Domestic violence in India clusters in persistent district hotspots, five-year map shows

Recorded domestic violence against women in India is not scattered randomly across the country’s districts. Instead, it forms a stable geography of concentrated hotspots, with a contiguous belt running through northern and central India and a smaller cluster in the south, according to a new district-level spatiotemporal analysis covering 2020 to 2024. The study, published in Frontiers in Public Health, is the first to combine global and local spatial statistics with directional trend classifications for a composite measure of domestic violence spanning all Indian districts over the most recent five-year reporting period.

Researchers led by KSK Susmitha and JBJ Balamurugan of Vellore Institute of Technology, together with Upali Weerakoon of the University of Peradeniya, built their analysis from the National Crime Records Bureau’s annual Crime in India reports. They extracted five crime categories that specifically codify violence within marital or domestic relationships: Dowry Deaths, Abetment to Suicide, Cruelty by Husband or Relatives, Dowry Prohibition Act offenses, and cases registered under the Protection of Women from Domestic Violence Act. Because district boundaries and names changed between the 2011 Census and NCRB reporting, the team constructed a crosswalk that matched every NCRB-reported district to its Census 2011 counterpart, merging post-2011 subdivisions back into parent districts and excluding island territories and non-geographic units. The result was a harmonized set of 636 mainland districts analyzed on consistent geography.

To combine the five categories into a single measure, the researchers converted each district’s rate for each category into a z-score relative to that year’s national distribution and averaged the five scores with equal weighting, producing a composite Domestic Violence Index. Rates were expressed per 100,000 female population, with district populations estimated by applying state-level projection ratios from official 2011–2036 population projections to Census 2011 female counts. The authors acknowledge that this assumes every district within a state grew at the same rate, an assumption that cannot account for migration or uneven urbanization and that could shift rates, and potentially hotspot classifications, in opposite directions for fast-growing and slow-growing districts in the same state.

The spatial signal was strong and consistent. Global Moran’s I, a statistic measuring whether similar values cluster geographically, indicated statistically significant positive spatial autocorrelation in the DV Index in every study year, with values ranging from 0.360 in 2021 to 0.525 in 2024, all with p-values at or below 0.0001 based on 9,999 random permutations. The index rose from 0.396 in 2020 to its peak in 2024, but the increase was not continuous: it dipped in 2021, climbed in 2022, slipped slightly in 2023, and then reached its highest point. Taken together, the authors write, recorded domestic violence formed spatial clustering that was stronger at the end of the period than at the start.

Local Getis-Ord Gi* statistics, calculated for every district in every year with the same permutation framework, identified between 78 significant hotspots in 2021 and 97 in 2024. Sorting districts by how many years they met the hotspot criterion revealed 30 districts, or 4.7 percent, as persistent hotspots, significant in all five years, and 52 districts, or 8.2 percent, as recurrent hotspots, significant in three to four years. Together these 82 districts, 12.9 percent of the total, formed a contiguous belt across northern and central India, mainly in Uttar Pradesh, Madhya Pradesh, and Haryana, with smaller clusters in Telangana, including Hyderabad, Nalgonda, and Medak, as well as in Kerala, Bihar, and Jharkhand. Hotspots in northeastern India were sporadic only, comprising 13 districts in Assam and one in Meghalaya.

The directional analysis added a time dimension that the authors say national studies have not captured before. Using Mann-Kendall trend statistics on each district’s five annual Gi* values, they classified 14 districts as new hotspots, significant for the first time in 2024; 31 as emerging hotspots, not significant in 2020 but significant by 2024 with a rising trajectory and at least one significant intervening year; 38 as intensifying hotspots, persistent or recurrent districts with a strengthening trend; and 33 as diminishing hotspots, significant in 2020 but no longer by 2024 with a weakening trend. Notably, five of the 30 persistent hotspots showed a weakening Gi* trend and another five showed no directional change, meaning local clustering weakened in some districts even as they remained significant hotspots every year.

The study is unusually candid about the fragility of some of its classifications. Because the Gi* testing involved 636 simultaneous district-level tests per year, the researchers applied a Benjamini-Hochberg false discovery rate correction as a check. Under this stricter adjustment, no district remained a significant hotspot in all five years, so the strict persistent hotspot classification did not survive. However, the correction hit the early years hardest: from 2022 onward, 21 to 30 districts remained significant hotspots each year, suggesting a smaller core withstands the stricter threshold. Sensitivity analyses also tested the equal-weighting assumption. A principal component analysis showed the five categories explain only 25.6 percent of variance on the first component, indicating they do not collapse into a single dominant pattern, and a PCA-weighted alternative index for 2020 identified 135 hotspots against 91 under the equal-weighted index, with 60.4 percent overlap. Repeating the classification with a k-nearest-neighbors spatial weights specification instead of queen contiguity kept 86.7 percent of persistent hotspots at least at recurrent status, indicating the core hotspot geography is not an artifact of the chosen neighborhood definition.

The descriptive trends underneath the spatial pattern were mixed. Median district rates declined for two of the five categories over the period: Dowry Deaths fell 37.7 percent, from 0.63 to 0.39 per 100,000 female population, and Abetment to Suicide fell 23.2 percent, from 0.44 to 0.34. The median rate for Cruelty by Husband or Relatives, the largest recorded category, rose slightly by 2.1 percent, from 8.01 to 8.18, with its interquartile range widening from 15.44 to 18.60, pointing to greater dispersion across districts. Dowry Prohibition Act and PWDVA cases had median district rates of zero in both endpoint years, so percentage change could not be calculated. For the composite index, district-level direction was split almost evenly, with 277 districts showing an increasing trend, 275 a decreasing trend, and 84 no clear trend, even though the national median remained comparatively stable.

The authors stress that these maps show where formal registration of domestic violence is concentrated, not necessarily where prevalence or unmet need is highest. Survey evidence points to a vast gap between lived experience and recorded cases: NFHS-5 found that 32 percent of ever-married women had experienced physical, sexual, or emotional violence from their spouse, yet only 14 percent sought help to stop it, and earlier research found fewer than 2 percent of women experiencing domestic violence had approached the police to file a complaint. Geographic differences in police accessibility, reporting norms, and the functioning of PWDVA mechanisms may therefore partly account for the observed hotspots independently of true victimization levels. The authors also flag that the apparent decline in Abetment to Suicide may partly reflect registration practices rather than a genuine fall in incidence, citing Global Burden of Disease estimates far exceeding NCRB-reported suicide rates in Bihar and Uttar Pradesh.

One possible explanation for the dip in clustering strength between 2020 and 2021 is pandemic-related disruption to reporting rather than a genuine reduction in risk, though the study did not directly test this and presents it as a hypothesis. Global evidence links COVID-19 restrictions to sharp declines in recorded crime followed by post-restriction rebounds, and an analysis of complaints to the National Commission for Women found domestic violence complaints rose sharply in districts with the strictest lockdown rules. The authors suggest the 82 persistent and recurrent hotspot districts offer a concrete basis for directing existing protective infrastructure, such as One Stop Centres, PWDVA protection officers, and women helplines, toward areas with sustained recorded burden, while cautioning that hotspot status should be treated as one input among several. Districts with high recorded burden may have strong reporting systems, weak service infrastructure, or both, and districts with low recorded rates may conceal high hidden prevalence. Because every analysis was conducted at the district level, the results show broad patterns across areas and cannot speak to risk to individuals or to cause and effect. The authors call for triangulation with survey-based prevalence data, finer spatial resolution, and formal confounding-adjustment methods to establish whether recorded crime hotspot mapping can serve as a public health surveillance tool.

Subject of Research: Spatiotemporal mapping of police-recorded domestic violence against women across Indian districts from 2020 to 2024

Article Title: Persistent and emerging district-level hotspots of domestic violence against women in India: a spatiotemporal analysis, 2020–2024

Article References: Susmitha, K., Balamurugan, J., & Weerakoon, U. (2026). Persistent and emerging district-level hotspots of domestic violence against women in India: a spatiotemporal analysis, 2020–2024. Frontiers in Public Health, 14, Article 1994555. https://doi.org/10.3389/fpubh.2026.1994555

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1994555

Keywords: domestic violence, India, hotspot analysis, spatial epidemiology, NCRB, public health surveillance, Getis-Ord Gi*, Moran's I, district-level analysis, crime against women, spatiotemporal analysis, Frontiers in Public Health