Home > Uncategorized > What are prisons for?

What are prisons for?

July 8, 2015

Yesterday i had the good fortune to interview Sonja Starr, who has been researching evidence-based sentencing models and wrote this New York Times op-ed on the subject.

If you haven’t heard about them, they are “recidivism risk” models being widely used by judges – in 20 states at least – to help determine sentencing lengths. Different judges will use them differently, but think of them as one factor in deciding how long someone convicted of a crime goes to jail. Recidivism refers to the concept of returning to jail, so high recidivism risk means someone is deemed by this model to be likely to return to jail, and when a judge sees that, they typically put them in jail for longer.

Moreover, the attributes which go into recidivism models are often proxies for race and class, which means that, hidden underneath the computer code, a given judge is essentially putting someone away for longer because they are poor.

So, there are lots of issues, but one thing I asked her to talk to me about was whether recidivism is the right question in the first place. After all, when we consider recidivism we are judging someone more harshly if “people like them,” however that is defined, have recommitted crimes after being in jail. In other words, we are judging them more harshly because we suspect they might commit a crime in the future. This is Minority Report type stuff.

The other side of the argument is that judges are explicitly told to decide on sentencing based on multiple factors, and always have been. Namely:

  1. Punishment for the crime that was committed (“just deserts”)
  2. deterrence for others who will want to avoid prison,
  3. the public good – we want to protect people from criminal acts, and
  4. rehabilitation, where we prepare the prisoner to be a well-functioning member of society after they leave prison.

The third goal, that of protecting the public, is where recidivism comes in. If we have good reason to believe the person will commit yet another crime, we want to keep them away from doing so, or even better decrease their recidivism risk with various methods. A few problems, however:

  • The models don’t measure the effect of prison on recidivism risk.
  • In fact they don’t measure how recidivism risk changes at all.
  • The recidivism risk models are probably amplifying that third factor in judges’ sentencing opinions.
  • Drug offenders have higher recidivism risks than people who commit violent crime, but we really only care about the latter in terms of the public good.

After thinking this through, I’m pretty convinced that, aside from using problematic inputs in recidivism models, we need to carefully examine how we want to weight those four factors.

Categories: Uncategorized
  1. Josh
    July 8, 2015 at 9:03 am

    There are many things we could do to reduce recidivism. Rehabilitation, job training, providing a basic income or a job are all possibilities — and probably cheaper than keeping someone locked up.

    In Norway, one of their guiding principles are: “someday, he will be somebody’s neighbor”. “Here it is more normal so you can act more normal when you get out”.

    The US is not Norway. Everything there may not be transferable. But treating the prisoners like human beings, most of whom will be returning to society, would be a big improvement over the current system.


  2. July 8, 2015 at 10:09 am

    For an interesting evolutionary take on punishment, try “The Punisher’s Brain” by Morris Hoffman.


  3. Josh
    July 8, 2015 at 10:23 am

    It seems to me that a naive observer could conclude that our prison system (and our treatment of people released from prison) is designed to increase the recidivism rate.


  4. July 8, 2015 at 10:33 am

    I call so much of this the “excess” scoring of America, way beyond just sentencing and this is a big issue I think with way too much confidence place in “scoring” people. See what pharmacy benefit managers do to you every time you fill a prescription, you’re scored on medication adherence predictions. Are you a male, ding, a male seeing a female doctor, ding, or not using mail order prescription services, ding. Over 300 metrics, and most of which look like they have nothing to do with taking your prescriptions. These scores are bought by health insurers who use the scores to determine your overall risk and price of your premium as well as purchased by drug companies. The scores are proprietary and nobody can verify for accuracy and Express Scripts right on their site brags about it.


    Prison terms or prescriptions, we have too much excess “secret” scoring going on in the US.


  5. someone
    July 8, 2015 at 2:28 pm

    Well, if you are concerned about recidivism you should always sentence someone to life imprisonment, that should be obvious. 🙂
    But perhaps some measurement of “cost” and proportionality also needs to be taken into the equation?
    Perhaps to you it is obviously subsumed into one of the points you mentioned, but I don’t see that.


  6. captain obvious
    July 8, 2015 at 5:48 pm

    “…proxies for race and class.”

