In graduate school I was captivated by the power of agent-based models. In a few lines of code, I could simulate virtual societies of actors responding to abstract costs and benefits imposed by the environment or by their interactions with each other. After a few minutes away from my computer, this virtual population would reach an equilibrium, from which it was easy to draw sweeping conclusions.
While these models smoothed away much of the real-world complexity of human behavior, they captured an important truth: people respond to incentives. Greater rewards tend to lead to more of a behavior; greater punishment tends to lead to less.
At the same time, it was easy to highlight the absurdity of overly reductive models. For example, imagine that we wanted to simulate the effect of different levels of punishment on speeding, holding constant the probability of getting caught. A population of perfectly ‘rational’, payoff-maximizing agents will respond in a predictable way. If you remove all punishment, agents in equilibrium will ‘choose’ to speed when it benefits them. If you ramp up the punishment to the max, nobody will speed.
Given that speeding causes accidents, which lead to injury and death, we could infer from this analysis that the maximum-punishment world is ideal. The fewest people die from accidents, and nobody actually experiences the harm of getting punished, either, because nobody dares to speed. The policy implications are clear: we should impose a death penalty for speeding.
An imperfect world
I’m kidding, of course. People aren’t incentive-maximizing robots. At best, they’re noisy maximizers who generally respond to incentives, but also make mistakes. Moreover, some people are much better maximizers — that is, less noisy — than others. Indeed, perhaps the kind of rational framework championed by Gary Becker doesn’t apply at all to a subset of the population.
And so, what might look like a crime-free utopia under the assumptions of universal rationality becomes a dystopia once we account for a little bit of decision noise. Indeed, almost nobody would find a death penalty for speeding palatable even if it saved many more lives than it sacrificed. Part of this intuition is driven by the fact that speeding is not the kind of crime that we think ever deserves such a draconian punishment, whether or not it’s done fully intentionally. But I want to turn attention to a different issue: in this hypothetical world where we impose a death penalty for speeding, speeding would never be done intentionally. The only speeders — and the only people killed for speeding — would be those who sped accidentally: those whose speedometers broke, who experienced momentary lapses in attention, or who were incapable of responding to the threat of punishment. And that would make the policy even more monstrous.
Selection bias from the threat of punishment
This contrived example is meant to illustrate what I take to be a genuine dilemma for designing punitive policy: the more that a threatened punishment succeeds in deterring ‘rational’ (i.e., incentive-sensitive) actors from committing crimes or civil violations, the more that the people who actually experience punishment will be unresponsive to the incentives that inspired these policy. So if we focus our attention on those who enter the justice system, deterrence will appear ineffective, and the penalties imposed on these individuals may seem unnecessarily cruel. But this is the tragic irony of living in a world where incentives mostly work to keep people from being punished in the first place. The people who actually get punished are the ‘failures’ of the system insofar as they weren’t successfully deterred.
Lest you get the wrong idea, I don’t mean to suggest that our current justice system is optimal or that it would be more optimal with harsher punishment. In fact, I first began thinking about this kind of selection bias last year when I was auditing a course on neuroscience and the law. This course, which was taught by a retired federal judge who seemed to regret having sentenced many criminals to long stints in prison, explored many ways that our current justice system imposes unfair or downright evil harms on convicts, many of whom suffered from brain damage, mental illness, or environmental trauma that should mitigate retributive justifications for punishment. These are obviously important concerns, which I hope inspire compassion for inmates and encourage reform.
Nevertheless, it felt natural to come out of the course wondering what the point of all this punishment was for in the first place. Sure, for violent crimes, maybe we need to incapacitate people who are dangerous even when we don’t think they deserve to suffer. (Ideally, we would do this in a more humane way than is done now.) But beyond this rationale, the justice system seemed to be applying an antiquated notion of retribution to a population of inmates who were largely ill-equipped to understand the nature of their crimes.
This is a deeply unfortunate state of affairs. But consider the flip side, which is easy to miss because it’s hypothetical and uncertain. If we reformed the penal system to be substantially more lenient than it currently is — thereby making it kinder to many of the undeserving people currently suffering — might it be the case that some of the people who don’t currently commit crimes because they want to avoid the harsh punishments would change their behavior?
I realize that the empirical evidence on this question is somewhat mixed, and that in general, would-be criminals are more deterred by the certainty of punishment than the severity of it. But, for practical reasons, most of this evidence concerns relatively small changes in policy (e.g., changes in a few years of prison) — not the kind of dramatic changes that some prison-reform advocates have in mind. At the very least, if we made punishment much more lenient for crimes where the incentives are clear, such as tax evasion, I think it is quite likely that many more financially savvy people would try to get away with it. Likewise, if we put speeders in prison, I suspect that plenty of people would drive more cautiously than they do now. I know I would!
