A RAND Research Brief Offers a New Model for Estimating Recidivism Risk
Key Takeaways :
The risk of recidivism declines the longer a person is in the community and does not commit a crime. Therefore, past criminal records are rarely a good predictive measure of future convictions.
The reset principle sets forth the requirement that employers can and should estimate a person's risk when a background check is conducted rather than at the time of their last interaction with the criminal justice system. This then takes into account the time spent in the community and their declining risk for recidivism.
After sufficient time without a new conviction, even people originally estimated as high risk for reoffending (people with a more extensive criminal background) move to risk levels similar to those who were initially estimated to be at the lowest risk level.
Criminal background checks are often conducted by employers to identify and filter out candidates who are considered too risky to hire. Results from these background checks are often used to justify barring people with convictions from obtaining gainful employment for a set period of time.
A RAND research report supports previous findings that the likelihood of an individual reoffending declines the longer they are in the community. The authors describe the “reset principle” - and risk-prediction models that satisfy it - to account for the time a person has spent free in the community without a conviction. This time signals declining recidivism risk that may contribute to their employability.
To read the full RAND study, please visit:
The research team developed a model based on a North Carolina data set of convictions including more than 1 million people. They used high standard statistical and machine-learning techniques to analyze the data and assess the adequacy of its model in order to estimate recidivism risk. The authors encourage employers to use the reset principle as a tool to fill vacancies with low risk candidates who are not able to be identified using current models.