    Actually, Starr’s research found that for the odds of conviction and incarceration, the race and class coefficients are in the opposite direction than implied in the popular discourse about “race and class”.

    http://ssrn.com/abstract=2144002 (Table 2) purports that

    – it’s better to be a HS graduate than a College graduate
    – people with GED’s or College degree don’t get lower sentences than people with minimal education levels (no HS diploma, no GED)
    – blacks have 10 percent lower odds (vs whites) for Non-Petty Conviction and for Incarceration. This finding is replicated in Starr’s earlier study on racial disparity in sentencing, using a male-only sample.
    – it’s better for every measured outcome to *not* be a US citizen

    By far Starr’s strongest finding was that

    – women are overwhelmingly favored at every stage of the process over men, no matter what subgroups and controls are considered. The effects have very large magnitude and much larger than anything related to race, education or other background variable. As Starr discusses in the paper, it is hard to account for these results without gender discrimination (in favor of women) being the predominant factor.

    This is not to say that every finding went against popular expectations. Blacks get 10% longer sentences after conviction. But in light of the other findings this may partly (or entirely) reflect some selection effect on which cases that go forward.


    • captain obvious
      July 8, 2015 at 5:56 pm

      (correction: the statement about the US citizen coefficient is wrong. It is better for some outcomes and worse for others. The point still holds, that Starr’s study confounds the popular notions on discrimination, including the idea that immigrants always are treated worse.)


      • July 8, 2015 at 8:37 pm

        Interesting. I’m pretty sure it’s worse to not be a high school graduate than to be one, and I’d also guess that the fact that blacks are more likely to be arrested for drug convictions in regions of “broken windows” policing has the opposite effect as well.


  7. captain obvious
    July 8, 2015 at 9:50 pm

    It’s worse in life generally to not graduate high school, but (as Starr’s regression suggests) that is not necessarily true for criminal defendants, after controlling for some other relevant factors.

    For instance, there could be a sort of ability selection in crime as in most other pursuits, that makes less functional people more prone to spend their time on smaller-scale violations. Or the no-HS people may be likelier to do exactly what their lawyer advises. Or the judges might actually be a bit more lenient with people who have difficult life stories.

    Starr’s results are an existence proof. The prediction models that operate in judges’ brains, which do use all kinds of additional information about the defendant, do not seem to systematically disadvantage the usual suspects (except men). Hence a computer model might not do so, either, and a judge who uses a computer model as one more source of input would not necessarily have his decision biased against those who are poor, or black. (If the article is any indication, taking things a bit out of the judge’s hands might actually make things more equitable for men!)


    • July 8, 2015 at 9:56 pm

      What is your evidence that judges don’t discriminate? I have certainly seen evidence that they do.

      In any case I don’t think we disagree with the main point, which is that there’s no reason to think a model is inherently more fair or objective than a judge. It is at most more consistent.


  8. AG
    July 10, 2015 at 2:51 am

    According to Kwame Anthony Appiah, whose lecture on moral revolutions I attended on 5 July, the US has 4% of the total world population and 24% of the world prisoners (he even made a remark, which I fund difficult to believe, that the percentage of inmates in the US today is higher than in the USSR during Stalin’s times. )

    The treatment of inmates, is very likely, according to Appiah, to be the subject of a forthcoming moral revolution (see his book “The Honor Code: How Moral Revolutions Happen”).

    PS. See also



  9. Andy
    July 19, 2015 at 7:40 am

    Of course, prison has very little rehabilitative impact and probably has negative, so we should remove the last principle as it is just a fig leaf for people to feel good about sending others to live in cages (“he’ll see the light and become a better person!”). Also — the 3rd principle of incapacitation is horrible as it leads directly to the reasoning that one should imprison someone else just for the likelihood of committing a crime without actually having committed it. The 2nd principle is cruel as well because we might as well hurt anybody for any reason on the grounds that it might inspire the others to behave better. And with regards to the 1st, I’m an atheist and don’t understand this “just deserts” philosophy (as if there is some cosmic Justice) waiting to be meted out, so that doesn’t appeal to me either. So — I’m really stumped on how we choose to put people in prison and for how long other than, of course, the powerful wanting to incapacitate an annoying segment of the population and scare the others into behaving.


  1. No trackbacks yet.
Comments are closed.
%d bloggers like this: