“For the eight years I was locked up, I never even heard the word ‘reentry,’” recalled Marcus Bullock, the prisoner-turned-CEO who was incarcerated in 1996 at the age of 15.

Today, Marcus is speaking to a group of academics and policy advocates from across the United States as they prepare for four days of reentry program tours in jails and prisons throughout Nashville, Tennessee. The two of us were part of that group. Many in the criminal justice reform movement are concerned with a perceived lack of reentry programing across the country, but the reality is that we have come a long way since Marcus was released in 2004. Indeed, in most of the facilities we toured, dozens of incarcerated individuals were participating in reentry programming opportunities, and dozens reported that their particular programs were playing a key role in keeping them crime-free after release. Similarly, the facilitators of these programs consistently lauded the low recidivism rates that participants had experienced relative to some measured average. Yet their consistent positivity stands in direct conflict with the large body of academic research suggesting that most programming, including many of the types used around Nashville, fails to improve outcomes upon release.

To be sure, the individuals with whom we spoke were not a random sample of program participants but instead chosen representatives expected to put each program in the best possible light. Moreover, at least part of the gap between the statistics trumpeted by program directors and those found in academic studies comes from selection bias. This error implies that those who opted to participate in reentry programming may have already been more likely to experience different outcomes than the people who did not participate, even if the program itself had no effect on the outcome of interest. In a prison setting, selection bias means that individuals volunteering to participate in a particular program may simply be those who have a real desire to improve their lives. Given that this desire to change is likely a key element of success after release, it should not be surprising that those voluntarily choosing to participate in any type of program are less likely to return to incarceration after release than those who choose not to participate.

Despite these concerns with the story presented by program leaders and participants, we left Nashville convinced that many academic studies in this area are also failing to measure the full impact of prison programs. To avoid selection bias, academic researchers often rely on randomized control trials (RCTs). RCTs have long been considered the gold standard in academic policy research, including research into the effectiveness of prison program interventions. The basic idea is that participants in the study are divided randomly into two groups: treatment and control. The treatment group is given the option to participate in a particular program while the control group is not. By allowing only a randomly chosen subset of the population to participate in a program, researchers can avoid spuriously attributing benefits to programming when those benefits are really coming from differences between those who choose to participate and those who do not.

A month ago, we believed that by eliminating selection bias using an RCT, researchers could fully evaluate whether each program was effective in reducing recidivism. Now, however, we believe that RCTs necessarily eliminate something that is likely a key element of reentry success: choice. By definition, prisons dramatically limit individual autonomy and freedom. In other words, incarcerated people are allowed few meaningful choices. This is the case even with choices that could affect them long after release. This state of affairs, in turn, makes the choices they get to make even more valuable than they would have been otherwise.

In a typical RCT, members of the treatment group simply choose whether to participate (or, if participation is mandatory, they choose the level of effort with which they participate). Members of the control group have no choice and are instead informed they will not be allowed to participate in any reentry program. This structure limits selection bias but also prevents potentially valuable self-sorting. For instance, in the final prison we toured — Metro-Davidson County Detention Facility — incarcerated people were given the discretion to choose among a wide range of programs including education, entrepreneurship, vocational training, dog training and religious training. This creates opportunities for prisoners to find the program that worked for them. Entrepreneurship programs, for example, may be hugely beneficial for individuals hoping to start their own business upon release, but are largely useless for others even among those with a real desire to make positive changes in their lives. In this way, each of these programs may be causing real improvements in outcomes that are not driven by selection bias. However, by eliminating individual agency, a randomized control trial may not identify these benefits.

The research question that most prison program RCTs are equipped to answer is: Will this program improve outcomes among everyone eligible to participate? This is the right question to ask if we are willing to assign programming and further limit prisoner autonomy and choice. But we should recognize that in limiting prisoner autonomy, we are making it much more difficult for any program to be successful or cost-effective.  Instead, we should allow individuals themselves to identify the programs that will work for them whenever possible. In that case, the question we would need to ask to make good policy becomes: Will people who select into this program be better off given program participation than they would have been otherwise? While more difficult to answer, RCTs and other tools can be leveraged to do just that. By harnessing the power of choice and self-selection, we can likely improve prisoner outcomes while identifying and expanding the programs that are having a real, positive impact on people’s lives.

We have certainly come a long way from reentry being a foreign concept to people like Marcus Bullock, but now we must ensure that reentry programming is effective. One way to do that is to ensure that it provides agency to individuals who barely feel like humans.

Logan Lee is an Assistant Professor of Economics at Grinnell College. You can follow him on Twitter at @norsemanlee. Arthur Rizer is a DPhil candidate at the Centre of Criminology and the Criminal Justice Policy Director at the R Street Institute. He is also an adjunct Professor of Law and George Mason University and a former police officer and federal prosecutor. You can follow him at @arthurrizer.