The study, conducted by researchers from the New York City Department of Health and Mental Hygiene and reviewed by the Centers for Disease Control and Prevention, examined daily hospital visit data across six summers, from 2016 through 2021. Published in the journal Environmental Advances, the research provides one of the most detailed portraits to date of how short-term temperature changes relate to nonfatal violence in a dense urban environment, drawing on healthcare utilization records rather than the police reports and mortality data that have dominated prior work in this field.

The data source was the New York Statewide Planning and Research Cooperative System, known as SPARCS, an all-payer database that captures every inpatient hospitalization and emergency department visit in New York State. The research team filtered these records for nonfatal injuries carrying assault-related diagnosis codes under the International Classification of Diseases, Tenth Revision, encompassing every mechanism of interpersonal harm — firearms, blunt force, sexual assault and more — while deliberately excluding self-harm. Because the analysis focused on the intent behind the injury rather than the weapon or method used, it captured the full spectrum of violence serious enough to require hospital-based care, whether the patient was treated and released or admitted as an inpatient. The study window began in 2016 in part because the 2015 transition from ICD-9 to ICD-10 coding introduced discontinuities in injury classification that would have complicated longer-term comparisons.

On the exposure side, the researchers relied on daily maximum temperatures recorded at the National Weather Service station at LaGuardia Airport, chosen for its complete records and its high correlation with the city’s other major stations in Central Park and at John F. Kennedy Airport. Although New York exhibits considerable intra-urban temperature variability — Central Park sits in vegetated parkland, JFK on a cooler coast, LaGuardia amid pavement and large structures — day-to-day temperature swings during the warm season correlate at above 0.9 across all three sites. LaGuardia’s maximums ran one to two degrees Fahrenheit higher on average, making it a sensible sentinel for the city’s hottest readings. Maximum temperature was also strongly correlated with daily minimum (r = 0.95) and mean temperature (r = 0.98), allowing it to serve as a proxy for overall daily heat load. Notably, maximum temperatures tend to occur during daylight hours, when people are most active outdoors and most exposed.

The statistical architecture of the study centered on a distributed lag nonlinear model, a technique that allows researchers to estimate both nonlinear exposure-response relationships and delayed effects across time. The team fitted the model with negative binomial regression, using a natural cubic spline with three degrees of freedom for daily maximum temperature and allowing each of the lags from zero to five days to carry its own response function. Temporal trends were controlled with an 18-degree-of-freedom spline over sequential study days — three per year across the six-year period — effectively absorbing both within-season patterns and calendar-year effects. The reference temperature was set at 9.4°C, the lowest observed warm-season daily maximum, equivalent to the zeroth percentile of the May-through-September distribution. Cumulative rate ratios were then summed across the six lag days.

The results were striking. Across the six warm seasons, 126,304 nonfatal assault hospital visits were recorded, with annual counts ranging from 19,074 in 2017 to 23,105 in 2019. Mean daily maximum temperatures across the warm seasons hovered between 26.7°C and 28.2°C. The cumulative rate ratio for assault visits climbed with temperature, peaking at 1.76 (95% confidence interval: 1.53 to 2.03) at 32.2°C — the 82nd percentile of the warm-season temperature distribution — relative to the reference. Beyond that peak, the estimate declined, but remained approximately 1.65 times the reference level even as temperatures approached the city’s extreme of 37.7°C, though uncertainty widened at those upper reaches. Sensitivity analyses using lag windows of zero to six and zero to seven days, spline degrees of freedom from three to seven, and reference temperatures re-centered at percentiles from zero to 100 all preserved the same inverted U-shape and the same peak temperature.

The lag structure proved revealing. Rate ratios were highest at lags of one to three days — 1.66 at a one-day lag, declining to 1.54 by five days and 1.45 at seven days, with confidence intervals staying above 1.0 throughout. This suggests that a hot day’s influence on violence is not confined to that day alone; the behavioral and social ripples of extreme heat appear to persist for several days afterward. Age stratification added further texture: people aged 25 to 44 accounted for 47.2 percent of assault hospitalizations and showed the highest peak rate ratio at 1.83, followed by those aged 0 to 24 (26.4 percent of cases) and those 45 and older (24.6 percent).

The data also captured the shadow of larger events. Assault counts peaked around the July 4th holiday in every year of the study, and a distinct decline occurred between March and June of 2020, coinciding with New York State’s COVID-19 lockdown order that closed nonessential businesses and discouraged gatherings. The models adjusted for these factors — along with federal holidays, school days, precipitation in the previous 24 hours, and day of the week — all of which shape routine activities and, with them, opportunities for interpersonal conflict.

Why would violence peak at 32.2°C rather than continue rising with the mercury? The authors point to two dominant theoretical frameworks. The first, Routine Activities theory, holds that warmer temperatures draw more people into public spaces and extend the time they spend outdoors, multiplying the interpersonal encounters in which conflicts can escalate. On this reading, the decline beyond the peak temperature may reflect behavioral adaptation: when heat becomes truly oppressive, people retreat indoors, into shade, or into air conditioning, reducing the pool of public interactions even as physiological stress remains high. The second framework, Heat-Aggression theory, posits that uncomfortably hot conditions directly stimulate hostility and aggression — a mechanism that may better explain temperature-related increases in violence occurring indoors, such as domestic or partner violence, sexual violence, and child abuse. The inverted U-shaped curve observed in New York, the authors note, lends some support to the Routine Activities interpretation, though the two mechanisms are not mutually exclusive, and the study’s observational design cannot adjudicate between them.

The shape of the curve matters for public health planning, because it diverges sharply from the relationship between heat and heat-related mortality, which is typically linear or exponential, with deaths climbing ever higher at extreme temperatures. Violence, by contrast, appears to be mediated through a complex interplay of social, economic, and behavioral factors rather than a direct physiological response to heat stress. That distinction suggests that cooling centers, expanded green space, and air-conditioning access — the standard toolkit of heat preparedness — may not, on their own, address heat-linked violence. Instead, the authors argue, interventions likely need to target the root causes of violence itself, alongside continued surveillance of assault patterns during warm months.

The stakes are rising with the climate. New York City projects a temperature increase of 2.3°C to 3.2°C by the 2050s, a change that will interact with existing heat vulnerability shaped by housing type, access to green space and air conditioning, and socioeconomic conditions across a population of more than eight million. Prior research has estimated that 37 percent of warm-season heat-related deaths across 43 countries between 1991 and 2018 were attributable to anthropogenic climate change, and modeling studies have predicted increases in U.S. violent crime through 2099 if warming continues. The new findings align with a national analysis of firearm violence across 100 U.S. cities, which found risk peaking around the 84th percentile of local temperature before a modest decline — a pattern echoed in earlier work from Dallas, Texas, where violent crime peaked at 90°F (32.2°C), the very same temperature identified in the New York study.

The authors are careful about the limits of their analysis. A single weather station cannot capture neighborhood-scale temperature variation, and outdoor maximum temperature is an imperfect proxy for personal exposure, ignoring time spent indoors and air-conditioning use — sources of misclassification that may bias estimates toward the null. The study did not model humidity-derived metrics such as the Heat Index, nor overnight minimum temperatures or heatwave duration, all of which could modulate both physiological stress and social behavior. Hospital claims data capture only assaults serious enough to be medically treated, contain no information about aggressors or incident locations, and cannot distinguish residents from tourists or commuters. Residual confounding from unmeasured factors — mass outdoor events, pandemic-era changes in policing and mobility, neighborhood socioeconomic conditions — also cannot be excluded, and results from New York may not generalize to cities with different climates, infrastructure, or demographics.

Even with those caveats, the study extends the heat-violence literature into an important gap: nonfatal assaults, which comprise the large majority of violence-related events but have been studied far less than homicides and violent deaths recorded in criminal justice data. The consistency of the inverted U-shape — in New York hospital data, in Dallas crime records, in firearm violence across a hundred American cities — suggests a robust and repeatable pattern in how human behavior responds to heat. As summers grow hotter, understanding where that curve peaks, for whom, and why may become an essential input for emergency departments, public health agencies, and violence-prevention programs alike.