for mean values for 2007 for the fraction of DV arrests prosecuted, defined for the prosecutorial district in which the arrest for DV occurred, and conviction conditional on prosecution, defined by judicial district of the arrest, as covariates. Prosecutorial and judicial districts overlap but are not identical.

3.5 Instrumental variables

A major econometric problem in microanalysis of recidivism is that researchers do not observe many characteristics of offenders observable to participants in the criminal justice system—police officers, prosecutors, judges, and others that are likely to affect the probability of recidivism. To deal with this issue, we adopt an instrumental variables strategy.

Our key IVs refer to the relative stringency of prosecutors and judges. Although prosecutorial districts have policies regarding prosecution of specific offenses, in the end, the decision of whether or not to prosecute is the individual prosecutor’s decision. Identification is achieved under assumptions that cases are randomly assigned to prosecutors and judges, and prosecutors and judges employ different criteria in deciding whether or not to prosecute and convict persons of DV.

In 2010–2011, there were 681 prosecutors (14.0 per district) and 270 district court judges (6.8 per district) (North Carolina Administrative Office of the Courts 2011). If the cases are randomly assigned, we expect that for prosecutors who decline fewer cases or judges who convict and sentence defendants to jail more often, the marginal case demonstrates less evidence of guilt and offense severity (observed by the prosecutor or judge but unobserved by us) when the prosecutor or judge is stricter. Persons assigned a stricter prosecutor or judge will randomly be assigned a higher probability of a penalty, which introduces a quasi-experimental research design. However, if cases are assigned on the basis of the quality of evidence on guilt and offense severity, variation in fraction of cases leading to prosecution or penalty may just reflect omitted heterogeneity on the offense or offender.

We have no direct evidence on assignment protocols. But several institutional factors make endogeneity of prosecutors or judges improbable. On average, prosecutors in North Carolina spend 19 minutes prosecuting a misdemeanor case other than driving while intoxicated and drug charges (North Carolina Office of Indigent Defense Services 2011). The high volume and low time spent per case speak against much deliberation about choice of prosecutor. Change in venue for criminal cases is highly unlikely as it is in some civil litigation. Nor is it likely that defendants can change the person assigned to prosecute the charge(s). It is possible in North Carolina for a defendant to request a change in the judge, but this must be for a valid reason, such as judge conflict of interest or illness. There are no data on the frequency of judge reassignments, in general or for domestic violence. One reason to suspect that reassignments are not frequent is that only a fifth of DV defendants retain a private attorney. As a matter of state policy, public defenders and court-appointed attorneys are randomly assigned. Court-appointed attorneys in North Carolina are paid for court-appointment defense work on a fixed budget set by the judge. A change in judge is difficult, and requires substantive proof of judicial bias, e.g., financial or familial relationship with the victim. Unlike other states (see e.g., Henning and Feder (2005), which analyzes data from a domestic violence court in Memphis, Tennessee), North Carolina does not have domestic violence courts which assign DV cases to judges who specialize in DV.3

Hence, at the margin, the threshold of evidence on defendant guilt leading to a specific penalty is likely to vary among prosecutors or judges within a district. This introduces variation in the probability that an individual case at a fixed level of evidence of guilt will be prosecuted, convicted or jailed. This approach has been used in previous work (Doyle 2007; Chang et al. 2008; Doyle 2008; Hjalmarsson 2009; Green and Winik 2010). We test for weak instruments; if there are many errors in the data on the identity of prosecutors and judges trying individual cases, the F-values will be low.

ACIS data include a table listing both the prosecuting attorney and the defense attorney for the case. To obtain information for the IV for index arrests resulting in prosecution, we use prosecutor names from this file. Unique identifiers are constructed using a Soundex code for last name, first initial, and prosecutorial district.4 Since prosecutorial districts changed between 2004 and 2010, each attorney is assigned to his or her 2004 district so that the prosecutor identifier would not change during the observational period. We compute the mean fraction of DV arrests leading to prosecution by prosecutor using data from 2004 to 2007. We require that each mean value be based on a minimum of 10 arrests per prosecutor. On average, there are 86.2 arrests per prosecutor in our sample.

The IV for index arrests resulting in prosecution is Zpkd constructed according to
$$ Z{p_{kd }}=\left( {\frac{{{P_{kd }}}}{{{N_{kd }}}}} \right)-\left( {\frac{{{P_d}}}{{{N_d}}}} \right), $$
(4)
where Pkd is the number of DV cases prosecuted by prosecutor k in district d and Nkd is the number of DV arrests assigned to the kth prosecutor in district d. We subtract the rate of prosecution for DV in the district from the kth prosecutor’s DV prosecution rate. The rate of prosecution in the district for DV is an explanatory variable representing p1.

We also construct IVs for the share of prosecuted DV cases that result in convictions by judge and share of convictions with jail sentences by individual judges within districts. ACIS data include an identifier for the judge assigned to a particular case. The identifier consists of the judge’s initials (two or three initials). To increase accuracy of the judge identifier since there may be judges in the same district with the same initials, we link ACIS data to North Carolina election data by district court or superior court district. The election data include the full name of elected judges in the state. Using the election data, we compute initials for each judge and compare the results to judges’ initials from ACIS. If a judge is the only judge in the district with the initials listed, we consider the ACIS identifier to be valid. All valid identifiers for judges within districts are combined with a district identifier to create a unique judge identifier across the state. As with the prosecutor IV, the judge identifiers are defined for 2004 and the criterion of a minimum of 10 observations per mean value is applied here as well. We compute an IV for judges Zjkd using a method analogous to the one defined for prosecutors (Eq. 4). There are 71.0 prosecuted arrests per judge in our sample. Especially since the instruments are normalized by corresponding shares for the districts, it is extremely unlikely that the prosecutor- and judge-specific variables reflect the extent of DV in the district.

