Hide table of contents

This essay was submitted to Open Philanthropy's Cause Exploration Prizes contest.

If you're seeing this in summer 2022, we'll be posting many submissions in a short period. If you want to stop seeing them so often, apply a filter for the appropriate tag!



This piece makes the case for crime reduction as a cause area, as a complement to general criminal justice reform. The argument is structured as follows. The first section lays out the total cost of crime in the United States, chosen as a relatively tractable example in the developed world, and argues that reductions are desirable under any reasonable social welfare function. The second section sets out reasons why crime may be more tractable than suspected, with particular emphasis on the distribution of offences and characteristics of offenders, and potential alterations to the physical environment to deter crime. The final section addresses existing work in this field, and makes the case for Open Philanthropy joining the fray.

Disclaimer: There are clear areas of uncertainty within this analysis, reflecting its status as a first overview of a new cause area. These will be flagged where they occur.


In 2019 - the final year of pre-pandemic data - there were just over 1.2 million violent crimes reported to law enforcement agencies in the USA. This included 16,425 murders, nearly 140,000 rapes, 268,000 robberies, and 821,000 aggravated assaults. In the same year, there were just under 7 million property crimes, consisting of 1.1 million burglaries, 5.1 million instances of larceny-theft, and 722,000 motor-vehicle thefts. This total increases when variation in reporting is taken into account. The police almost always learn about murder; conversely, many rape cases are unreported. 

Crime imposes a substantial welfare burden on citizens. There are the obvious tangible direct costs - deaths and injuries, medical bills, damage or loss of property, lost income through inability to work through crime victimisation or incapacitation as a result of sentencing. There are obvious intangible costs; anticipation, worry, fear, suffering, pain, and anguish. There are defensive expenditures at societal, individual, and business levels: spending on things like police forces, jails, security cameras, fences, locks, reinforced doors, or security guards to reduce the probability of becoming a victim of crime.

There are also less obvious costs through avoidance behaviour and opportunities forgone. People tend to move away from high crime areas, many of which are in cities. Levitt and Cullen found this reduces urban density; people forgo the productivity and welfare gains of being in dense city centres and move to suburbia, increasing their carbon footprints in the process. Crime erodes social capital and trust in institutions, making for less effective governance. And ultimately, people do not do things they would otherwise do; a Samsung advert showing a woman jogging alone at 2am with her earbuds in was called “unrealistic” and “insensitive”, but absent crime and the fear of crime this would be a totally reasonable activity.

The number of people who might wish to literally go running at 2am is probably relatively small. The number of people who have experienced anxiety at the thought of walking home alone, altered their plans and indeed the location where they live to avoid doing so, taken circuitous routes to avoid bad neighbourhoods, made efforts to keep nice possessions from public display, or felt stress and fear when approached at night is likely to be substantially higher.

One approach to estimating the total cost of crime is to simply ask people what they would be willing to pay to prevent it. Aggregated across the population, this method provides an estimate of willingness to pay (WTP) to avert crime, and in turn of the costs of crime to its victims.

A 2004 study estimating willingness-to-pay for crime prevention found that typical US households would be willing to pay somewhere between $100 and $150 per year ($156.86 and £235.29 in 2022 dollars) to reduce specific crimes by 10% in their communities. Extrapolated collectively, this indicated a willingness to spend $25,000 ($39,215) to prevent a burglary, $70,000 ($109,802) to prevent serious assaults, $232,000 ($363,918) to prevent armed robberies, $237,000 ($371,761) to prevent rape or sexual assaults, and $9.7 million ($15.2m) to prevent a murder. 

Applying the inflation adjusted figures for these five offences - using disaggregated figures for armed robbery and noting that ‘serious assaults’ defined in the paper line up with the FBI category of ‘aggravated assault’ - gives an annual cost in the region of $481 billion, or $538 billion when adjusted for population growth. As an estimate of victim costs - so not including the cost of the criminal justice system, or the opportunity cost of jail sentences - this is reasonably in line with other estimates. The total aggregate burden has previously been estimated at 3-4 times victim costs, implying an overall burden of these crimes to the US in the region of $1.6-2.2 trillion - or 7-11% of GDP. This is comparable to estimates of the benefits from land use reform, an existing Open Philanthropy grantmaking area. 

