Domestic abuse / Evidence / Better protection
No two domestic relationships are the same,
so neither should our response be to them
Why we should rethink guidance on arresting at every domestic incident
Identify the danger within the relationship. Match the intervention to it. Judge the response by the protection it delivers.
Conceptual illustration: connections between homes and relationships. No locations, cases or risk scores are represented.
Key takeaways · The two-minute read
Four steps through
the evidence.
Evidence
Published meta-analyses associate particular abusive behaviours with partner homicide. An offence label cannot describe the whole relationship.
Associations, not individual predictions.
Matias and colleagues, 2020Read the chapter →Understanding
Evidence about partners cannot automatically tell us the danger between a teenager and a stepfather.
Age and direction limit what transfers.
Bracewell and colleagues, 2022Read the chapter →Options
Require officers to explain what danger their chosen response addresses and what protection follows.
An evidence-informed proposal, not a validated instrument.
Read the chapter →Next steps
Judge a pilot through subsequent harm, victims’ safety and access to justice. Arrest totals are a process measure.
Evaluate outcomes before changing thresholds.
Read the chapter →The response should follow the danger
Police should organise their response to domestic abuse around the danger within the relationship, with the investigation of the reported offence running alongside it. The aim is to identify and interrupt serious harm earlier, and use police powers more effectively.
A common assault, an assault occasioning actual bodily harm (ABH), and a communications offence describe different allegations. None gives a complete account of what one person is doing to another over time. Published meta-analyses identify associations between partner homicide and behaviours including threats, stalking and control. Those findings give police a reason to look beyond the offence that brought them to the address. Matias and colleagues, 2020.
The decision facing policing is whether guidance on arrest gives that understanding enough influence over what happens next. This article argues for reviewing the guidance and testing a more explicit connection between relationship risk and intervention. Forces should be accountable for protection that continues after the initial attendance.
Efficiency here means directing the right work towards preventing harm. A shorter incident log or a lower arrest count cannot establish that police have become better at protecting people.
This is a policy argument for England and Wales, drawing on existing meta-analyses and British research. It is not a new meta-analysis, a risk calculator or authority to depart from current law and policy.
The guidance strongly favours arrest, but the law requires a reason
The College of Policing’s suspect-strategy guidance, revised in March 2026, states:
“When an offence has been committed in a domestic abuse case, officers should arrest the suspect.”
The accompanying text says arrest will normally be necessary. That is a strong starting point. It is also more specific than saying every domestic incident automatically requires an arrest: the wording concerns an offence. College of Policing, suspect strategy; revision record, 30 March 2026.
The statutory power remains subject to its own conditions. Under section 24 of the Police and Criminal Evidence Act 1984, the officer needs the relevant grounds concerning an offence and reasonable grounds for believing arrest is necessary for a specified purpose. Code G requires an individual assessment, considering the victim, offence, suspect and investigation. Preventing injury, protecting a child or vulnerable person, and enabling a prompt and effective investigation are among the possible grounds. PACE, section 24; Code G, paragraphs 2.1–2.9.
A standard-risk assessment does not cancel investigative necessity. Equally, a domestic-abuse flag is not a substitute for explaining necessity. Arrest, charge and the eventual case outcome are separate decisions.
There is a serious reason to preserve a strong expectation of action. HMICFRS’s A duty to protect report identified confusion about positive action and weaknesses in the use of protective measures. Removing an expectation without improving supervision and the available alternatives could leave victims with less protection. HMICFRS, 2021.
The case for revision therefore has to answer a harder question than whether arrest is sometimes unnecessary. It must explain how officers would make a better decision and what action would follow. Supervisors also need to recognise a dangerous decision disguised as discretion.
What the meta-analyses actually tell us
A meta-analysis combines results from several studies. Its value depends on what those studies measured and whom they compared.
Spencer and Stith’s review, published online in 2018 and in a 2020 journal issue, examined 17 studies and 148 effect sizes concerning male perpetrators and female intimate-partner victims. It compared homicide or, in some samples, homicide and attempted homicide with non-fatal partner violence. Eleven studies were American; the remainder included Britain and other countries. These are international associations, not British individual-risk estimates. Spencer and Stith, 2020.