We also include an IV for whether or not the county has a no-drop policy for DV. This variable comes from a survey of North Carolina district courts conducted in 2007 (Kim and Starsoneck 2007). Finally, we include an IV for whether it is district policy for the judge to refer defendants to an abuser treatment program.

4 Results

4.1 Descriptive statistics

4.1.1 Disposition of arrests

A total of 29,700 persons were arrested for a DV violation during 2007. Of those persons with an index arrest, i.e., arrested at least once during 2007, the state prosecuted 9,256 persons (31.2%, Fig. 1). Among DV-related arrests leading to prosecutions, the vast majority, 8,304, resulted in a conviction on one or more charges. Of those persons who were convicted, 6,709 received a fine and/or a jail term. In total, 22.6% of those arrested ultimately paid a fine and/or were sentenced to jail. In view of our estimate that only 0.15 to 0.2% of cases involving DV lead to an arrest, the probability of DV resulting in a fine or jail is slightly under 0.04. Community service, and regular and supervised probation without a fine or jail time was extremely rare for DV index arrests (< 1%—not shown in table).
https://static-content.springer.com/image/art%3A10.1007%2Fs11166-012-9159-z/MediaObjects/11166_2012_9159_Fig1_HTML.gif
Fig. 1

Domestic violence, from arrest to penalty

4.1.2 Penalties following conviction

Focusing hereafter on index arrests (first arrest on a DV-related charge in 2007) rather than on total arrests, the majority of persons convicted of an arrest including a domestic violence charge received some jail time at sentencing (Table 1).5 Imposition of fines for DV is relatively rare. The values in the table are penalties as ordered by the court, not fines actually paid or jail time actually served.6 Convicted individuals may receive credit for time served before verdict.
Table 1

Penalties imposed at conviction

 

Penalty conditional on having penalty

Expected penalties

Jail (days)

Offense category

Penalty (%)

Mean

95% confidence interval

Felony category 1 (66)a

89.4

287.3

263.4–311.2

256.8

Felony categories 2/3 (34)

79.4

1047.8

604.3–1491.3

832.0

Misdemeanor category 1 (1,080)

62.7

35.3

31.2–39.3

22.1

Misdemeanor category 2 (6,097)

80.4

80.1

77.6–82.5

64.4

 

Fine ($)

 

Felony category 1 (66)

12.1

187.50

114.2–260.8

22.7

Felony categories 2/3 (34)

17.6

275.00

100.2–449.8

48.4

Misdemeanor category 1 (1,080)

26.6

86.72

78.8–94.6

23.1

Misdemeanor category 2 (6,097)

22.3

123.82

118.8–128.9

27.6

aNumber of observations in parentheses

The probability of any time sentenced to jail decreases with the severity of the felony. Fines for DV are trivial; for example the expected fine for a misdemeanor category 2 conviction is only $27.61. By contrast, the expected jail sentence following arrest, prosecution, and conviction is 64.4 days or about 2 months. The distribution of penalties is highly concentrated around the mean values, as indicated by the 95% confidence intervals. The 95% confidence intervals for misdemeanors are particularly small. Penalties for felonies exhibit more variation, but there are few felonies for DV.

Table 2 describes the most frequent charges associated with index arrests by category of charge. For index offenses with multiple charges and/or convictions, the index offense is classified in the most serious category for which the person was initially charged or, in some analyses, convicted. The most serious charge and/or conviction associated with an index arrest may not have been a DV-related offense. Thus, for example, 10 persons charged with a DV-related offense were also charged with a common law robbery, which appears in the table because this offense is more serious than DV-related offenses as reflected in sentencing guidelines.
Table 2

Most common index charges leading to convictions*

Most serious charge

Most serious conviction

Offense description

Freq.

Percent

Offense description

Freq.

Percent

Felony category 1 (379)

Felony category 1 (66)

 Assault by strangulation

264

69.70

 Habitual misdemeanor assault

37

56.10

 Habitual misdemeanor assault

41

10.80

 Assault by strangulation

14

21.20

 Common law robbery

10

2.64

 DV protective order violation

4

6.06

 DV protective order violation

7

1.85

 

 Felony possession of a schedule 2 substance

6

1.58

 Larceny after breaking and entering

6

1.58

Felony category 2 (188)

Felony category 2 (25)

 Assault with a deadly weapon - serious injury

67

35.60

 Assault with a deadly weapon - serious injury

8

32.00

 Second degree kidnapping

54

28.70

 Assault - serious bodily injury

7

28.00

 Assault - serious bodily injury

25

13.30

 Indecent liberties with a child

6

24.00

 Assault with a deadly weapon - intent to kill

12

6.38

 

 Felonious restraint

12

6.38

 Discharge weapon - occupied property

4

2.13

 Abduction of children

3

1.60

 Attempted first degree burglary

3

1.60

 Indecent liberties with a child

3

1.60

Felony category 3 (48)