This estimate lines up with a 2021 attempt to provide total cost estimates for personal and property crimes in 2017. This study combined estimates of victim costs, law and order (police investigations and incarcerations), but excluded the cost of avoidance behaviours and defensive expenditures. Rather than using crimes reported to the police, the paper attempted to estimate total incidence for a more accurate summary of the cost of crime. It’s worth noting that this paper did not consider the redistributive effects of crime, treating losses through larceny-theft or burglary as pure costs. 

A more comprehensive welfare approach ought to take the redistributive effects of property related crime into account. However, it seems likely that those benefits are outweighed by the costs of crime. Each small transfer of wealth through property crime in general causes far more damage in financial terms to actual and potential victims of such crime, given the additional distress and avoidance costs.

In addition to being a ‘leaky’ form of transfer it is likely to be a highly inefficient one; wealthier individuals and households can afford defensive expenditures which direct property crime elsewhere, and selection by property prices means that low income criminals are likely to have other low income people as their immediate neighbours and most obvious targets. This is backed by the observation that the burden of crime falls disproportionately on low income households.

Returning to the efficiency of the transfer mechanism, it is highly likely that if property crime were reduced to zero, social welfare would increase. In this circumstance, the minor redistributive benefits of such crimes could be achieved much more efficiently through taxation and direct payments to those most in need. For violent crimes without transfers of wealth, there are no redistributive benefits at all except insofar as they may deter or eliminate individuals who were otherwise harming those on low incomes. These seem likely to constitute only a small fraction of violent crime.

Major contributors to the cost of crime included the loss of 10.9 million quality-adjusted life years from violence and murder, 3% of GDP spent on emergency services, victim services, the judicial system, and punishment, and of course property losses. The headline estimate for the total cost of crime was $2.6 trillion, of which 85% was attributed to violent crime. Of the headline figure, $1.95 trillion came in the form of pain, suffering, and lost quality of life, in line with the victim costs estimate obtained through the willingness to pay method above. This would correspond to a cost of 9.3% of US GDP.

There are also costs which are rarely considered; Ben Southwood notes that crime tends to increase urban sprawl, with individuals moving out of the city centre. His back of the envelope calculation based on prior research has one murder in New York equalling 600 people moving to the suburbs. Suburban New Yorkers tend to produce an additional 50 tonnes of CO2 annually relative to those in the city centre. With 600 people producing an additional 50 tonnes of CO2 per year, and a social cost of carbon around $50 per ton, the climate change externality generated by sprawl resulting from a single murder could be in the region of $45m. 

This is one example of the wider costs of crime. Other losses are likely to include reduced productivity through sprawl and income losses to individuals and businesses resulting from decisions to avoid investment in high crime areas. The combination of these broader effects and the difficulties in existing methodologies means that there are clear uncertainties in putting a precise figure on the cost of crime. Estimates vary by methodology, but consistently show that crime is highly costly in welfare terms, with lower end estimates reaching figures between 2% and 6% of US GDP. My contention is that decreasing crime within the United States is clearly desirable under almost any reasonable social welfare function.


My basis for believing that it is possible to decrease crime in the United States is the replication of certain basic facts across developed countries.

The distribution of criminal behaviour is highly skewed, with a very small proportion of the population accounting for a significant share of total crime. Incapacitation of this group through imprisonment results in a significant reduction in societal exposure to criminal activity.

(i) A study of the Swedish population born between 1958 and 1980 found that 3.9% of the cohort was convicted of a violent crime. A group of persistent offenders accounting for 1% of the total population accounted for 63.2% of all convictions. These offenders were relatively likely to commit offences early, use drugs, and display personality disorders. 