Figure 01 / Evidence
The history contains warning signs.
Selected associations from partner-homicide research
Previous non-fatal strangulation
7.23odds ratio
95% confidence interval
4.61–11.34
5 effect sizes
Previous threat with a weapon
7.36odds ratio
95% confidence interval
2.99–18.11
5 effect sizes
Stalking
3.13odds ratio
95% confidence interval
2.58–3.81
4 effect sizes
Read the figures as a table
| Marker | Odds ratio | 95% interval | Effects |
|---|---|---|---|
| Previous non-fatal strangulation | 7.23 | 4.61–11.34 | 5 |
| Previous threat with a weapon | 7.36 | 2.99–18.11 | 5 |
| Stalking | 3.13 | 2.58–3.81 | 4 |
An odds ratio describes an association in the research. It does not mean that seven out of ten victims will be killed, give a probability for next week, or prove that removing a marker would reduce risk by the same amount. These estimates are unadjusted: other differences between the compared groups can contribute. The markers cannot be multiplied together to score a relationship.
Matias and colleagues’ separate 2020 meta-analysis included 28 empirical articles, with comparison groups that included non-fatal abuse, other homicides and homicide-suicide. Its findings also drew attention to abusive relationship dynamics. Kim’s later three-level meta-analysis included 29 studies and 473 effects and found that results varied with the sex of those involved and the comparison group. Those are reasons to examine precisely which question each result answers. Matias and colleagues; Kim, 2023, online 2022.
British evidence adds context. Chopra and colleagues examined 263 partner-homicide reviews published in England and Wales between July 2011 and November 2020, highlighting separation, stalking and new relationships. Because these were fatal cases, the study cannot establish the risk among all separating couples. Separation should prompt questions about behaviour and changing access, not a conclusion based on relationship status alone. Chopra and colleagues, 2022.
The reviews are not three independent populations to add together. They can reuse earlier studies. Their findings support more searching enquiries about behaviour; they do not establish a reliable ordering of every domestic relationship or tell us which intervention will work for an individual.
Nor is recognising the pattern a new ambition within policing. DARA was designed with greater attention to coercive control, and the NPCC confirmed in August 2025 that it remained the recommended frontline assessment tool. The proposed improvement is to make the assessed danger more visibly determine the protection, investigative priorities and follow-up. NPCC, 2025.
The offence and the relationship answer different questions
The practical implication is that the seriousness of the current allegation and the danger of future harm deserve separate attention. One concerns what has happened; the other concerns what the available evidence suggests could happen next.
Figure 02 / Understanding
Two questions.
Both need an answer.
An offence label alone cannot choose the response.
What happened?
Establish the alleged offence, injury, evidence and immediate needs.
Investigate the incident.What danger continues?
Establish the pattern, threats, access, fear, dependencies and what has changed.
Understand the relationship.Explain arrest necessity, protection and follow-up. Neither answer cancels the other.
A report initially described as common assault can justify urgent protective attention if enquiries uncover a dangerous pattern. Calling it common assault must not close off investigation of other offences. Conversely, an ABH allegation still demands an appropriate investigation even where a careful assessment identifies fewer indicators of continuing danger.
“Communications only” is particularly unhelpful shorthand if it ends the enquiry. Police need to establish the content, repetition, context and effect of the contact. The statutory definition of domestic abuse expressly includes threatening, controlling, coercive and psychological abuse; physical contact is not required. That definition does not itself prove any particular criminal offence. Domestic Abuse Act 2021, section 1.
This is not a proposal to create a new low-risk category from an officer’s first impression. A missing history may mean nobody asked, the victim could not speak privately, records were not available, or earlier abuse was never reported. Those possibilities require enquiries. They should not silently become reassuring answers.
The victim’s account needs room in the decision. What are they frightened will happen? What has changed? Officers need to understand what would make seeking help unsafe. The proposed supervisory record should distinguish what is known, what remains unknown and who will obtain the missing information.
Serious harm is also broader than predicted homicide. Ongoing coercion and psychological injury deserve a response in their own right; protection should not depend on a prediction that someone will be killed. CPS domestic abuse guidance.
