Title: The Effects of Arrest on Intimate Partner Violence: New Evidence
From the Spouse Assault Replication Program

---------------------------

Figures, charts, forms, and tables are not included in this ASCII plain-text
file. To view this document in its entirety, download the Acrobat Document by clicking here.
---------------------------

U.S. Department of Justice
Office of Justice Programs
National Institute of Justice

The Effects of Arrest on Intimate Partner Violence: New Evidence From
the Spouse Assault Replication Program

By Christopher D. Maxwell, Joel H. Garner, and Jeffrey A. Fagan

After nearly 20 years of research designed to test the effects of arrest on
intimate partner violence, questions persist on whether arrest is more
effective at reducing subsequent intimate partner violence than such
informal, therapeutic methods as on-scene counseling or temporary
separation. The most important research efforts addressing this question
were six experiments known collectively as the National Institute of
Justice's (NIJ's) Spouse Assault Replication Program (SARP).[1] These
field experiments, carried out between 1981 and 1991 by six police
departments and research teams, were designed to test empirically whether
arrests deterred subsequent violence better than less formal alternatives.

In the first of the six studies, the Minneapolis Domestic Violence
Experiment (MDVE), Sherman and Berk found that arresting batterers
reduced by half the rate of subsequent offenses against the same victim
within a 6-month followup period.[2] Subsequently, after five replication
experiments were completed, [3] Schmidt and Sherman conducted a
qualitative synthesis of MDVE and the five replications.[4] They reported
that in three studies, offenders assigned to the arrest group had higher
levels of repeat offending (recidivism) and that in the other three studies, a
statistically significant but modest reduction was found among batterers
assigned to arrest. Thus, rather than providing results that were consistent
with MDVE, the published results from the five replication experiments
produced inconsistent findings about whether arrest deters intimate partner
violence.

Because of the inconsistent and conditional findings generated by the five
replication experiments, scholars interested in the validity of deterrence
theories [5] and policymakers working to reduce intimate partner violence
have become less confident about relying on arrest as the primary response
to violence between intimates.[6] The development of a coherent
evaluation of the effectiveness of arrest based on the five experiments with
published results was complicated by the differences across the
experimental sites in case selection, incident eligibility rules, statistical
analysis, and outcome measurements. With these differences, prior
attempts to synthesize and understand the substantive diversities among
and within the experiments proved difficult. Thus, the full potential of
SARP to answer questions about the specific deterrent effect of arrest and
the safety of victims has not been realized.

We have previously reviewed and compared the published data from the
five replication sites that had reported final results to NIJ by 1993.[7] In
concluding our review, we cautioned readers not to use our synthesized
results as the final conclusion on whether arrest deters repeat spouse
assault. We pointed out that the comparisons were based on information
drawn from different outcome measures, analytical models, and case
selection criteria. Furthermore, we asserted that the inconsistency between
sources and measures across sites was not necessarily because of
limitations in the experimental designs, but because the SARP design
called for multiple data sources and measures that could capture variations
in the nature of the deterrent effect. We argued that conclusions about the
deterrent effect of arrest therefore should wait until a more careful
statistical analysis was completed, one based on data pooled from all five
sites and using standardized measures of intervention and outcome. This
Research in Brief summarizes the findings of such a statistical analysis.

We studied the deterrent effect of arrest, using an approach that addressed
many problems faced by prior efforts to synthesize the results from SARP.
Supported by NIJ and the Centers for Disease Control and Prevention
(CDC), the project pooled incidents from the five replication experiments,
computed comparable independent and outcome measures from common
data intentionally embedded in each experiment, and standardized the
experimental designs and statistical models. Using the increased power of
the pooled data, this study provides a more consistent, more precise, and
less ambiguous estimation of the impact of arrest on intimate partner
violence. Key results of this study include the following:

o Arresting batterers was consistently related to reduced subsequent
aggression against female intimate partners, although not all comparisons
met the standard level of statistical significance.

o Regardless of the statistical significance, the overall size of the rela-
tionship between arrest and repeat offending (i.e., the deterrent effect of
arrest) was modest when compared to the size of the relationship between
recidivism and such measures as the batterers' prior criminal record or age.

o The size of the reduction in subsequent intimate partner aggression did
not vary significantly across the five sites. In other words, the benefit of
arrest was about equal in regards to reducing aggression in all five sites.

o Regardless of the type of intervention, most suspects had no subsequent
criminal offense against their original victim within the followup period,
and most interviewed victims did not report any subsequent victimization
by their batterer.

o This research found no association between arresting the offender and
an increased risk of subsequent aggression against women.

About SARP

Historical background. In 1968, the New York Times Magazine reported
on an innovative program in which some New York City police officers
were trained to use psychology to handle family crisis calls. A subsequent
report published by the National Institute of Law Enforcement and
Criminal Justice asserted the value of this alternative over traditional law
enforcement approaches to domestic violence.[8] However, by the late
1970s, many law enforcement officials and domestic violence advocates
started believing that this nonpunitive and therapeutic practice of
responding to violence against women was ineffective. States enacted
arrest laws and police departments implemented policies that authorized
police officers to make an arrest even if they did not witness a domestic
violence incident or saw no evidence of a felonious act. These changes,
largely fueled by groups seeking better protection of victims through the
sanction and control of offenders,[9] enabled the criminal justice system
to initiate a more forceful response to less severe family problems.