Felony category 3 (9)

 Assault with a deadly weapon - intent to kill or seriously injure

11

22.90

 Attempted first degree murder

2

22.20

 First degree kidnapping

7

14.50

 Assault with a deadly weapon - intent to kill or seriously injure

2

22.20

 First degree burglary

6

12.50

 Second degree rape

2

22.20

 First degree sexual offense

6

12.50

 

 Second degree rape

6

12.50

 Attempted first degree murder

2

4.17

 First degree rape

2

4.17

 First degree sexual offense with child

2

4.17

Misdemeanor category 1 (2,968)

Misdemeanor category 1 (1,080)

 Simple assault

2,235

75.30

 Simple assault

713

66.00

 Assault and battery

493

16.60

 Assault and battery

79

7.31

 Harassing phone call

100

3.37

 Resisting a public officer

79

7.31

 Threatening phone call

53

1.79

 Injury to personal property

32

2.96

 Resisting a public officer

29

0.98

 Simple affray

28

2.59

 Second degree trespassing

19

0.64

 Second degree trespassing

27

2.50

 Simple affray

11

0.37

 Harassing phone call

25

2.31

 Disorderly conduct

6

0.20

 Disorderly conduct

24

2.22

 First degree trespassing

5

0.17

 Threatening phone call

19

1.76

 Cyberstalking

4

0.13

 Possession of marijuana up ≤0.5 oz.

7

0.65

Misdemeanor category 2 (19,607)

Misdemeanor category 2 (16,097)

 Assault on a female

12,882

65.70

 Assault on a female

3,902

64.00

 DV protective order violation

2,527

12.90

 DV protective order violation

916

15.00

 Communicating threats

1,554

7.92

 Communicating threats

405

6.65

 Assault with a deadly weapon

973

4.96

 Assault with a deadly weapon

192

3.15

 Domestic criminal trespassing

578

2.95

 Domestic criminal trespassing

171

2.81

 Interfering with emergency communication

243

1.24

 Interfering with emergency communication

85

1.40

 Assault - inflict serious injury

213

1.09

 Assault - inflict serious injury

83

1.36

 Assault by pointing a gun

182

0.93

 Injury to personal property

53

0.87

 Misdemeanor stalking

92

0.47

 Injury to real property

49

0.80

 Injury to real property

75

0.38

 Assault by pointing a gun

47

0.77

 Breaking or entering (M)

68

0.35

 Breaking or entering (M)

31

0.51

 Assault on a child under 12

57

0.29

 Possession of drug paraphernalia

22

0.36

 Misdemeanor larceny

37

0.19

 Assault on a government official/employee

21

0.34

 Unauthorized use of a motor vehicle

22

0.11

 Assault on a child under 12

21

0.34

 Assault on a government official/employee

19

0.10

 Misdemeanor larceny

20

0.33

 Assault on a handicapped person

17

0.09

 False imprisonment

13

0.21

 False imprisonment

13

0.07

 Misdemeanor stalking

12

0.20

*Not all charges and convictions shown. Thus, percents do not sum to 100. Also not shown- traffic offenses

The most frequent most serious charge among index arrests by far is assault on a female, N = 12,882, of which 3,902 resulted in conviction. The next most frequent charge is a DV protection order violation, N = 2,527, with 916 convictions. While much less frequent, felonies are more severe crimes, which in North Carolina are tried in superior court with a jury present. Misdemeanors are tried in district court without juries. The most common felony charge is assault by strangulation (N = 264). However, very few of these charges result in conviction (N = 14). In total, there are only 100 felony convictions in our sample. An explanation for the low variation in penalties is that there are minimum and maximum penalties for each type of offense, and a few types account for a high proportion of total DV-related charges. Judges have more discretion about conviction/no conviction and on the arrest charges than on the penalty amount conditional on a conviction on a given charge.

4.1.3 Characteristics of the analysis sample

Probability of re-arrest and conviction during follow-up

Re-arrest rates on a DV-related charge for the two-year follow-up period are 22, 24, and 25% in the all arrest, prosecuted, and convicted samples, respectively (Table 3). Conviction rates for the follow-up are 8.3, 12, and 13% in the three samples, respectively.
Table 3

Descriptive statistics: analysis of recidivism

Sample

Arrest

Prosecuted

Convicted

Variable

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev.

Dependent variables

 Re-arrested during follow-up

0.22

0.42

0.24

0.43

0.25

0.43

 Re-convicted during follow-up

0.083

0.28

0.12

0.32

0.13

0.34

Penalty: District-level

 Prosecuted (prosecutorial district)a

0.35

0.14

 Any conviction (district court district)b

0.83

0.071

 Any jail (district court district)c

0.82

0.094

Characteristics of the arrest

 Prosecuted

0.38

0.49

 Any conviction

0.83

0.38

 Any jail

0.80

0.40

 Court-appointed attorney

0.28

0.45

0.26

0.44

0.28

0.45

 Public defender

0.23

0.42

0.22

0.41

0.22

0.42

 Attorney waived

0.29

0.45

0.33

0.47

0.33

0.47

 Private attorney (omitted)

