Heat & Crime
Analyzing the Relationship Between Temperature and Crime in Los Angeles
Contributing Authors: Camila Cortes Rodriguez, Angela Zhang, Junhong Duan, Sung Eun Cho, Mckailey Walters
Crime and heat
Crime is a public health challenge
Crime is a growing issue of public health concern. Violence, particularly gun violence, has been identified as a major public health crisis in the United States (Freire-Vargas, 2018; Murthy, 2024). Elevated crime levels contribute to the deterioration of communities by instilling fear, fostering social fragmentation, and increasing residential mobility (Boggess & Hipp, 2010). These dynamics weaken residents’ attachments to their neighborhoods, reduce neighborhood satisfaction, and discourage active community participation (Taylor, 1995).
Background
A growing body of research on heat and crime
Research consistently highlights the detrimental effects of crime on individuals, communities, and society at large. Accordingly, previous research has investigated the relationship between temperature and crime by examining various temporal scales of measurement.
Seasonal patterns
Seasonal patterns, in particular, play a significant role in shaping this relationship. For instance, studies have consistently shown that crimes against persons tend to peak during summer, while property crimes are more prevalent in winter (Cohen, 1941). This pattern varies across regions, reflecting broader climatic differences. While earlier studies predominantly focused on broader temporal scales, recent research has shifted toward examining narrower scales, such as monthly variations.
Monthly fluctuations
For example, Cohn (1990) analyzed monthly weather fluctuations and found that warmer temperatures were associated with increases in violent crimes, including assault and homicide, as well as property crimes.
Daily weather
Expanding the focus to daily temporal scales, recent studies have explored the nuanced relationship between heat and crime. Schinasi and Hamra (2017) analyzed daily heat indices in Philadelphia and found a linear association between higher temperatures and increases in violent crimes and disorderly conduct, particularly during colder months when temperatures are more “comfortable.” Similarly, Baryshnikova, Davidson, and Wesselbaum (2021) identified a significant impact of daily temperature on violent and property crimes, noting a nonlinear relationship where crime rates intensified during periods of extreme heat.
Current gaps in the literature
Characterizing the dynamic relationship between heat and crime
Despite the growing body of research on heat and crime, most studies have overlooked the role of temporal lag. Heat’s physiological and psychological impacts begin within minutes to hours, peak within 1–2 days, and can persist for weeks with prolonged exposure (Anderson & Deneve, 1997; Laurent et al., 2018). As such, focusing solely on fixed-time effects may fail to capture the true dynamics of the heat-crime relationship. Furthermore, understanding the lagged effects between temperature and crime can inform crime prevention strategies during critical periods
Categorization of crime
Another notable limitation in the existing literature is the narrow scope of crime types analyzed. Most studies primarily focus on violent crimes (e.g., homicide, assault, domestic violence) and property crimes (e.g., burglary, theft, motor vehicle theft), while others aggregate these into general crime rates. This narrow focus on violent and property crimes leaves a significant gap in understanding how heat influences crimes against society or morality. These crime types are critical for assessing the broader public health and behavioral impacts of temperature changes, highlighting the need for more comprehensive research in this area.
What is this study hoping to answer?
Research Question
What is the time delayed effect of heat on crime in Los Angeles from 2010-2019?
Hypothesis
Our hypothesis is that temperature has a lagged effect on crime which differs for particular types of crime.
Methods
Data Name | Source | Source | Years | Spatial Unit | Temporal Unit |
|---|---|---|---|---|---|
Crime | Crime Open Database (CODE) | 2010-2019 | Geographic coordinates, aggregated at the census tract level | Daily | |
Temperature | Daily Surface Weather and Climatological Summaries (DAYMET) | 2010-2019 | 1 km x 1 km gridded surface, aggregated at the census tract level | Daily | |
Demographic variables | 5-year estimate American Community Survey (ACS) | 2010-2019 | Census tract | Yearly |
Table 1. Data Sources
Temperature
Maximum daily temperature raster data was overlaid with Los Angeles City census tract boundaries to extract census tract-specific values.
Crime
The study utilized crime data from the Crime Open Database (CODE), which contains georeferenced crime records collected and aggregated from the municipal websites of ten major U.S. cities, whose records satisfied specific criteria including four consecutive years of detailed crime data with geographic (e.g., geographic coordinates) and temporal (e.g., date) information (Ashby, 2022) . The data was accessed through the “crimedata” package in RStudio (Ashby, 2022) . We organized crime into three main categories predefined in the CODE database: crimes against people, property, and society.
