Policing Domestic Violence in the Post-SARP Era

Abstract

During the Spousal Assault Replication Program, the city of Charlotte, North Carolina, was identified as a site where arrest did not deter misdemeanor domestic violence. Shortly after these findings were published, the Charlotte-Mecklenburg Police Department developed a Domestic Violence (DV) unit to combat the problem of intimate partner violence. The mission of the Charlotte DV unit is to reduce future offending through intensive investigation and victim assistance. The current study evaluates the impact of the Charlotte DV unit versus standard patrol on official accounts of offender recidivism in a random sample of 891 domestic violence cases. Controlling for offender demographics, prior criminal history, case severity, and additional criminal justice responses, suspects processed through the DV unit had significantly lower rates of re-offending across an 18- to 30-month follow-up period. Theoretical explanations for the DV unit effect are proposed.

Keywords

intimate partner violence domestic violence police unit recidivism

Introduction

Sherman and Berk’s (1984) Minneapolis Domestic Violence Experiment (MDVE) is one of the most influential studies in policing, and it is arguably one of the most influential studies in all of social science (Fagan & Browne, 1994). The study found that offenders who were arrested for misdemeanor domestic violence had lower recidivism rates than those given advice or ordered away from the scene (Sherman & Berk, 1984). With its central finding that “arrest worked best,” the MDVE has been cited more than 700 times,¹ has received extensive coverage in the popular media, has influenced the way many police departments respond to intimate partner violence, and eventually prompted the National Institute of Justice to sponsor a series of similar studies collectively known as the Spousal Assault Replication Program (SARP; Buzawa & Buzawa, 1996; Sherman, 1992).

Results from the SARP proved to be inconsistent; the replications in Colorado Springs and Miami affirmed the deterrent properties of arrest, whereas those in Charlotte, Milwaukee, and Omaha did not (Sherman, 1992). Attempts to offer global conclusions from SARP (e.g., Berk, 1993; Schmidt & Sherman, 1993) were challenging in part because of the methodological and analytical differences found across the studies (Garner, Fagan, & Maxwell, 1995). Yet when data from the five replications are pooled, standardized measures of police interventions are used, and recidivism is measured through official records, arrest is associated with a 4% decrease in the prevalence and an 8% decrease in the frequency of re-offending (Maxwell, Garner, & Fagan, 2001, 2002). Neither of these reductions is statistically significant (although larger and significant effects were found in victim interview data²).

In sum, (1) MDVE found that arrest worked best, (2) SARP found that arrest works sometimes, and (3) a synthesis of SARP results found that arrest did not work at all—at least, when using official measures of recidivism. Given these paradoxical conclusions regarding the deterrent properties of arrest, it is not surprising that many police departments have sought additional and more progressive strategies to curb domestic violence (Buzawa & Buzawa, 1996; Clark, Burt, Schulte, & Maguire, 1996; Corcoran & Allen, 2005; Hovell, Seid, & Liles, 2006; Townsend, Hunt, Kuck, & Baxter, 2005). The current study describes how one former SARP city—Charlotte, North Carolina—changed its approach to policing domestic violence shortly after the city’s null effects of arrest were announced. More specifically, the study describes the development of the Charlotte-Mecklenburg Police Department’s (CMPD’s) domestic violence (DV) unit and evaluates its impact on official measures of offender recidivism.

Literature Review

Background: The Charlotte SARP study

In the Charlotte SARP study, persons suspected of misdemeanor domestic assault were randomly assigned to receive one of three police responses: (1) advising or separating the couple, (2) issuing a citation that required the offender to appear in court, or (3) arresting the offender (Hirschel, Hutchinson, & Dean, 1992; Hirschel, Hutchison, Dean, Kelley, & Pesackis, 1991). At the conclusion of the study’s 6-month follow-up period, arrested persons did not differ statistically from the other groups of offenders in terms of their prevalence and frequency of re-offending, prompting Hirschel, Hutchinson, and Dean (1992) to write

The results of this experiment are decisive and unambiguous, and indicate that arrest of misdemeanor spouse abusers is neither substantively nor statistically a more effective deterrent to repeat abuse than either of the other two police responses examined in this location. (p. 29)

Hirschel, Hutchinson, and Dean (1992) offered possible explanations for why arrest may have failed to deter domestic violence in Charlotte (and also in other SARP cities; see Hirschel & Hutchison, 1996; Hirschel, Hutchison, Dean, & Mills, 1992). For example, they noted that a large percentage of offenders in the study had a prior criminal record and were therefore familiar with (and perhaps immune to) the effects of arrest. Additionally, Hirschel, Hutchinson, and Dean (1992) suggested that the lack of “follow-through” on the part of courts may have diminished any deterrent effects of arrest (see Davis, Smith, & Nickles, 1998; Ventura & Davis, 2005; Wooldredge & Thistlethwaite, 2002). Specifically, 66% of arrested offenders had their cases dismissed, and of the 95 offenders who were prosecuted and eventually convicted, 91 received suspended sentences (Hirschel et al., 1991). Given that the average jail time of those arrested in the Charlotte SARP study was just 9 hours (Hirschel, Hutchinson, Dean, et al., 1992), the typical arrest in Charlotte was likely viewed as nothing more than an inconvenience, especially among those with a prior criminal history.³

In addition, Sherman (1984) argued that the deterrent effects of arrest may be conditioned on the offender’s social bond. In a reanalysis of the Milwaukee and Omaha SARP data, offenders with greater “stakes in conformity”—that is, those who are employed and/or married—were found to be more deterrable through arrest (Sherman, Smith, Schmidt, & Rogan, 1992). Comparable results were found in the Colorado Springs and Miami SARP data. A subsequent reanalysis of the Milwaukee data also found that arrest failed to deter the frequency of recidivism among African American suspects, a finding that may be attributed to racial differences in employment rates (Sherman, Schmidt, Rogan, et al., 1992). Unfortunately, the Charlotte SARP data were not included in these reanalyses because they were not available at the time (Sherman, Smith, et al., 1992); thus, it is difficult to assess how the demographic features of the Charlotte sample may have affected the study’s results. Yet it is noteworthy that 23% of the Charlotte offenders were unemployed, less than half (48%) were married to their victims, and 70% were African American (Maxwell et al., 2002).

Although we do not know for certain why arrest failed to deter domestic violence in Charlotte, the above research indirectly suggests that the suspects’ past offending history, the city’s low rates of prosecution, and the abusers’ race and/or their low stakes in conformity may have contributed to the null findings. Given these factors were—and still are—beyond the control of CMPD, the ability of CMPD officers to curb domestic violence through arrest alone remained in doubt. Thus, the need for a new approach to policing domestic violence in Charlotte was born.

The Domestic Violence Police Unit

In the 1990s, as scholars looked to the future of domestic violence policy in America, some foresaw (and others recommended) an increase in the number of specialized domestic violence policing units (Ford, Reichard, Goldsmith, & Regoli, 1996; Schmidt & Sherman, 1993). By the year 2000, 48% of local law enforcement agencies and 10% of state agencies had DV units (Reaves & Hickman, 2004; see Townsend et al., 2005). Although there appears to be no uniform structure or set of responsibilities for domestic violence police units (Uekert, Miller, DuPree, Spence, & Archer, 2001), Ford et al. (1996) envisioned these units as collaborations between police and other criminal justice/social service agencies with the responsibilities of ensuring the implementation of departmental policies for domestic violence, coordinating the service and enforcement of warrants and protection orders, identifying relationships at risk for intimate partner and family violence, using data to identify “hot spots” of domestic violence, and engaging in proactive criminal justice and/or community intervention strategies. As such, these specialized police units embody many of the fundamental elements of a coordinated community response (CCR) to domestic violence (see Bledsoe, Sar, & Barbee, 2006; Pence & Shepard, 1999).

