Man being placed in handcuffs
VoxEU Column Gender Health Economics

Arrests are effective in breaking the cycle of domestic violence

Domestic violence, which affects roughly a third of women worldwide, is rarely a one-time event. Most victims are caught in a cycle of violence, repeatedly abused by the same partner. This column examines the impact of a controversial policing strategy aimed at breaking the cycle in a region of England: arresting suspects on the spot. This approach reduced repeat victimisation over the following year by around 50%. Data suggest that an immediate arrest facilitates a cooling down period in the short run and deters further abuse in the long run.

Domestic violence is a serious crime affecting roughly one-third of women worldwide (WHO 2021). This global problem increased after the onset of the Covid pandemic, likely due to increased time spent at home and stress caused by job loss (e.g. Bhalotra 2020, Anderberg et al. 2021, Yamamura and Tsutsui 2020).

A key aspect of domestic violence is that it is rarely a one-time event: many women are abused repeatedly by the same partner (Tjaden and Thoennes 2000, Aizer and Dal Bo 2009). A central question is how best to police this crime to break the cycle of violence. A highly controversial police response is to make a provisional arrest of suspects on the spot.

Deterrence or backlash?

Advocates of arrest as a response to domestic violence argue that it not only temporarily incapacitates the offender but also deters future offenses by signalling a high cost for repeat incidents (Berk 1993). Critics of arrest fear repercussions (Schmidt and Sherman 1993, Goodmark 2018) and argue that while an arrest provides immediate relief, it triggers an escalation of violence in the longer term. Arrest might also impact whether victims will report future incidents of domestic violence: some argue that it empowers women to do so while others suggest that it prevents future calls to the police since women fear their partners’ reactions.

Although there is a burgeoning empirical literature on the causes and consequences of domestic violence (e.g. Bhuller et al. 2023, Bhalotra et al. 2021), it has been challenging to assess the impact of arrest on repeat victimisation. Most of the evidence base stems from the Minneapolis Domestic Violence Experiment of 1981 (Sherman and Berk 1984) and its replications in five other US cities and counties. In these experiments, the research design called for patrol officers who encountered a violent situation to randomly implement one of three responses: arrest, separate the victim and offender, or provide advice. Although this approach was innovative, the studies did not yield consistent results, struggled with small sample sizes, and were plagued by noncompliance in treatment assignment (i.e. patrol officers frequently deviated from the randomly assigned response). Related work has looked at the impact of law enforcement on re-victimisation along other margins, including the effects of ‘no drop’ policies (Aizer and Dal Bo 2009), specialised domestic violence courts (Golestani et al. 2021), and criminal charges (Black et al. 2022).

Overcoming data and selection bias challenges

Estimating the consequences of arrest on future victimisation is challenging for two reasons: data availability and the fact that arrests are not random. In recent work (Amaral et al. 2023), we tackle both of these challenges to ascertain whether arrest causes deterrence or backlash.

On the data front, we use over 124,000 domestic violence emergency calls (999 calls) for West Midlands, the second most populous county in England. We track incidents from the time of an emergency call to the arrival of police officers and any intervention on the crime scene. Our data includes information on whether these first-response officers provisionally arrest a suspected offender. We merge these records with information on whether an investigating officer subsequently initiates a criminal investigation and, if so, whether the offender is charged with a crime.

To create a linked panel of violent incidents in the same household over time, we use information on the geocoordinates of the incident’s location. This approach exploits the fact that most domestic violence occurs at home and that most police responses occur through a 999 emergency call. The main advantage of our approach is that we can track repeated acts of violence even if no formal criminal charge is filed. For this reason, it is likely that we capture a much higher proportion of repeat cases compared to other panel datasets that only contain information on formal charges.

To deal with the fact that arrests do not occur randomly, but are likely to occur more often in cases where the probability of repeat victimisation is high, we exploit two features of our setting. First, police officers are randomly assigned to emergency calls conditional on a set of observable characteristics (time, location, and priority assigned by the call handler). Second, police officers differ in their tendency to make arrests. We construct the average arrest propensity in the other cases officers have handled and use this as an instrumental variable for whether an arrest will be made in the current case. The intuition is that officers with a higher propensity to arrest in other cases will also be more likely to arrest in the current case. But since officers are randomly assigned to cases, this higher arrest propensity should not be related to the characteristics of the current case. Indeed, while the instrumental variable empirically predicts whether there will be an arrest in the current case, it does not correlate with observable case characteristics.

A reduction in repeat emergency calls

Our main result is that an arrest reduces the probability of a repeat domestic violence call within the following 12 months by 49 percentage points. This corresponds to a 51% reduction in repeat offenses. Our analysis also explores the timing of the drop. Without an arrest, one-fourth of perpetrators become violent again within 96 hours. An arrest prevents almost all of these immediate recurrences (see Figure 1). In addition to this short-run effect, further reductions in repeat offenses can be observed over the following year, indicating a longer-term effect as well (see Figure 1).

