rates, poverty levels, and unemployment rates across all states over the study period (2000–2022). No publicly available source provides harmonized annual PM 2.5 averages at the state level for this time span. Air pollu- tion monitoring in the U.S. is conducted pri- marily at the station and county levels by the U.S. Environmental Protection Agency (U.S. EPA) Air Quality System (AQS), with spatial and temporal coverage varying substantially across locations (Table 1). To address this limitation, the most polluted counties nationwide were identified using 2000–2022 U.S. EPA-based pollution rankings. Each county was mapped to its corresponding state, yielding a sample of 10 highly polluted states: California (CA), Oregon (OR), Okla- homa (OK), Georgia (GA), Illinois (IL), Texas (TX), Pennsylvania (PA), Ohio (OH), Nevada (NV), and Arizona (AZ) (Table 2). For coun- ties with incomplete historical PM 2.5 records, a fallback selection strategy was applied, in which the next most polluted county within the same state was used to maintain tempo- ral continuity. Although a single county can- not fully represent statewide exposure, this approach enabled the construction of a con- sistent state–year longitudinal PM 2.5 series in the absence of published statewide averages. Additionally, crime data presented chal- lenges due to incomplete coverage in the Fed- eral Bureau of Investigation (FBI) Uniform Crime Reporting (UCR) system, particularly during the transition toward the National Incident-Based Reporting System (NIBRS). To ensure continuity, supplementary data were obtained from state crime repositories, along with historical crime statistics from Macrotrend, USAFacts, and other validated sources. Crime rates were cross-checked across sources and standardized as incidents per 100,000 population. Furthermore, data for state-level poverty and unemployment rates were obtained from publicly available sources from the U.S. Cen- sus Bureau and the U.S. Bureau of Labor Sta- tistics. All variables were aligned temporally and spatially to ensure consistency across states and years. Descriptive statistics and raw state–year values are reported in the Sup- plemental Table S1. Data Structure and Sample Definition The final dataset is structured as a state–year panel, with each observation corresponding
FIGURE 1
Flow Chart of Methodological Steps
Data Collection
Data Cleaning
Variable Rates Calculation
Regression Analysis
Visualization
Interpretation
Policy Implications
thermal discomfort (Heo et al., 2021), noise (Manukyan, 2022), or pollution (Weitekamp & Hofmann, 2021), can heighten physiologi- cal arousal and frustration, thereby predis- posing individuals to aggression. Likewise, the cognitive resource model suggests that chronic exposure to pollutants can deplete cognitive energy, in turn reducing the capac- ity for rational decision-making and emo- tional regulation (Zhang et al., 2018). Despite increasing evidence linking air pol- lution to mental and behavioral health out- comes, few empirical studies have quantified these relationships in an urban U.S. context using longitudinal data that combine envi- ronmental and social indicators (Humphrey et al., 2019). Most existing research remains cross-sectional or focused on physiological endpoints. Furthermore, the psychosocial dimensions of air pollution, specifically its role in crime and suicide, have been largely overlooked in policy discourse. Hence, this study addresses these gaps using online-available data to investigate whether fluctuations in PM 2.5 correspond with changes in annual crime rates across 10 U.S. states. Moreover, this study statistically inves- tigates the effect of the two most-linked factors to crime in the U.S.: the percentage of poverty and the unemployment rate per state. The
study objective was to determine if increases in PM 2.5 are associated with higher crime lev- els and to consider how socioeconomic condi- tions might influence this relationship.
Methods
Study Design and Analytical Framework
This study follows a structured, stepwise ana- lytical framework to examine the association between ambient PM 2.5 levels, socioeconomic conditions, and crime rates in selected U.S. states (Figure 1). The empirical strategy was based on a state–year panel design, which exploits both cross-sectional and temporal variation while minimizing bias arising from data aggregation and shared national trends. Regression models were estimated sequen- tially, progressing from pooled specifications to fixed effects models, to assess the robust- ness under increasingly stringent controls of the relationship between PM 2.5 and crime.
Data Acquisition Procedure and Challenges
Constructing a consistent longitudinal data- set posed several nontrivial challenges, par- ticularly in obtaining continuous annual measures of PM 2.5 concentrations, crime
19
October 2026 • Journal of Environmental Health
Powered by FlippingBook