Performance Based Contracting for Colorado Community Corrections

The Urban Institute conducted a baseline outcome assessment and developed a performance-based contracting plan for the Colorado Department of Public Safety, Office of Community Corrections. This repositoty contains the code for the analysis. The full report, including a detailed technical appendix, is available here.

Purpose of code

1. Develop baseline outcome assessment that is stratified by risk level

Use FY14-18 for successful completion, employment, LSI change, technical violation, and escape outcomes
Use FY14-17 for recidivism outcomes
Create tables and graphs showing means by risk level
Note outcomes are calculated at two levels:
- Average of all clients released of a given risk level (state-level)
- Average of all programs of a given risk level, based on their average risk score (program-level)

2. Evaluate program performance in FY19 against those baseline measurements (FY18 for recidivism)

Requires categorizing programs by risk level
Create tables and graphs showing percentage of programs meeting those baseline targets by risk level
Note program risk level is calculated two ways:
- Whether 50% or more of clients in a program are high/very high risk
- Whether the average risk score of the program falls into the high/very high range (29+)

3. Examine reliability of outcomes and other assessment tools used by state (PACE and Core)

Correlation analysis for outcomes
Cronbach’s alpha analysis for outcomes, PACE, and Core
Create tables of results

R packages

Analysis completed in R 3.6.3. The packages used are listed below. Note that the urbnthemes package is used to format the graphs according to the Urban Institute style.

library(tidyverse)
library(haven)
library(readxl)
library(ltm) 
library(Hmisc)

install.packages("devtools")
devtools::install_github("UrbanInstitute/urbnthemes")
library(urbnthemes)

set_urbn_defaults(style = "print")

Data

All data were provided by the Colorado Department of Public Safety, Office of Community Corrections. There are three files for analysis:

file of individual-level data on community correction clients
program scores on the PACE assessment
program scores on the Core Security Audit

Recidivism data come from Colorado court data and were merged into individual-level data.

None of the data are publicly accessible.

Code files

01_baseline_targets_co.R

create state-level and program-level baseline targets
conduct significance tests on each outcome by risk group
graph targets by risk
save table of baseline targets

02_eval_performance_co.R

calculate how many programs met baseline targets
administrative outcomes in FY19
recidivism outcomes in FY18
graph percent meeting mean by outcome and risk group
save data set of performance by outcome and risk group

03_reliability_co.R

descriptive statistics for PACE, Core, state-level outcomes, and program-level outcomes overall (mean, median, n)
Cronbach’s alpha and correlation analysis

Creators

Ashlin Oglesby-Neal and Libby Doyle

UrbanInstitute / co-comm-corrections-pbc