Police reform is often assessed through visible acts of compliance. Was a complaint registered? Was the prescribed procedure followed? These are necessary measures of institutional performance, but they reveal little about how an officer exercises judgement once a case is underway.
The distinction is particularly important in gender-based violence (GBV), where officers exercise discretion over how a complaint is assessed and pursued. An officer may register a complaint and still give greater weight to the accused than the victim, treat competing accounts differently, or decide which aspects of a case warrant investigative attention. The formal rule may therefore be followed even as the quality of justice experienced by the complainant varies considerably.
Compliance Can Hide Discretion
Evidence from Madhya Pradesh shows why this distinction matters. An experiment involving 323 police officers examined how officers said they would handle fictitious GBV cases. Of those who completed a follow-up case one week later, 93 percent of the control group said they would register a formal complaint. Yet registration was only one decision. In the same case, 84.4 percent of control officers prioritised the victim’s statement.
The two figures capture different layers of police action: registration measures compliance with a formal requirement, while prioritising statements involves professional judgement.
Judgement Within the Rules
The experiment also tested what happened when officers were asked to confront bias in their own responses. A senior police officer privately showed officers assigned to the intervention evidence of bias in how they had handled an earlier fictitious GBV case. The intervention was intended to help them recognise and correct that bias. There was no statistically significant average effect on how officers handled the follow-up case, but the aggregate result concealed different responses.
Among female officers, confrontation increased prioritisation of the victim’s statement by 23 percentage points, equivalent to a 33 percent increase over the control mean. Male officers moved in the opposite direction: prioritisation of the victim’s statement fell by about 13 percentage points, while prioritisation of the accused’s statement increased by about 9 percentage points. The analysis by officer gender was exploratory, so these differences should not be read as evidence that gender itself determined the response.
Yet the evidence suggests that confronting bias can itself produce different behavioural responses. Effective reform must ask whether officers can translate formal expectations into fairer discretionary decisions. The institutional challenge is to identify consequential discretionary behaviour, understand the conditions that shape it, and improve it without generating defensive responses.
Reform Must Govern Discretion
Police reform should govern discretion at the points where it affects a complainant’s treatment. Four changes follow.
The first step is to measure judgment calls directly. Complaint-registration rates are useful, but insufficient. Police organisations should examine whose statement receives priority, which investigative leads are pursued, how competing accounts are assessed, and where cases are dropped or deprioritised. Looking across comparable cases can reveal patterns that aggregate compliance indicators leave hidden.
The second is to structure consequential case-by-case choices more deliberately. Where an officer has legitimate reasons to depart from an expected investigative sequence, those reasons should be recorded and capable of supervisory review. The objective is to preserve professional judgement while ensuring that important discretionary choices leave an evidentiary trail.
The third is to rethink the assumption that the same intervention should be delivered to everyone. In the experiment, 28 percent of female officers and about half of male officers displayed strong pre-treatment bias. But initial bias, officer age, gender, and confrontation intensity varied together, so the experiment cannot establish which factor explains the difference.
This points to a more diagnostic approach to behavioural reform: test whether feedback alone works for some officers while others need supervision or structured decision-making. These alternatives should be assessed by how they change case handling, rather than assumed to work because they are more intensive. The experiment also found little evidence that the intervention changed deeper attitudes, even as it changed some officers’ responses to the subsequent case. Institutions need not wait for beliefs to change before seeking better behaviour. They can change the conditions under which professional decisions are made and scrutinised.
The fourth is to evaluate reform through the decisions it is meant to improve. Attendance, completion and immediate responses to training exercises tell us little if the relevant problem emerges later, during an investigation. This requires examining whether victim statements receive appropriate weight, whether investigative decisions are consistent across comparable cases, and whether departures from established procedures are documented and reviewed. These measures are harder to assess than training completion, but they are closer to the experience of justice.
Measure What Compliance Cannot See
The Madhya Pradesh experiment does not establish that every police reform programme will produce the same pattern. It was conducted in one state and measured responses to structured cases rather than outcomes in actual investigations. It does, however, show that an intervention intended to correct bias can prompt different responses that an average result would conceal.
The next generation of police reform should focus less on eliminating discretion than on governing it better. The goal should be a system in which consequential judgements are structured, recorded, open to scrutiny and evaluated by what they produce for people seeking justice.
The measure of reform ultimately lies in what happens to the person who depends on the system for justice, including whether formal rules translate into fair and consistent decisions.