0.20

0.40

0.19

0.39

0.18

0.38

 Victim witness

0.57

0.50

0.58

0.49

0.57

0.49

 Any felony

0.026

0.159

0.03

0.17

0.011

0.11

 Misdemeanor category 1

0.11

0.31

0.029

0.17

0.15

0.36

 Misdemeanor category 2 (omitted)

0.78

0.41

0.74

0.44

0.73

0.44

 Unclassified misdemeanor

0.077

0.27

0.15

0.36

0.001

0.038

 Any traffic offense/infraction

0.0014

0.037

0.0031

0.056

0.005

0.069

 Any felony in look-back

0.0094

0.096

0.01

0.1

0.12

0.32

 Misdemeanor category 1 in look-back

0.0098

0.098

0.0065

0.08

0.01

0.10

 Misdemeanor category 2 in look-back

0.17

0.37

0.18

0.38

0.11

0.31

 Unclassified misdemeanor in look-back

0.03

0.17

0.038

0.19

0.052

0.22

 Any traffic offense/infraction in look-back

0.0001

0.009

Characteristics of the arrestee

 Female

0.19

0.39

0.13

0.34

0.12

0.32

 Age under 21

0.17

0.38

0.17

0.38

0.17

0.37

 Age 21–25

0.14

0.35

0.14

0.35

0.14

0.35

 Age 26–34

0.28

0.45

0.27

0.45

0.28

0.45

 Age 35–44

0.25

0.43

0.25

0.43

0.25

0.43

 Age 45–54

0.12

0.33

0.13

0.33

0.13

0.33

 Age 55–64

0.027

0.16

0.028

0.16

0.027

0.16

 Age 65+ in look-back

0.01

0.10

0.011

0.11

0.009

0.095

 White

0.48

0.50

0.44

0.5

0.42

0.49

 Black

0.43

0.50

0.46

0.5

0.47

0.50

 Hispanic

0.068

0.25

0.076

0.26

0.082

0.27

 Other race

0.025

0.16

0.025

0.16

0.025

0.16

County-level characteristic

 Arrests per 10,000 populationd

0.012

0.011

0.012

0.010

0.012

0.010

Instrumental variables excluded from second stage

 Prosecutore

−0.019

0.12

 Judge (conviction)f

−0.067

0.19

 Judge (jail)g

0.018

0.11

 No drop policy

0.79

0.41

0.85

0.36

0.84

0.36

Observations

23,985

 

9,240

 

7,648

 

Ranges by sample (arrested, prosecuted, convicted):

aArrested min: 0.10, max: 0.72

bProsecuted min: 0.66, max: 0.93

cConvicted min: 0.48, max 0.99

dArrested min: 0.001, max: 0.12; Prosecuted min: 0.001, max: 0.12; Convicted min: 0.001, max: 0.12

eArrested min: −0.64, max: 0.69

fProsecuted min: −0.88, max: 0.40

gConvicted min: −0.47, max: 4.06

Index case characteristics

Thirty-eight percent of index arrests are prosecuted.7 Only a fifth of persons retain a private attorney for the index arrest (20%). Slightly lower shares of persons arrested and prosecuted and convicted retain a private attorney (19 and 18%, respectively). Waiving the right to an attorney occurs almost twice as frequently. Court-appointed attorneys and public defenders are only available to indigent defendants. Twenty-nine percent of defendants waived their right to an attorney. Defendants in North Carolina are entitled to an attorney if the minimum sentence involves some jail time. However, defendants may be assessed for the cost of the public defender or court-appointed attorney. Defendants may elect to defend themselves for this reason. The victim is a witness to the allegation in 57% of index arrests. Slightly over 21% of persons with an index arrest had been arrested for a DV-related offense in the past 3 years. Forty-three percent of index arrestees are black and 6.8% are Hispanic. By contrast, in 2007, 22% of NC residents were black and 7.0% were of Hispanic origin (United States Census Bureau 2011).

Instrumental variables

The sample mean values for the prosecutor and judge variables are around zero since these variables are normalized by the corresponding district rates. The majority of NC counties had official policies of no-drop in 2007. Whether or not no-drops increase DV prosecution rates depends in large part on how the policy is applied in practice.

4.2 First-stage results

The prosecutor and judge IVs are positively and significantly related to the probabilities of case prosecution, conviction, and jail (Table 4). The parameter estimates are appreciably larger in the analysis for prosecution and jail sentences (jail/no jail) than for convictions, which plausibly reflects more heterogeneity in policies among prosecutors and judges as they pertain to prosecution and jail than for judges in assessing defendant guilt. The F-statistic on the exclusion restrictions for endogenous variables for prosecution and jail exceed 10, which implies that these IVs are not weak (Staiger and Stock 1997). However, the F-statistic for the endogenous conviction variable is only 7.19.
Table 4

First stage regressions

Variables

Prosecuted

Any conviction

Any jail

1

2

3

Penalty: District-level

 Any prosecution (prosecutorial district)

0.870*** (0.023)

 Any conviction (district court district)

0.810*** (0.062)

 Any jail (district court district)

1.030*** (0.053)

Characteristics of the arrest

 Court-appointed attorney

−0.021** (0.008)

0.061*** (0.011)

0.100*** (0.014)

 Public defender

0.010 (0.009)

0.038*** (0.013)

0.063*** (0.015)

 Attorney waived

0.073*** (0.008)

0.052*** (0.011)

0.072*** (0.014)