The Charlotte DV Unit

In 1995, CMPD developed its own DV unit staffed with one sergeant, one administrative assistant, five detectives, and four counselors (Lord et al., 2004). This central team has at its disposal a group of citizen volunteers who can assist victims as they fill out paperwork, call victims to remind them of upcoming court dates, and escort victims within courthouse. The DV unit team also has access to an assortment of government and nonprofit organizations that can provide individualized resources for victims, such as crisis intervention, shelter, and food and gas cards.

Operating within CMPD’s “preferred arrest” policy for domestic violence,⁴ the mission of the DV unit is to reduce future offending through intensive investigation (e.g., conducting detailed follow-up interviews, identifying/correcting missing information in patrol reports, preparing case materials for the district attorney, etc.) and through victim assistance (e.g., filing restraining orders). As part of this endeavor, all domestic violence cases in the county are forwarded to the DV unit sergeant, who reviews the reports and decides whether to return the case to standard patrol or to accept it in the DV unit. The primary basis for this decision is the sergeant’s subjective assessment of case history/severity, with more chronic and/or violent cases designated for the DV unit (Lord et al., 2004). It is possible that some acts of family violence may be assigned to the DV unit, although this appears to be a rare exception to the rule. Most DV unit cases involve current or former intimate partners.

Each DV unit case is assigned a lead detective trained to conduct intensive investigations and to charge the suspect based on the evidence of the case rather than on the wishes of the victim (Lord et al., 2004). Such police practices appear to be the central elements that distinguish the DV unit from standard patrol. On a case-by-case basis, the DV unit detectives and/or counselors also provide victims with intervention information, provide gift cards for food, make referrals to social agencies, help victims develop a safety plan, and better explain relevant criminal justice procedures such as how to obtain a restraining order.

Empirical research on DV units

As noted earlier, DV units are conceptually similar to CCRs to domestic violence. Generally speaking, research on domestic violence CCRs finds that they are well received by victims (Lane, Greenspan, & Weisburd, 2004) and that they increase the probability of formal sanctioning for the offender (Bledsoe et al., 2006; Corcoran & Allen, 2005; Salazar, Emshoff, Baker, & Crowley, 2007). However, evidence of their ability to reduce re-offending is weak/inconsistent (Davis, Maxwell, & Taylor, 2006; Garner & Maxwell, 2008; Hovell et al., 2006; Muftić & Bouffard, 2007).

Research examining DV police units specifically (as opposed to DV CCRs) is scarce, and most of these published studies/reports are process evaluations (Butzer, Bronfman, & Stipak, 1996; DuPree, 2003; Lord et al., 2004; Pennell & Burke, 2002), do not include measures of recidivism (Balenovich, Grossi, & Hughes, 2008), and/or do not include a comparison group (Willson, McFarlane, Malecha, & Lemmey, 2001; Wing, 2004). In our review of this literature, we found just two studies of sufficient methodological rigor to provide a meaningful evaluation of a DV unit’s impact on future offending: one occurring in the United States (Whetstone, 2001) and the other in the United Kingdom (Farrell & Buckley, 1999).

Using a quasi-experimental pre–post design, Whetstone (2001) compared measures of domestic violence in a police district containing a DV unit (i.e., treatment district) with an “equivalent” district without a DV unit. The two districts were compared on indicators such as domestic violence calls for service, domestic violence arrests, and domestic violence arrests per incident of intimate partner violence. Results revealed that the treatment district experienced a statistically significant increase in the outcome measures from predesign to postdesign, whereas the comparison district saw no statistical change. However, because these are aggregate-level measures, it is impossible to discern if the treatment district’s increase in domestic violence activity is attributable to an increase in repeat offending or a rise in the number of new domestic assaults in the area.

In their evaluation, Farrell and Buckley (1999) focused on the number of annual repeat calls for domestic incidents before and after the implementation of a DV unit in one of seven police divisions. Results indicated that the DV unit division was the only division to see a decline in repeat calls, whereas all others reported an increase. This finding suggested that the unit was effective in reducing recidivism; however, subsequent analyses revealed that offenders with a history of chronic abuse—that is, those generating eight or more calls for service in the preceding year—were unfazed by the domestic violence unit. Thus, the unit appeared to have its greatest preventive effect among persons less entrenched in the cycle of violence.

Based on the contradictory findings from these two studies, the impact of DV units on offender recidivism remains unclear. Additional research in this area is needed. Given Charlotte’s historical connection to SARP and CMPD’s ongoing commitment to combat domestic violence, we believe that Charlotte is uniquely situated to serve as the testing grounds for a more rigorous evaluation of specialized domestic violence police units. The current study uses official police/court data to conduct a retrospective evaluation of the Charlotte DV unit. The study examines the effect of the DV unit versus standard patrol on offender recidivism while controlling for relevant suspect, incident, and criminal justice variables.

Method

Sample

In 2003, the CMPD filed approximately 6,000 domestic violence–related police incident reports; approximately 8% of these cases were eventually assigned to the DV unit. From this population of 6,000 offenses, a preliminary sample of 1,000 cases was randomly selected for this evaluation. Given their low base rate in the population, we purposefully oversampled DV unit cases (225 DV unit and 775 standard patrol cases). Many cases in this preliminary sample contained multiple victims, multiple suspects, and/or dual aggressors. These cases present complex problems associated with data coding, statistical analyses, and the interpretation of the findings; as a result, we excluded them from the current study. Therefore, the final unweighted sample consists of 891 domestic violence cases (n = 220 DV unit cases), each involving one victim and one suspect. Prior to data analyses, cases were weighted by the inverse of their selection probabilities.

Data

Unless otherwise noted, all data come from electronic incident reports maintained in CMPD’s computerized database.

Recidivism measures

Offender recidivism—measured as both the prevalence and frequency of future domestic violence acts against any victim—is our dependent variable. To determine offender recidivism, an electronic search for the name of each suspect in the sample was conducted in June of 2005, and incident reports for domestic violence–related offenses occurring after the triggering incident (in 2003) were identified. Note that because we searched for offenses committed as of June 2005, the length of time between the triggering offense and review date ranges from 18 to 30 months. Given the large number of suspects and the amount of data to be extracted from each case, it was not feasible to count, review, and code information from every incident occurring during the follow-up period.⁵ Therefore, our measure of recidivism frequency is a trichotomy denoting whether suspects have 0, 1, or 2+ future offenses.

Unit assignment

Unit assignment for the triggering offense is our primary independent variable and is expressed as a dichotomy (0 = assigned to a standard patrol unit; 1 = assigned to the DV unit). Note that CMPD incident reports do not include information on the quantity or quality of services provided by the assigned unit, and so we cannot identify the specific processes in which cases were handled. Thus, we are only able to evaluate the impact of unit assignment (globally) as opposed to the effect of individual services provided.

Suspect characteristics

Prior research finds age, race, gender, and marital status to be important predictors of domestic violence offending (Caetano, Field, Ramisetty-Mikler, & McGrath, 2005; Caetano, Vaeth, & Ramisetty-Mikler, 2008; Herrera, Wiersma, & Cleveland, 2008; Smith, 1990; Straus, Gelles, & Steinmetz, 1980; Tjaden & Thoennes, 2000). We therefore control for these suspect characteristics using data taken from the police report for the triggering offense.⁶ Furthermore, given the relationship between past and future offending (Kingsnorth, 2006; Piquero, Farrington, & Blumstein, 2003), we also control for suspects’ prior history of domestic violence using a dichotomous measure (0 = no DV priors; 1 = prior DV offender).

Severity of the triggering offense

Prior research indicates that the severity of an offender’s crime is positively related to his or her frequency of offending (Piquero, 2000; Piquero, Farrington, Blumstein, 2007). We therefore control for the severity of the triggering incident using one of three indicators. First, we recorded the National Incident-Based Reporting System (NIBRS) charge associated with the triggering offense (U.S. Department of Justice, 2000). Note that CMPD officers report the only one NIBRS charge per incident, reflecting what they perceive to be the most serious act committed during the incident. After collapsing similar charges into aggregate categories (e.g., theft from building and theft from motor vehicle were collapsed into a general “theft” category), we identified 10 possible NIRBS categories: trespassing, intimidation, vandalism, theft, breaking/entering, simple assault, aggravated assault, sexual assault, kidnapping, and an assortment of “other” offenses with extremely low base rates (e.g., false pretenses/swindling, arson, etc.,).⁷ Rather than attempting to rank-order these offenses in terms of their severity, we treat these NIBRS categories as nominal-level measures.