(ii) A paper comparing convictions from a UK longitudinal study and a US dataset of self-reported delinquency found that in both cases, criminal behaviour was well described by a power law. 

(iii) Some 70% of custodial sentences in England and Wales are handed out to offenders with at least seven previous convictions or cautions; 50% to those with at least 15.

(iv) Roughly 73% of 2016 US federal offenders had previous convictions. Among this group, the average number of convictions was ~6. Around 39.5% of these offenders had prior violent offences.

Offenders often possess characteristics which require greater support both in prison and in general society, and which may make standard models of rational choice a poor fit for their decision-making processes.

(i) A study of the Swedish population linking measures of time preference at 13 to administrative data found a strong relationship between high discount rates and criminal involvement.

(ii) A study of individuals prosecuted for murder, aggravated robbery, and burglary in the state of New South Wales found that “the great majority” of prisoners displayed high discount rates.

(iii) A study using a quasi-experiment based on a collective pardon in Italy found that prisoners displayed “significant discounting behaviour”

(iv) Some 53,109 adults in British prisons were being treated for alcohol or drug abuse issues in 2017-18, out of a total prison population of around 82,000.

(vi) The National Institute on Drug Abuse states that ”85% of the [US] prison population has an active substance use disorder or were incarcerated for a crime involving drugs or drug use”

(vi) The UK does not have up-to-date figures on mental health in prisons, but the Institute of Psychiatry estimated over half of prisoners have common mental disorders - PTSD, anxiety, or depression - while another 15% have specialist mental health needs, and 2% acute and serious problems.

(vii) The APA estimates that “at least half of prisoners have some mental health concerns”, and 10-25% of US prisoners serious mental illnesses such as schizophrenia. 

The deterrence effect of jail time appears to be relatively weak, in part due to these commonalities.

(i) The classical economic model of criminal behaviour can be represented by a simple cost-benefit decision for the agent: criminal behaviour is profitable if

            Benefit(B,R) - (P)(Punishment(B,S)) - O > 0

Where P refers to the probability of being ‘caught’, O is the ‘outside option’ - the utility gained through the non-criminal action - and the benefit and punishment are functions transforming the individual discount rate (B), immediate return to crime (R), and severity of punishment (S) into appropriately discounted flows of utility. Generally speaking, we expect the benefit of crime to accrue primarily in the short term, and the costs over a longer period. 

(ii) Under this framework, when the discount rate is high the majority of any deterrent effect results from the initial period of punishment.

(iii) Similarly, when the outside option - the economic return to non-criminal activity - is low value the return to crime may be disproportionately high.

(iv) The characteristics of the criminal population listed in the previous section indicate that both of the preceding statements are likely to be true. In addition, it is not clear whether classical rational behaviour is a good approximation to the behaviour of individuals suffering from mental illness or substance abuse issues. 

(v) These suspicions are born out by David Roodman’s excellent analysis of deterrence, which notes that punishments designed under the presumption that increases in the certainty of punishment better dissuades criminal behaviour than the threat of greater sanction did not produce the desired effects.

(vi) Roodman also notes that US studies on increased severity have shown little in the way of strong deterrence effects.

(vii) The existence of significant procedural delays in trials and sentencing may exacerbate this, pushing the eventual punishment further into the discounting window.

Conversely, the deterrence effect of visible policing is relatively high

(i) A study using variation caused by the terror alert level set by the Department of Homeland Security found that crime levels decrease significantly when police presence is increased. 

(ii) An Argentine study using variation in policing following terror attacks found a strong deterrent effect of observable police in the immediate vicinity.

(iii) Evidence from the United Kingdom shows that victim-reported crime drops with higher numbers of police, while police recorded crime may increase - indicating a deterrent effect alongside improved reporting.

(iv) Variation in police spending in sync with electoral cycles shows that increases in police force numbers have a strong effect on violent crime.

(v) A review of the literature on hotspot policing found that focusing police attention on high crime areas caused an appreciable decrease in crime, with some evidence that high visibility patrols were more effective than normal patterns of behaviour. 

