Anne Velazquez (2022-11-06)
This capstone research project was completed as part of the 2022 Data and Policy Summer Scholar Program (DPSS) held through the University of Chicago Harris School of Public Policy.
The purpose of this analysis is to estimate the impact of the Safe Passage Program, a public policy designed and introduced by the the City of Chicago (see here for more details on the program). The main purpose of the program is “to provide a positive, trusted adult presence for students as they travel to and from school.”
This project used the five datasets outlined below relating to the Safe Passage Program, attendance rates at Chicago Public Schools and crime statistics near these schools.
-
“Attnd” (metrics_attendance_2021.xlsx): Attendance data for Chicago public schools. This dataset is publicly available and can be found here under “Attendance Over Time”.
-
“control” (“control.csv”): List of Chicago schools without the Safe Passage Program and geographic information
-
“treatment” (“treated.csv”): List of Chicago schools with the Safe Passage Program, when it was implemented and geographic information
-
“crimes” (“crime-data.csv”): Information on Chicago crimes within 500 yards of schools including crime type, location, and date of crime.
-
“schools_v_crime” (“school-vs-crime-distances.csv”): Chicago crimes that happened within 500 yards of school and their distance from the relevant school(s)
Let’s look at where the schools with the Safe Passage Program are located compared to other schools in the Chicago area.
From the map of the treated schools compared to control schools, the treated schools do not appear to be evenly distributed throughout the Chicago area. Most schools with the safe passage program (treated) are in the southern part of Chicago. There is a higher concentration of treated schools near the South Side, New City, and Englewood areas specifically. Treated schools generally appear more concentrated near the center of the city and are less prevalent towards the edges of the city.
Now lets look at the Chicago crime data. The map below shows crimes from 2013-2018 that took place within a 50 yard distance from schools. Based on the map, crimes appear more highly concentrated near the center of the city.
Given we have panel data across several years for various schools, there may be unobserved variables impacting crimes or attendance that are specific to the locations/schools being studied or to the time frame the data was collected. To control for these time invariant and entity invariant unobservable variables and better understand the impact of the Safe Passage Program, we can use a fixed effects regression model with ‘School ID’ and ‘Year’ as fixed effects.
However, in order to use fixed effects regression analysis to produce an unbiased estimate of the impact of the Safe Passage Program on crime near schools, first we must confirm if the ‘parallel trends’ assumption holds. In other words, we will confirm if crime rates near schools with the Safe Passage Program were trending similarly to those without the program prior to 2015 when the program was implemented. If this is true, we can use fixed effects regression to analyze the impact of the program.
First, lets evaluate the trends of crimes within 50 yards of schools. The graph below shows that, on average, all schools had decreasing rates of crime within 50 yards prior to 2015. After 2015, schools with the Safe Passage Program continued to see nearby crime decrease on average through 2018, while schools without the program saw nearby crimes increase slightly and hold relatively flat through 2018. In short, nearby crime rates were trending similarly at all schools before the the Safe Passage Program was implemented in 2015, but not perfectly parallel.
If instead we expand the radius to evaluate trends of crimes within 200 yards of schools, this appears to strengthen the parallel trends assumption between schools with and without the Safe Passage Program.
Now look at trends for school attendance rates before and after the Safe Passage Program went into effect. From the graph below, it does not look like the parallel trend assumption holds for school attendance rates. The schools that did implement the program had attendance rates trending slightly upward prior to 2015, while control schools had fairly flat attendance rates. Given the parallel trends assumption does not hold, we cannot produce an unbiased estimation of the impact of the Safe Passage Program on school attendance rates using a fixed effects regression model.
Now, we can use a fixed effects regression model to determine if the Safe Passage Program had a significant impact on crimes nearby to schools. The table below outlines the results of two fixed effects regression models that evaluate the impact of the Safe Passage Program on total crimes within 50 and 200 yards of schools. Both models include fixed effects for ‘school’ and ‘year’.
Looking at the model for crimes within 50 yards of schools, the coefficient for the indicator variable for the Safe Passage Program is -8.24, which means that schools with the Safe Passage Program in effect had about 8 fewer nearby crimes on average each year compared to schools that did not implement the Safe Passage Program. The coefficient has a p-value significant at the .001 level. This means that the Safe Passage Program has a statistically significant relationship with the rate of crime within 50 yards of schools.
Looking at the model for crimes within 200 yards of schools, the coefficient of the indicator variable is -18.69 and has a p-value significant at .001 level as well. Thus, the Safe Passage Program also has a statistically significant relationship with the number of crimes within 200 yards of schools.
By including fixed effects for schools and each year in the models, this eliminates the risk of bias due to omitted factors that vary across schools and that vary over time.