A risk grade should lead somewhere. If an assessment identifies stalking, the protection plan should address access and unwanted contact. If the concern is danger around separation, the plan should address that period. A generic record that simply repeats “high risk” leaves the central decision unexplained.
A teenager and a stepfather are not a substitute for a partner-abuse study
The domestic-abuse definition can cover a 16-year-old and a stepfather as well as separating spouses. The statutory age threshold is 16, and the rules on being personally connected include qualifying relatives and stepfamily relationships. That breadth is necessary for recognition; it does not make these relationships interchangeable for research or response. Domestic Abuse Act 2021, sections 1–2; Family Law Act 1996, section 63.
Bracewell and colleagues examined 66 adult-family domestic homicide reviews in England and Wales, drawn from a larger collection of 317 reviews. They identified intertwined circumstances involving care, economic pressures, trauma, mental health and substance use. These fatal-case reviews illuminate circumstances and opportunities for intervention. Without comparable non-fatal cases, they cannot tell us how much a particular circumstance changes an individual’s likelihood of being killed. Bracewell and colleagues, 2022.
An adult child threatening a parent, a parent abusing a teenager and a former partner stalking an ex-partner require different enquiries. Who depends on whom for housing or care? Who is frightened? Who is controlling whom? Has apparent aggression occurred in response to earlier abuse? Officers must establish direction and context rather than infer them from age, sex or who made the call.
A 16-year-old also needs a response that recognises they are a child. CPS guidance explicitly addresses teenage relationship abuse and the overlap with child safeguarding. An adult partner-homicide model cannot settle those decisions. CPS domestic abuse guidance.
None of this supports a presumption that stepfamilies, older people or people with mental illness are dangerous. Nor does limited research on a group establish that its victims are safe. Relationship-specific questions should improve access to protection, including for people poorly represented in the evidence.
The arrest evidence does not justify a universal promise
Recognising danger and reducing it are separate research problems. A study showing an association with homicide cannot establish whether arrest will prevent the next assault.
Connor and colleagues’ systematic review, published online in 2024 and in a 2025 issue, included 34 studies of adults in the United States comparing arrest with a non-arrest response. Its meta-analysis did not find a consistent reduction in repeat violence. Results varied, including according to the comparison response. Much of the evidence was old, and outcomes beyond repeat violence received limited attention. Connor and colleagues, 2025.
That finding challenges an unconditional claim that arrest reliably prevents recurrence. It does not prove arrest is ineffective in every case, measure its value for a particular investigation, or establish which English or Welsh relationships should receive a non-arrest response. Different legal systems, services and operational practices limit transfer. The review studied adults; it cannot settle the response to a 16-year-old.
The appropriate policy question is therefore what a complete intervention achieves. Arrest can be necessary to interrupt immediate danger or progress an investigation under Code G. What protection then follows, including on release, requires a further decision. PACE Code G.
The strongest objection to greater discretion is that it could revive dismissal of abuse as a private dispute. That objection deserves an answer in the design. A non-arrest decision should remain accountable, based on lawful grounds and an evidenced plan. The victim must not carry responsibility for making the police decision, and reduced arrest must never become a performance target.
There is also a cost. Checking the history and speaking privately with the victim take officer time. Obtaining specialist advice and reviewing the case take more. Specialist services need capacity to accept referrals. Changing guidance while leaving those tasks unfunded would create an alternative on paper.
Make the assessment change the work
The College, NPCC and participating forces should test a revised decision standard with specialist domestic-abuse services and survivor involvement. The proposal is to require an explanation of the relationship’s danger, the legal purpose of any arrest, and the protective action that follows. It should supplement validated assessment processes, not introduce points copied from a meta-analysis.
Figure 03 / Proposed practice
Make the risk assessment
change the work.
Recognise
Record who is harming whom, the history and the gaps.
Act
Explain the legal grounds and how the intervention addresses danger.
Follow through
Name the person responsible for protection, review and the next contact.
Test
Check subsequent harm, survivor safety and investigation outcomes.