Local interventions can also reduce crime by altering the physical environment

(i) Just as visible policing deters crime, alterations to the local environment can also dissuade criminal activity.
(ii) An analysis of the impact of street lighting on crime in the UK found that targeted interventions tended to reduce crime, and also made residents less fearful, potentially reducing costly behavioural alteration. 

(iii) A broader review backs this conclusion, finding that street lighting significantly reduces crime.

(iv) A 2009 review of the literature on the installation of CCTV cameras found that it resulted in a small but significant decrease in criminal activity, particularly in car parks. 

(v) A similar review conducted in 2019 replicated these findings, with the additional note that active monitoring resulted in a larger reduction than passive systems.

(vi) One study suggests that the bulk of the benefit is deterrence of automobile theft.

The incapacitation effect of imprisonment is relatively strong

(i) Almost by definition, an individual in prison is highly unlikely to commit a crime against a member of the general public. 

(ii) Set against this, they are also more likely to commit crimes within prison. Greater monitoring of behaviour must be set against the lack of possibility for effective further punishment (longer sentences may not act as sufficient deterrent), and an environment which is likely to be criminogenic (see below). 

(iii) The cost of crime within prison is likely to be lower to society than the cost of crime without; it is not committed against random members of the population, results in less defensive expenditure or avoidance behaviour, and is focused on a smaller group of individuals in an environment designed to reduce criminal activity. 

(iv) Roodman’s review of incarceration notes that multiple studies find strong evidence that incapacitation works to prevent crime. This should be expected.

Certain currently common forms of sentencing are likely to increase reoffending 

(i) Turning again to Roodman’s review, there is a fairly solid bed of evidence that US prisons are criminogenic. This is not sufficient to outweigh the benefits of incapacitating serious offenders, but is cause for reappraisal for marginal offenders.

(ii) Harsher prison conditions are associated with greater recidivism. This is unsurprising, as brutalising conditions are unlikely to provide effective conditions for self-improvement.
(iii) A study of Italian jails shows that worse prison conditions does not reduce future criminal activity, instead driving increases in future criminal behaviour.

(iv) A study of US jails produced a similar conclusion, with a suggestion that harsher conditions increased crime post-release.

(v) A UK Ministry of Justice report concluded that short-duration sentences if anything increased reoffending relative to court orders. 

(vi) Norwegian prisons consistently display low rates of recidivism, and are marked by their civilised treatment of prisoners.


Combining these observations suggests the following agenda for work:


I have yet to come across research which seriously tests the tenets of the ‘high probability short sentence’ model of sentencing against the observed characteristics of the prison population and the known effects of prison. 

This suggests that a fruitful avenue of research - qualitative and quantitative - would be in understanding the degree to which prisoners actually respond to high probability short-term sentences, and assessing the potential criminogenic effects of these sanctions. Determining the characteristics on the upper part of the power law distribution for criminal behaviour should also be a matter of priority, both in order to better identify high probability reoffenders and to understand their susceptibility to different forms of deterrence.

More importantly, evidence on the burden of different crimes should be collated and used to push for resource allocation according to actual costs. With the evidence base we have to hand, it appears that effectively prosecuting rape cases may be a very high return activity. Under each study rape accounts for a huge share of the cost of crime - 56% in the 2021 dataset - and also one of the highest ‘per crime’ costs. While the general deterrent effect of prison sentences may be low, current clearance rates for the overall number of rape cases are extremely low. Addressing the backlog of forensic kits accrued in US states could be an extremely effective use of funds if it results in sufficient convictions to dent behaviour.

However, further research is needed to establish both the improved deterrent effect of detection, the best way of improving it, and indeed whether there are more effective areas to direct resources for the aim of reducing violent crime. 

Generally, one takeaway from the process of pulling together this note has been the lack of joined-up policy thinking in this space. There are admirable policy evaluations available on the general efficacy of individual policing approaches, or on small scale programmes for prison reforms. There are very few comprehensive cost-benefit analyses identifying the best use of funds in targeting crimes to improve overall welfare, and it is very possible that such an exercise would identify low-hanging fruit.

