by Daniel Hadley
People cycling in and out of the justice system is one of our country's most persistent problems. The term “recidivism” is a common term we hear in our work with governments and providers, and many Pay for Success projects attempt to lower the recidivism rate for groups of people, including Massachusetts’ pioneering collaboration with Roca. I was surprised, therefore, when I was explaining to my in-laws at a family dinner what I was doing at work and received only blank stares: “what is recidivism?”
“Well,” I started, “recidivism is the rate at which people return to prison.” Then I remembered, “or it could be the rate that they are rearrested, or return to jail, or violate parole, or are convicted of a new crime.” The more I researched the term, the more I realized that it is highly specific to what is being measured and hard to unpack.
The standard US estimate for recidivism is 55% at 60 months, meaning that approximately one in two (!) former inmates is incarcerated again within five years.
However, the recidivism estimate varies: another analysis found that only one in three will return to prison when measured across time rather than by cohort.
Confused? Try answering these questions: if a state were to lower its recidivism rate by 10%, how would that change the number of jail beds they filled each day over a five year period? And on the other end of the spectrum, how many additional arrests would take place if recidivism increased by 30% over five years?
More confused? You’re not alone. Everyone from politicians to Supreme Court Justices seem to get the math wrong.
There’s an App for That
In my work as a data scientist helping to quantify social outcomes, I have found that these types of questions seldom lead to intuitive answers or are solved by simple analysis. Granted, we want them to be—simple solutions are more desireable than complex ones. However, truthful, fact-based analysis should not be eschewed for convenience. What’s more, I’ve reviewed economic models that missed the mark by a wide margin, ascribing benefits to recidivism reductions that were not warranted by the data.
In an effort to help remedy this, I worked with a brilliant University of Utah student named Sam Nelson to develop an interactive model that shows how changes to recidivism rates could impact more concrete outcomes, like arrests and prison time.
The result is an agent-based stochastic model that takes user inputs as parameters and then visualizes outcomes. In non-geek-speak, this is a simplified version of SimLife for former prisoners, where each of 1,000 “sims” (i.e., simulated people) is tracked over five years to see how changes to the recidivism rate impact the time they spend behind bars.
As the late British statistician, George Box, wryly observed, “all models are wrong, but some are useful.” We know that the simulation we created is an oversimplification of a very complex system. The outcomes of former prisoners are influenced by any of a million different variables. We do hope, however, that this model at least proves useful for preliminary analysis, when translating recidivism to more concrete measures.
Daniel Hadley is the Chief Data Scientist at Sorenson Impact Center, where he and his team mine the data behind social challenges. To read more about his thoughts on data and recidivism, see this post on this personal blog.