This is not just a thought experiment; this trade off is playing out now in the domain of eviction policy. Housing is one of the clearest domains in which rational incentives from strategic actors mix with the universal necessity of shelter. On one hand, we don’t want to give landlords too much power to surge rents or demand immediate payment from those who are trying their best to pay and would experience enormous hardship from losing their homes. On the other hand, if policy shifts too much in favor of tenants’ rights, opportunists who want to avoid paying rent will be better able to exploit the system. These opportunists will not only take advantage of their landlords, but potentially deter future landowners from renting out their places, thereby raising rents for future renters.
This is a genuinely difficult trade off. But I again believe it would be a mistake to devote exclusive attention to the immediate victims of the policy (i.e., those who are actually evicted). In states with liberal eviction laws that favor landlords relatively more, we should expect that the population of evicted tenants will be more skewed towards the genuinely needy because of the selection bias we’ve been discussing. The strategic actors won’t find it worthwhile to try to take advantage of the system to save a little bit of rent. To the extent this is true, the policy is working, and we shouldn’t infer from the composition of those evicted that this policy is necessarily more harmful to renters in aggregate than a policy that makes eviction more difficult for landlords. We need to weigh the harms of actual punishment with the benefits of deterring people who would be punished under the more lenient policy.
Of course, sometimes it’s possible to mitigate these selection issues by evaluating violations on a case-by-case basis. The legal system recognizes degrees of intent for serious crimes like murder and even allows for an insanity defense in a small minority of cases in which there’s overwhelming evidence that the defendant lacked an understanding of their behavior. Judges can likewise moderate their sentences of convicted defendants based on perceived hardship or necessity. Yet, based on what I learned in the law and neuroscience seminar, these seem to be weak remedies for a more general problem. And malingering and exaggeration are genuine issues: it’s easy to craft arguments, bolstered by possibly unscrupulous expert witnesses, that the defendant lacked agency due to mental or physical disability. If the legal system makes such defenses more available, we again need to worry about the deterrent effects on rational actors who are not currently breaking the law.
Conclusion
As I argued in a previous post, selection bias is an important issue for the legal system to grapple with. It’s well known that the people most involved in the justice system are not a random sample of the general population and that they tend to be less privileged on a variety of socioeconomic, racial, and cognitive dimensions. But, here, I’ve hoped to highlight a kind of selection that emerges from the process of deterrence itself. The more that the system ‘succeeds’ in deterring rational, incentive-responsive actors, the less these people will become involved in the justice system. The people who are left to be punished will be disproportionately drawn from a group who doesn’t, or can’t, respond reliably to abstract incentives. Thus, perhaps we shouldn’t be too surprised if ex-convicts reoffend at high rates and don’t dramatically change their behavior in response to longer potential prison sentences. On its own, this result is expected even in a world where deterrence generally works.
In essence, the Becker-inspired perspective of law and economics and the more up-close perspective of the justice system that I encountered in my law and neuroscience seminar are more compatible than they initially appear. The former perspective takes an outside look at society, like my agent-based models. From this perspective, the threat of punishment largely works to deter. The latter perspective takes an inside look at prisons and courtrooms where an unusual — but very important— subset of people are subjected to consequences that are largely ineffective and borderline cruel. This difference in perspectives is a matter of how far we zoom in on one cut of society versus zoom out to consider the broader population. Both perspectives are essential to take into account when crafting thoughtful and effective laws.
Moreover, this selection ‘paradox’ poses both practical and ethical challenges. Should we favor harsher sanctions to keep more people out of the justice system and redirect potential criminals towards more productive activities? Or should we favor more lenient penalties to avoid unnecessarily harming people who are potentially undeserving of their punishments and unlikely to respond to them? I realize this is a gross simplification of many issues, and there are likely to be ways that we can manipulate punishment to be both less cruel and more effective as a deterrent. But I suspect that this broader trade off will be difficult to avoid completely. While it would be morally convenient to believe that criminals are ordinary rational actors who fully foresaw the consequences of their choices, I am skeptical of approaches that try to frame all behaviors, including mental illness and classic addictions as purely incentive-based. Reality is far messier. As I enter law school in the fall, I hope to explore how best to balance the societal need for broad-based deterrence with the tragic suffering that befalls many of the people who enter the justice system.