In 1980, the Police Foundation received permission from the Minneapolis
Police Department to test the efficacy of actions its police officers could
take when responding to a domestic dispute that did not require an
arrest.[10] (See "From the Minneapolis Domestic Violence Experiment to
the Five Replications.") Using two sources of data--official police records
of new offenses and interviews with victims--and several statistical tests,
the researchers reported that arrest reduced by nearly 50 percent the rate of
subsequent assaults during the 6-month followup period.[11] These results
were subsequently argued by several scholars as among the most
influential results ever generated by social science.[12]

After considerable public discussion about the findings, NIJ announced in
1986 that it would fund a multisite replication of the Minneapolis
experiment. Five NIJ-funded sites completed their experiments: Charlotte,
North Carolina; Colorado Springs, Colorado; Dade County, Florida;
Milwaukee, Wisconsin; and, Omaha, Nebraska. Between 1986 and 1990,
NIJ provided about $750,000 for each of these research projects; in
addition, the local police departments contributed substantial resources to
these efforts over a several-year period.

Methodological background. NIJ and CDC supported the project described
in this Brief to develop a more reliable estimation of the overall deterrent
effect of arrest on intimate partner violence. The project design called for
pooling incident-level data from the five completed replication
experiments, computing comparable outcome and control measures across
the five replication sites, and providing consistent analytical models that
tested the various intervention protocols across different measures.

While we faced many problems completing each task, one of the most
difficult was deciding which outcome measures to use from a database
containing more than 300 potential outcome measures. After assessing the
literature on the nature of intimate partner violence, each site's raw data,
and the published results from the five sites, we decided to employ two
outcome measures, one from the criminal history database and one from
the victim interviews. We believe these two outcomes provide the best
available indicators of the overall extent and breadth of aggression by the
batterer against the original victim during the followup period (6 months
to 3 years).

In addition to physical assaults, the two aggression measures captured
incidents involving damage of property owned by the victim or the
common household. The aggression measure from the victim interviews
also captured verbal threats of physical or property damage made by the
batterer against the victim. We calculated several dimensions of
aggression within each of the two data sources. For the outcome measure
based on criminal history, we computed a dichotomous recidivism
measure (yes or no) that captured any incident within the first 6 months
after the experimental incident, a count of the number of days that elapsed
between the experimental incident (i.e., the one that included the suspect
in the study) and the first subsequent police-recorded incident, and a
measure of the annualized frequence of reported incidents of aggression
(i.e., the number of incidents, adjusted to represent a per-year rate). We
also computed two outcome measures based on the victim interviews: (1)
a dichotomous (yes or no) measure that captured aggression occurring
between the experimental incident and the last victim interview (which
typically covered at least a 6-month window) and (2) the frequency of
aggression between the experimental incident and the last interview.

We also collapsed the assigned intervention categories into two groups:
nonarrest and arrest. The nonarrest group contains all suspects who were
randomly assigned to one of seven alternative interventions. Exhibit 1
provides descriptive information on the experimental implementation by
site, as well as on the interview completion rates and on the suspect and
incident characteristics.

These and other methodological issues are discussed in greater detail in
the sidebar "About the Sample."

What victim interviews suggest about whether arrest deters subsequent
aggression

Exhibit 2 presents results from our statistical analysis of the relationship
between arrest and several dimensions of intimate partner aggression. The
first analysis (prevalence) uses victim interview data to test for the
association between arrest and any subsequent aggression during the
period between the experimental incident and the last time the victim was
interviewed. This model estimated that if their batterers were arrested,
about 25 percent fewer female victims than expected reported one or more
incidents of aggression. In other words, when the likelihood of failure
(reoffending) is estimated for the typical case, about 36 percent of suspects
in the arrest group reoffended, compared with 48 percent of suspects in the
nonarrest group. This difference was statistically significant while
controlling for differences among sites, the length of time the researchers
tracked the victims, and characteristics of the suspect and incident. When
examining the rates or frequency of aggression, we again found a
statistically significant reduction in subsequent aggression that is related to
arrest. On average, female victims whose batterers were arrested reported
about 30 percent fewer incidences of subsequent aggression than expected
over the followup period. Thus, we found a sizable reduction in
subsequent aggression reported by victims whose batterers were assigned
to the arrest group. However, because these results are based on a
subsample of interviewed victims, rather than on the entire sample of
eligible cases, the results from the victim interviews alone should be used
with some caution because victims not interviewed may have been
involved with suspects who responded differently to their intervention.

What other factors are related to aggression?

Besides the consistent deterrent relationship between arrest and
aggression, other factors were consistently related to aggression, but some
factors were not. First, compared with the Omaha victims, a significantly
smaller percentage of victims from the other sites (except Milwaukee)
reported one or more victimizations by the suspect. On average, victims
from these three sites also reported less frequent victimization. These
differences in the base rates of aggression across the sites, however, did
not translate into significantly different relationships between arrest and
aggression in the different sites. In other words, the reduction we find in
aggression reported by victims whose batterers were assigned an arrest is
of about equal size in each site.

In addition to the comparisons we made across the sites, we looked for
differences in aggression reported by the victims across several suspect
characteristics. These comparisons found that the suspect's age and race
were consistently and significantly related to the frequency of subsequent
aggression as reported by the victims. These victims reported significantly
less aggression when the suspect was older and nonwhite. The suspects'
prior arrest records and their marital status with the victim were also
consistently related to aggression, but only the prior record was significant
in all but one of the analyses. Finally, several other suspect characteristics,
such as employment and the use of intoxicants, were inconsistent in the
direction of their relationship across the two dimensions of aggression
(prevalence and frequency). For example, about 2 percent more victims of
employed suspects reported one or more incidents of aggression, though
these same victims simultaneously reported about 21 percent fewer
incidents of aggression over the followup period.

