EAs should do more rough policy modelling: UK prisons as an example

What is Effective Altruism in 2025? My go-to definition is that it is a community of people that want to do as much good as we can in the world. But what unites us? As cause areas and approaches naturally fragment, it is not so much any shared substantive belief, but how we reach our judgements, I think. As much as anything else, EA is a mode of reasoning, a process for working out how to do good.

I expect the features of this reasoning process will be familiar. Comfort with probabilistic thinking. A desire to quantify uncertainties. A focus on identify the crux of arguments, and carefully distinguishing moral premises from empirical questions. I think that toolkit has been deployed to great effect in ways that have had a positive influence on things like development economics. But I think it remains under-utilised in politics and policy, even increasing numbers of EAs seem to get involved in those fields.

That’s a great shame, because my view, having spent around a decade in and around British politics, is that we could do with a lot more of such reasoning to clarify problems and identify the right way to think about them. What follows is an attempt to exemplify the sort of thing I’d like to see more of: rough cost-benefit or cost-effective analyses, using back of the envelope calculations, transparent about uncertainties and inviting debate over key parameters. The sort of thing that organisations like GiveWell and Open Philanthropy and Rethink Priorities do a lot of in other fields, but which aren’t exactly commonplace on policy questions (with the possible exception of AI policy). 

There are some positive examples. I think of George McGowan’s post on UK housing policy from a couple of years ago. Or more recently, Adam Corlett’s post on how impact assessments should quantify animal welfare changes. But I’d love to see more of this sort of thing, so I’m cross-posting a piece from my blog.

Last October, the UK government launched a review of criminal sentencing with a remit to reduce reoffending and establish how to make better use of non-prison alternatives. Its objective is to ensure that the country never runs out of prison space again, though the government has also committed to adding 14,000 prison places by the mid-2020s. Meanwhile, a high profile campaign, which seems to be aimed more at Conservatives than the government, says we can “crush crime” by “up to 90%” by sending more prolific offenders to prison. Yet Great Britain already has the highest incarceration rate in Western Europe.1 So who has it right? Are we too lenient or too severe?

I found Scott Alexander’s summary of the evidence on prison and crime - and David Roodman’s evidence review which it features prominently - helpful in thinking through these issues. But they are very US focused - what might they tell us about the UK? 

In this post I want to ask the question of what the optimal prison population would be in order to cost-effectively minimise crime. The first thing to say is that this is in some ways the wrong question. Preventing crime is not the only - or even primary - objective of the criminal justice system. Certainly the criminal legal process seems oriented more towards individual-level justice than it is towards keeping us collectively safe. The appropriate balance between these objectives is something we really ought to debate more explicitly.

The second thing to say is that if we want to reduce crime, prison is probably the wrong place to start. There is enough ambiguity around its effects, and enough other levers to pull - policing, technology (eg home and car alarms), environmental (eg lead reduction), policy (eg alcohol) - that promise greater benefits for less cost. 

With those two caveats aside, here are the three striking stylised facts from the reviews:

1. The deterrence effect of harsher prison sentences is relatively modest, particularly for violent crimes that tend to be less calculated
2. The conventional wisdom is that “nothing works” when it comes to rehabilitation, and to the contrary sending someone to prison is more likely to increase the risk of reoffending
3. What is effective is incapacitation: people can’t commit crimes against the general public while locked up (though they can victimise other people in prison)

These conclusions are remarkably counter-cultural. Rehabilitation is likely to be at the heart of the sentencing review, and has been a significant part of successive governments’ rhetoric. Meanwhile, incapacitation seems to happen mostly as a side effect of the justice system, rather than as a deliberate strategy (this is what I meant when I said we should have more of a debate between preventing and avenging crime. Would greater focus on incapacitation be a pragmatic response to ‘what works’, or an unjust use of individuals as a means to an end?). It is notable that maybe the most prominent example of detention for incapacitation - imprisonment for public protection - was highly controversial and quickly reversed. 