Reconsidering the role of prisons

Incapacitation is perhaps our best tool for reducing crime through the legal system, with the vast majority of the effect results from locking up the small percentage of repeat offenders. Given the relative burden of property and violent crime, it is clear that the latter should be prioritised. A simple suggestion would be research and advocacy centred around stiffer sentences for violent crime, alongside trying to reduce the number of people in prison for minor property crimes, or placed in on ‘X strikes’ laws that do not sufficiently account for severity of offending. 

Other organisations are already focused on reducing the brutalising element in prison treatment. This is good, as it does not seem warranted to expect any rehabilitative effect while putting human beings into a deeply traumatising environment. Lending low cost support may be an effective use of resources on occasion. 

Given the characteristics of the prison population with regards to time preference, drug use, and mental illness, expecting certainty of punishment to create deterrent effects may be optimistic. Similarly, standard prison experiences are likely to be stressful and cruel. I strongly suspect that a model closer to ‘assisted living’ - in line with the Norwegian approach - would pay strong dividends. However, I do not have sufficient evidence to state this with confidence. Research on this may prove a highly effective use of funds. 

The use of short-term sentences may not be an optimal approach for low-level offenders, with the potential criminogenic effects of their sentences possibly outweighing the benefits from short-term incapacitation.

Learn from best practice in policing 

Evidence suggests that putting police on the beat in visible locations has a strong effect in reducing crime. Such patrols - relatively common in European countries - may also help to build community relationships and trust through the assignment of regular beats, allowing repeated contact with the inhabitants of an area. 

More generally, there is a broad body of evidence available from the National Institute of Justice listing evaluations in policing practice. An analysis of these evaluations and provision of cost-benefit assessments may be useful in attempting to sway police forces towards tactics which actually work.

Altering the local environment 

The evidence on CCTV cameras and street lighting suggests that targeted schemes for installation could prove to be a cost-effective approach for reducing criminal activity. Directly sponsoring the installation of street lighting or surveillance cameras in high crime areas likely to be susceptible to these interventions could prove an effective use of funds.

Upstream intervention may prevent some criminality

The share of criminals with learning disabilities, mental health issues, or drug addiction problems strongly suggests that while not responsive to deterrence, they may well be responsive to upstream interventions designed to provide them with necessary assistance in day to day life. 

In particular, investment in drug addiction programmes and mental health interventions is likely to produce strong effects in crime reduction at a cost substantially lower than that of imprisonment. 


Combining these observations, a tentative approach to reducing criminal behaviour and incarceration would look something like this: reduce short sentences in brutalising prisons in favour of greater use of suspended sentence or punishments akin to UK community orders, scale jail terms significantly for repeat offenders to remove them from circulation, and place an emphasis on humane treatment within prison environments. Policing should emphasise visibility and contact with communities as a central point of practice. 

More generally, a research programme which collates evidence across all strands may prove highly useful. Across points (2) and (3), a major standout from my brief research was the lack of efficient policy advocacy.


US criminal justice reform is not a classic neglected problem. Think tanks, prisoner advocacy organisations, and politicians have invested significant effort in identifying and promoting strategies for the reduction of prison populations. Open Philanthropy itself contributed $138.8 million to the area over a 6 year period, including a $50 million grant to launching Just Impact as an organisation advocating for “safely reducing incarceration”.

However, most of these organisations focus on racial disparities, reducing capital punishment, providing legal assistance, or decarceration. In comparison, few focus on crime prevention. The relative lack of interest in this topic may be due to the assumption that the large and well-funded organisations dedicated to crime prevention have already optimised their approaches; that police forces adopt the best techniques, sentencing is based on risk reduction, and jails seek to limit reoffending. Unfortunately - perhaps unsurprisingly given what we know about government behaviour in other fields - this does not appear to be true.