 Victim witness

0.003 (0.006)

0.015** (0.0076)

0.002 (0.009)

 Any felony

−0.005 (0.019)

−0.075*** (0.026)

−0.200*** (0.046)

 Misdemeanor category 1

−0.270*** (0.010)

−0.730*** (0.020)

−0.120*** (0.016)

 Unclassified misdemeanor

0.330*** (0.010)

0.130*** (0.007)

−0.230* (0.140)

 Any traffic offense/infraction

0.440*** (0.057)

−0.067 (0.056)

−0.240***(0.081)

 Any felony

−0.014 (0.031)

−0.008 (0.034)

0.062 (0.080)

 Misdemeanor category 1 in look-back

0.008 (0.025)

0.046 (0.039)

0.059 (0.069)

 Misdemeanor category 2 in look-back

−0.001 (0.008)

0.015 (0.010)

−0.003 (0.079)

 Unclassified misdemeanor in look-back

0.042** (0.018)

0.035** (0.017)

0.003 (0.018)

 Any traffic offense/infraction in look-back

−0.270*** (0.019)

Characteristics of the arrestee

 Female

0.017* (0.010)

0.047*** (0.011)

−0.110*** (0.018)

 Age 21–25

−0.014 (0.010)

−0.006 (0.013)

0.013 (0.016)

 Age 26–34

−0.007 (0.009)

0.005 (0.011)

0.015 (0.013)

 Age 35–44

0.016* (0.009)

−0.004 (0.011)

0.023* (0.014)

 Age 45–54

0.016 (0.011)

−0.023* (0.014)

0.014 (0.016)

 Age 55–64

0.021 (0.019)

−0.011 (0.025)

−0.070** (0.032)

 Age 65+

0.037 (0.029)

−0.072 (0.044)

−0.029 (0.051)

 Black

0.001 (0.006)

0.021*** (0.008)

0.045*** (0.010)

 Hispanic

0.043*** (0.012)

0.067*** (0.013)

0.031* (0.017)

 Other race

−0.005 (0.019)

0.019 (0.024)

0.038 (0.031)

County-level characteristic

 Arrests per’0000 population

0.200 (0.280)

1.150*** (0.340)

0.520 (0.430)

Instrumental variables excluded from second stage

 Prosecutor

0.800*** (0.022)

 Judge (conviction)

0.068*** (0.022)

 Judge (jail)

0.450*** (0.098)

 No-drop policy

−0.0003 (0.007)

−0.026** (0.010)

−0.001 (0.012)

Constant

0.073*** (0.013)

0.098* (0.057)

−0.140*** (0.049)

Observations

23,985

9,240

7,648

R2

0.20

0.17

0.13

F-test of excluded instruments

656.7

7.19

10.8

Robust standard errors in parentheses

***p < 0.01; **p < 0.05; *p < 0.1

The probability of conviction conditional on prosecution is lower in districts with a no-drop policy. We find no significant effect of no-drop on the probability that the case was prosecuted. Possibly some weaker cases that have a lower a priori chance of resulting in a conviction are prosecuted in counties with a no-drop policy but, judging from our results, the number of such cases is probably not large.

As anticipated, measures of pj, district-wide rates of prosecution, conviction, and jail are positively related to prosecution, conviction, and jail outcomes in individual cases. Cet. par., defendants with private attorneys tend to fare better in terms of avoiding prosecution, conviction, and jail. Cases with a victim witness are more likely to result in a conviction, conditional on prosecution. Prior DV arrests do not generally affect the probability of adverse outcomes for defendants.

Females have a higher probability of being convicted of a DV offense, but are less likely to be sentenced to jail, conditional on being convicted. Cet. par., there are no differences in outcomes for black individuals. Hispanics fare worse on average in terms of all three dependent variables.

4.3 Second-stage results

Overall, the results do not support the view that penalties as meted out by the criminal justice system reduce repeat arrests for DV or convictions for DV in the two-year follow-up period (Table 5). Although four of six signs on parameter estimates for the pj are negative, none of the parameter estimates are statistically significant at the 5% level or better. Only one parameter estimate on the variables for sj is both negative and significant at better than the 10% level, the estimate for being convicted for the index arrest in the prosecuted sample. Lack of significance applies to the district-level sanction threat variables as well as the sanction variables for individual arrests.
Table 5

Two stage least squares estimates of penalties on arrest and conviction at follow-up

Sample

Arrested

Prosecuted

Convicted sample

Penalty

Prosecuted

Convicted

Any jail

Variables

Arrest

Conviction

Arrest

Conviction

Arrest

Conviction

1

2

3

4

5

6

Penalty: Individual

 Prosecuted

0.046 (0.03)

0.05*** (0.02)

 Any conviction

0.28 (0.33)

−0.56* (0.30)

 Any jail time

0.15 (0.10)

0.025 (0.08)

Penalty: District-level

 Any prosecution (prosecutorial district)

−0.048 (0.03)

0.11*** (0.02)

 Any conviction (district court district)

−0.22 (0.27)

0.55** (0.25)

 Any jail (district court district)

−0.11 (0.12)

−0.034 (0.09)

Characteristics of arrest

 Court-appointed attorney

0.048*** (0.01)

0.031*** (0.01)

0.041* (0.03)

0.079*** (0.02)

0.041** (0.02)

0.044*** (0.01)

