Statistics for the Recidivism Patterns Explorer (RPE) are sourced from three separate national studies BJS conducted. States were included in these studies if their state department of corrections provided prison records through the National Corrections Reporting Program (NCRP). These records were needed to collect the criminal history data for released prisoners. For each of these studies, BJS used prisoner records from the NCRP and national criminal history data from the FBI and state repositories to produce three different recidivism measures: a new arrest, a new conviction, and return to prison following release. These measures include recidivism that occurred both within and outside of the state where the persons were imprisoned.

The three cross-sectional studies include separate and independent samples of prisoners released in 2005, 2008, and 2012. Data from these studies were used to produce the recidivism estimates featured in this tool. Follow the appropriate link for additional findings and details on the methodologies of these studies.

Recidivism of Prisoners Released in 34 States in 2012: A 5-Year Follow-Up Period (2012–2017)

In this study, BJS randomly sampled about 92,100 released prisoners to represent the approximately 408,300 state prisoners released across 34 states in 2012. While the arrest measure was produced for persons released in all 34 states, the conviction measure was based on persons released in the 31 states with the necessary court data, and the return-to-prison measure was based on persons released in the 21 states with the necessary prison data. All 3 measures included a 5-year follow-up period.

Recidivism of Prisoners Released in 24 States in 2008: A 10-Year Follow-Up Period (2008–2018)

In this study, BJS randomly sampled about 73,600 released prisoners to represent the approximately 409,300 state prisoners released across 24 states in 2008. While the arrest measure was produced for persons released in all 24 states, the conviction measure was based on persons released in the 22 states with the necessary court data, and the return-to-prison measure was based on persons released in the 18 states with the necessary prison data. All 3 measures included a 10-year follow-up period.

2018 Update on Prisoner Recidivism: A 9-Year Follow-up Period (2005-2014)

In this study, BJS randomly sampled about 68,000 released prisoners to represent the approximately 401,300 state prisoners released across 30 states in 2005. While the arrest measure was produced for persons released in all 30 states, the conviction measure was based on persons released in the 29 states with the necessary court data, and the return-to-prison measure was based on persons released in the 23 states with the necessary prison data. The arrest and conviction measures included a 9-year follow-up period, and the return-to-prison measure included a 5-year follow-up period.

Accuracy of estimates

To conduct these studies, BJS selected separate stratified random samples to represent all persons released from prison in 2005, 2008, and 2012 across the states included in the studies. Because these studies were based on samples rather than complete enumerations of the released prisoner population in these states, the recidivism estimates are subject to sampling error. One measure of the error associated with an estimate is the standard error, which measures the variation between a sample estimate and the average of all possible estimates that could be obtained by taking different samples from the same population. The standard errors in the RPE were generated using the “survey” package in R. Since the RPE does not include tests of statistical significance, users should not draw inferences when comparing the results in the tables and graphs. However, the standard errors can be used to generate a confidence interval around the estimate as a measure of the margin of error.

Users can calculate 95% confidence intervals around an estimate by multiplying the standard error by ±1.96 (the t–score of a normal, two–tailed distribution that excludes 2.5% at either end of the distribution). The result can then be added to the estimate to produce the upper confidence interval and subtracted from the estimate to produce the lower confidence interval. The primary test procedure was to examine whether confidence intervals of the estimates overlapped. This check was performed by comparing the upper bound of the smaller estimate to the lower bound of the larger estimate. A statistically significant difference is detected when the upper bound of the smaller estimate is less than the lower bound of the larger estimate. For example, the upper bound estimate for females arrested within 5 years following release in 2012 is 63.6% (63.1% + 0.5%), and the lower bound estimate for males arrested within 5 years following release in 2012 is 71.2% (71.7% - 0.5%). In this example, there is a statistically significant difference at the .05 level because the upper bound estimate for females is lower than the lower bound estimate for males.

Recommended citation

Bureau of Justice Statistics. [Insert chart/table title]. Generated using the Recidivism Patterns Explorer at bjs.ojp.gov/recidivism-patterns-explorer. Report generated [Insert date generated].