Second, CMPD officers denote the level of threat/harm the victim experienced using one of five descriptors: none, threatened (not gun-related), gun-related threat, minor injury, and major injury. Note that this assessment of harm is subjective and left to the discretion of the reporting officer. We use these descriptors as additional (nominal) indictors of offense severity. Finally, CMPD officers provide a narrative account of the offender’s actions, which includes qualitative data not easily captured elsewhere. To extract contextual information regarding offense severity from these narratives, we created a “check list” of non–mutually exclusive abusive acts that may occur during episodes of domestic violence (e.g., verbal threats, pushing/shoving, hitting/slapping, etc.). Each police narrative was reviewed to determine which of these events had occurred (0 = unknown; 1 = present). Features not mentioned in a narrative were coded as “unknown” as opposed to “did not occur” given the possibility the reporting officer may not have listed every detail of the offense.

Criminal justice responses

Using police records, we determined if the triggering offense was cleared by an arrest (0 = no; 1 = yes). Additionally, jail records were examined to determine the number of days the suspect served for the triggering offense (including pretrial confinement).

Results

Descriptive and Bivariate Results

Table 1 summarizes the study variables across unit assignment in the weighted sample (n = 6,139). Offenders in the study are 33 years old on average, predominately African American (70%), predominately male (86%), and predominately single (56%). Approximately one third of offenders have a prior history of domestic violence. With respect to the severity of the triggering offense, the modal NIBRS charge was “simple assault” and the modal injury level was deemed to be “minor” by the reporting officer. The most commonly occurring offense features listed in the police narratives were acts of “hitting/slapping/biting/choking,” followed by “pushing/shoving/pulling hair,” and “verbal threats.” Approximately 46% of suspects were arrested, and they spent an average of 11 days in jail for the triggering offense. Although most suspects (64%) did not re-offend during the follow-up period, approximately 21% were found to have one future domestic violence charge, with 15% having two or more.

Table 1.

Study Variables by Unit Assignment (Weighted Sample).^a

	Weighted Sample			Effect Size		Test Statistic
	Standard Patrol (n = 5,607)	DV Unit (n = 532)	Total (N = 6,139)	V	d	χ²	t
Suspect characteristics
Mean age in years at incident (SD)	33.14 (10.57)	33.93 (8.74)	33.21 (10.43)		0.08		1.93^†
Race (%)
White	29.9	21.1	29.1	0.06		25.39***
African American	69.1	78.9	69.9
Other	1.1	0.0	1.0
Male (%)	85.6	86.1	85.7	0.004		0.08
Marital status (%)
Married	23.8	31.6	24.5	0.07		17.53***
Separated	5.6	7.2	5.7
Divorced	2.9	4.9	3.1
Single	56.7	48.0	56.0
Other/unknown	11.0	8.2	10.7
Percentage with prior DV offense	35.0	45.4	35.9	0.06		22.85***
Offense severity
NIBRS classification (%)
Trespassing	1.2	1.5	1.2	0.42		1097.38***
Intimidation	11.1	12.3	11.2
Vandalism	4.5	0.0	4.1
Theft	2.5	1.5	2.5
Breaking and entering	0.7	5.3	1.1
Simple assault	67.4	20.8	63.3
Aggravated assault	6.8	35.1	9.3
Sexual assault	0.1	5.8	0.6
Kidnapping	1.8	10.2	2.5
Other	3.9	7.5	4.2
Injury level (%)
No injury or threat	25.0	11.2	23.8	0.33		664.32***
Threatened—not gun-related	13.6	17.9	14.0
Gun-related threat	0.6	6.6	1.1
Minor injury	60.0	49.7	59.1
Major injury	0.8	14.6	2.0
Percentage of cases known to involve . . .^b
Telephone hang-ups	5.1	12.9	5.8	0.09		53.62***
Verbal threats	18.5	36.6	20.0	0.13		99.04***
Pushing/shoving/pulling hair	34.4	27.9	33.8	0.04		8.93***
Hitting/slapping/biting/choking	56.1	50.0	55.6	0.03		7.22***
Open wounds	2.1	14.1	3.1	0.19		229.30***
Hostage taking	0.4	9.3	1.2	0.23		315.84***
Hospital transport	4.8	25.9	6.6	0.24		346.37***
Sexual assault/rape	0.1	8.6	0.9	0.25		395.43***
Criminal justice responses
Case cleared with an arrest (%)	43.8	71.1	46.2	0.15		145.18***
Mean days in jail—arrested suspects only (SD)	7.64 (29.80)	31.42 (73.57)	10.81 (39.44)		0.42		6.21***
Recidivism
Prevalence (% recidivated)	36.9	29.3	36.2	0.05		12.16***
Frequency
With 0 future charges (%)	63.1	70.9	63.8	0.05		12.7***
With 1 future charge (%)	21.3	17.3	20.9
With 2 or more future charges (%)	15.6	11.8	15.3
Follow-up period
Mean number of months (SD)	22.56 (3.42)	22.55 (3.26)	22.56 (3.41)		0.003		0.06

Note. DV = domestic violence; NIBRS = National Incident-Based Reporting System.

Percentages within columns are shown.

Offense types are coded from police narratives and are not mutually exclusive.

†

p < .10. *p < .05. **p < .01. ***p < .001.

Most of the variables in Table 1 vary significantly across unit assignment; however, in such a large sample, weak relationships can become statistically significant. To put the relationships in Table 1 into perspective, we include an estimate of each relationship’s effect size computed as either Cramer’s V or Cohen’s d. As seen in the table, all the suspect characteristics are very weakly related to unit assignment, whereas measures of offense severity—especially NIBRS classification and victim injury level—are more moderately related. Arrest is weakly related to unit assignment, but among arrested suspects, there is a much stronger relationship between unit assignment and jail time. Finally, assignment to the DV unit is associated with a lower prevalence and frequency of recidivism, albeit weakly. Given the preexisting differences across DV unit and standard patrol cases, multivariate analyses are needed to examine this apparent DV unit effect more rigorously.

Multivariate Results

Recidivism prevalence

In Table 2, Column 2 summarizes a logistic regression in which recidivism prevalence is regressed onto the length of the follow-up period, suspect characteristics, and offense severity. Because of high levels of multicollinearity across the three indicators of severity, we include only the NIBRS measure in the model at this time. As seen in the table, goodness-of-fit indicators are mixed. The likelihood ratio chi-square is statistically significant and suggests a good fit; however, the Hosmer–Lemeshow test is also significant, suggesting a poor fit. The Nagelkerke R² is weak. Within this model, offenders who have longer follow-up periods, those who are male, who are nonwhite, and who have a prior history of domestic violence are significantly more likely to have recidivated. Compared with those who committed a simple assault, offenders who engaged in trespassing, intimidation, and “other” crimes are more likely to have recidivated, whereas those who committed breaking and entering and aggravated assault are less likely to have re-offended.

Table 2.

Logistic Regression Predicting Recidivism Prevalence.

