Catalyzing Police Reform with Data

In 2018, the National Neighborhood Indicators Partnership (NNIP) and Microsoft launched a partnership to spur more data-driven and community-led criminal justice reform efforts, with the goal of building trust with law enforcement and improving public safety. As one of the activities under this partnership, we selected one city – Los Angeles, California – to explore how publicly-available police data can be analyzed to create a comprehensive measure of community-police engagement.

We used cluster analysis to develop a typology that elucidates the relationship between resident-initiated and police-initiated activity in the City of Los Angeles, as well as how that relationship varies across neighborhoods, by synthesizing data sources on calls for service, stops, arrests, and crime. We find that resident calls for service and police stops and arrests generally increase together, and that neighborhoods with high amounts of activity tend to have a greater proportion of violent crime and serious calls for service. The neighborhoods with high activity also tend to have wider racial disparities in stops and arrests, and more economic hardship.

This repository provides the code behind this analysis, which was published by the Urban Institute in the Catalyzing Police Reform with Data: Policing Typology for Los Angeles Neighborhoods brief in May, 2020.

More information on the methodolgy used in this analysis can be found in the Catalyzing Police Reform with Data: Technical Appendix published by the Urban Institute.

This analysis was sponsored by Microsoft. We collaborated closely in the design, analysis, and interpretation of the findings with the Microsoft Criminal Justice Reform team, the Microsoft Data Science team, and the local NNIP partner at University of Southern California’s Sol Price Center for Social Innovation.

This analysis was conducted in RStudio using R v 3.6.1.

Data Sources

We use the following data sources for this project:

LAPD open data on reporting districts, crime, arrests, calls, and stops:

• Reporting District Boundaries
• Crime: 2010-2019
• Arrests: 2010-Present
• Calls for Service: 2018
• Stops: 2010-Present

American Community Survey data on neighborhood demographics:

• USC Neighborhood Data for Social Change project.

HUD USPS data on residential and business addresses:

• HUD USPS vacancy data

All of the raw source data files that the authors used to create the analytical data set are available from the authors by request, with the exception of the HUD USPS data which is restricted access and must be downloaded from the HUD website at the link above. Requesting the static datasets will enable replication of our exact results as the LAPD open data sources are frequently updated on the LA Open Data Portal. We observed that these updates can alter historical data so pulling the data on different dates has yielded different 2018 totals. The data/ repository includes the following files:

• Analytical data set: data/analysis_ready_by_district_final.csv
• Tract crosswalk: data/rd_tract_xwalk.csv
  • This file includes the tract-to-reporting district weights produced by crosswalking of the 2010 Census Tracts and the LAPD Reporting Districts using ArcGIS based on the methodology described in the technical appendix.
• Reporting District Exclusions: data/Repdist_exculsions.csv
  • Includes a dummy variable indicating whether each reporting district is a LAPD reporting district.
• LA Times Neighborhood Shapefiles: data/LA Times Neighborhood Shapefiles/LAC_Neigh.shp
  • Neighborhood boundaries in Los Angeles from LA Times.

If you want to replicate this analysis with the most up-to-date LAPD data, you can access the data via the LA Open Data Portal API at the source links given above.

Getting Started

Prior to running the analysis, you will need to install R and RStudio on your computer if you do not have this software installed.

This analysis requires that the following R packages be installed:

"skimr", "lubridate", "janitor", "knitr", "tidyverse", "psych", "cluster", "factoextra", "gridExtra", "here", "rlang", "sf", "devtools", "urbnthemes", "mclust", "dbscan", "data.table", "R.utils"

Installation and import for these required packages is handled in the beginning of 01_lapd-data-processing.R and the first code chunk of 02_final-cluster-analysis.Rmd. The data visualizations for this analysis are produced using the ggplot2 package, styled using the urbnthemes package. Note that the Urban Institute uses Lato font for publications. After installing urbnthemes, submit urbnthemes::lato_test() to see if Lato is imported and registered.

If Lato isn’t imported and registered, install Lato and then submit urbnthemes::lato_install(). If you are on a Windows, you may need to install ghostscript and then submit Sys.setenv(R_GSCMD = "link to the ghostscript .exe") before running urbnthemes::lato_install().

Always load library(urbnthemes) after library(ggplot2) or library(tidyverse). This is already handled in the 02_final-cluster-analysis.Rmd file.

Replicating the Analysis

To run the analysis, perform the following steps:

1. Clone this repository on your local machine.
2. Download the HUD USPS Vacancy data from the HUD website and save the data as usps_data.csv.gz in the data/ directory. Note that the HUD USPS data is restricted access and requires registration. From the HUD USPS data registration page:

Under the current agreement with the USPS, HUD can only grant access to the vacancy data to governmental and non-profit organizations. In order to be granted access as a user, a requestor must be verified as a member of a non-profit or governmental organization.

3. Open the R project la-policing-typology.Rproj in RStudio.
4. Open the analysis/02_final-cluster-analysis.Rmd file in RStudio and knit to html. When kniting, the code will check whether the necessary packages are installed and install any packages that are not already installed. This file reads in the analytical data set data/analysis_ready_by_district_final.csv that is produced by running etl/01_lapd-data-processing.R and conducts the cluster analysis. Results tables will be written to the results/ directory and plots will be written to the images/ directory. Note that analysis/02_final-cluster-analysis.Rmd uses a global trial_name variable to enable the researchers to test multiple different models, each defined by a separate trial_name during the analysis process. This variable is set to final by default. All outputs in the results/ and images/ directory will use the trial_name as a prefix.

If you want to replicate the production of the analytical data set, you will need to complete the following additional steps:

1. Request the raw static data files for reporting districts, American Community Survey, crimes, arrests, stops, and calls for service from the authors or download the data files from data sources linked above. Place all files in the data/ directory.
2. Open the R project la-policing-typology.Rproj in RStudio.
3. Open the etl/01_lapd-data-processing.R file in RStudio and run the file to create the analytical dataset data/analysis_ready_by_district_final.csv. Note that this file will not sucessfully run if you have not completed step 1.

The etl/ directory contains the key file (01_lapd-data-processing.R) used for preprocessing the raw data sources to create to create the analytical data set.

The analysis/ directory contains the key files used for analysis which are:

• 02_final-cluster-analysis.Rmd: file containing data pre-processing, cluster analysis, and analysis of cluster results.
• cluster-functions.R: file containing functions for cluster analysis including variable weighting, cluster modeling, and stability analysis.
• data_viz_functions.R: file containing functions for data visualization of cluster analysis results.

Getting help

Contact Alena Stern with feedback or questions.

License

This software is licensed under the terms of the MIT License.