 Public defender

0.027*** (0.01)

0.017*** (0.01)

0.009 (0.02)

0.034* (0.02)

0.013 (0.02)

0.008 (0.01)

 Attorney waived

0.0022 (0.01)

0.005 (0.01)

−0.019 (0.02)

0.026 (0.02)

−0.028* (0.02)

−0.01 (0.01)

 Victim witness

−0.013** (0.01)

0.013** (0.00)

−0.039*** (0.01)

−0.004 (0.01)

−0.043*** (0.01)

−0.018** (0.01)

 Any felony

−0.008 (0.02)

0.004 (0.01)

−0.01 (0.03)

−0.063** (0.03)

−0.11** (0.04)

−0.066** (0.03)

 Misdemeanor category 1

−0.017 (0.01)

−0.008 (0.01)

0.18 (0.24)

−0.45** (0.22)

−0.034* (0.02)

−0.035** (0.01)

 Unclassified misdemeanor

0.048*** (0.02)

0.036*** (0.01)

0.043 (0.05)

0.12*** (0.04)

0.31** (0.14)

0.23 (0.14)

 Any traffic offense/infraction

−0.11* (0.06)

−0.10*** (0.01)

−0.17*** (0.06)

−0.060 (0.04)

−0.13*** (0.05)

−0.10*** (0.02)

 Any felony in look-back

0.19*** (0.03)

0.051** (0.02)

0.15*** (0.05)

0.083* (0.05)

0.12 (0.14)

0.11 (0.12)

 Misdemeanor category 1 in look-back

0.18*** (0.03)

0.024 (0.02)

0.25*** (0.07)

0.076 (0.05)

−0.02 (0.13)

−0.047 (0.11)

 Misdemeanor category 2 in look-back

0.18*** (0.01)

0.073*** (0.01)

0.17*** (0.02)

0.10*** (0.01)

0.052 (0.14)

0.019 (0.12)

 Unclassified misdemeanor in look-back

0.046** (0.02)

0.06*** (0.02)

0.052* (0.03)

0.087*** (0.03)

0.056** (0.03)

0.015 (0.02)

 Any traffic offense/infraction in look-back

−0.30*** (0.05)

−0.073*** (0.01)

Characteristics of the arrestee

 Female

−0.096*** (0.01)

−0.044*** (0.01)

−0.12*** (0.02)

−0.028 (0.02)

−0.072*** (0.02)

−0.049*** (0.01)

 Age 21–25

0.004 (0.01)

0.001 (0.01)

0.007 (0.02)

−0.003 (0.02)

0.014 (0.02)

0.007 (0.01)

 Age 26–34

−0.013 (0.01)

−0.004 (0.01)

−0.006 (0.01)

−0.001 (0.01)

−0.01 (0.02)

−0.005 (0.01)

 Age 35–44

−0.0002 (0.00)

0.002 (0.01)

−0.006 (0.01)

−0.007 (0.01)

−0.019 (0.02)

−0.006 (0.01)

 Age 45–54

−0.042*** (0.01)

−0.015** (0.01)

−0.042** (0.02)

−0.036** (0.02)

−0.048*** (0.02)

−0.02 (0.01)

 Age 55–64

−0.064*** (0.02)

−0.043*** (0.01)

−0.096*** (0.03)

−0.075*** (0.02)

−0.13*** (0.03)

−0.072*** (0.02)

 Age 65+

−0.13*** (0.02)

−0.051*** (0.01)

−0.12*** (0.04)

−0.12*** (0.04)

−0.13*** (0.04)

−0.097*** (0.02)

 Black

−0.007 (0.01)

−0.006 (0.00)

−0.033*** (0.01)

−0.008 (0.01)

−0.031*** (0.01)

−0.022** (0.01)

 Hispanic

−0.10*** (0.01)

−0.043*** (0.01)

−0.14*** (0.03)

−0.037 (0.02)

−0.14*** (0.02)

−0.090*** (0.01)

 Other race

−0.016 (0.02)

0.002 (0.01)

0.004 (0.03)

0.014 (0.03)

0.004 (0.03)

0.0003 (0.03)

County-level characteristic

 Arrests per’0000 population

0.40 (0.26)

0.38** (0.17)

0.60 (0.57)

1.00* (0.51)

0.93 (0.49)

0.25 (0.39)

Constant

0.21*** (0.01)

0.016** (0.01)

0.21*** (0.06)

0.072 (0.06)

0.27*** (0.05)

0.16*** (0.04)

Observations

23,985

23,985

9,240

9,240

7,648

7,648

R2

0.058

0.042

0.023

−0.41

0.039

0.037

Hansen’s J-statistic (p-value)

0.082

0.04

0.52

0.34

0.34

0.90

Robust standard errors in parentheses

*** p < 0.01; ** p < 0.05; * p < 0.1

Parameter estimates on the arrest ratio are consistently positive, suggesting that recidivism, if anything, is higher, not lower, in counties with relatively high arrest rates. Prior DV offenses are predictive of re-arrest and conviction during follow-up. This is in contrast to our finding from the first stage that prior offenders do not have statistically different rates of prosecution, conviction, or jail sentences.

However, other results provide indirect evidence for deterrence. Four of the six parameter estimates on the binary variable for victim witness available at trial are negative and statistically significant at the 5% level or better. A fifth parameter estimate is negative and significant at better than the 10% level. Defendants who retain private attorneys are less likely to be charged with a DV offense and are less likely to be convicted of a DV offense during follow-up. Such persons incur the full legal expense in defending their cases, which the vast majority of arrestees for DV who benefit from publicly supplied legal defenses or waive their right to an attorney do not.

Female and Hispanic individuals are less likely to recidivate. There are no statistical differences in recidivism rates between white and black individuals, although black persons are, in proportion of the population, much more likely to be arrested for DV than are whites.

The results for the Hansen J-test indicate that in all regressions except one, prosecution/conviction in the follow-up period, the null hypothesis that the instrumental variables can be excluded from the main equation is accepted at conventional levels of statistical significance. Overall, the evidence on the exclusion restriction is stronger in analysis based on samples of prosecuted and convicted persons.

Conceptually, repeat offenders should learn less from a single arrest than first-time offenders do. To examine whether or not there is a difference in the recidivism response to the index arrest experience and to probabilities of sanctions in the districts in which the person is arrested, we split the sample into (1) persons with index arrests but with no DV-related arrests in the three-year look-back period and (2) persons with one or more index arrests during the look-back period. The results for each sub-sample are essentially the same as those presented in Table 4. Since the vast majority of index arrests are in the first sub-sample, the results for this sub-sample are closer to those for the entire sample in Table 4.

Finally, to assess whether our results are sensitive to the exclusion restrictions, we add the prosecution rate variables to the first stage of the analysis of whether or not the person was convicted and the conviction rate and the prosecution rate variables to the first stage analysis for jail conditional on conviction. The second stage parameter estimate and standard errors on the deterrence variables are similar to those we present.

5 Discussion and conclusions

The role of sanctions in an economic model of crime is to deter criminal activity. Overall, the criminal sanctions for domestic violence in North Carolina are light. This may be a sufficient condition for our finding that the threat of sanctions does not lower the probability of re-arrest.

The probability of incurring a penalty in the form of a fine or a jail term conditional on committing a DV-related charge is about 0.04. The low probability mainly reflects two factors, the low probability of arrest—which is typical of crimes more generally but higher for DV than many other crimes, in part because the denominator of the DV arrest rate is the number of DV victims rather than single DV events—and the high probability that prosecutors decline such cases. Factors underlying the low prosecution rate for DV include the high burden of proof, the lack of availability of admissible evidence, and low participation of victims in the judicial process even those victims who report having been abused to public law enforcement. If arrested, prosecuted, and convicted, a person can expect a sentence of about two months.8 In North Carolina, probation is rarely substituted for fines and jail following conviction for a DV-related offense.

Fines for DV-related offenses are particularly low. The lack of reliance on fines runs counter to the economic theory of public enforcement (Polinsky and Shavell 2007), which predicts that the criminal justice system would rely on fines rather than incarceration since the administrative cost of the former is far lower than of the latter. A limitation in the use of fines is that offenders may be incapable of paying the fine, be “judgment proof,” thus reducing the intended deterrent effect (Gilles 2006). The reasoning may be that jail time is a stronger deterrent not only because of the convict’s time lost in jail, but also because of the reputational loss which, in turn, may cause difficulty in finding employment (Western et al. 2001; Pager 2003). Whether or not these assumptions are valid is, in the end, an empirical question. Another possibility is that judges may believe that fines punish the victim as well as the offender because total household income is reduced. Variation in fines is much lower than are financial penalties in civil litigation,9 in large part because criminal penalties for DV are scheduled in North Carolina and a few types of charges account for a large share of arrests on DV-related charges.

Conceptually, the framework of Bayesian updating of subjective beliefs is highly appropriate in this context. People presumably learn about the workings of the criminal justice system from direct experience with it. They are likely to have a prior belief about probabilities of arrest, prosecution, conviction, and penalties conditional on conviction. In this study, we assume that these prior beliefs reflect district-wide outcomes following arrest for DV. The offender updates his/her beliefs based on his/her own experience with the system. We hypothesize he/she learns differently from encounters with strict prosecutors and judges than from more lenient ones.

A Bayesian framework underlies concepts of general and specific deterrence (Stafford and Warr 1993). General deterrence is achieved if the threat of sanctions, as reflected in our study by the district-specific probabilities of sanctions, succeeds in preventing individuals or organizations from committing illegal acts. The rationale for specific deterrence, individualized punishment, is that general deterrence sometimes fails to discourage such acts and hence the expectation is that by directly experiencing punishment, individuals will be discouraged from committing illegal acts in the future. Of course, to be credible, threats must reflect reality. For there to be general deterrence, threats must be carried out sometime. Failure to find either general or specific deterrence does not represent rejection of Bayesian updating in the context of criminal activity, but rather there may be no empirical support for such updating when the law is not enforced. There is some empirical evidence in support of updating of risk perceptions, which we reviewed above.

The low level of penalties is one reason for the failure of penalties to deter. The learning that occurs from a conviction for DV may be that the probability of sanctions is low and/or learning ways to avoid an arrest and conviction the next time; some of these methods may succeed in avoiding a re-arrest for DV even if the arrestee does not alter his or her abusive behavior.10 Iyengar (2009) finds that implementation of mandatory arrest laws for domestic violence actually leads to a perverse effect, an increase in intimate partner homicides. A reason is that mandatory arrest leads to retribution by the abuser.