What official records suggest about whether arrest deters subsequent
aggression

We next examined data collected by police departments to measure
aggression by the suspect against the victim. The approach to testing
whether arrest was related to officially recorded aggression follows the
approach to the victim interviews, except we added a statistical analysis
that examined the timing of the first new aggressive incident. Overall,
the results based on the police data regarding the effectiveness of arrest are
consistent in direction with those based on the victim interview data:
A consistent deterrent relationship exists between arrest of the suspect and
later aggression while controlling for the differences across the sites,
the victim interview process, and suspect characteristics (see exhibit 2).
However, the police data show a far smaller reduction in aggression
because of the arrest treatment than what was detected using victim
interview data, and none of these relationships reached the traditional level
of statistical significance. Specifically, in the first analysis (prevalence),
we found about 4 percent fewer than the expected percentage of male
suspects in the arrest group with one or more incidents of subsequent
aggression during the first 6 months of followup. The second analysis,
which tested for the relationship between the intervention and the annual
rate of aggression, found a reduction of about 8 percent from the expected
number of incidents per year for suspects assigned to the arrest group.
Finally, the last analysis, which examined the relationship between arrest
and the timing of the first new incident, found that the expected risk of a
new incident on any given day after arrest or nonarrest is reduced nearly
10 percent among the arrested suspects. Thus, depending on the dimension
of the outcome, the average amount of reported aggression by the suspects
dropped by between 4 and 10 percent if they were assigned to the arrest
group.

Focusing more closely on the timing of the first subsequent incident of
aggression, exhibits 3 and 4 display two "survival" graphs. Exhibit 3
displays, by site, the proportion of suspects with no officially recorded
aggression against their intimate partner beyond a specified time (i.e.,
cumulative survival). The average survival rate throughout the followup
period varied substantially by site. On the high end was Omaha, where
nearly 90 percent of the suspects had not reoffended by the end of their
observation period. On the low side was Dade County, where that figure
(the cumulative survival rate) was slightly less than 60 percent. These
differences between sites, however, did not result in differences in survival
rates by intervention group when the five sites were pooled together.
Exhibit 4 shows that throughout the followup period, which for some
suspects lasted nearly 3 years, batterers who were assigned an arrest had a
consistently greater rate of survival (nonoffending) than did those assigned
an informal intervention.

This consistent, but small, difference in the survival rate by intervention is
important because earlier analysis using data from Milwaukee suggested
that arrest may have a significant long-term criminogenic effect.[13] Our
more detailed statistical analysis supports the visual evidence presented in
these exhibits. During no particular observation period were the suspects
assigned to an arrest more likely to batter their intimate partner than those
in the control (nonarrest) group. Thus, among this larger sample of male
intimate partner abusers, the survival rate for aggression among those
assigned an arrest was never less than that of the control group, as earlier
statistical analysis in one site had suggested.

Our statistical analysis also showed that the suspects' age, race,
employment status, and use of intoxicants at the time of the experimental
incident were consistently and significantly related to subsequent
aggression against the victim. Contrary to what we found with the victim
interviews, white and employed suspects had lower levels of repeat
offending according to the police records. Furthermore, suspects who were
intoxicated at the time of the experimental incident and those with prior
arrests for any crime had, on average, a greater likelihood of aggression
recorded by the police. Only the measure of the suspect's marital status
with the victim was not consistently or significantly related to aggression.
Similar to what we found with the victim interview data, marriage did not
appear to provide notable protection against subsequent levels of
aggression. Finally, we found that the longer the researchers were able to
track the victims for followup interviews, the more initial failures were
reported to the police.

In addition to our findings about the relationship between arrest and
aggression, we observed some patterns in the pooled data. First, we found
a general pattern of cessation or termination of aggression that was only
moderately related to the suspects' assigned intervention. According to
officially recorded data, less than 30 percent of the suspects, arrested or
not, aggressed against the same victim during the followup period.
Furthermore, only about 40 percent of the interviewed victims reported
subsequent victimization of any measured type by the suspects. Other
studies that specifically estimated the rate of desistance from intimate
violence have also found similar rates over a 1- to 2-year period.[14]

A second pattern concerns the high concentration of repeat aggression
among a small number of batterers. During the 6-month followup, the
3,147 interviewed victims reported more than 9,000 incidents of
aggression by the suspects since the initial incident. While most victims
reported no new incidents of aggression, about 8 percent of them reported
a total number of incidents that represented more than 82 percent of the
9,000 incidents. The same 8 percent also accounted for 28 percent of the
1,387 incidents recorded by the police that involved an interviewed victim.

Conclusion and policy implications

In 1998, the National Academy of Sciences report Violence in Families
concluded that "arrest in all misdemeanor cases will not on average
produce a discernable effect on recidivism."[15] Our early substantive
assessment of the published reports was similar to their conclusion, but we
also argued that there was insufficient evidence in the site-specific and
multisite publications to assess the effectiveness of arrest as a deterrent to
spouse assault.[16] Our multisite pooled analysis of the five replication
experiments found good evidence of a consistent and direct, though
modest, deterrent effect of arrest on aggression by males against their
female intimate partners. The victim interviews indicate that the arrest of
the suspect and any subsequent confinement, when compared with the
alternative interventions collectively (see "About the Sample"),
significantly reduced the expected frequency of subsequent aggression by
30 percent. Similarly, arrest may have reduced by a smaller amount the
number of times the police responded to subsequent domestic violence
incidents involving the same victim and suspect and may have extended
the time between the initial incident and the first subsequent incident.