	Column 1			Column 2			Column 3
	b	SE	Exp(B)	b	SE	Exp(B)	b	SE	Exp(B)
Intercept	−2.75***	0.24	0.06	−2.85***	0.25	0.06	−2.86***	0.25	0.06
Follow-up period (in months)	0.05***	0.01	1.05	0.05***	0.01	1.05	0.05***	0.01	1.05
Suspect characteristics
Age	−0.004	0.003	1.00	−0.003	0.003	1.00	−0.004	0.003	1.00
Male	0.31**	0.09	1.36	0.33***	0.09	1.39	0.34***	0.09	1.41
Nonwhite	0.58***	0.07	1.78	0.60***	0.07	1.82	0.60***	0.07	1.83
Married^a	−0.02	0.09	0.98	−0.02	0.09	0.98	−0.004	0.09	1.00
Prior DV offender	1.22***	0.06	3.38	1.21***	0.06	3.36	1.23***	0.06	3.43
Highest NIBRS classification^b
Trespassing	0.79**	0.28	2.21	0.78**	0.28	2.18	0.83**	0.28	2.29
Intimidation	0.44***	0.09	1.56	0.46***	0.09	1.59	0.51***	0.09	1.67
Vandalism	0.16	0.15	1.18	0.19	0.15	1.21	0.19	0.15	1.21
Theft	−0.01	0.19	0.99	0.004	0.19	1.00	0.02	0.19	1.02
Breaking and entering	−0.86**	0.29	0.42	−0.87**	0.30	0.42	−0.70*	0.30	0.50
Aggravated assault	−0.24*	0.10	0.78	−0.24*	0.11	0.79	−0.08	0.11	0.93
Sexual assault	−0.31	0.37	0.73	−0.19	0.38	0.82	0.24	0.39	1.28
Kidnapping	0.14	0.19	1.15	0.12	0.18	1.13	031	0.19	1.36
Other	0.45**	0.15	1.57	0.51**	0.15	1.67	0.59***	0.15	1.80
Criminal justice responses
Suspect arrested	—	—	—	0.14*	0.06	1.15	0.17**	0.06	1.19
Days in jail	—	—	—	−0.003*	0.001	1.00	−0.002^†	0.001	1.00
DV unit assignment	—	—	—	—	—	—	−0.62***	0.12	0.54
Model statistics
Likelihood ratio χ²	683.43 (df = 15)***			692.42 (df = 17)***			720.72 (df = 18)***
Hosmer–Lemeshow	66.08 (df = 8)***			40.32 (df = 8)***			40.14 (df = 8)***
Nagelkerke R²	0.15			0.15			0.16

Note. DV = domestic violence; NIBRS = National Incident-Based Reporting System.

Reference group consists of all relationships other than “Married.”

Reference category is “Simple Assault.”

†

p < .10. *p < .05. **p < .01. ***p < .001.

In Table, Column 2, we control for arrest and time served for the triggering offense, the latter serving as a proxy for the offender’s “time on the street” during the follow-up period.⁸ Goodness-of-fit indicators remain mixed, and the Nagelkerke R² is unchanged. All the significant predictors from Column 1 retain their significance. Furthermore, arrest is positively associated with recidivism, jail time is negatively related, and both are statistically significant. The apparent criminogenic effect of arrest is noteworthy, especially in light of the findings from the MDVE and SARP studies; however, as we will show, this effect disappears when propensity score weighting is used to further control for preexisting differences in case severity across DV unit and standard patrol group. Thus, the positive coefficient for arrest in Table 2 may be selection effect and attributed to the lack of statistical equivalence in the DV unit and standard patrol cases. (By design, the MDVE and SARP studies minimized selection effects.)

As seen in Table 2, Column 3, when DV unit assignment is included in the set of predictors, only a few changes emerge. For example, the coefficient for aggravated assault is no longer significant, and the effect of jail time is now significant at p < .10. Most important, however, the coefficient for DV unit assignment is statistically significant and negative. With an odds ratio of 0.54, the likelihood of DV unit suspects recidivating is approximately half that of those assigned to standard patrol.

Given the limitations of the NIBRS charge as a measure of offense severity, we ran the models in Table 2 a second time using victim injury level as our indicator of severity, and then again using the case features reported in police narratives. Regardless of how we control for the severity of the triggering offense, the findings remain largely the same, including the significant reduction in recidivism for those assigned to the DV unit. For example, when victim injury levels are included in the model, the odds ratio for the DV unit assignment is 0.57. When case features from police narratives are included, the odds ratio becomes 0.53.

Recidivism frequency

Recall that our measure of recidivism frequency is not a continuous count of all future domestic violence offenses, but is instead an ordinal ranking measured as “0,” “1,” or “2+” offenses. We therefore conducted a series of generalized ordinal regressions paralleling the prevalence analyses in Table 2, but for simplicity, we present only the model that included unit assignment. Table 3 summarizes this regression. Goodness-of-fit indicators are again mixed, and the Nagelkerke R² is weak.

Table 3.

Generalized Ordinal Regression Predicting Recidivism Frequency.

	b	SE	Exp(B)
Threshold (0 future offense)	2.59***	0.24	13.33
Threshold (1 future offense)	3.89***	0.24	48.91
Follow-up period (in months)	0.40***	0.01	1.49
Suspect characteristics
Age	−0.10**	0.003	0.90
Male	0.45***	0.09	1.57
Nonwhite	0.65***	0.07	1.92
Married^a	0.09	0.09	1.09
Prior DV offender	1.27***	0.06	3.56
Highest NIBRS classification^b
Trespassing	0.44	0.23	1.55
Intimidation	0.32***	0.09	1.38
Vandalism	0.38**	0.15	1.46
Theft	−0.03	0.18	0.97
Breaking and entering	−0.57	0.30	0.57
Aggravated assault	−0.11	0.11	0.90
Sexual assault	0.09	0.38	1.09
Kidnapping	0.44*	0.18	1.55
Other	0.80***	0.14	2.23
Criminal justice responses
Suspect arrested	0.14*	0.06	1.15
Days in jail	−0.004**	0.001	1.00
DV unit assignment	−0.57***	0.12	0.57
Model statistics
Likelihood ratio χ²	789.31 (df = 18)***
Deviance χ²	9493.18 (df = 1664)***
Nagelkerke R²	0.16

Note. DV = domestic violence; NIBRS = National Incident-Based Reporting System.

Reference group consists of all relationships other than “Married.”

Reference category is “Simple assault.”

p < .05. **p < .01. ***p < .001.

Although some unique significant effects emerge (e.g., age), in general, the findings for recidivism frequency replicate those for recidivism prevalence. The length of the follow-up period, the offender’s gender, race, and prior offending history significantly predict re-offending. Furthermore, most of the NIBRS charges that predicted prevalence also predict frequency, as do the measures of arrest and jail time. Most notably, DV unit assignment is negatively and significantly related to the frequency of future offending, and this finding remained unchanged when we reran the model using the two other indicators of severity.

However, given that our measure of recidivism frequency is a trichotomy in which 85% of suspects have either 0 or 1 re-offense, it is possible the DV unit’s apparent reduction in frequency may actually be a reflection of its effect on prevalence. To examine this possibility, we treated the frequency variable as categorical rather than ordinal and used multinomial regressions (not shown) to determine the DV unit’s ability to predict re-offending. These models included the same set of predictors as in the generalized ordinal regression.

When predicting 1 versus 0 re-offenses, the DV unit had a negative and significant effect (b = −0.61, p < .001, Exp(B) = 0.71). When predicting 2+ versus 0 re-offenses, the DV unit had a similar effect (b = −0.61, p < .001, Exp(B) = 0.55). These findings are consistent with the prevalence analyses and indicate that offenders processed through the DV unit are less likely to recidivate (at any level) than those processed through standard patrol. However, when predicting 2+ versus 1 re-offense, the DV unit exerted a trivial and nonsignificant effect (b = .01, p > .05, Exp(B) = 1.01). Thus, the DV unit does not appear to be effective at curbing the frequency of re-offending among those offenders who eventually recidivate. Collectively, these findings suggest that the DV unit is better able to curb offending discretely rather than linearly.

Propensity Score Analysis

As shown in Table 1, there are fundamental differences between the DV unit and standard patrol cases, especially with respect to the severity of the triggering offense. In the preceding analyses, we have attempted to control for these preexisting differences using multivariate regression techniques. However, as Ridgeway (2006) has noted, there is reason to question the use of regression analyses when the treatment and control groups differ “substantially” on some set of independent variables, x . As an alternative statistical approach, we adopted a propensity score weighting technique, which helps to create treatment and control groups that are comparable on a broad range of measured factors (McCaffrey, Ridgeway, & Morral, 2004). Such an approach can construct a better “apples-to-apples” comparison by weighting the distribution of x among the standard patrol cases in such a way that it approximates the distribution of x in DV unit cases. Standard patrol cases whose x is similar to that of the DV unit cases will receive greater weight in the analyses, whereas those whose x is very different will receive very little weight or will be excluded altogether. Ideally, what remains is a set of DV unit and standard patrol cases that are statistically balanced on x .