To the extent that learning occurs, one would expect there to be more learning among first-time than among repeat offenders. However, our separate analysis for first- versus repeat-offenders does not differ according to whether or not there were prior offenses. Another possibility is that while sanctions do not deter arrests for DV, they actually do deter acts of DV. We cannot rule out this possibility with the data we have.

While the empirical evidence shows that sanctions for DV do not deter re-arrest, there is some reason for hope if the low prosecution rate could be overcome. First, first-time arrests for domestic violence are much more common than re-arrests, which viewed in isolation suggests that penalties are effective in deterring recidivism. Second, the presence of a victim witness consistently reduces the probability of arrest on a DV-related charge or a conviction during follow-up. This is consistent with a previous finding that women who are less tolerant of abuse are less likely to be victims of repeat domestic violence (Hanson and Wallace-Carpretta 2004). Possibly offenders learn from the experience that the injury victim will stand up for his/her rights and this is more important than any punishment by the criminal justice system, especially within the range of sanctions that apply to DV.

Third, defendants who retain private attorneys are less likely to be arrested or convicted during follow-up. One interpretation is that persons who are arrested for DV and retain a private attorney bear a higher cost of the arrest, and hence opt not to commit DV in the future. Another interpretation is that such persons incur higher costs in terms of loss of employment, reputation, or in terms of payments in the event of separation or divorce. Still another view, not mutually exclusive with the others, is that having a private attorney is a proxy for an unobserved personal characteristic affecting recidivism.

One strength of our study is the use of a longitudinal state-wide database on arrests for DV and our treatment of endogeneity of sanctions, as applied to individual arrests. While there is a substantial amount of previous literature, most studies have been small-scale, both in terms of sample sizes and geographic areas included (Gross et al. 2000; Grann and Wedin 2002) and/or include a limited amount of time for follow-up (Williams and Houghton 2004; Yates et al. 2008). This study contributes to research on deterrence of DV by: (1) making a special effort to control for factors other than penalties, as well as endogeneity of penalties, that might affect recidivism; (2) use of a longer follow-up period than many studies; and (3) a much larger sample covering a broader geographic area than in almost all previous studies.

A deficiency of administrative data on arrests is the lack of information on arrestees. It would have been desirable to have data on such objective characteristics as income, educational attainment, family structure, mental health, and use of addictive substances as well as information on preferences, including motivations for committing DV, and subjective beliefs about probabilities of adverse consequences from such behavior—not only what the probabilities are but also on what information individuals rely to form these probabilities. Ideally, such information would be collected prospectively.11

In sum, using administrative data from one state, sanction levels for DV-related offenses are insufficient to deter repeat arrests. A weak link in the chain of sanctions is in the low probability of prosecution. Judging from the literature, this problem is not unique to North Carolina. While some jurisdictions have implemented no-drop policies to raise the prosecution rate, even this policy has not been entirely successful. If criminal sanctions are to be used to prevent domestic violence, considerably more attention should be focused on this weak link in the process.

Footnotes
1

See, e.g., Robinson (2000), Simpson et al. (2006), Hirschel et al. (2007), and Simon et al. (2010). An estimate in Eitle (2005) of 49% is higher, but the data came from police departments primarily subject to mandatory arrest policies.

 
2

We suppress subscripts for individuals here and elsewhere.

 
3

Durham, North Carolina at one time had a domestic violence court consisting of special judges and special prosecutors assigned to domestic violence cases. Even in this case, the judges and prosecutors may have been randomly assigned within the specialized court, but we did not examine this.

 
5

The arrest may be for several charges including a DV-related charge. The categories in the table are for the DV-related charge, but the penalties include those for any charge on the DV arrest for which the person was convicted.

 
6

North Carolina maintains an electronic database on time served by persons sent to state prison. Relatively few persons convicted of a DV sentence are sent to prison in this state. The vast majority of incarcerations for DV are in county jails. In a survey of county jails we conducted in early 2012, we were only able to identify three North Carolina counties that maintained electronic records of the incarcerations under their management. Thus, the relationship between jail time at sentencing and actual jail time is unknown.

 
7

Comparing these values with those in Fig. 1 indicates that re-arrests within the year of the index arrest are substantially less likely to be prosecuted than is the index arrest.

 
8

We acknowledge that even a night in jail may be viewed by at least some potential offenders as costly. See Polinsky and Shavell (1999) for a conceptual discussion of the disutility of jail time. How the marginal disutility of jail time changes as a function of days in jail is unknown.

 
9

For example, compare the variation in criminal penalties shown in Table 2 with those for civil penalties in medical malpractice litigation (e.g., Sloan and Hsieh (1990)).

 
10

Another possibility is that standard economic models of behavior do not apply in the context of domestic violence. Rejecting such models opens up nearly an infinite number of possibilities. DeRiviere (2008) discusses complexities not considered by conceptual and empirical economic research to date. Unfortunately, the data needed to study the types of complexities she mentions empirically, e.g., the dynamics of interpersonal relationships and the endogeneity of human capital, are well beyond data sets currently available to researchers.

 
11

As opposed to retrospectively as in the National Longitudinal Survey of Youth 1997 (see Lochner (2007)). However, retrospective data are far better than nothing.

 

Copyright information

© Springer Science+Business Media New York 2013