An arrest total measures activity. The evaluation must establish what happened afterwards.
For a relationship showing serious warning signs, the test should be whether police accelerate the work those signs require. That could include prompt specialist review, enquiries into the wider pattern, attention to children and other affected people, and an explicit plan for contact or release. Protective measures should be selected under their own legal tests. A victim should be able to understand who owns the next action and when it will happen.
Where immediate danger appears lower, the officer should still explain the investigation and safeguarding response, identify missing information and specify what would trigger reassessment. The current offence retains its seriousness. Lower assessed danger is not a finding that no crime occurred or that the victim needs no help.
Supervisors should review a sample of arrests as well as non-arrests. Both decisions can conceal poor reasoning. Review should ask whether the available history was considered, whether the person believed to pose the danger was correctly identified, and whether the action addressed that danger. An arrest count cannot answer those questions.
The design should also follow the perpetrator across relationships where lawful information sharing permits it. Keeping each attendance in a separate box risks missing information relevant to another victim. That is a proposed recording and review requirement; this article does not claim a quantified benefit from it.
A better response must demonstrate better outcomes
A pilot should begin with a clear account of current practice, then compare the revised approach with usual practice across participating teams or forces. Researchers should specify the outcomes before implementation and account for initial risk and differences between the groups. Simply comparing arrested and non-arrested people would be misleading if those arrested began with greater danger.
Measure serious repeat harm, victims’ reported safety and continuing coercion, access to support, and the progress of investigations. Keep homicide, serious non-fatal injury and any repeat report distinct. Audit cases initially assessed as lower risk that later escalate. Check results separately for partner and family abuse, age groups and other populations where the data permit a credible comparison.
Follow-up at three, six and twelve months would be a practical starting proposal, subject to the study’s design and resources. An independent evaluation should examine the number of missed dangerous cases and whether less-studied groups lose access to protection. Homicide is too rare to be the only short-term test.
Police records alone are insufficient to interpret success. Fewer repeat reports could reflect less abuse, reduced confidence in reporting or a change in recording. Confidential survivor follow-up and scrutiny of case handling are needed to help distinguish those explanations. The proposed evaluation should never treat a victim’s silence as proof of safety.
Efficiency belongs in that evaluation too: how quickly did relevant protective action happen, and what officer and partner-agency resources did it require? Time saved has value if it produces capacity for effective protection. A fall in workload without improved safety would not meet the purpose of the reform.
The decision for police leaders is whether they can show how the response follows the danger. Reviewing arrest guidance is justified as part of that work, with the legal safeguards retained and the alternatives properly tested. Police should understand the relationship well enough to explain why their intervention fits the danger. They should then be able to show whether people are safer afterwards.
Read the evidence and test the definition
This is a targeted synthesis of existing reviews, not an exhaustive systematic search or a newly pooled analysis. Spencer and Stith’s full text and its Table 3 were checked; the selected numerical estimates are reproduced with confidence intervals. Its table and narrative differ on some other estimates, so those disputed numbers are not reproduced. Matias, Kim and Connor are used at the level supported by their retrieved abstracts. Full-text appraisal of those reviews remains a publication-review task.
The College’s current arrest wording and March 2026 revision were checked through its indexed official text; the full page could not be retrieved during this review. The legal distinction is independently supported by Code G. No claim is made that all forces or officers currently respond identically.
- Spencer and Stith: partner-homicide risk factors · 17 studies; international; male perpetrators/female victims. Full-text copy, Table 3, PDF page 9.
- Matias and colleagues: intimate partner homicide · 28 articles; mixed comparison groups.
- Kim: three-level meta-analysis · 29 studies; comparison-dependent findings.
- Connor and colleagues: outcomes associated with arrest · 34 included studies; United States adults.
- Chopra and colleagues: partner homicide in England and Wales · 263 reviews; no non-fatal comparison group.
- Bracewell and colleagues: adult-family homicide · 66 reviews; England and Wales; no non-fatal comparison group.
- College suspect strategy, PACE Code G and CPS domestic abuse guidance.
For the related question of what a risk grade actually communicates, read Why medium risk is the safe option.
Optional: six questions about the domestic-abuse definition
The first five scenarios are hypothetical; the sixth concerns a real judgment. They test legal classification, not future risk. They include a housing-law judgment, which must not be confused with a criminal conviction. The answers link to the relevant authority.
Choose one answer in each. Scenarios 1–5 are hypothetical training examples; scenario 6 summarises a real judgment. This is a learning exercise, not a risk-assessment tool. Answers stay in your browser.
1. One incident between spouses
Show a hint
Separate the general definition from offences that require a repeated pattern.
Show answer and legal explanation
Answer B. Sections 1 and 2 of the Domestic Abuse Act 2021 cover this relationship and conduct. Section 1(3) expressly permits a single incident. Whether an offence can be proved remains a separate question.
Why the other options do not fit
A. Section 1 allows a single incident; previous abuse is not a prerequisite.
C. Visible injury is not a condition of the statutory domestic-abuse definition.
D. Marriage provides the connection; sharing an address is not required.
2. A shared address without a qualifying relationship
Show a hint
Ask what creates the personal connection: an address or a relationship recognised by the Act?
Show answer and legal explanation
Answer C. A domestic setting is not enough. Section 2 specifies the qualifying connections. This classification does not decide whether a threat is criminal or how urgently the person needs protection.
Why the other options do not fit
A. Living together alone does not establish a section 2 personal connection.
B. Threatening behaviour is included in section 1; the missing element here is the qualifying connection.
D. The location does not replace the age, relationship and behaviour conditions.
3. A 16-year-old stepson threatens his stepfather
Show a hint
Check both the exact age threshold and the statutory meaning of a relative.
Show answer and legal explanation
Answer A. Section 2(1)(g) includes relatives, using the Family Law Act 1996 section 63 definition. Direction still matters for understanding harm and selecting research. Meeting the definition neither makes this equivalent to spousal abuse nor assigns a homicide-risk grade.
Why the other options do not fit
B. Section 1 requires both people to be at least 16, not 18.
C. The statutory definition of relative includes stepfamily relationships without requiring adoption.
D. The definition does not restrict abuse to one direction within the qualifying relationship.
4. A child experiences and is directly threatened with abuse
Show a hint
The adult definition, recognition of affected children and prosecution policy serve different purposes.
Show answer and legal explanation
Answer D. Section 3 applies to an under-18 who sees, hears or experiences the effects of domestic abuse and is related to either adult. The CPS expressly applies its guidance more broadly than section 1’s age limit. Assess the direct threat and the child’s safety as well as the abuse between the parents.
Why the other options do not fit
A. Section 3 recognises a related child who sees, hears or experiences the effects of the qualifying adult abuse.
B. Section 3 recognises children as victims in specified circumstances; it does not rewrite section 1’s two-person age condition.
C. CPS says prosecutors should apply its domestic-abuse guidance regardless of the victim’s or suspect’s age.
5. Coercive control after separation
Show a hint
Check when the cohabitation rule changed, then distinguish an offence’s ingredients from proof of them.
Show answer and legal explanation
Answer B. The cohabitation restriction was removed from 5 April 2023; that change is not retrospective. Section 76 requires repeated or continuous behaviour, the qualifying connection, a serious effect, and knowledge or what the accused ought to know. These events are expressly in 2024.
Why the other options do not fit
A. For behaviour from 5 April 2023, former partners and qualifying family members need not live together.
C. Serious alarm or distress substantially affecting usual day-to-day activities is an alternative route to serious effect.
D. A relationship alone does not prove every element, and criminal liability must be established on the evidence.
6. Using a judgment without overstating it
Show a hint
Identify the legal question the court decided and the order it actually made.
Show answer and legal explanation
Answer C. The case concerned section 177 of the Housing Act 1996 as it then stood. See Lady Hale at paragraphs 35–37. It illustrates why non-physical abuse must not be dismissed, but does not determine guilt or interpret the later 2021 Act.
Why the other options do not fit
A. This was a housing appeal, not a criminal prosecution.
B. The Court rejected the narrow physical-contact interpretation.
D. The judgment did not create that automatic entitlement.