New study reveals spatial patterns of crime rates and media coverage across Chicago

A spatial analysis of crime in Chicago suggests that certain environmental and socioeconomic factors, such as building density, are linked with crime rates, but these relationships differ between neighborhoods. The study also reveals discrepancies between media coverage and actual crime patterns, with coverage disproportionately concentrated in certain areas. Yuxuan Cai of the University of Chicago, U.S., and colleagues present these findings in the open-access journal PLOS One on November 26, 2025.

Numerous prior studies have investigated factors associated with crime in Chicago and other cities. In addition, evidence suggests that media coverage strongly influences public perceptions of crime in urban areas.

However, most prior studies of crime in Chicago have been limited in scope, such as by focusing on only a few socioeconomic factors at a time, or not distinguishing between different crime types. To gain a more comprehensive view, Cai and colleagues used computational tools to analyze associations between rates of different crime types and various environmental and socioeconomic factors in different parts of Chicago, as well as spatial patterns in media coverage of crime. Their dataset included 234,918 crime incidents that occurred from 2017 to 2023.

Using the natural language model GPT-3.5-Turbo, the researchers analyzed 3,507 local news reports on crime in Chicago from 2023 to 2024. They found that the most frequent crimes—theft, criminal damage, and battery—were significantly underrepresented in news reports, while homicides and assaults were overrepresented. Media coverage tended to focus more on downtown and certain north and south side areas, while neglecting other neighborhoods.

The analysis also uncovered links between certain spatial factors and crime. For instance, more walkable areas had higher rates of theft but lower rates of criminal damage and battery. Dense road networks and higher building density were associated with higher rates of all crime types. In downtown and northern areas of the city, higher housing density and economic activity were associated with lower crime rates, but this pattern was not seen for south side areas.

These findings could inform efforts to improve safety and media crime coverage. However, further research is needed to determine the mechanisms underlying these spatial patterns.

The authors add: "Our analysis shows a troubling disconnect between media crime coverage and actual crime patterns in Chicago. While the most common crimes include theft, criminal damage, and battery, media outlets disproportionately focus on homicides and assaults. This bias not only distorts public perception but may also influence policy decisions in ways that don't address the most prevalent public safety concerns."

"What makes this study unique is our integration of large language models with traditional spatial analysis methods. By using GPT-3.5 to analyze over 3,500 news articles, we were able to systematically compare media narratives against actual crime data on a large scale. This computational approach opens new possibilities for understanding how media shapes public perception of urban safety."

"Our geographically weighted regression analysis shows that the same environmental factors, such as road density or vacant buildings, can have completely different relationships with crime across neighborhoods. This means one-size-fits-all urban planning solutions won't work. Policymakers need spatially-targeted interventions that account for local conditions, and the media needs to report more balanced stories."

In your coverage, please use this URL to provide access to the freely available article in PLOS One: https://plos.io/49YJ8Eq

Citation: Cai Y, Chen A, Tang Z, Wang Y, Song Y (2025) Advancing urban management: Integrating GIS, LLMs, and media narratives into environmental and socio-economic analyses for enhanced urban crime analysis. PLoS One 20(11): e0331788. https://doi.org/10.1371/journal.pone.0331788

Author countries: U.S.

Funding: The author(s) received no specific funding for this work.