Roodman argues that the evidence shows that the ‘criminogenic’ effect of prison (the increased risk of crime due to the influence of fellow prisoners, and the disruption to one’s life and career) more or less exactly offsets the incapacitation effect. So the marginal prisoner commits about as many additional crimes after leaving prison as they are prevented from doing inside. 

But his argument relates to the US, which has a much higher incarceration rate than the UK. We might assume that means that the marginal British prisoner is more of a risk to the public than the marginal American prisoner. Think of it this way: to be in a British prison, you have to be in the top 0.13% of the population in terms of criminality, whereas in the US you only need to be in the top 0.7%. Indeed, studies conducted in Europe seem to find that prison has a stronger effect on reducing crime than studies in the US. 

Is more prison cost-effective? The direct cost of a year in prison in England and Wales is around £52,000. That is equivalent to the government’s estimate (uprated for inflation) of the social and economic cost of a single rape incident, three violent attacks or robberies, 7 burglaries or 28 thefts. David Roodman’s ‘devil’s advocate’ estimates for the US imply that a single year of prison there prevents at best seven crimes in expectation, and most of them are thefts. Applying his numbers to the UK valuations, the expected benefit of a year’s prison is just over £44,000 - so a bit less than the direct cost. But remember that is based on the assumption that prison is as ineffective on the margin here as it is in the US.

An alternative approach uses ‘elasticities’. A few different studies converge on the estimate that a 1% increase in prison population would reduce crime by around 0.3%. Interestingly, this study in Italy finds the incapacitation effect to be in the range of 0.2-0.35%, so not too far out of line with the US, despite the smaller Italian prison population. Given the UK government’s valuation of the cost of crime, a 0.3% reduction would be carry societal benefits valued at £177 million.2 By comparison, the cost of a 1% increase in prison population would be £45 million. Even if we try to account for the negative effect on prisoners - their lost earnings and reduced wellbeing - that only increases the cost of prison to £89 million. 

I’m not sure how exactly to square the difference between these two methods of estimating the benefits of prison. A 0.3% elasticity would imply each year of prison would avert 16 incidents of crime, more than double Roodman’s figure. That feels plausible, given the difference between the British and American margins. As a sense check: this study of the distribution of violent convictions in Sweden finds that the 0.10th percentile offended three times as often as the 0.69th percentile (12 vs 4). This is all very rough and speculative, but I’m inclined to provisionally conclude that additional prison would reduce crime in the UK. 

That’s just a first pass answer, and I welcome anybody who wants to pick holes in my model or propose refinements. You can see it here. But it makes a certain intuitive sense: if there is genuine equipoise in the US with its much bigger prison population as to whether prison reduces crime on the margin there, it stands to reason that prison ought to be effective here. 

Comparing international prison and crime statistics raises some doubts in my mind. First, what are we to make of the fact that other countries have much lower prison populations than the UK with similar or lower levels of crime? My instinct is to say this reflects the fact that prison population is only one of a number of factors that influence crime rates, and suggests those countries could have even lower crime rates with higher incarceration rates. But this is speculation. Second, how do we square the pessimism about the effectiveness of rehabilitation with the reductions in recidivism achieved by countries like Norway, where two year reoffending fell from 70% to 20%? That needs more investigation. But it’s worth pointing out something a little paradoxical here. The strongest case against prison - the central pillar of Roodman’s case for decarceration - is its criminogenic effect, the fact that it makes some people more likely to commit crime on release. If that criminogenic effect is weakened, or reversed, prison becomes less societally costly. Thus, the better prisons are at rehabilitation, the more people we ought to imprison, so that more can benefit. 

Finally, while putting more people in prison might be effective at reducing crime and pass a cost-benefit test, that does not mean it is the optimal approach or the right priority. There may be more humane and less costly ways of achieving incapacitation (for example, electronic tagging). And more active prevention could be a better use of time, resources and political capital. I haven’t considered these issues explicitly here. But I’m struggling to avoid the conclusion that at least in a narrow sense, Michael Howard was right: prison works.