A detailed discussion of propensity scores and propensity weighting is beyond the scope of this article (see instead McCaffrey et al., 2004; Nichols, 2007; Rosenbaum & Rubin, 1983, 1985; Rubin, 2006). Briefly, we created propensity score weights by first regressing unit assignment onto the suspects’ age, race, gender, marital status, prior DV history, and all three indicators of offense severity.⁹ We then used the predicted probabilities (p) from this regression to create weights for the standard patrol group, computed as p/(1 − p). After applying these weights, we were still unable to achieve complete balance across all measures of offense severity. These remaining differences appeared to be driven by a handful of cases involving sexual assault and/or hostage taking, which were assigned almost exclusively to the DV unit. Given these cases account for such a small percentage of the sample, we decided to eliminate them from the analyses and reexamine the comparability of the DV unit and standard patrol cases again. Table 4 summarizes the results from this propensity weighted sample (n = 871).

Table 4.

Study Variables by Unit Assignment (Propensity Weighted Sample).^a

	Propensity Weighted Sample			Effect Size		Test Statistic
	Standard Patrol (n = 434)	DV Unit (n = 437)	Total (N = 871)	V	d	χ²	t
Suspect characteristics
Mean age at incident (SD)	33.24 (10.63)	33.83 (8.81)	33.53 (9.77)		0.06		0.38
Race (%)
White	18.9	19.7	19.3	0.06		3.09
African American	80.4	80.3	80.4
Other	0.7	0.0	0.3
Male (%)	82.7	83.4	83.1	0.01		0.08
Marital status (%)
Married	27.0	27.7	27.4	0.07		2.15
Separated	4.5	6.2	5.3
Divorced	3.3	5.0	4.1
Single	57.8	52.9	55.3
Other/unknown	7.4	8.3	7.8
Percentage with prior DV offense	46.2	47.6	46.9	0.01		0.17
Offense severity
NIBRS classification (%)
Trespassing	1.6	1.8	1.7	0.06		2.60
Intimidation	12.0	14.0	13.0
Vandalism	0.0	0.0	0.0
Theft	1.6	1.4	1.5
Breaking and entering	5.8	5.7	5.7
Simple assault	21.2	23.6	22.4
Aggravated assault	40.6	38.2	39.4
Kidnapping	9.2	7.3	8.3
Other	8.1	8.0	8.0
Injury level (%)
No injury or threat	11.5	13.5	12.5	0.05		1.80
Threatened—not gun-related	16.8	18.3	17.6
Gun-related threat	7.6	6.3	6.9
Minor injury	46.8	46.2	46.5
Major injury	17.3	15.8	16.5
Percentage of cases known to involve . . .^b
Telephone hang-ups	12.4	14.0	13.2	0.02		0.44
Verbal threats	33.9	36.4	35.1	0.03		0.60
Pushing/shoving/pulling hair	25.3	25.4	25.4	0.001		0.00
Hitting/slapping/biting/choking	45.6	47.0	46.3	0.01		0.17
Open wounds	18.9	16.9	17.9	0.03		0.57
Hospital transport	29.7	26.8	28.2	0.03		0.90
Criminal justice responses
Case cleared with an arrest (%)	63.8	71.6	67.7	0.08		6.06*
Mean days in jail—arrested suspects only (SD)	29.99 (77.43)	24.64 (54.20)	27.15 (66.13)		0.08		0.96
Recidivism
Prevalence (% recidivated)	43.3	31.5	37.4	0.12		12.99***
Frequency
With 0 future charges (%)	56.7	68.5	62.6	0.14		17.79***
With 1 future charge (%)	20.0	18.5	19.3
With 2 or more future charges (%)	23.3	13.0	18.1
Follow-up period
Mean number of months (SD)	23.25 (3.43)	22.42 (3.25)	22.83 (3.36)		0.05		3.66***

Note. DV = domestic violence; NIBRS = National Incident-Based Reporting System.

Percentages within columns are shown.

Offense types are coded from police narratives and are not mutually exclusive.

p < .05. **p < .01. ***p < .001.

In stark contrast to Table 1, the findings in Table 4 show that the DV unit and standard patrol cases are now statistically comparable on all the control variables except for two: arrest and length of follow-up. Even so, the effect sizes associated with these two statistically significant relationships are very weak. Most important, DV unit offenders continue to have significantly lower recidivism rates than those processed through standard patrol, and the corresponding effect size estimates are even larger than those reported in Table 1. However, because the propensity weights were not able to achieve total balance across all control variables, we do not have a perfect apples-to-apples comparison. We therefore examined the prevalence and frequency of re-offending using statistical controls for arrest and length of follow-up. Table 5 summarizes these models.

Table 5.

Prevalence and Frequency Results with Propensity Weighted Sample.

	Column 1: Prevalence			Column 2: Frequency
	b	SE	Exp(B)	b	SE	Exp(B)
Intercept	0.84	0.51	2.33	—	—
Threshold (0 future offense)	—	—	—	−1.39**	0.53	0.25
Threshold (1 future offense)	—	—	—	−0.34	0.53	0.71
Follow-up period (in months)	−0.05*	0.02	0.95	−0.06**	0.02	0.94
Suspect arrested	0.09	0.15	1.09	−0.12	0.16	0.89
DV unit assignment	−0.56***	0.14	0.57	−0.73***	0.15	0.48
Model statistics
Likelihood ratio χ²	19.15 (df = 3)***			30.32 (df = 3)***
Hosmer–Lemeshow	42.53 (df = 8)***			—
Deviance χ²	—			358.53 (df = 97)***
Nagelkerke R²	0.03			0.03

Reference group consists of all relationships other than “Married.”

Reference category is “Simple Assault.”

p < .05. **p < .01. ***p <.00.

Table 5, Column 1 summarizes the logistic regression predicting recidivism prevalence. The results largely replicate the earlier prevalence analyses, with suspects in the DV unit significantly less likely to recidivate (odds ratio = 0.57). Table 5, Column 2 summarizes the findings from a generalized ordinal regression predicting the frequency of re-offending. As in our earlier analysis of recidivism frequency, the coefficient for the DV unit is negative and significant. In a series of follow-up analysis, multinomial regressions indicated that the DV unit had a negative and significant effect when predicting 1 versus 0 re-offenses (b = −0.38, p < .05, Exp(B) = 0.68), and when predicting 2+ versus 0 re-offenses (b = −0.76, p < .001, Exp(B) = 0.47). However, when predicting 2+ versus 1 re-offenses, the effect of the DV unit remained negative but failed to reach statistical significance (b = −0.38, p = .10, Exp(B) = 0.68). These results reaffirm our earlier conclusion that the DV unit does not curb the frequency of offending among those who eventually go on to recidivate.

Discussion

In light of inconsistent evidence regarding the ability of law enforcement to deter domestic violence through arrest alone, many police departments have initiated more innovative strategies to combat the problem (Buzawa & Buzawa, 1996; Clark et al., 1996). The current study evaluated one such strategy—a specialized domestic violence police unit—in the city of Charlotte, North Carolina. Prior evaluations have found DV units to be associated with both an increase in domestic violence activity (Whetstone, 2001) and a decrease in repeated activity (Farrell & Buckley, 1999). In contrast, the current study examined the impact on the Charlotte DV unit on different measures of recidivism using different multivariate models and different analytic techniques—and the findings were robust.

Suspects processed through the DV unit had a significantly lower prevalence rate of recidivism across the 18- to 30-month follow-up window. The odds of recidivating for DV unit offenders were approximately 50% less than that of offenders processed through standard patrol, and this effect was net of offenders’ demographics, prior history of domestic violence, case severity, arrest, and jail time. Furthermore, DV unit offenders also had significantly lower recidivism frequency rates net of the control variables; however, this reduction in frequency appears to be largely attributed to the overall reduction in prevalence. Of those DV unit offenders who eventually go on to re-offend, the frequency rate is comparable to that of the recidivists processed through standard patrol. This finding is similar to Farrell and Buckley’s (1999) contention that offenders who are entrenched in a pattern of intimate partner violence are unfazed by a domestic violence unit. Be that as it may, the overall results from this evaluation are positive and suggest that the comprehensive approach of the DV unit has succeeded where arrest alone has failed—in significantly reducing future acts of domestic violence in Charlotte.

Explaining the DV Unit Effect

Based largely on police incident reports, the data set for the current study does not contain rich measures of theoretical constructs or the quantity/quality of the police interventions provided. As a result, we are unable to identify exactly how the DV unit exerts its effect. Nevertheless, we can identify at least four possible explanations. First, recall that one mission of the Charlotte unit is to engage in intensive investigation, which in turn may lead to more cases cleared through arrest. From a deterrence perspective, greater certainty, swiftness, and/or severity of formal punishment should lead to a decrease in the offender’s future criminal activity (Beccaria, 1963; Blumstein, Cohen, & Nagin, 1978). Second, recall that the Charlotte unit targets offenders who commit more serious acts of domestic violence—offenders who should therefore be at greater risk for incarceration and serve longer jail sentences. As a result, the DV unit may also reduce recidivism through greater incapacitation effects (Blumstein et al., 1978; Spelman, 2000).

Another mission of the Charlotte DV unit is to provide victim assistance, such as helping victims obtain restraining orders and/or develop safety plans. Scholars debate whether such forms of advocacy/assistance are actually “empowering” victims (McDermott & Garofalo, 2004), and whether victim empowerment as a whole increases or decreases victim safety (Finn, 2003; Ford, 1991; Goodman & Epstein, 2005; Waites, 1985). Nevertheless, from a routine activities perspective (Cohen & Felson, 1979), such forms of victim assistance may decrease target suitability, increase guardianship, and/or minimize the temporal–spatial convergence of offenders and victims in the absence of guardianship (Farrell, Phillips, & Pease, 1995).

Finally, the Charlotte DV unit may also reduce recidivism through the process of procedural justice. Proponents of procedural justice contend that compliance with the law will be enhanced when offenders perceive (in part) an adequate opportunity to state their case to the authorities and feel that the authorities are basing their decisions on the facts of the case (Tyler, 1990; Tyler & Lind, 1992). In a reanalysis of the Milwaukee SARP data, arrested offenders who found the arresting officers to be impartial and respective to offenders had lower recidivism rates than those who felt their cases were handled less fairly (Paternoster, Bachman, Brame, & Sherman, 1997). Procedural justice has also been invoked to explain the reduction in rearrests among offenders processed through a specialized domestic violence court (Gover, Brank, & McDonald, 2007). Given that the Charlotte DV unit detectives are trained to conduct in-depth investigations (which presumably allows for greater opportunity to listen to the offender’s description of the incident) and to charge the suspect based on the evidence rather than the wishes of the victim, DV unit offenders may perceive greater levels of procedural justice than offenders processed through standard patrol.

The current study is unable to test these four explanations directly. Yet it is noteworthy that the DV unit was associated with lower recidivism rates after controlling for the effects of arrest and jail time. Although future research is needed, this finding suggests that the DV unit’s effect may not be attributed to simple deterrence or incapacitation effects. Furthermore, a separate evaluation of the Charlotte DV unit’s impact on repeat victimizations found DV unit victims to be just as likely to experience future attacks (by any offender) as those victims whose cases were processed by standard patrol (Friday, Lord, Exum, & Hartman, 2006). In other words, the DV Unit is helping offenders break their cycle of violence (as suggested by the current study) but not the victims, who may be moving from one abusive relationship to another. This conclusion is counter to the notion that the DV unit reduces crime by enhancing victim guardianship and/or making them less suitable targets.

Although the data in the current study do not permit us to assess whether the Charlotte DV unit affords greater procedural justice than standard patrol, the unit’s reliance on intensive investigation and its goal of letting the facts of the case dictate police response (rather than the wishes of the victim) is consistent with the principles of procedural justice (Tyler, 1990; Tyler & Lind, 1992). As future research is conducted on domestic violence police units, researchers should include measures of procedural justice as well as deterrence, incapacitation, and routine activities to determine what role, if any, each theory has in the implementation of the DV unit. Until then, theoretical explanations for the DV unit effect remain tentative.

Strengths and Limitations

Findings from the current evaluation should be interpreted within the context of the study’s strengths and weaknesses. For example, one strength of the study is that, unlike the MDVE and many of its replications, we do not limit our consideration of future offenses to only those crimes committed against the same victim—a practice that has been criticized for failing to account for violence against “new” intimates and for failing to account for displacement effects (Reiss, 1985; Sherman, Schmidt, & Rogan, 1992). Additionally, our follow-up period was three to five times longer than the 6-month window typically used in the SARP studies, allowing for a richer assessment of the unit’s impact on recidivism. However, by limiting our count of future offenses to a trichotomy, we fail to distinguish between those who re-offended twice from those who re-offended even more. This has bearing on our analysis of recidivism frequency rates. Fortunately, only 15% of the sample reached this artificial limit of two or more future offenses, and arguably, a portion of this group is likely to have had no more than two offenses. Thus, the percentage of offenders in the present study who had more than two re-offenses should be relatively small.

The current evaluation does not record data from victim interviews, a feature common to the MDVE and SARP studies. By relying on only official records, we may be overlooking acts of domestic violence that do not come to the attention of the police. This is especially problematic if victims who experience more serious acts of violence (and are therefore assigned to the DV unit) are less likely to report their victimizations. However, in five of the six MDVE/SARP studies, the conclusions derived from victim interviews were consistent with those based on official records (Sherman, Schmidt, & Rogan, 1992; but see Maxwell et al., 2002). This suggests that although official data may not necessarily uncover the same degree of recidivism as victim data, these nonetheless provide a reliable estimate of the general trends in re-offending. Furthermore, given DV unit victims have already experienced a severe assault that has come to the attention of the police (thereby resulting in their assignment to the DV unit), we have little reason to believe that these same victims would report future acts of severe violence differently from those victims whose triggering incidents were assigned to standard patrol.

In a similar vein, because our recidivism data were obtained solely from CMPD records, the current study overlooks any re-offending that occurred in other jurisdictions. This problem of undercounting will weaken the study’s internal validity to the degree that it affects offenders in the DV unit and standard patrol groups differently. Yet we have no reason to believe either group of offenders is any more or less likely to re-offend outside CMPD’s jurisdiction, and thus, problems associated with undercounting would appear to be minimal.

Conclusion

Despite the widespread popularity of domestic violence units among police departments across the United States, their impact on recidivism remains largely unknown. The current study evaluated the Charlotte DV unit, and the results are promising. Net of other criminal justice responses (arrest and jail time), offenders processed through the Charlotte DV unit are half as likely to recidivate than those processed through standard patrol. As the number of these specialized units continues to grow and if future evaluations find them to be as successful as the one in Charlotte, domestic violence units may very well become the most effective strategy for policing domestic violence in the post-SARP era.

Postscript

After becoming fully operational in 1997, the Charlotte DV unit remained a staple of the CMPD for more than a decade. In 2008, the city of Charlotte hired a new police chief. As policing priorities were refocused and resources reallocated, the Charlotte DV unit was disbanded. As of this writing, there are no plans to reinstate the program.

Footnotes

Acknowledgements

We are indebted to the Charlotte-Mecklenburg Police Department (CMPD) for their support during this evaluation. We especially wish to thank Sergeant Vicky Suarez, Major Tim Danchess, and Ms. Veronica Sorban for answering our (sometimes redundant) questions with extraordinary patience and diligence. Finally, we are grateful to Dr. Robert Brame and the anonymous reviewers for their comments on an earlier version of this article. All errors remain our own.

