A Zombie in the Wild

I have long thought fairly highly of the Atlantic, both as a magazine and as a blog. So the following article by Richard Gunderman¹ is disheartening to read. It is a perfect example of the very zombie I am trying so hard to kill: the "Standard Story" that unquestioningly accepts the generally-incorrect conventional explanations without (for obvious reasons) providing data to back them up. So I thought I'd spend this post attacking it point by point, just so it is clear how deeply flawed the conventional story is, and to highlight the dubious arguments that are so often made in favor of it.

Gunderman starts with the standard it-isn't-crime explanation:

Why have U.S. incarceration rates skyrocketed? The answer is not rising crime rates. In fact, crime rates have actually dropped by more than a quarter over the past 40 years.

His statement that crime has dropped by 25% over 40 years is wrong in several ways. As the graph below (taken from here) shows, crime has only been dropped since 1991, which is 24 years ago. Between 1974 (that's 40 years ago) and 2011 (the last year for which the FBI has data), violent crime has risen by 23%, and property crime has falled by just over 2%. The net change: + 0.1% (since there is about 10 times as much property crime as violent crime). So he is just factually wrong.²

But looking at the graph reveals another, deeper problem with his analysis. Given that crime soared from 1960 to 1991 (with a little pause for violent crime in the early 1980s), why present just a single percent-change number? If we want to understand why prison populations have risen sharply since the mid-1970s, we can't just ignore the unprecedented rise in crime that accompanied the first 20 years of prison growth.

Furthermore, if we want to understand why crime remains such a politically powerful issue, just note that despite the crime drop since 1991, violent crime is still 100% higher than it was in 1960, which were the formative years of the politically-powerful Baby Boom cohort. And much of the drop since 1991 has come through self-protective measures that don't necessarily make us actually feel safer (security systems, not going out at night, etc.). So we are still a relatively violent country by historical standards for a large bloc of voters. 

Gunderman's conceptually and factually misleading number misses all of this, and thus understates the direct and indirect roles that crime can play. Sadly, this is not an unusual problem in the literature.

Next he moves on to sentencing:

New sentencing guidelines have been a key factor. They have reduced judges' discretion in determining who goes to jail and increased the amount of time convicts sentenced to jail spend there. A notable example is the so-called "three-strikes" law, which mandates sentences ranging from 25 years to life for many repeat offenders. 

First, let's start with the strike laws. While a majority of states have them, according to Frank Zimring about 90% of all strike sentences are handed down in California. So states have them, but don't use them. 

Moreover, guidelines are used in a minority of states, and some data suggests that guidelines are negatively correlated with prison growth: states adopted them to rein in prison populations. Now if we are talking about the federal guidelines, maybe Gunderman is right. But the story is much different in the remaining 89% of the system.

Even more important, it simply isn't the case that longer sentences has caused prison growth. This is the biggest zombie idea of them all, and I will be dedicating several posts to it down the line. But it simply is not the case. I'll give Gunderman a pass on this claim, though. It is almost accepted as gospel inside and outside the academy. I've had people tell me that my results must be wrong because of the conclusions I reach, a complete inversion of the (social) scientific process, and one that must make Thomas Bayes and Pierre-Simon Laplace spin in their graves.

Up next, an oldie-but-goodie:

Perhaps the single greatest contributor has been the so-called "war on drugs," which has precipitated a 12-fold increase in the number of incarcerated drug offenders. About 1.5 million Americans are arrested each year for drug offenses, one-third of whom end up in prison. Many are repeat offenders caught with small quantities of relatively innocuous drugs, such as marijuana, a type of criminal activity often referred to as "victimless."

Do I even need to say anything more at this point? Maybe just two small things. First, the ratio of drug inmates to drug arrests is about 23%, not 33% (see here and here). And 1.5M arrests is a large number, but keep in mind we arrested almost 12.5M people in 2011. It would be surprising if just 12% of all arrests drove everything. The back-of-the-envelope calculations don't seem to work.

Gunderman then turns his attention to crack/powder sentencing in particular:

Some sentencing laws seem little less than perverse. For example, in the 1980s, crack cocaine received a great deal of public attention. In response, the U.S. Congress passed legislation imposing a 100 to 1 sentencing ratio for possession of crack cocaine, as compared to its powdered form. ... From a medical point of view, this makes little sense.

First--again!--we should focus on state sentencing, not federal. And apparently most states do not punish crack and powder differently, and those that do use lower ratios.

Moreover, the medical argument is tricky. What matters is not the chemical form but the method of delivery: oral ingestion is more addictive than smoking or IV use, and smoking and IV use are more addictive than inhaling. Since crack is generally smoked and powder frequently inhaled, the form did make a difference. Moreover, there were real social costs associated with the introduction of crack, though these were almost certainly linked more to the destabilizing effects of the crack markets, not the drug itself, since crack use appears to remain at about 70% of its peak use level.

Now perhaps targeting form rather than method of distribution, or targeting the drug rather than the social ills directly, were bad policy decisions. But the issue is far more complex than the glib "little less than perverse" implies.

Next, Gunderman turns to the costs of prisons, arguing:

The costs of incarceration are high. For example, the state of California spends approximately $9,000 per year for each public school student it educates but over $50,000 per year for each inmate it keeps incarcerated. The proportion of the state budget devoted to imprisonment has been increasing at a rate much faster than that for education. Moreover, despite California's huge prison expenditures, its prisons recently held 140,000 prisoners in facilities designed for only 80,000.

