The Minneapolis Domestic Violence Experiment (MDVE) tested the deterrent power of police actions upon future violence in domestic relationships. Its major finding was that persons who were arrested were violent again at a much lower rate than those who were dealt with in other ways.

The MDVE is noteworthy for three reasons. (1) It was the first controlled field experiment in which police officers’ responses to a particular problem (misdemeanor-level assault of a spouse or co-habitating adult) were dictated by random assignment rather than officer discretion. (2) It contributed to a nationwide trend toward proarrest policies in major city police agencies. (3) Because the findings were counterintuitive to prevailing ”common sense” on the issue, the National Institute of Justice (NIJ) funded its only major initiative to confirm research, the Spousal Abuse Replication Project (SARP).

The MDVE took place during a time of changing attitudes toward violence within families and between intimates. A strong push for the criminal justice system to treat ”domestic” violence on an equal footing with ”stranger” violence had already resulted in a number of changes in the law that made the experiment possible. Most important was a change in the procedural rules binding police arrest powers. Ordinarily, police are forbidden to make a ”probable cause” arrest for misdemeanor crimes they do not witness; they may act on probable cause only for felony offenses. Unless violence was occurring when the police arrived (and in the vast majority of cases it was not), officers could only advise the victim to swear out a warrant before a magistrate, in order to give the police the legal power to arrest the batterer. Many other circumstances thwarted that option—economic and emotional dependence, the need to care for children, transportation difficulties, and hassles at the courthouse—and most victims failed to follow through after the police left.

New laws created a time-specific window during which the police could make an arrest on probable cause in domestic violence cases: in Minnesota, it was four hours after the assault. Effectively, ”probable cause” meant the victim’s complaint, with or without visible injuries or corroborative testimony. Removal of the assailant from the household was presumed to protect the victim from immediate violence once the police left, though long-term consequences were not as clear.

Changing the law was justified by the greater potential for continued violence in ongoing relationships. Victims’ advocates asserted that lack of action by the police provided affirmation that violence against women was permissible. Police responded that many victims not only failed to follow through but actively sought dismissal of charges when the batterer was arrested. Advocates in turn pointed to the many cases when women tried to extricate themselves from abusive conditions but were actively discouraged from doing so by the police.

The Experiment

The MDVE is the most famous of a series of studies stemming from a collaborative relationship between Minneapolis Chief of Police Anthony Bouza and Lawrence W. Sherman, of the Police Foundation and later the Crime Control Institute. Funded by a grant from NIJ, the experiment consisted of a volunteer team of police officers who responded to eligible incidents of misdemeanor assault. Thirty-eight officers were trained at the outset, and another eighteen later joined the experiment. Upon confirming an incident’s eligibility, officers would take action according to a randomized sequence communicated to them by a researcher. The assailant would be arrested, sent from the scene, or ”advised” and left at the scene with the victim.

A number of circumstances might exempt a case. If the offender had an outstanding arrest, an arrest would have to be made regardless of what the experimental protocol directed. An offender who assaulted a police officer would be arrested for that offense, negating the protocol as well. Police-delivered treatment was different from the protocols in 18% of the cases, primarily by arrests of resistant suspects, made when protocols indicated another action. ”Arrest” in Minneapolis meant spending a night in jail, which may have created additional deterrent effect.

A total of 330 cases were entered into the experiment, and 314 deemed eligible for analysis. Two hundred of the cases were entered by a small group of twelve officers; seven officers entered only one case each, and some contributed none. Only one suspect designated for arrest was not arrested, but the “advised” and “sent” categories achieved only 78% and 73% fulfillment, respectively.

Sixty percent of the victim and suspect cohorts were unemployed; more than 40% of each group were high school dropouts. Most were cohabitating, some apparently only for a short time. Only 40% had been married at any time. Sixteen percent were Native American, 35% African American, and the remainder European American. Only 5% of the suspects had prior arrests for domestic violence, though almost 60% had criminal records. Overall, both victims and suspects in the experiment represented the “have-nots” of the city.

Victims were asked about subsequent violence at two-week intervals after the initial incident. Only 62% could be located for the first follow-up, and fewer than one-third were still involved at the six-month mark. Their responses were compared to official police records of repeat incidents, which contained all reports involving victims and suspects during the twelve-month follow-up period.


The results indicated that 10% of arrested suspects committed subsequent acts of violence against the same partner, compared to 19% of those advised and 24% of those sent from the scene. (Analysis of the results ”as assigned” by the protocols rather than “as delivered” by police produced repeat violence rates of 7%, 14%, and 18% in those categories.) The more complete official data demonstrate a similar pattern of fewer assault cases, but victim interviews indicated more violent behaviors, including threats and destruction of property. Several reanalyses of the experiment’s data have been conducted, revealing a calling bias (a majority of calls came from third parties, not the victim) and a potential erosion of deterrence by the end of the six-month follow-up period.


When announced in the context of the larger social movement, the experiment’s results were widely cited by advocates pushing for more serious police attention to domestic violence. Sherman and Berk (1984b) cautioned against assuming that arrest worked best in all situations until replication could explore additional issues, including the potential for a ”backfire effect” of arrest to increase future violence among the unemployed.

The publicity and the policy response generated considerable debate among researchers but succeeded in highlighting the need for replication. Other questions were raised, including whether Minneapolis was truly representative of American urban demographics, whether violence might have been displaced to other victims, and whether the erosion of victim follow-up data masked tactical displacements in violence against the victims. Concerns about interviewer effects, ineffective police mediation skills, and the effects of jail time (rather than the arrest itself) have also been voiced as problematic areas.


The importance of the issue and unanswered concerns spurred the National Institute of Justice to fund additional experiments in different cities: Omaha, Nebraska, Miami, Florida, Charlotte, North Carolina, Colorado Springs, Colorado, and Milwaukee, Wisconsin, whose replication project was directed by Sherman. Despite slight variations based in local conditions at each site, all five SARP experiments followed the Minneapolis model closely enough to allow valid comparisons.

No official data from the other sites confirmed a deterrent effect at the six-month mark (though Miami did find deterrence in one of the two official measures employed). Three (including a ”borderline” finding in Omaha) observed deterrence effects in victim interviews during that period. An ”escalation effect”— increased incidence of violence—was found in the official records from six months to a year after the initial contact in three cities, but no city had that effect with victim interviews. Results were slightly more positive for a period of thirty to sixty days after the initial contact, with four sites finding significant or borderline reductions in violence in the short term.

In the three cities that examined the issue (Milwaukee, Omaha, and Colorado Springs), the suspect’s employment status had a significant and important impact. Suspects who were employed were more likely to be deterred by arrest, but unemployed suspects were more likely to engage in escalating violence.

The Minneapolis Domestic Violence Experiment provided the first statistically significant evidence for the effectiveness of any police policy. During the following decade and a half, many more randomized, controlled field experiments were conducted, including the Drug Market Analysis Program (DMAP) and experiments in proactive gun law enforcement. The experiment was a tipping point in police research, helping to bring policing out of the realm of craft and into that of scientific inquiry.

The experiment also helped to raise public awareness of the issue of domestic violence and prompted the police to reconsider their long-held assumptions. While many of the larger issues still remain, most notably the tendency of victims to seek temporary relief but avoid longer-term measures, one of the legacies of Minneapolis has been a sustained effort to address the problem effectively. ”Preferred arrest” policies have replaced mandatory arrest, providing more latitude to officers, but victimless prosecution, mandatory counseling programs for batterers, and a growing network of shelters and assistance programs for victims continue the work begun in Minneapolis.

Next post:

Previous post: