Hide table of contents

I wrote this post on my blog, and given that it's smack bang in the middle of what this forum deals with, I thought I'd share it here.

If anyone with a background in this sort of analysis wants to work with me to make a detailed cost-benefit analysis of incarceration a reality, get in contact

Cost-Benefit Analysis in a nutshell

Cost-Benefit Analysis is a methodology for evaluating the aggregate costs and benefits of programs. Roughly speaking, it works as follows. For some intervention, X, we look at how much the supporters of X would be willing to pay to make it happen, and how much the opponents of X would pay for it not to happen. We add in the amount it costs the government to carry out the policy on the costs side, and if:

Benefits - Costs > 0

The program is in some sense “recommended” by the procedure. Later in the essay, we will describe how to make a Cost-Benefit Analysis approximate an assessment of whether or not a policy is good from a utilitarian point of view.

Do you notice anything missing in this assessment of the costs of imprisonment? A certain glaring absence?

Let us review an extract from "Does prison pay?" Revisited by Piehl & Dilulio, which is exceedingly representative of the field as a whole. Notice that something very glaring is missing in the mix.

[Feel free to skip this section]

Estimating the social costs and benefits of competing transportation or environmental policies is no analytical picnic. But estimating them for imprisonment and other sentencing options is a certain analytical migraine.

For starters, it is widely asserted that it costs $25,000 to keep a prisoner behind bars for a year. But the latest Bureau of Justice Statistics figures for average annual spending per prisoner are $15,586 for the states and $14,456 for the federal Bureau of Prisons (which holds about 10 percent of all prisoners). These figures are calculated by dividing the total spent on salaries, wages, supplies, utilities, transportation, contractual services, and other current operating expenses by the average daily inmate population.

But hidden and indirect costs of running prisons might bring the $25,000 figure closer to reality than the official spending averages would allow. For example, some tiny but nontrivial fraction of government workers outside of corrections (human services, central budgeting offices) spend time on matters pertaining to prisoners. And Harvard economist Richard Freeman and others suggest that incarceration decreases post-release employability and lifetime earnings potential. Thus an ideal estimate of the social costs of imprisonment would include any relevant spending by other government agencies, plus whatever public unemployment compensation, welfare, and health expenditures result from the negative short- and long-term labor market effects of imprisonment on ex-prisoners.

Also, there is wide inter- and intra-system variation not only in what it costs to operate prisons, but in how prison dollars are allocated as between security functions (uniformed custodial staff), basic services (food, heat, medical supplies), treatment programs, recreational facilities, plant maintenance, and other expenditures. Whatever the best estimate of prison operating costs, such cost differences suggest that efficiency losses are occurring in some places and that efficiency gains are possible in others.

The cost-effectiveness of prisons, however, is by no means strictly determined by correctional administrators. Over the past 25 years the courts have had a major impact on both the total costs of operating prisons and the distribution of prison dollars between security and other needs. For example, in the wake of a sweeping court order, prison operating costs in Texas grew from $91 million in 1980 to $1.84 billion in 1994, a tenfold increase in real terms, while the state's prison population barely doubled. Texas is now one of at least 20 states that spends less than half of every prison dollar on security.

Finally, it is worth remembering that barely a penny of every federal, state, and local tax dollar goes to support state prisons and local jails. State and local governments spend 15 times what the federal government spends on corrections. But state and local spending on prisons and jails amounts to only $80.20 per capita a year, or $1.54 per capita a week.

Nowhere in the above is there any acknowledgment that prison inflicts major non-economic costs (or more accurately, harms) on prisoners.

A curious reluctance

There are, in the literature, relatively few Cost-Benefit Analyses of incarceration compared to, say, studies of the costs and benefits of the provision of water. When these Cost-Benefit Analyses do happen, they are often unwilling to tally all costs and benefits. In particular, almost all studies exclude costs to prisoners. Certainly, there are some CBAs that include these costs, but far fewer than one would like. Even those authors who make a heroic attempt to include all the costs and benefits which need to be included all seem to fail- this is not their fault, but rather reflects the underdeveloped state of the literature on this topic, which forces authors to do their CBA almost from the ground up.

Why is there so little CBA on prisons? My guess is moral and political reservations, but that’s an odd excuse, given the ruthless, economizing logic of policy analysis after 1970. CBA was meant to be the cure to government waste, inefficiency, and unintended harm. It betrays a certain timidity and evokes a certain suspicion that economists and criminologists haven’t applied it widely in this field. Might it reveal among economists a certain sense that some things are just too important for cost-benefit analysis, conveying an almost sacral quality to punishment?

Prisons give lie to a view of neoliberalism sometimes held by both its supporters and detractors- that it has baptized everything in the “icy waters of egotistical calculation”. Prison has in no sense been rationalized, even while so much else has because prison, the infliction of suffering on the hated, gives a kind of baleful warmth to the arctic world of post-80s governance. There is no alternative, but don’t you worry, we’re gonna torture that sicko. Cost-Benefit analysis has always been, in practice, used in highly selective and strategic ways. War, policing and intelligence are lightly scrutinized while schooling, healthcare and the like are put under the electron microscope.

If there is a lack of Cost-Benefit Analysis in academic studies of incarceration, there is even less by bureaucrats with real access to decision-making processes. In almost all areas of US federal government, Cost-Benefit Analysis is ubiquitious. Existing standards require cost-benefit analysis as a preresquite for any time of regulation for example, yet strangely, in the worlds of incarceration and prosecution, it is absent.

This paper notes:

Despite that leeway, criminal law currently does not use CBA to inform its choices. Criminal law is a massive government social policy intervention on a scale with the largest regulatory agendas and, broadly conceived, it does the same sorts of things regulatory agencies do. For in- stance, it aims to reduce specific types of risks to bodily harm, property loss, and associated psychological-emotional injuries. We have tremendous amounts of data on crime rates, demographic information on crimes and offenders, as well as extensive statistical accounts of, inter alia, arrests, prosecutions, and case dispositions at state, local, and federal levels. What we lack is extensive analysis of the sort we have for federal regulatory projects. We have a woefully thin account of whether the particular approaches we employ for criminal law enforcement cause more harm one in respect than good in another, and little sense such knowledge into criminal law administration.

Instead of methods like cost-benefit analysis, we use sounder methods to assess the quality of criminal justice, such as rhyming and sports puns (“Do the crime, do the time”, “Three strikes and you’re out”).

But despite this lack of existing literature- academic or grey, I want to talk about CBA of prison. I’m pretty sure that given the enormous costs of prison to society and to the imprisoned, imprisonment will only pass a CBA only where there is no alternative. Regardless of whether I’m right or not, I want to commend thoroughgoing utilitarian CBA of marginal incarceration as a vital project for political reformers, effective altruists, humanitarians, and criminologists.

And I want you to feel something about this somewhat abstruse topic. In fact, I want you to be indignant about it.

If we are going to be utilitarian, we need to do it properly

My proposal is to approach CBA of incarceration from a utilitarian perspective. We will be trying to assess whether increasing incarceration by a small amount increases or decreases aggregate preference fulfillment. For this, I propose using cost-benefit analysis altered to measure aggregate preference fulfillment. Personally, I favor an alternative conceptual framework to traditional CBA I call Subjective Wellbeing Effect Evaluation (SWBEE) as outlined in my draft doctoral thesis. However, I will outline my preferred methodology in terms of CBA, because it’s the standard language of these things.

Does approaching this analysis from a utilitarian perspective mean that we must believe in utilitarianism? Not at all. It just means that we’re looking to see if, in total, it helps or harms people more. That’s a good start to a moral conversation but it doesn’t have to be the end- it just seems like, at minimum, a baseline of knowledge we should have in discussing prison.

Once we have established this baseline, opponents of incarceration can introduce further considerations (e.g., the inherent value of society minimizing vengeance, the cruelty of a carceral system that primarily affects those born into the lower class etc.) and supporters of incarceration can introduce their own further considerations (the moral value of punishment, the reduction in the moral weight of prisoners needs due to their wrongdoing etc.) Rights-based arguments, complex considerations from political and legal philosophy etc. can be marshaled- but I suggest gathering information on harms and benefits before we have this broader conversation. This call for a wider discussion after we have the data isn’t just a concession on my part. I believe in a richer form of consequentialism than can be captured by the Cost-Benefit methodology I describe. So I genuinely think we need to have a conversation about broader moral considerations, but let’s get some data on human welfare first.

How will we measure what people want? Largely by how much money they are willing to pay to secure an outcome, or avoid it. However, as is appropriate for this kind of cost-benefit analysis, we will factor in differences in income. A wealthy man might be willing to pay millions to avoid prison, whereas a poor man might only be able to scrape together a pittance. The best available figures, based on hypothetical gambles, suggest that for every one percent a person’s income rises, the value to them of a dollar falls by about 1.2 to 1.5%, so this is the range of figures I recommend using.

It would be wholly inappropriate, for example, to use willingness to pay for bail alone as a measure of desire to avoid incarceration, without considering that many prisoners are credit constrained- as at least one researcher has, bizarrely, done.

Adopting this approach to measuring preferences does not mean that we have to grant human welfare=weighted willingness to pay. We’re using weighted willingness to pay because it’s a reasonably accessible proxy for something that is much richer- human wellbeing. This is just a first pass.

What if you think there are important effects on human welfare that go beyond people’s own desires and preferences? To pick an obvious example, some people don’t really care about preferences, they just care about happiness. Others pick more rarefied qualities- autonomy, dignity, capabilities and so on. I can only reply that people’s preferences are probably correlated with their own welfare- whatever welfare may be.

With difficult-to-conceive situations like being incarcerated, being the victim of a crime etc. we should be extremely reluctant to take people’s assessment of costs and benefits to themselves ex-ante unless there is no other choice. It is likely that people’s estimates of how bad a year in prison for them would be, if they’ve never experienced prison, or how bad a burglary would be, if they’ve never had their house robbed are worthless. Hence we should derive our willingness to pay to avoid misfortune estimates from those who have suffered the misfortune in question. Given the politically contested nature of this topic, we should be careful that subjects aren’t choosing their willingness to pay values strategically, to try and influence the outcome of the analysis.

Perhaps the primary characteristic of a utilitarian analysis is that it involves performing cost-benefit calculations in a way that includes all significant costs and benefits. Simply having done the wrong thing doesn’t exclude you from the calculus. We might exclude costs and benefits from analysis if they seem both likely to be small, and very difficult to calculate, but we exclude no costs or benefits for ‘moral’ reasons.

Objectionable work in this area has been a problem in the past. Cost-benefit Analysis is a well-understood methodology, perhaps its biggest no-no is arbitrarily excluding large costs or benefits for moral reasons. If you think there is a case for excluding certain costs or benefits for moral reasons, you should create two models, one that excludes the cost, and one that doesn’t- so readers can choose for themselves.

Another problem is excluding costs and benefits likely to be large purely for ease of calculation. This can render analyses all but useless. There are good alternatives, viz, to take multiple estimates of the value of hard-to-quantify variables and providing different models based on the valuations as a kind of sensitivity test.

Obiter dicta, I think it is not just bad practice but wickedness to perform a cost-benefit analysis of incarceration, without valuing the loss of liberty and other harms to incarcerated people. It is treating the experience of some of the most vulnerable people in society as worthless. I regard it as morally on par with colonialist intellectuals who whitewashed what their nations were doing overseas. It’s one thing to say that people’s experience of a lack of liberty should be discounted because they’ve done the wrong thing and therefore ‘deserve’ it, but to not even bother assessing what we are discounting before tossing it aside, is, to use something of a cliché, epistemic injustice.

Also, we approach incarceration ‘as is’. We do not consider what the optimal rate of incarceration would be under a hypothetical, better and more humane prison system. Recommending higher incarceration, and then putting in a little caveat that of course you only mean to recommend higher incarceration on the condition of better prisons, is moral cowardice.

Our assessment is to be at the margin, in the first instance- that is to say, assessing the costs and benefits of increasing total incarceration by an arbitrarily small amount. However, it would be good to go back from the margin- to be able to make some guesses about what halving or doubling aggregate incarceration would look like.

The major costs and benefits of incarceration

With all that said, let’s survey the cost and benefits of incarceration. I haven’t got the resources or capability to do an analysis myself, but at least we can start thinking about the types of costs involved.


The non-economic costs to the incarcerated of incarceration at the time of incarceration.

Loss or significant reduction the incarcerated person’s economic productivity- at the time of imprisonment

Loss or significant reduction of the incarcerated person’s economic productivity after release

The marginal costs to the state of an additional person incarcerated

Long-term health impacts of incarceration

The costs of additional crime in prison caused by an additional person incarcerated (e.g., prison violence)

Psychic costs to the incarcerated person’s family and/or partner

Increase in welfare expenditure required after release (e.g. due to increased unemployment)


Crimes outside prison prevented because prisoners cannot commit them

The deterrence effect on non-prisoners of an additional person incarcerated

The psychic satisfaction of victims and their relatives gained through incarceration

Reduction in welfare expenditure during incarceration


The effect of incarceration on the likelihood of committing crimes after the incarcerated person is released (could be positive- through reform or negative through the so-called “university of crime” effect, although most evidence indicates that incarceration increases the likelihood of committing crimes after prison.)

The effect on society of promoting the kind of values incarceration reflects. Incarceration reflects a preference for vengeance over mercy. There are presumably other values based effects. It may also have political economic effects- e.g. creating families and communities invested in incarceration. These factors are intangible- and some might argue they’re simply too difficult to account for in Cost-Benefit Analysis, but given their possible magnitude, my vote is that we should at least try. Some sub-factors here are clearly negative- e.g. the possibility of entrenching and promoting both structural and attitudinal racism through widespread incarceration (at one point, 70% of male Black Americans without a high school diploma had gone to prison at some point before their 30s).

Excluded costs

It is justifiable to exclude some costs that seem likely to be both minor and difficult to calculate. Minor costs that are easy to calculate should be included, and excluding major costs that are difficult to calculate makes the analysis worthless.

Here are some minor and difficult-to-calculate factors I would recommend excluding:

The impacts of incarceration on the incarcerated person’s friends as opposed to family.

Likewise, in some cases, the effects of incarceration on victims and their families may be negative (not all victims support lengthy sentences), but because this seems likely to be a fairly rare and secondary effect, I would suggest discounting it.

Limitations of the current literature

There are studies that perform CBA analysis of incarceration, but unless I’ve missed any, they all leave out major costs and benefits.

One of the worst is: The Incapacitation Effect of Incarceration: Evidence from Several Italian Collective Pardons

In the abstract they confidently state:

“A cost-benefit analysis suggests that Italy’s prison population is below its optimal level.”

Yet they only include the cost to the government of incarceration and the marginal cost of crime (and the marginal cost of crime is calculated rather poorly).

One of the better ones is: The impacts of incarceration on crime an excellent paper. But although this study takes a broad and responsible approach, especially compared to much of the literature, nonetheless the study considers as a benefit of incarceration only the prevention of crimes. It considers as costs only the value of liberty to the incarcerated person, the cost of running the prison, and the reduction of economic activity by the incarcerated person before and after incarceration. There are many other important costs and benefits as we listed above.

The study itself notes in relation to other costs:

“Other impacts on prisoners, families, and communities. These are not counted, because of the challenges of valuing them. The President’s Council of Economic Advisers (2016, pp. 48–51) lists additional consequences of incarceration for inmates, families, and their communities. While a well-developed literature has attempted to put dollar values on crime, these other considerations have garnered little attempt at valuation. Crime is committed and suffered in prison as well as beyond. Families are sundered, with more children missing parents and higher rates of divorce (though some families also benefit from a dangerous loved one being locked up). For released convicts, a felony record can cut off access to public housing and other benefits, and the right to obtain a driver’s license. More generally, mass incarceration can engender deep distrust of government, especially in poor and minority communities.”

I also have concerns about this study’s derivation of a figure for the value of liberty. The study suggests that a QALY is worth 100,000 and that a year in prison reduces the value of that year by 50%. Personally, knowing what little I know about prison, I would prefer to simply ‘skip’ a year of life than spend a year in prison, but this is simply my uninformed opinion. A more systematic study of the welfare of prisoners, especially in willingness to pay terms, is needed.

I’d also like to give a special shoutout to this noble study:

“While these economic analyses of incarceration offer important insights, they suffer from a near-universal flaw: They fail to account for crime that occurs within prisons and jails. Instead, when scholars and policymakers measure the benefits of incarceration, they look only to crime prevented "in society." Similarly, when they measure the costs, they ignore the pains of victimization suffered by inmates and prison staff. This exclusion is significant, as prison crime is rampant, both in relative and absolute terms.”

This shows, I think, that despite what sometimes seems like a mountain of depressing evidence to the contrary, being an economist needn’t mean lacking compassion.

I’d like to end by thumping the table and reiterating that it is disciplinary malpractice to perform a cost-benefit analysis of incarceration without considering what happens when factoring in at least costs to prisoners- even if it’s just a simple figure like “50,000 a year”. It’s easy to do this. If one is unsure whether it is the right thing to do, one can create a model that includes these costs and a model which doesn’t. Thus anyone who doesn’t even bother performing an analysis which includes costs to prisoners, has decided, ahead of time, that prisoners don’t matter- and that this is so obvious that it’s not even worth considering what would follow if they did matter. It’s malpractice for an economist in much the same way as supporting quack treatments in public would be malpractice for a doctor.





More posts like this

Sorted by Click to highlight new comments since: Today at 10:12 AM

Hey, thanks for writting this. You might want to check out the work by Just Impact, and the estimations by Open Philanthropy when they were working on this area.

Thanks! I was aware of Roodman's article (which as I indicate in the text, I regard as one of the better prior attempts) but was not aware of its roots in EA research.

Curated and popular this week
Relevant opportunities