Our conclusion of a direct deterrent effect from arrest contradicts at least
one assessment of findings from the original SARP publications. Berk, for
example, argued that "the current balance of scientific evidence from the
particular sites studied suggests that although arrest is not superior to a
variety of other criminal justice interventions, one can on average do no
better."[17] There are, however, various reasons the current statistical
analysis should be preferred over prior individual and multisite analyses of
SARP experiments for the following reasons:

o The consistent use of eligibility criteria across sites--we include only
male offenders and female victims of intimate violence.

o The use of a consistent measure of repeat offending across all sites.

o The use of additional statistical controls for site and suspect differences
between arrest and nonarrest groups.

o The use of longer followup periods and statistical controls for
variability in followup periods.

o Increased statistical power from pooling cases from five sites.

o The consistent comparison of arrest with all other treatments combined.

The findings of this research have several implications for policy. First,
our findings provide systematic evidence supporting the argument that
arresting male batterers may, independent of other criminal justice
sanctions and individual processes, reduce subsequent intimate partner
violence. The size and statistical significance of the effect of arrest varied
depending on whether the subsequent aggression was measured by victim
interviews or police records; even so, in all measures (prevalence,
frequency, rate, and time-to-failure), arrest was associated with fewer
incidents of subsequent intimate partner aggression. This finding exists
during the first several days after the experimental incident regardless of
the period of detention, as well as beyond 1 year. The arrested suspects
were detained an average of 9 days, but the reduction in aggression
associated with arrest did not vary by the length of the suspect's detention.
Thus, our research finds no empirical support for the argument that arrest
may eventually increase the risk for violence against women.

Second, our research showed that a minority of suspects continued to
commit intimate partner violence, regardless of the intervention they
received. While arrest reduced the proportion of suspects who reoffended
and the frequency with which they reoffended, arrest did not prevent all
batterers from continuing their violence against their intimate partners. In
fact, we found a small number of victims who have chronically aggressive
intimate partners. Future research needs to build on preliminary efforts to
accurately predict high-rate repeat offenders and to find methods of
helping their victims before they are victimized further.

Third, our research showed that a majority of suspects discontinued their
aggressive behaviors even without an arrest. This suggests that policies
requiring arrest for all suspects may unnecessarily take a community's
resources away from identifying and responding to the worst offenders and
victims most at risk. Our research has documented the size of the specific
deterrent effects of arrest, which, although consistent across sites and time,
appeared modest compared with the overall percentage of suspects
desisting from intimate partner violence. Although there may be other
benefits from policies requiring arrest that this research has not measured
(including general deterrence), there are also likely costs of using arrests
every time the police respond to an incident of intimate partner violence.
Future research in this area needs to assess the benefits and costs of
arresting all suspects before there can be a systematic conclusion of
preferred or mandatory arrest policies.

Finally, it is unlikely that any single study can provide definitive answers
to scientific questions or policy debates. Rather, a program of rigorous
research involving many studies over time and place is necessary to
provide sound bases for generating knowledge and improving policy.

---------------------------

Notes

1. Garner, Joel, Jeffrey A. Fagan, and Christopher D. Maxwell, "Published
Findings from the Spouse Assault Replication Program: A Critical
Review," Journal of Quantitative Criminology 11 (1) (1995): 3-28; and
Sherman, Lawrence W., "The Influence of Criminology on Criminal Law:
Evaluating Arrests for Misdemeanor Domestic Violence," Journal of
Criminal Law and Criminology 83 (1) (Spring 1992): 1-45.

2. Sherman, Lawrence W., and Richard A. Berk, "The Specific Deterrent
Effects of Arrest for Domestic Assault," American Sociological Review
49 (1) (1984): 261-72.

3. A sixth site was funded, but the experiment was never completed.

4. Schmidt, Janell D., and Lawrence W. Sherman, "Does Arrest Deter
Domestic Violence?" American Behavioral Scientist 36 (5) (May/June
1993): 601-10.

5. Berk, Richard A., "What the Scientific Evidence Shows: On the
Average, We Can Do No Better Than Arrest," in Current Controversies on
Family Violence, ed. Richard J. Gelles and Donileen R. Loseke, Newbury
Park, CA: Sage Publications, Inc., 1993: 323-36; Fagan, Jeffrey A., "The
Criminalization of Domestic Violence: Promises and Limits," paper
presented at the Conference on Criminal Justice Research and Evaluation,
Washington, DC, January 1996; and Sherman, "The Influence of
Criminology on Criminal Law," 1992 (see note 1).

6. Chalk, R.A., and P.A. King, eds., Violence in Families: Assessing
Prevention and Treatment Programs, Washington, DC: National Academy
of Science, Committee on the Assessment of Family Violence
Intervention, Board on Children, Youth, and Families, National Research
Council and Institute of Medicine, 1998; Clark, Jacob R., "Where to Now
on Domestic Violence? Studies Offer Mixed Policy Guidance," Law
Enforcement News 30 (April 1993): 1, 17; Frisch, Lisa A., "Research That
Succeeds, Policies That Fail," Journal of Criminal Law and Criminology
83 (1) (Spring 1992): 209-16; Lerman, Lisa G., "The Decontextualization
of Domestic Violence," Journal of Criminal Law and Criminology 83 (1)
(Spring 1992): 217-40; and Mitchell, David B., "Contemporary Police
Practices in Domestic Violence Cases--Arresting the Abuser: Is It
Enough?" Journal of Criminal Law and Criminology 83 (1) (Spring 1992):
241-49.