Figure 1 Effect of arrest on repeat victimisation

Figure 1 Effect of arrest on repeat victimisation

A change in abuse or a change in reporting behaviour?

Whether the reduction in the number of repeat emergency calls for domestic violence after an arrest is good or bad depends on whether it reflects an actual decrease in abuse or simply a change in reporting behaviour by victims. To unbundle these alternative explanations, we develop a simple threshold reporting model. In this model, victims who experience a backlash after an arrest increase the amount of abuse they are willing to tolerate before reporting it in the future, while victims who are empowered by the deterrent effect of an arrest report future abuse at a lower threshold.

Using a measure for the severity of repeat emergency calls, we find that reporting thresholds on average decrease after an arrest: there is a sharp drop in severe domestic violence calls and an increase in less severe domestic violence calls. This compositional shift is statistically significant. Interpreted through the lens of our threshold reporting model, this means that the decrease in domestic violence emergency calls is not due to a change in reporting behaviour but rather to a decrease in actual violence.

As a second test, we use the differences between calls made by victims themselves and those made by a third party, such as a neighbour. The idea behind this comparison is that the reluctance to report a repeat offense should be much greater for the victim due to fear of retaliation than for a third party. However, we find the opposite: a larger (but statistically insignificant) decrease in reports by third parties compared to victims. This is also consistent with a decrease in actual abuse.

Possible mechanisms

Our analysis explores several mechanisms which could explain the drop in repeat victimisation. The near-elimination of repeat domestic violence in the first four days after an arrest mentioned above suggests a cooling-off period when the perpetrator is temporarily detained and removed from the scene. However, further reductions over the following year also indicate a longer-term deterrent effect. Consistent with a deterrent effect, we find that an arrest increases the likelihood of the offender being formally charged with a crime. For non-arrested offenders, the probability of being formally charged with a crime is only 2%, while this probability increases to 12% for arrested offenders. These results contest the claim that an arrest has weak consequences and is therefore ineffective.

Concluding remarks

Our research provides evidence that arrest can help break the cycle of domestic violence. The findings argue against recent calls for the complete decriminalisation of domestic violence. Instead, they suggest the optimal police response in our setting is to lower the police threshold for arrests and take strong action against perpetrators. We caution, however, that this conclusion does not necessarily apply to countries such as the US, where the current arrest rate is already much higher than in the UK.

Authors’ note: This column is based on research by the authors (Amaral et al. 2023), which uses data generously provided by the West Midlands Police.


Aizer, A and P Dal Bo (2009), “Love, hate and murder: Commitment devices in violent relationships”, Journal of Public Economics 93(3–4): 412–428.

Amaral, S, G B Dahl, V Endl-Geyer, T Hener and H Rainer (2023), “Deterrence or Backlash? Arrests and the Dynamics of Domestic Violence”, NBER Working Paper 30855.

Anderberg, D, H Rainer and F Siuda (2021), “Assessing the magnitude of the domestic violence problem during the COVID-19 pandemic”,, 20 November.

Berk, R A (1993), “What the scientific evidence shows: On average, we can do no better than arrest”, in Current Controversies on Family Violence, Sage Publications.

Bhalotra, S (2020), “A shadow pandemic of domestic violence: The potential role of job loss and unemployment benefits”,, 13 November.

Bhalotra, S, D Britto, P Pinotti and B Sampaio (2021), “Job displacement, unemployment benefits and domestic violence”, CEPR Discussion Paper DP16350.

Bhuller, M, G Dahl, K Løken and M Mogstad (2020), “The effects of domestic violence on the mental health and well-being of victims and their children”,, 27 February.

Black, D A, J Grogger, K Sanders and T Kirchmaier (2022), “Criminal charges, risk assessment, and violent recidivism in cases of domestic abuse”, NBER Working Paper 30884.

Golestani, A, E Owens and K Raissian (2021), “Specialization in criminal courts: Decision making, recidivism, and re-victimization in domestic violence courts in Tennessee”, SSRN Working Paper 3923832.

Goodmark, L (2018), Decriminalizing domestic violence, University of California Press.

Schmidt, J D and L W Sherman (1993), “Does arrest deter domestic violence?”, American Behavioral Scientist 36(5): 601–609.

Sherman, L W and R A Berk (1984), “The specific deterrent effects of arrest for domestic assault”, American Sociological Review 49: 261–272.

Tjaden, P G and N Thoennes (2000), Full report of the prevalence, incidence, and consequences of violence against women: Findings from the National Violence Against Women Survey, US Department of Justice, Office of Justice Programs, National Institute of Justice.

WHO (2021), Violence against women prevalence estimates, 2018.

Yamamura, E (2020), “COVID-19, mental health, and domestic violence: Evidence from Japan”,, 22 June.