# External evidence and analytic scope

Research update: 24 September 2026. This supplement adds independent source documentation and separately defined public datasets. It does not refit or replace the published arrest-record models. Acquisition dates, source hashes, processing code and limitations accompany each dataset.

## Questions the added sources can address

1. **Reporting completeness.** Compare supplied record activity with official reporting qualifications and separately tabulated agency activity. A difference between unlike units is a diagnostic, not a completeness percentage.
2. **Outcome definition.** Distinguish calls for assistance, reported offenses, domestic-violence indicators and administrative arrest groups. They represent different stages of reporting and enforcement; their counts must not be added or treated as interchangeable.
3. **Exposure measurement.** Extend observed-station coverage and establish a temperature baseline preceding the 2021-2024 model period. Station observation days and subdaily sampling must be audited before substituting them for civil-day reanalysis.
4. **Geographic interpretation.** Confirm service jurisdiction and population context before comparing locality totals. A recorded city is not a harmonized population at risk.

## Primary research

**Dey AK, Rao N, Thomas EE, et al. (2025).** Extreme Heat and Calls to Law Enforcement Related to Domestic Violence. *JAMA Network Open*, 8(8), e2530530. [doi:10.1001/jamanetworkopen.2025.30530](https://doi.org/10.1001/jamanetworkopen.2025.30530).

The New Orleans study used 2011-2021 DV-related calls and a time-stratified case-crossover design. Five consecutive days above a local historical thermal-index threshold were associated with higher call odds (OR 1.07, 95% CI 1.03-1.12). The exposure was a thermal-comfort index, not daily maximum air temperature. Its outcome, climate and exposure definition differ from this study. It supports investigation of sustained heat; it does not establish a San Diego arrest effect. The paper also documents a later change in local call classification, illustrating why longitudinal source definitions require review.

**Heilmann K, Kahn ME, Tang CK (2021).** The urban crime and heat gradient in high and low poverty areas. *Journal of Public Economics*, 197, 104408. [doi:10.1016/j.jpubeco.2021.104408](https://doi.org/10.1016/j.jpubeco.2021.104408).

This Los Angeles study used geographically disaggregated daily crime records and reported variation in the heat association by neighborhood poverty. It provides regional context and a rationale for examining exposure and neighborhood heterogeneity. Crime outcomes and threshold comparisons differ from the supplied arrest-record outcome. The published article supersedes an earlier working-paper version whose estimates and policing interpretation differ; the versions must not be combined.

**Sanz-Barbero B, Linares C, Vives-Cases C, González JL, López-Ossorio JJ, Díaz J (2018).** Heat wave and the risk of intimate partner violence. *Science of the Total Environment*, 644, 413-419. [doi:10.1016/j.scitotenv.2018.06.368](https://doi.org/10.1016/j.scitotenv.2018.06.368); [PubMed](https://pubmed.ncbi.nlm.nih.gov/29981991/).

The Madrid summer time-series study examined intimate-partner femicides, police reports and helpline calls during 2008-2016. Its heat-wave and lag analyses motivate explicit temporal hypotheses. These different outcomes and setting preclude transferring the reported associations directly to local arrest counts.

**Zhu Y, He C, Bell ML, et al. (2023).** Association of Ambient Temperature With the Prevalence of Intimate Partner Violence Among Partnered Women in Low- and Middle-Income South Asian Countries. *JAMA Psychiatry*, 80(9), 952-961. [doi:10.1001/jamapsychiatry.2023.1958](https://doi.org/10.1001/jamapsychiatry.2023.1958).

This population survey study linked annual mean temperature to self-reported past-year intimate-partner violence in India, Nepal and Pakistan. Its exposure timescale, survey outcome and design differ substantially from daily administrative records. It offers complementary evidence rather than direct replication.

## Interpretation and next-analysis requirements

External studies reporting positive associations and a local confidence interval crossing the null can both be correct. The target populations, exposure contrasts and measured outcomes differ. Statistical nonsignificance does not establish no effect; statistical significance elsewhere does not resolve local reporting or linkage uncertainty.

Before estimating effects with any newly acquired outcome, freeze its counting unit, inclusion criteria, reporting-completeness exclusions, exposure assignment, heat-wave definition, lags and multiplicity family. Retain the current models as a versioned reference. A targeted sensitivity excluding independently documented reporting gaps should be labeled retrospective and reported regardless of its result. Annual or monthly benchmarks cannot establish a daily heat association or be used to fill missing daily arrests. Missing reporting is not zero activity.

This is a targeted primary-source review, not a systematic literature review or meta-analysis. Sources were selected for local applicability, outcome measurement and exposure methods; they were not selected to alter the published p-values.