Authors’ Note

Points of view expressed in this document are those of the authors and do not necessarily represent the official position of the National Institute of Justice.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the authorship and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was supported by a grant from the National Institute of Justice (Grant No. 2004-WG-BX-0004).

Notes

Author Biographies

M. Lyn Exum is an associate professor in the Department of Criminal Justice and Criminology at the University of North Carolina at Charlotte. His research centers largely on testing criminological theories.

Jennifer L. Hartman is an associate professor of criminal justice and criminology at the University of North Carolina at Charlotte with an adjunct professor appointment in the Women’s and Gender Studies program. Her recent publications have focused on gendered effects of drug users in a community setting, exploring violence against women issues, and issues related to student learning and retention.

Paul C. Friday is a professor of criminal justice at the University of North Carolina at Charlotte and serves as Treasurer and Executive Committee member of the World Society of Victimology. He has published more than 70 professional articles and 4 books on crime, delinquency, comparative criminology, victimology, and social policy.

Vivian B. Lord is a professor of criminal justice and criminology and interim Chair of the Department of Public Health Sciences at the University of North Carolina at Charlotte with adjunct professor appointments in the Public Policy Doctoral Program and the International Studies Departments. Her recent publications focus on suicide by cop, role of mental health in police interventions, student retention, and criminal justice interviewing.

References

Balenovich

Grossi

Hughes

(2008). Toward a balanced approach: Defining police roles in responding to domestic violence. American Journal of Criminal Justice, 33, 19-31.

Beccaria

(1963). On crimes and punishments. Englewood Cliffs, NJ: Prentice Hall.

Belknap

Hartman

J. L.

Lippen

V. L.

(2009). Misdemeanor domestic violence cases in the courts: A description of the cases. In Garcia

Clifford

(Eds.), Female victims of crime: Reality reconsidered (pp. 259-278). Upper Saddle River, NJ: Prentice Hall.

Berk

R. A.

(1993). What the scientific evidence shows: On the average, we can do no better than arrest. In Gelles

R. J.

Loseke

D. R.

(Eds.), Current controversies on family violence. Newbury Park, CA: Sage.

Bledsoe

L. K.

Sar

B. K.

Barbee

A. P.

(2006). Impact of coordinated response to intimate partner violence on offender accountability. Journal of Aggression, Maltreatment and Trauma, 13, 109-129.

Blumstein

Cohen

Nagin

(1978). Deterrence and incapacitation: Estimating the effects of criminal sanctions on crime rates. Washington, DC: National Academy of Sciences.

Butzer

Bronfman

L. M.

Stipak

(1996). The role of police in combating domestic violence in the United States: A case study of the domestic violence reduction unit, Portland police bureau. In Pagon

(Ed.), Policing in central and eastern Europe: Comparing firsthand knowledge with experiences from the west (pp. 161-172). Ljubljana, Slovenia: College of Police and Security Studies.

Buzawa

E. S.

Buzawa

C. G.

(1996). Domestic violence: The criminal justice response (2nd ed.). Thousand Oaks, CA: Sage.

Caetano

Field

C. A.

Ramisetty-Mikler

McGrath

(2005). The 5-year course of intimate partner violence among White, Black, and Hispanic couples in the United States. Journal of Interpersonal Violence, 20, 1038-1057.

10.

Caetano

Vaeth

P. A. C.

Ramisetty-Mikler

(2008). Intimate partner violence victim and perpetrator characteristics among couples in the United States. Journal of Family Violence, 23, 507-518.

11.

Clark

S. J.

Burt

M. R.

Schulte

M. M.

Maguire

(1996). Coordinated community responses to domestic violence in six communities: Beyond the justice system. Washington, DC: Urban Institute.

12.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activities approach. American Sociological Review, 44, 588-608.

13.

Corcoran

Allen

(2005). The effects of a police/victim assistance crisis team approach to domestic violence. Journal of Family Violence, 20, 39-45.

14.

Davis

R. C.

Maxwell

Taylor

(2006). Preventing repeat incidents of family violence: Analysis of data from three field experiments. Journal of Experimental Criminology, 2, 183-210.

15.

Davis

R. C.

Smith

B. E.

Nickles

L. B.

(1998). The deterrent effect of prosecuting domestic violence misdemeanors. Crime & Delinquency, 44, 434-442.

16.

DuPree

(2003). Chicago Heights, Illinois, domestic violence unit: A process evaluation. Washington, DC: U.S. Department of Justice.

17.

Fagan

J. A.

Browne

(1994). Violence between spouses and intimates: Physical aggression between women and men in intimate relationships. In Reiss

A. J.

Jr. Roth

J. A.

(Eds.), Understanding and preventing violence (Vol. 3, pp. 115-292). Washington, DC: National Academy Press.

18.

Farrell

Buckley

(1999). Evaluation of a UK police domestic violence unit using repeat victimisation as a performance indicator. Howard Journal, 38, 42-53.

19.

Farrell

Philips

Pease

(1995). Like taking candy: Why does repeat victimization occur? British Journal of Criminology, 35, 384-399.

20.

Finn

(2003). Effects of victims’ experiences with prosecutors on victim empowerment and re-occurrence of intimate partner violence: Final report. Washington, DC: U.S. Department of Justice.

21.

Ford

D. A.

(1991). Prosecution as a victim power resource: A note on empowering women in conjugal relationships. Law & Society Review, 25, 313–334.

22.

Ford

D. A.

Reichard

Goldsmith

Regoli

M. J.

(1996). Future directions for criminal justice policy on domestic violence. In Buzawa

E. S.

Buzawa

C. G.

(Eds.), Do arrests and restraining orders work? (pp. 243-265). Thousand Oaks, CA: Sage.

23.

Friday

P. C.

Lord

V. B.

Exum

M. L

Hartman

J. L.

(2006). Evaluating the impact of a specialized domestic violence police unit: Final report. Washington, DC: U.S. Department of Justice.

24.

Garner

J. H.

Fagan

J. A.

Maxwell

C. D.

(1995). Published findings from the spouse assault replication program: A critical review. Journal of Quantitative Criminology, 11, 3-28.

25.

Garner

J. H.

Maxwell

C. D.

(2008). Coordinated community responses to intimate partner violence in the 20th and 21st centuries. Criminology & Public Policy, 7, 525-535.

26.

Goodman

Epstein

(2005). Refocusing on women: A new direction for policy and research on intimate partner violence. Journal of Interpersonal Violence, 20, 479-487.

27.

Gover

A. R.

Brank

E. M.

McDonald

J. M.

(2007). A specialized domestic violence court in South Carolina: An example of procedural justice for victims and defendants. Violence Against Women, 13, 603-626.

28.

Herrera

V. M.

Wiersma

J. D.

Cleveland

H. H.

(2008). The influence of individual and partner characteristics on the perpetration of intimate partner violence in young adult relationships. Journal of Youth & Adolescence, 37, 284-296.

29.

Hirschel

J. D.

Hutchison

I. W.

(1996). Realities and implications of the Charlotte spousal abuse experiment. In Buzawa

E. S.

Buzawa

G. G.

(Eds.), Do arrest and restraining orders work? (pp. 54-82). Thousand Oaks, CA: Sage.

30.

Hirschel

J. D.

Hutchison

I. W.

Dean

C. W.

(1992). The failure of arrest to deter spouse abuse. Journal of Research in Crime and Delinquency, 29, 7-33.

31.

Hirschel

J. D.

Hutchison

I. W.

Dean

C. W.

Kelley

J. J.

Pesackis

C. E.

(1991). Charlotte Spouse Assault Replication Project: Final report. Washington, DC: National Institute of Justice.

32.

Hirschel

J. D.

Hutchison

I. W.

Dean

C. W.

Mills

(1992). Review essay on the law enforcement response to spouse abuse: Past, present and future. Justice Quarterly, 9, 247-283.

33.

Hovell

M. F.

Seid

A. G.

Liles

(2006). Evaluation of a police and social services domestic violence program: Empirical evidence needed to inform public health policies. Violence Against Women, 12, 137-159.

34.

Kingsnorth

(2006). Intimate partner violence: Predictors of recidivism in a sample of arrestees. Violence Against Women, 12, 917-935.