7. Garner, Fagan, and Maxwell, "Published Findings from the Spouse
Assault Replication Program," 1995 (see note 1).

8. Bard, M., and J. Zacker, "The Prevention of Family Violence:
Dilemmas of Community Intervention," Journal of Marriage and the
Family 33 (1971): 677-82; and Bard, Morton, Training Police as
Specialists in Family Crisis Intervention, Washington, DC: U.S.
Department of Justice, National Institute of Law Enforcement and
Criminal Justice, 1970.

9. Fagan, Jeffrey A, "Contributions of Family Violence Research to
Criminal Justice Policy on Wife Assault: Paradigms of Science and Social
Control," Violence and Victims 3 (3) (1988): 159-86.

10. Bouza, Anthony V., The Police Mystique: An Insider's Look at Cops,
Crime, and the Criminal Justice System, New York: Plenum Press, 1990.

11. Sherman, L.W., and R.A. Berk, The Minneapolis Domestic Violence
Experiment, Police Foundation Reports, no. 1, Washington, D.C.: Police
Foundation, 1984.

12. Fagan, Jeffery A., and Angela Browne, "Violence Against Spouses and
Intimates," in Understanding and Controlling Violence, ed. A.J. Reiss, Jr.,
and J.A. Roth, vol. 3. Washington, DC: National Academy Press, 1994;
Gelles, Richard J., "Constraints Against Family Violence: How Well Do
They Work?" American Behavioral Scientist 36 (5) (1993): 575-86;
Lempert, Richard, "Humility is a Virtue: On the Publication of Policy
Relevant Research," Law & Society Review 23 (1) (1989): 145-61; and
Sherman, Lawrence W., and Ellen G. Cohen, "The Impact of Research on
Legal Policy: The Minneapolis Domestic Violence Experiments," Law &
Society Review 23 (1) (1989): 117-44.

13. Sherman, Lawrence W., et al., "The Variable Effects of Arrest on
Crime Control: The Milwaukee Domestic Violence Experiment," Journal
of Criminal Law and Criminology 83 (1992): 137-69.

14. Feld, Scott L., and Murray Straus, "Escalation and Desistance of Wife
Assault in Marriage," Criminology 27 (1) (February 1989): 141-61;
Langan, Patrick A., and Christopher A. Innes, Preventing Domestic
Violence Against Women, Special Report, Washington, DC: U.S.
Department of Justice, Bureau of Justice Statistics, 1986; and Quigley,
Brian M, and Kenneth E. Leonard, "Desistance of Husband Aggression in
the Early Years of Marriage," Violence and Victims 11 (4) (1996): 355-70.

15. Chalk and King, Violence in Families, 1998 (see note 6).

16. Garner, Fagan, and Maxwell, "Published Findings from the Spouse
Assault Replication Program," 1995 (see note 1).

17. Berk, "What the Scientific Evidence Shows," 1993 (see note 5).

---------------------------

Issues and Findings

Discussed in this Brief: An analysis of 4,032 incidents in which males
assaulted their female intimate partners, comparing the number of repeat
offenses when batterers are and are not arrested. The data in this study
were obtained from five jurisdictions included in the National Institute of
Justice-sponsored Spouse Assault Replication Program. This multisite
analysis was cosponsored by the National Institute of Justice and the
Centers for Disease Control and Prevention.

Key issues: Analysis of 314 incidents in the 1984 Minneapolis Domestic
Violence Experiment found that when the assaulter was arrested,
statistically significant reductions in subsequent offending were reported
both in victim interviews and in official police records. Replication
experiments began in the early 1990s. Five jurisdictions that used a
diverse set of incidents and a variety of outcome measures reported that
the use of arrest was only occasionally associated with statistically
significant reductions in subsequent repeat offending. The results of the
experiments varied by measures used and by the jurisdiction studied.

Key findings: Using consistent definitions of eligible cases across all five
jurisdictions, a consistent set of five measures of repeat offending and
appropriate statistical analyses for the combination of data from a multisite
experimental study, this research finds that

o Arrest is associated with less repeat offending in all five measures of
repeat offending.

o Reductions in repeat offending are larger and statistically significant in
the two measures that are derived from interviews with victims.

o Reductions in repeat offending are smaller and not statistically
significant in the three measures that are derived from official police
records.

o The effectiveness of arrest does not vary by jurisdiction.

o The size of the reduction in repeat offending associated with arrest is
modest compared with the effect of other factors (such as the batterer's age
and prior criminal record) on the likelihood of repeat offending.

o Regardless of whether or not the batterer was arrested, more than half of
the suspects committed no subsequent criminal offense against their
original victim during the followup period.

o A minority of suspects continue to commit intimate partner violence
regardless of whether they were arrested, counseled, or temporarily
separated from their partner. Future research needs to focus on identifying
such offenders and the policies and practices that will prevent their
partners from being victimized further.

Target audience: Criminal justice and public health researchers and
practitioners; police managers; advocates for victims of domestic violence;
and legislators, policymakers, and domestic violence intervention planners
at all levels of government.

From the Minneapolis Domestic Violence Experiment to the Five
Replications

A 1970 report published by the National Institute of Law Enforcement and
Criminal Justice [a] recommended training police officers to calm down
domestic violence situations--separating the parties, listening to the
concerns of each disputant, and attempting to address the immediate
problem underlying the current dispute--and to provide the victim with
phone numbers for a variety of social services. Arresting one or both
parties was not part of this approach, which was touted as integrating the
psychologist's knowledge of human behavior with the coercive authority
of the law in a manner that promoted collaboration among the police and
other social service agencies.[b]

By the early 1980s, the effectiveness of this nonpunitive approach was
being questioned, and police departments began introducing policies that
changed the ways their officers responded to domestic violence by
switching from the therapeutic models to more formal, certain, and
punitive responses. Rooted in the assumptions of specific deterrence and
incapacitation, these changes emphasized expanding the police officers'
legal powers and codifying when arrests could and should be made.

