This is the first of three posts about the INFORMS Annual Meeting. Check back in the next few days for the other two posts.
I thoroughly enjoyed the Threats to Life and Limb session. Thanks to Arnie Barnett for organizing a great session!
In this session, Al Blumstein gave a provocative talk about criminal recidivism. His research is motivated by employe background checks. Since most records are electronic, arrest records from the distant past survive. In many cases, a reformed individual could be denied employment for a minor indiscretion committed decades previously. The data indicates that more than 50% of the population will have a non-traffic related arrest at some time in their lifetime, making this a huge problem. However, most of these arrests are for youthful indiscretions.
Blumstein proposes developing hazard curves (based on age of first arrest and type of crime committed). The hazard rates quantifies the likelihood of someone being arrested for the first time after a given amount of time has passed. Blumstein uses these hazard rates to determine when a distant, initial arrest should be ignored after someone has been clean long enough. Blumstein’s research assumes that being “clean” means not being arrested, since background checks only see if arrests have occurred (not whether someone is committing crimes that are undetected).
Blumstein’s results are provocative and generated quite a stir within the session. Based on the age of first arrest and the type of crime, it takes about 4-8 years of a “clean” record to have a hazard rate similar to the general population. Since many employers ignore criminal records older than seven years, and they would be on track.
The general population, however, contains many criminals that we probably don’t want to hire. A better solution may be to compare those with a criminal record to a law-abiding population instead (those who have never been arrested). The probability of being arrested in a given year is about 10% for the general population and about 1% for those who have never had a prior arrest. Of course, a former criminal has to stay clean much longer to be “good as new.” In fact, the hazard functions of the former criminals and the never arrested never overlap. You can compare the two hazard rates when they get “close enough.” The confidence intervals here are pretty wide, but Blumstein’s results suggest 10-20 years of being clean.
Blumstein’s research will hopefully be used to create guidelines for employer background checks (in terms of when they can ignore distant arrests). Criminal justice needs good OR.