35.

Lane

Greenspan

Weisburd

(2004). The second responders program: A coordinated police and social service response to domestic violence. Washington, DC: U.S. Department of Justice.

36.

Lord

V. B.

Arentsen

Bean

Bryley

Carver

Clifton

Erdogdu

Fitch

Kula

Marker

Powell

Ross

Sazak

(2004). Process evaluation of Charlotte-Mecklenburg Police Department’s domestic violence unit. Prepared for the Charlotte-Mecklenburg Police Department.

37.

Maxwell

C. D.

Garner

J. H.

Fagan

J. A.

(2001). The effects of arrest on intimate partner violence: New evidence from the Spouse Assault Replication Program (Research in brief). Washington, DC: U.S. Department of Justice.

38.

Maxwell

C. D.

Garner

J. H.

Fagan

J. A.

(2002). The preventative effects of arrest on intimate partner violence: Research, policy and theory. Criminology & Public Policy, 2, 51-80.

39.

McCaffrey

D. R.

Ridgeway

Morral

A. R.

(2004). Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychological Methods, 9, 403-425.

40.

McDermott

M. J.

Garofalo

(2004). When advocacy for domestic violence victims backfires: Types and sources of victim disempowerment. Violence Against Women, 10, 1245-1266.

41.

Muftić

L. R.

Bouffard

J. A.

(2007). An evaluation of gender differences in the implementation and impact of a comprehensive approach to domestic violence. Violence Against Women, 13, 46-69.

42.

Nichols

(2007). Causal inference with observational data. Stata Journal, 7, 507-541.

43.

Paternoster

Bachman

Brame

Sherman

L. W.

(1997). Do fair procedures matter? The effect of procedural justice on spouse assault. Law & Society Review, 31, 163-204.

44.

Pence

E. L.

Shepard

M. F.

(1999). An introduction: Developing a coordinated community response. In Shepard

M. F.

Pence

E. L.

(Eds.), Coordinating community responses to domestic violence: Lessons from Duluth and beyond (pp. 3-24). Thousand Oaks, CA: Sage.

45.

Pennell

Burke

(2002). Centralized response to domestic violence: San Diego county sheriff. Washington, DC: U.S. Department of Justice.

46.

Piquero

A. R.

(2000). Assessing the relationships between gender, chronicity, seriousness, and offense skewness in criminal offending. Journal of Criminal Justice, 28, 103-116.

47.

Piquero

A. R.

Farrington

D. P.

Blumstein

(2003). The criminal career paradigm. In Tonry

Michael

(Ed.), Crime and justice: A review of research (Vol. 30, pp. 359-506). Chicago: University of Chicago Press.

48.

Piquero

A. R.

Farrington

D. P.

Blumstein

(2007). Key issues in criminal career research: New analyses of the Cambridge study in delinquent development. New York, NY: Cambridge University Press.

49.

Reaves

B. A.

Hickman

M. J.

(2004). Law enforcement management and administrative statistics, 2000: Data for individual state and local agencies with 100 or more officers. Washington, DC: U.S. Department of Justice.

50.

Reiss

A. J.

Jr. (1985). Some failures in designing data collection that distort results. In Burstein

Freeman

H. E.

Rossi

P. H.

(Eds.), Collecting evaluation data: Problems and solutions (pp. 161-177). Beverly Hills, CA: Sage.

51.

Ridgeway

(2006). Assessing the effect of race bias in post-traffic stop outcomes using propensity scores. Journal of Quantitative Criminology, 22, 1-29.

52.

Rosenbaum

Rubin

D. B.

(1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70, 41-55.

53.

Rosenbaum

Rubin

D. B.

(1985). Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. American Statistician, 39, 33-38.

54.

Rubin

D. B.

(2006). Matched sampling for causal effects. New York, NY: Cambridge University Press.

55.

Salazar

L. F.

Emshoff

J. G.

Baker

C. K.

Crowley

(2007). Examining the behavior of a system: An outcome evaluation of a coordinated community response to domestic violence. Journal of Family Violence, 22, 631-641.

56.

Schmidt

J. D.

Sherman

L. W.

(1993). Does arrest deter domestic violence? American Behavioral Scientist, 36, 601-610.

57.

Sherman

L. W.

(1984). Experiments in police discretion: Scientific boon or dangerous knowledge? Law and Contemporary Problems, 47, 61-81.

58.

Sherman

L. W.

(1992). Policing domestic violence: Experiments and dilemmas. New York, NY: Free Press.

59.

Sherman

L. W.

Berk

R. A.

(1984). The specific deterrent effects of arrest for domestic assault. American Sociological Review, 49, 261-272.

60.

Sherman

L. W.

Schmidt

J. D.

Rogan

D. P.

(1992). Policing domestic violence: Experiments and dilemmas. New York, NY: Free Press.

61.

Sherman

L. W.

Schmidt

J. D.

Rogan

D. P.

Smith

D. A.

Gartin

P. R.

Cohn

E. G.

. . . Bacich

(1992). The variable effects of arrest on criminal careers: The Milwaukee domestic violence experiment. Journal of Criminal Law & Criminology, 83, 137-169.

62.

Sherman

L. W.

Smith

D. A.

Schmidt

J. D.

Rogan

D. P.

(1992). Crime, punishment and stake in conformity: Legal and informal control of domestic violence. American Sociological Review, 57, 680-690.

63.

Smith

M. D.

(1990). Sociodemographic risk factors in wife abuse: Results from a survey of Toronto women. Canadian Journal of Sociology, 15, 39-58.

64.

Spelman

(2000). What recent studies do (and don’t) tell us about imprisonment and crime. In Tonry

(Ed.), Crime and justice: A review of research (Vol. 27, pp. 419-494). Chicago, IL: University of Chicago Press.

65.

Straus

M. A.

Gelles

R. J.

Steinmetz

S. K.

(1980). Behind closed doors: Violence in the American family. Garden City, NY: Anchor Books.

66.

Tjaden

Thoennes

(2000). Full report of the prevalence, incidence, and consequences of violence against women. Washington, DC: National Institute of Justice and Centers for Disease Control and Prevention.

67.

Townsend

Hunt

Kuck

Baxter

(2005). Law enforcement response to domestic violence calls for service: Final report. Washington, DC: U.S. Department of Justice.

68.

Tyler

T. R.

(1990). Why people obey the law. New Haven, CT: Yale University Press.

69.

Tyler

T. R.

Lind

E. A.

(1992). A relational model of authority in groups. Advances in Experimental Social Psychology, 25, 115-191.

70.

Uekert

B. K.

Miller

DuPree

Spence

Archer

(2001). Evaluation of the STOP violence against women grant program: Law enforcement and prosecution components: Final Report. Washington, DC: U.S. Department of Justice.

71.

U.S. Department of Justice. (2000). National incident-based reporting system (Vol. 1). Washington, DC. Author. Retrieved from http://www.fbi.gov/ucr/nibrs/manuals/v1all.pdf

72.

Ventura

L. A.

Davis

(2005). Domestic violence: Court case conviction and recidivism. Violence Against Women, 11, 255-277.

73.

Waites

(1985). The criminal justice systems response to battering: Understanding the problem, forging the solutions. Washington Law Review, 60, 267-329.

74.

Whetstone

T. S.

(2001). Measuring the impact of a domestic violence coordinated response team. Policing: An International Journal of Police Strategies and Management, 24, 371-398.

75.

Willson

McFarlane

Malecha

Lemmey

(2001). A special police unit: Intimate partner violence and perceived helpfulness of services received. Journal of Police and Criminal Psychology, 16, 40-50.

76.

Wing

J. J.

(2004). Impact evaluation of the Nez Perce domestic violence unit. Retrieved from http://www.ncjrs.gov/policing/role161.htm

77.

Wooldredge

Thistlethwaite

(2002). Reconsidering domestic violence recidivism: Conditioned effects of legal controls by individual and aggregate levels of stake in conformity. Journal of Quantitative Criminology, 18, 45-70.