Neighborhood Socioeconomic Characteristics and Police Incident Rates in San Francisco
1 Background
This project examines whether neighborhood socioeconomic and demographic characteristics are associated with reported police incident rates across San Francisco census tracts. The analysis combines incident-level data from the San Francisco Police Department (SFPD) Incident Reports dataset, accessed through the DataSF API (City and County of San Francisco, DataSF 2026), with tract-level estimates from the U.S. Census Bureau’s American Community Survey (ACS) (U.S. Census Bureau 2023). The unit of analysis is the census tract, allowing reported incidents to be compared with neighborhood population, income, poverty, housing tenure, age, and density.
The project is informed primarily by Social Disorganization Theory, which emphasizes how structural neighborhood conditions may affect informal social control and exposure to crime (Shaw and McKay 1942; Kubrin and Weitzer 2003). Concentrated disadvantage and residential instability may reduce a community’s capacity to regulate public behavior or mobilize collective resources. Routine Activity Theory provides a complementary perspective by emphasizing the convergence of suitable targets, potential offenders, and limited guardianship in particular places (Cohen and Felson 1979).
The primary research question is:
Which neighborhood characteristics are associated with higher police incident rates across San Francisco census tracts?
The preliminary hypotheses are:
H1: Census tracts with higher poverty rates will have higher police incident rates.
H2: Census tracts with higher renter occupancy rates will have higher police incident rates.
H3: Census tracts with higher median household incomes will have lower police incident rates.
Median age and population density are included as additional neighborhood characteristics and statistical controls.
2 Data and Methods
The analysis uses initial SFPD incident reports recorded during 2025. Supplemental reports are excluded to reduce possible duplication, and repeated records associated with the same incident identifier are reduced to one observation. Incident coordinates are spatially joined to San Francisco census tracts, after which incidents are aggregated to the tract level.
The dependent variable is the number of reported police incidents per 1,000 residents. Neighborhood characteristics are drawn from the 2024 ACS five-year estimates. Census tracts with fewer than 1,000 residents are excluded because residential population provides an unstable denominator in primarily commercial, institutional, transportation, or other minimally residential tracts. Such tracts may have substantial daytime populations despite having very few residents.
Ordinary least squares (OLS) multiple regression is used to estimate the association between neighborhood characteristics and police incident rates. Because the data represent reported incidents, the findings describe patterns of police reporting and recording rather than the total amount of criminal activity.
3 Initial Findings
3.1 Figure 1: Geographic Distribution of Police Incident Rates
Figure 1 shows substantial spatial variation in reported police incident rates across San Francisco. Several tracts in the northeastern portion of the city have considerably higher rates than many western and southern residential areas. These differences suggest that police incidents are geographically concentrated rather than evenly distributed across the city. The pattern provides initial justification for examining whether tract-level socioeconomic and demographic characteristics help explain this variation.
3.2 Figure 2: Poverty and Police Incident Rates
Figure 2 indicates a positive bivariate association between neighborhood poverty and reported police incident rates. Census tracts with higher poverty rates generally have higher incident rates, although there is substantial dispersion around the fitted regression line. The variation among tracts with similar poverty levels indicates that poverty alone does not fully account for neighborhood differences in police incidents. This finding supports the use of a multivariate model that incorporates housing tenure, income, age, and population density.
3.3 Table 1: Preliminary Multiple-Regression Results
| Predictor | Estimate | Standard Error | p-value | Significance |
|---|---|---|---|---|
| Intercept | -144.939 | 78.099 | 0.06479 | |
| Median household income ($) | 0.000 | 0.000 | 0.24586 | |
| Poverty rate (%) | 4.083 | 1.172 | < 0.001 | *** |
| Renter occupancy rate (%) | 1.270 | 0.385 | 0.00113 | ** |
| Median age | 2.046 | 1.244 | 0.10133 | |
| Population density | 0.000 | 0.000 | 0.23443 |
Note. Dependent variable: reported police incidents per 1,000 residents.
* (p < .05); ** (p < .01); *** (p < .001).
Table 1 presents preliminary estimates from the multiple linear regression model. Holding the other neighborhood characteristics constant, poverty rate and renter occupancy rate are positively and statistically significantly associated with police incident rates. Each one-percentage-point increase in poverty is associated with an estimated increase of approximately 4.08 incidents per 1,000 residents, while each one-percentage-point increase in renter occupancy is associated with approximately 1.27 additional incidents per 1,000 residents.
Median household income, median age, and population density are not statistically significant at the conventional .05 level in the preliminary model. The model as a whole is statistically significant, (F(5, 224) = 9.87), (p < .001), and explains approximately 18.1% of the observed variation in police incident rates. The adjusted (R^2) is approximately .162, indicating that the selected neighborhood variables explain a meaningful but limited portion of geographic variation.
4 Preliminary Conclusion
These initial findings indicate that reported police incident rates vary substantially across San Francisco census tracts. The descriptive map shows clear geographic concentration, while the bivariate scatterplot identifies a positive relationship between poverty and incident rates. The preliminary multiple-regression model further suggests that poverty and renter occupancy remain significant predictors after accounting for income, median age, and population density.
The results provide initial support for the hypotheses concerning poverty and renter occupancy. However, the negative hypothesis concerning median household income is not supported in the preliminary multivariate model because its coefficient is not statistically significant. The findings also indicate that neighborhood socioeconomic conditions alone do not fully explain incident-rate variation.
Further analysis will evaluate regression diagnostics, influential observations, and possible multicollinearity among the explanatory variables. The final project will also discuss the limitations of using residential population to standardize incident counts, particularly in commercial and tourist-oriented tracts where the daytime population may substantially exceed the number of residents.
5 References
Allaire, J. J., Charles Teague, Carlos Scheidegger, Yihui Xie, and Christophe Dervieux. 2022. “Quarto: R Interface to ’Quarto’ Markdown Publishing System.” https://CRAN.R-project.org/package=quarto.
City and County of San Francisco, DataSF. 2026. “Police Department Incident Reports: 2018 to Present.” https://data.sfgov.org/Public-Safety/Police-Department-Incident-Reports-2018-to-Present/wg3w-h783.
Cohen, Lawrence E., and Marcus Felson. 1979. “Social Change and Crime Rate Trends: A Routine Activity Approach.” American Sociological Review 44 (4): 588–608. https://www.jstor.org/stable/2094589.
Kubrin, Charis E., and Ronald Weitzer. 2003. “New Directions in Social Disorganization Theory.” Journal of Research in Crime and Delinquency 40 (4): 374–402. https://doi.org/10.1177/0022427803256238.
R Core Team. 2022. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.
Shaw, Clifford R., and Henry D. McKay. 1942. Juvenile Delinquency and Urban Areas. Chicago, IL: University of Chicago Press.
U.S. Census Bureau. 2023. “American Community Survey 5-Year Estimates (2023).” https://api.census.gov/data/2023/acs/acs5.