Does incarceration work?
In response to this recent discussion on X, I thought it would be useful to summarise the evidence on incarceration, with an analysis of the mechanisms via which it reduces crime. My view is that incarceration works, likely more than a cursory glance of the literature implies, relative to other explanations, yet is far from the only relevant factor.
This is what Chat GPT says. Here are some additional points:
The 1960s crime surge occured despite relatively high rates of economic growth, full employment, low inequality, an expansion of social programs with Lyndon Johnson’s Great Society, and unprecedented levels of racial progress.
A relatively recent lottery study suggests that the link between poverty and crime is small at best.
Violent crime seems to lack any causal link to economic variables.
These two points downgrade any hypothesis rooted in socioeconomic explanations for crime. A psychological, incentive-based theory further supports the view that incarceration works.
Antisocial personality disorder is characterised by pervasive disregard for the rights of others, a disregard for social norms and laws, impulsivity, deficiency in empathy and remorse, and irresponsibility. Thought to occur in 1-4% of the population (depending on which studies you consult), inmates with ASPD are disproportionately represented in the penal establishment.
Hence, we can say that a disproportionate proportion of the most dangerous individuals in society are incapacitated. Noah Smith is indeed correct; crime follows a Pareto distribution. Moreover, recidivism rates are sky high. Most inmates are repeat offenders. Therefore, one might conclude that the prospects of rehabilitation for most offenders are slim (although ASPD symptoms may “burn out” with old age independent of incarceration, probation, or any rehabilitation program). “Lock ‘em up and throw away the key” really does seem like the best option for a large subset of criminals.
Suppose that an individual will commit crime if:
Utility(crime) - βp(caught)E[sentence length] > 0
where β represents time preference. If β=1 then the individual treats the future as equivalent to the present. If β=0 then the individual has no regard for the future at all, and just lives “in the moment”.
Of course, the impulsivity of ASPD is equivalent to a low β, which makes these individuals more crime-prone. Likewise, the lower β is, the less deterred an individual is to (length of) incarceration. However, as these individuals are expected to commit more crime, these individuals expose themselves to more opportunities to get caught. The likelihood of getting caught is cumulative over time, hence eventually, the most dangerous will end up in jail.
As such, incarceration will reduce crime, primarily via incapacitating the most dangerous individuals, which commit a disproportionate amount of the crime.
I also posit that conventional deterrence is likely to be effective for the prosocial majority of the population; those with a higher β. As most studies on crime and incarceration focus on actual offenders, the empirical literature underweights the counterfactual if, say, one was less likely to end up in prison or sentences were shorter. This counterfactual is unobserved, so studies that look at actual offenders rather than potential ones, will be biased towards understating the magnitude of the deterrence effect.
I wish to end with one vital caveat. The Brennan study (cited in the Chat GPT discourse from earlier) shows that approximately half of the variation in crime rates across time is an unexplained residual. Indeed, the precipitous fall in crime from its levels in the mid-20th century has occured across the West, regardless of whether governments have been tough on crime (expanding mass incarceration, such as in the US) or soft (e.g. Canada and much of Western Europe).
Likewise, the more perspicacious readers will notice that β itself is an exogenous residual, incorporated to model our ignorance regarding the neurology of crime. We cannot really explain, aside from “impulsivity” or reference to personality disorders, what drives one individual to be more likely to offend than the other. Even in diagnosing ASPD, there is a chicken and egg problem: how do we know if someone has ASPD? Because they keep getting into trouble. Why do they keep getting into trouble? Because they have ASPD!
So to conclude, whilst incarceration will dent the crime rate somewhat (and more than what much of the empirical literature summarised by Chat GPT implies), we do not really know what exactly motivates crime. Some factors are linked, yet much variation in crime remains an unexplained residual. Some epistemic humility is necessary here.