First, all fifty states are different, and when it comes to penal policy California is a distinct--albeit large--outlier. So it does not necessarily make sense to use California as a stand-in for the US. As my own work has shown, Census data on expenditures suggest that prison spending as a share of the budget has been flat since crime started dropping in the 1990s. States have become much richer during that time, and they have chosen to spend on everything. There may be some crowding out going on, but it is not immediately clear.

(In the figure, don't worry about the three lines: they are just different ways of thinking about a state's budget. But no matter how we define "budget," the same pattern holds.)

Moreover, at a national level, spending on schools greatly exceeds that on prisons. Perhaps on a per-student and per-prisoner basis the prisons get more, although the implications of that are not immediately clear--there are a host of assumptions about the correlation between spending and outcomes that underlie Gunderman's point. These assumptions may be true, but they need to be supported (or at least acknowledged).

Finally, note that the $50,000/prisoner number--which is one of the highest levels in the country--is just an average cost measure, not a marginal. Cutting one prisoner will not reduce costs by $50,000. After all, releasing one prisoner does not reduce heating, staffing, maintenance, or other costs at all. The best estimate of marginal costs that I have seen, using data from Maryland, suggests that marginal costs are half of average costs.

Then, he turns to the other side of the prison-crime problem:

Does prison do any good? This is a surprisingly difficult question to answer. 

He's absolutely right: given the endogenous nature of prison and crime, disentangling any sort of causal story is incredibly hard. But he uses this difficulty to basically just throw up his hands and admit that there may be some incapacitative and retributive benefits, but that's about it. Perhaps. But while complicated, there is a lot of data out there, and the more-methodologically sound studies do find that prison growth reduced crime. We may be well past a point of declining--maybe even negative--marginal returns, and our focus on prison likely distracted us from what would have been a much more efficient focus on police. But again, these are much more nuanced arguments than the usual "prison doesn't stop crime" argument that gets trotted out as part of the Standard Story.

He goes on:

Yet it is difficult to make the case that so-called correctional institutions do much in the way of correcting, reforming, or rehabilitating inmates. The recidivism rate at 3 years post-release is about two-thirds, of which over half end up back in prison. The most important factor in preventing recidivism is not the amount of time people serve in prison, but the age at which they are set free. The older inmates are at the time of their release, the less likely they are to return.

First, an "amen." The age-profile of offending is a hugely-overlooked issue in our criminal justice system. Offending does not occur randomly over the life-course. Those who offend repeatedly start in their early teens with property crime, graduate into violent crime in their late teens and early 20s, and start to ease out of offending in their late 20s and 30s. Of course, there is a lot of variation in these trends, and sadly we cannot seem to predict who will follow what trajectory in advance. 

But our policies clearly ignore this fundamental fact. Offenders generally get their harshest recidivist-enhanced sentences just as they are most likely to start aging out of criminal behavior. One could argue that such sanctions are the necessary evils of maintaining a credible threat, but the evidence about the deterrent effect of severe sanctions is weak. On the other hand, throwing the book at the young first-timer is hard because of the false-positive risk. Gunderman deserves credit for drawing our attention to this.

On the other hand, the 2/3 number is a really tricky one to understand. First, if 1/3 of all cancer patients receiving a chemotherapy treatment survived three years, would we call the therapy a "failure"? It depends on the baserate survival risk without the treatment--and we have no idea what that would be for the recidivists. If 1/3 would not have recidivated no matter what happened, then prison does not reform well. But if all would have recidivated but for incarceration, then maybe 1/3 is a remarkable success, given the challenges of changing human behavior later in life.

Instead, we should look to prison programs directly. And here there is a huge literature which suggestst that a lot actually works, although context, design, etc., etc. all matter significantly. Again, a much more complicated picture than the Standard Story is equipped to handle.

His next point is that incarceration hurts the families of inmates, and this is a good point to make; I don't really have anything to criticize. In our debates over prisoners-vs-victims, it is important to remember that many family members of inmates--particularly their children--are themselves now victimized by the process.

His turn to community harms, though, again reflects the unnuanced perspective of the Standard Story:

Incarceration also takes a big toll on communities. Its costs, both direct and indirect, are high, and it draws resources away from other equally or more worthy needs, such as education and healthcare. Some communities, particularly in inner cities, are devastated by incarceration.

True, but since crime is geographically concentrated, these are also the communities devastated by crime. As James Forman has pointed out, much of the demand for tougher sentencing laws during the 1980s came from inner-city black communities, which have also borne the brunt of their enforcement. Crime policy is not just some disinterested state imposing its will on politically powerless inner-city communities. 

That's enough (and the end of the Atlantic article). Sadly, this is what I am used to reading: this is the Standard Story in a nutshell. And it is wrong in so many ways. It undersells crime, it oversells harsh sentences, it focuses too much on drugs and not enough on the complicated politics of a disaggregated criminal justice system. It looks at the harms to inmates and families--perhaps because its focus on drug crimes leads it to think of average offender as someone who committe a "victimless" crime--but ignore the victims of crimes. And, in particular, ignores how the victims of crimes are generally the neighbors of the victimizers, making the community story a tough one to describe empirically.

And as long as we accept the Standard Story, it is unlikely we will implement reforms that really target the heart of the problem.

¹Gunderman is a pediatric radiologist who writes primarily about bioethics. So I have no idea why he feels qualified to write about prison growth, and why the Atlantic decided to publish his writings. His primary hook seems to be that mass incarceration has serious public health ramifications, which is indubitably true. But that does not automatically make a doctor qualified to write about such a complicated social process (nor an epidemiologist, for that matter). All I can think of is this xkcd cartoon.

²Violent crime has fallen by 37%, and property crime by 28%, since 1992. But that is just 24 years, not 40.