When police departments were beginning to make changes, there was little
systematic knowledge and little reason to believe that arrest or any
sanction could act as a specific deterrent or improve the safety of the
victims. Questions arose about whether new criminal penalties or civil
actions were the most appropriate remedies and whether police officers
should replace their peacekeeping efforts with formal sanctions.

Amidst this uncertainty, James Q. Wilson [c] recommended that police
departments systematically experiment with different methods of
"reducing the chance that a dispute will lead to an assault and an assault to
a homicide" within intimate settings.[d] This suggestion led to the
Minneapolis Domestic Violence Experiment (MDVE), the results of
which were published in 1984. The study design called for officers in the
Minneapolis Police Department (MPD) to carry out one of three responses
when they had probable cause to believe a misdemeanor assault had
occurred between cohabitants or spouses: (1) arrest the suspect, (2) order
one party out of the residence, or (3) advise the couple on how to solve
their problems at the scene.

The strength of MDVE was that the selection of a particular response in a
particular incident of domestic violence was determined by an
experimental design. This design made it easier to determine if differences
in police responses were responsible for any differences in subsequent
reoffending by the suspect.

In MDVE, researchers at the Police Foundation collected information on
subsequent offenses for a period of 6 months from both official police
records and from interviews with victims. Using data from 314 incidents,
the researchers reported that when the suspect was arrested, there were
statistically significant reductions in reoffending in the official records of
all the cases and in the cases with victim interviews. Based on these
results, the authors of MDVE recommended policies authorizing the use of
arrest in misdemeanor domestic violence offenses.

A 1989 survey of local police departments concluded that the published
results of MDVE may have substantially influenced over one-third of the
police departments responding to their survey to adopt a proarrest
policy.[e] At the national level, the 1984 Attorney General's Task Force on
Family Violence, citing the MDVE results, recommended that "chief
executives of every law enforcement agency establish arrest as the
preferred response in incidents of family violence."[f] The MDVE results
were published in the New York Times and in hundreds of other
newspapers in the United States; three television networks reported the
results during prime-time news programs; and numerous editorials and
nationally syndicated columnists featured the study and its findings.[g]

Support for replication of MDVE was widespread. The original authors
urged replication [h] and early praise for the study's design among
criminological scholars was tempered by a preference for replication.[i]
The Department of Justice task force recommending the adoption of a
pro-arrest policy nationwide also recommended replication of the
Minneapolis experiment.[j] A multisite replication of the Minneapolis
experiment was chosen because a single-site approach would provide only
one additional data point, only slightly improving the generalizability of
the Minneapolis findings.[k]

In designing a program of replications, the National Institute of Justice
(NIJ) required that each study involve (1) experimental comparisons of
arrest and alternative police responses to misdemeanor spouse assault
incidents and (2) measurements of victim safety using both official police
records and victim interviews.[l] Other aspects of the design were left to
the preferences of the local teams of researchers and implementing police
agencies. Seventeen law enforcement agencies competed to be part of the
replication program even though this program did not provide additional
financial resources to the department or to participating officers. The
replication effort was research, not a demonstration program, and there
were no Federal subsidies to the participating departments.

The characteristics of the five sites, the organizational structure of the
research projects, and the requirements of the solicitation led to similar
studies, but not to exact replications of each other or MDVE. For example,
each new study devised experimental designs that required the officers to
report whether an incident was eligible for the study before they were told
what the assigned treatment would be. The new studies included more
cases than the Minneapolis study and some of them broadened the study
eligibility to include female offenders, same-sex couples, and harassment
offenses. Omaha conducted victim interviews at 6 and 12 months. Dade
County used the same cases to conduct a second experiment studying
police officer followup with victims after the initial incident. Milwaukee
varied the number of hours arrested offenders would be detained in the
jail. NIJ encouraged these innovations and variations among sites, but it
also required that each site document the characteristics of victims,
offenders, and police behavior so that common analyses using consistent
eligibility requirements and outcome measures could be conducted.

a. Bard, Morton, Training Police as Specialists in Family Crisis
Intervention, Washington, DC: U.S. Department of Justice, National
Institute of Law Enforcement and Criminal Justice, 1970.

b. Garner, Joel H., and Christopher D. Maxwell, "What Are the Lessons of
the Police Arrest Studies?" in Program Evaluation and Family Violence
Research, ed. S.K. Ward and D. Finkelhor, Binghamton, NY: Haworth
Publishers, 2000.

c. Sherman, Lawrence W., "The Influence of Criminology on Criminal
Law: Evaluating Arrests for Misdemeanor Domestic Violence," Journal of
Criminal Law and Criminology 83 (1) (Spring 1992): 1-45.

d. Wilson, James Q., "Foreword," in Domestic Violence and the Police,
Washington, DC: Police Foundation, 1977: iii-vi.

e. Sherman and Cohen, "The Impact of Research on Legal Policy," 1989
(see note 11).

f. U.S. Attorney General's Task Force on Family Violence, Final Report,
Washington, DC: U.S. Department of Justice, 1984.

g. Sherman, Lawrence W., and Ellen G. Cohen, "The Impact of Research
on Legal Policy: The Minneapolis Domestic Violence Experiments," Law
& Society Review 23 (1) (1989): 117-144.

h. Sherman, L.W., and R.A. Berk, "The Specific Deterrent Effects of
Arrest for Domestic Assault." American Sociological Review 49 (1984b):
261-272.

i. Boffey, P.M., "Domestic Violence: Study Favors Arrest," New York
Times, April 5, 1983; and Lempert, R., "From the Editor," Law & Society
Review 18 (4) (1984): 505-513.

j. U.S. Attorney General's Task Force on Family Violence, Final Report
(see note f).

k. Garner, Joel H., "Two Three ... Many Experiments: The Use and
Meaning of Replication in Social Science Research," paper presented at
the Annual Meeting of the American Society of Criminology, Baltimore,
November 1990.

l. National Institute of Justice, Replicating an Experiment in Specific
Deterrence: Alternative Police Responses to Spouse Assault. Washington,
DC: U.S. Department of Justice, National Institute of Justice, 1985.

---------------------------

About the Sample

Sample of suspects. The design of the Spouse Assault Replication
Program (SARP) allowed each site to vary the eligibility of cases for its
experiment. For example, for most of its implementation, the Dade County
experiment only included married couples. The Milwaukee experiment
included same sex couples and violent disputes between siblings. Several
sites excluded incidents if the suspect had been included in a prior
experimental case. Other sites included these repeat suspects.[a]

Prior site-specific and multisite analyses of SARP data have not addressed
these differences, but this study does. In this analysis, we include only
cases where a male suspect committed violence against a female victim.
Because some of the sites excluded cases with repeat suspects, we exclude
all cases (N = 248) with repeat suspects from all sites to create a more
consistent sample.

Overall, cases were most likely excluded because of the victim's or
suspect's gender or because the suspect had appeared previously in the
experiment as a suspect (see exhibit A).

The site with the greatest percentage of cases excluded was Colorado
Springs (25 percent); the exclusions were due mainly to the site's inclusion
of some repeat suspects, male victims, female suspects, and incidents not
involving an assault of the victim. Dade County had the fewest cases
excluded (1 percent). The final count of incidents removed from our
analysis was 760 (16 percent), which left 4,032 unique suspects for the
cross-site analysis. We think these selected suspects are representative of
the majority of offenders reported to the police and represent cases of
intimate partner violence that are likely to interest policymakers and
victim advocates. Mainly, the cases represent violence by an adult male
suspect against a past or current female intimate partner.

Intervention comparisons. Our pooled analysis capitalizes on the features
of the experimental design implemented within each site by using the
suspects' assigned intervention, rather than the action actually taken by the
police. This method preserves the integrity of the random assignment. The
nonarrest group included such interventions as mediation counseling, a
citation to appear in court, an order to leave the scene, a restraining order,
or a warning about a future arrest if the officers were called back. The
arrested group comprises only those suspects from each site who were
randomly assigned to be arrested.[b]

The sites differed in the percentages of suspects assigned to the arrest
group and in the misassignment rates. These differences in the percentage
of suspects assigned an arrest were expected, given the different number
and types of interventions in each site. In Dade County, for example, only
two interventions were each assigned one-half of the incidents; in
Colorado Springs, suspects were assigned to one of four groups in equal
proportions. Thus, in Colorado Springs, only one-fourth (26 percent) of
the suspects were assigned to arrest and the other three-fourths to informal
intervention. Overall, 43 percent of suspects were assigned to the arrest
group.

The five sites differed in their rate at which suspects received an
intervention different from the one they were randomly assigned. The
misapplication of intervention occurred when the police officers, after
receiving the assigned intervention code from dispatch or the researcher
team, chose to take a different action. In MDVE and the five SARP
experiments, police officers could change the intervention while on the
scene if one or more specific circumstances arose (such as the suspect
assaulting or threatening the officers or the suspect assaulting or
significantly threatening the welfare of the victim in the presence of the
officers). Overall, misdelivery across the five sites occurred in less than 7
percent of the 4,032 cases, and in 90 percent of the misassigned cases, an
arrest was made when nonarrest had been assigned.[c]

Victim interviews. The rate at which initial and followup victim interviews
were obtained varied within and between sites. For example, the
completion rates of the final interview fluctuated from a high of nearly 80
percent in Milwaukee to a low of 42 percent in Dade. Nevertheless, in our
sample, about 70 percent of the victims were interviewed during the initial
followup period, and 63 percent were interviewed again after about 6
months. Although these completion rates are quite good considering the
challenges of interviewing victims of intimate personal violence (such as
locating them after 6 months), the less-than-100-percent completion rate
poses difficulties for the cross-site analysis. Some researchers have
suggested that the interviewing process itself could increase the likelihood
of officially recorded failures and bias the estimation of the deterrent effect
of arrest that is based on victim interview data.[d] Based on this concern,
our analysis includes an independent measure that partially captures the
effects of the interviewing process. This measure accounts for the
variations across the victims in the length of time between the
experimental incident and the last time they were interviewed, if at all.[e]

Suspect characteristics. Substantial differences in suspect characteristics
were found across the five sites for all demographic measures. Some
notable differences include the large percentage of African-American
suspects in Charlotte and Milwaukee but not in the other sites; the high
percentage of married suspects in Colorado Springs and Dade County but
not in the other sites; the small percentage of suspects in Dade County
without a prior arrest; and the high rate of unemployment among suspects
from Milwaukee. This substantial cross-site heterogeneity should be
considered an asset, because it increases the overall generalizability of our
findings. However, these variations also reinforce the need to control for
the suspects' characteristics as factors in the experimental model; in
previous criminal justice experiments, intervention effects may not have
been found due to the large uncontrolled differences across the subjects
among those deemed eligible for the experiment.[f]

To assess the characteristics of the suspects in the two intervention groups,
exhibit B provides the percentage of suspects assigned to and delivered an
arrest, by categories of the suspects' demographic characteristics. The table
also presents the proportions of suspects not receiving the assigned
intervention, by their demographic characteristics. The characteristics of
the suspects assigned to arrest versus the nonarrest control group and of
those arrested versus those not arrested differed across several suspect
characteristics. Among suspects assigned an arrest, more than expected
were using drugs or alcohol at the time of the incident; were not married,
white, or employed; and had prior arrests. Similar differences were also
found among those actually arrested when they were compared with those
suspects not arrested, except for the increase in the percentage arrested
among those using alcohol or drugs. This change in the assigned and
delivered arrest rates among suspects who used alcohol or drugs is further
confirmed when comparing suspects correctly assigned with those
receiving something other than their assigned treatment. Specifically,
nearly 9 percent of those using intoxicants were misassigned, compared
with 5 percent of those not using intoxicants.

The significant differences displayed in exhibit B, however, do not
necessarily mean that the experimental protocols were systematically
violated by the police officers. Rather, they suggest that when studies
randomly assign different proportions of their cases to arrest, and when the
demographically diverse sites were combined, the characteristics of
suspects in the arrest and nonarrest categories differ in many substantively
important ways (more high-risk suspects in the arrest group than in the
nonarrest group). This issue needs addressing in the outcome analysis.
Therefore, properly controlling for these characteristics in the regression
models is necessary to ensure that the intervention comparisons are fair.
Prior single-site and multiple-site analyses of the SARP experiments have
not controlled for variations in the timing of victim interviews, variations
in misapplication rates, or demographic differences between treatment
groups that occurred when data from multiple sites are combined.

a. For a complete list of the eligibility rules, see Maxwell, Christopher D.,
The Specific Deterrent of Arrest on Aggression between Intimates and
Spouses [diss.], Newark, NJ: Rutgers, the State University of New Jersey,
1998.

b. The comparison between arrest and nonarrest groups, instead of among
all the treatment groups, was suggested by Binder and Meeker in their
critique of the MDVE (Arnold Binder and James W. Meeker,
"Experiments as Reforms," Journal of Criminal Justice 16 (4) (1988): 347-
58). They argued that the best test of deterrence theory is one that
compares those who were punished with those who were not, rather than
individually comparing the differences between arrest and each of the
informal interventions. Besides testing for differences between arrest and
nonarrest, our regression analysis controlled for differences in the average
level of aggression in each site, the effect that the victim interviews may
have had on the outcome, and several suspect and incident characteristics.

c. Although this is a fairly low rate of misassignment, the fact that it is not
zero leaves the possibility that any effect of arrest is potentially biased
upward or downward.

See Richard A. Berk, Gordon K. Smyth, and Lawrence W. Sherman,
"When Random Assignment Fails: Some Lessons from the Minneapolis
Spouse Abuse Experiment," Journal of Quantitative Criminology 4 (3)
(1988): 209-33.

d. Berk, Richard A., and Lawrence W. Sherman, "Data Collection
Strategies in the Minneapolis Domestic Violence Experiment," in
Collecting Evaluation Data: Problems and Solutions, ed. Leigh Burstein,
et al., Beverly Hills, CA: Sage Publications, 1985: 35-48.

e. See Maxwell, 1998 (note a), for further information on how the
interviewing process was addressed in the outcome analysis.

f. Weisburd, David, "Design Sensitivity in Criminal Justice Experiments,"
in vol. 17 of Crime and Justice: A Review of Research, ed. Michael Tonry,
Chicago: University of Chicago Press, 1993: 337-80.

---------------------------

This study was conducted by Christopher D. Maxwell, Ph.D., assistant
professor, Michigan State University; Joel H. Garner, Ph.D., Joint Center
for Justice Studies; and Jeffrey A. Fagan, Ph.D., director, Center for
Violence Research and Prevention (Columbia University), and visiting
professor, Columbia University Law School. The authors would like to
acknowledge the important contribution made by the staff of the National
Archive of Criminal Justice Data and the technical assistance provided by
Dr. Jordan Leiter, former manager of NIJ's Data Resources Program, and
Dr. Angela Moore Parmley of NIJ.

Support for the study was provided by NIJ, the Centers for Disease
Control and Prevention (grant number R49/CCR210534), and the Harry
Frank Guggenheim Foundation. The writing of this Brief was funded
solely by NIJ.

Findings and conclusions of the research reported here are those of the
authors and do not necessarily reflect the official position or policies of the
U.S. Department of Justice.

The National Institute of Justice is a component of the Office of Justice
Programs, which also includes the Bureau of Justice Assistance, the
Bureau of Justice Statistics, the Office of Juvenile Justice and Delinquency
Prevention, and the Office for Victims of Crime.

This and other NIJ publications can be found at and downloaded from the
NIJ Web site (http://www.ojp.usdoj.gov/nij).

NCJ 188199

Support for this research was provided through a transfer of funds to NIJ
from the Centers for Disease Control and Prevention.

Quick Access to NIJ Publication News

For news about NIJ's most recent publications, including solicitations for
grant applications, subscribe to JUSTINFO, the bimonthly newsletter sent
to you via e-mail. Here's how:

o Send an e-mail to This e-mail address is being protected from spambots. You need JavaScript enabled to view it .

o Leave the subject line blank.

o Type subscribe justinfo your name.
    (e.g., subscribe justinfo Jane Doe) in the body of the message.

Or check out the "Publications and Products" section on the NIJ home
page (http://www.ojp.usdoj.gov/nij) or the "New This Week" section at
the Justice Information Center home page (http://www.ncjrs.org).