- Crimes against people: Homicide, Kidnapping/Abduction, Sex offences, etc.
- Crimes against property: Robbery, Arson, Burglary/Breaking & Entering, etc.
- Crimes against society: Prostitution offenses, Weapon law violation, Curfew/Loitering/Vagrancy, etc.
Demographic variables
Individual-level: Ratio of Males, Age Group Percentages (≤19 years, 20–69 years, and ≥70 years), and Simpson's Ethnic Diversity Index (Simpson, 1949).
Socioeconomic covariates: Unemployment rate and the percentage of families below the poverty line.
Housing-related covariates: Occupied Housing Unit Rate, showing the proportion of occupied housing units, and the Renter Occupied Unit Rate, indicating the proportion of units occupied by renters.
Household characteristics: Owner Occupied Household Size and Renter Occupied Household Size, representing the average household size occupied by owners and renters; building year characteristics are categorized into the Building Year Group Percentage, which includes the percentage of buildings constructed before 1979 (Pre-1979 Building Percentage), between 1980 and 2019 (1980–2019 Building Percentage), and whether the majority of buildings in a census tract were constructed post-1980 (Post-1980 Majority).
Statistical Analysis
The main model of our analysis is a non-linear quasi-poisson distributed lag model for each crime type with the covariates described above. We added a spline with 3 knots of maximum temperature (at the 10th, 70th, and 90th quantiles of max. temperature distribution), and a spline with 3 knots equally spaced in log scale to address lags. We estimated the relative risk of exposure to lagged temperature for each crime category. This function was obtained by assuming 30°C as a risk threshold for crime to surge significantly (Heilmann et al., 2021; Heo et al., 2024; Hou et al, 2023).
Contextualizing the Study Area
Mean daily temperature across Los Angeles
Examining mean daily temperature at the 1km level, we see that temperature values are spatially clustered. We also find that the farther North one goes, generally, the hotter it is in Los Angeles. Alternatively, the coastal areas are the coolest regions in the city.
Seasonality of temperature over time
Unsurprisingly, there is clear seasonality in the change in temperature over time which may have an impact on the seasonality of crime.
Average mean maximum daily temperature by census tract across study period (2010-2019)
Overall crime across Los Angeles
Crime occurs unevenly in Los Angeles. The average annual crime count in each census tract varies greatly from over one thousand crimes to zero crimes in a census tract. Annual crime count is not as spatially clustered as temperature, but some spatial autocorrelation is present.
Crime in Los Angeles across time
When we look at crime patterns over the study period (2010-2019), notably, we find that similar to the temperature data, seasonal patterns are visible for crimes against persons with crimes peaking mid-year.
For the other crime types, crime count fluctuates seasonally without a distinctive pattern.
Count of crimes against society, persons, and property crimes aggregated by month, respectively; Note missing data for 2013
Impact of Temperature on Crime
Same-Day Temperatures and Crime Rates
For all the crime types we evaluated, our distributed lag models indicate that temperatures from the same day of the crime were associated with an increased relative risk of crime in all categories.
For crimes related to persons and property, we find that temperatures between 10°C to 20°C (50°F to 68°F) and above 30°C (86°F) increase the relative risk of each of these crimes.
This means both at relatively cool temperatures and at relatively warm temperatures, there is an increase in the relative risk of crimes against persons and property.
In contrast, when examining the results for crimes against society, we find that the relative risk of crimes is higher at higher temperatures, reaching its maximum on the same day of the event at temperatures of 40 °C.
The lingering impact of past temperatures
Our results show the impact of past temperatures on crime. When a crime occurs, temperatures from five days prior slightly impact relative risk of a crime occurring for all three types of crime.
The relative risk of person and property crimes appears minimally influenced by temperatures five days before the crimes.
However, this rise in the relative risk is the most drastic for crimes against society. Society crime’s relative risk increases in the fifth day prior to a crime between temperatures from 10°C to 20°C, then it stabilizes, reaching 40°C.
Study Limitations
Due to the high presence of days where zero crimes were reported in a given census tract, our model residuals indicated poor model fit and predictiveness. As a result, our conclusions should be interpreted with caution.
Future work will model fixed effects at the level of the census tract to account for prior crime in the census tract.
Study Implications
- The relationship between heat and crime is complex and non-linear, the temporal space-time lag varies between types of crime
- The relative risk of crime does not only increase during high heat days, but also during cooler days for some types of crime
- Analyzing temporal lag can be applied to crime prevention efforts on days where the relative risk of crime is higher