A brief and non-exhaustive survey of institutions interested in crime prevention policy is given below. Those identified in the initial Open Philanthropy write-up of Criminal Justice Reform are excluded unless their remit has changed to include crime prevention.

The National Institute of Justice

The research and evaluation arm of the Department of Justice, the NIJ funds research to develop evidence for improving the justice system based on the priorities of the White House and Congress, with a particular emphasis on the development of technology. This research is disseminated to policymakers, but there is no lobbying effort to ensure uptake. Indeed, using NIJ funding to push for uptake is strictly forbidden. In addition, the NIJ’s status within the DoJ, administration-led prioritisation, and politically appointed head places it under political constraints that may not permit optimal policy suggestions. Total funding for the NIJ in combination with the Bureau of Justice Statistics varies, but is generally under $100m in a given year.

The US Department of Justice

The Department of Justice, in addition to its core work, offers a wide range of grants supporting the adoption of trialled strategies, including hot-spot policing, funding officers, resource provision for sexual assault kit processing, and training programmes. Uptake of these grants is dependent on applications, and it is very possible that this selection filter will mean that these grants are allocated to organisations that are already proactively seeking improvement. 

The Pew Public Safety Performance Project

The PPSPP is largely focusing on reducing prison, probation, and parole populations, with a particular emphasis on racial disparities. At least some work is spent on “strategies to ensure that people accused of crimes appear in court”. The total budget, and its breakdown, is not readily available. 

A large number of smaller and local trusts focus on particular strands of crime - the Isabel Allende Foundation, Channel Foundation, and AT&T foundation on violence against women and girls, the Allstate Foundation and Alphawood Foundation, on domestic violence, the David Bohnett Foundation on gun violence, and so on. Generalist organisations looking to improve policing and sentencing practices appear relatively rare.

This may be because crime is more salient to the poor than to high socioeconomic status people, who self-sort into expensive low-crime areas. As a result of this low salience, the cause area may be relatively easy to overlook as a cause of significant welfare losses relative to concerns more immediately applicable.

Combined with the points in the section on tractability, it seems plausible that there is a gap in the market for work directed at crime prevention, that there would be policy wins available, and that this could be achieved at reasonable cost. 


Briefly, crime imposes very large societal costs which are highly unevenly distributed. A reasonable estimate of the direct costs in terms of harm and crime prevention is in the region of 7-11% of US GDP. Once broader costs such as urban sprawl and foregone investment are factored in, the cost of crime is likely to be significantly higher. 

Current criminal justice practices do not seem to be optimised for crime reduction. The uneven distribution of criminal activity - highly concentrated among a handful of repeat offenders - and less effective nature of short centres suggest that a prison policy focused around incapacitation of the most serious offenders may produce large welfare gained, particularly if implemented alongside investments in elements of the physical environment and policing practices to deter crime in particular areas. 

While there are funding organisations working on criminal justice reform, there seem to be relatively few focusing on crime reduction specifically. There is likely to be scope for improvement in practices by evening quality of policing and sentencing.

Sorted by Click to highlight new comments since:

Tremendous stuff. One thing I would add is that higher crime, by its very nature, provokes worse policing. This happens through some obvious mechanisms (more brutalized, overworked, and trigger-happy cops), but also through some more subtle mechanisms. As crime reduces urban densities, police numbers relative to the area they cover start to fall, and cops become confined to their cars, since this is the only way they can cover their expanding territories. This kills off foot patrolling, one of the best and most proven anti-crime interventions, with a pedigree hearkening back to the Victorians. Some London police stations still have 19th-century maps on their wall of their territory divided up into individual officers' beats, which were often no more than a few streets of super-densely populated space. Solve the crime problem, and you may well find that America's policing problem fixes itself to a surprising degree.

Can we really say that crime reduces urban densities? Wouldn't the socio-economics come first, followed by crime as a multiplier on the impact on destruction of a community and its health.

Separately to the above, I think you make a really good point about foot patrolling. There is a link here to criminology and environmental psychology of why and where crime occurs to consider.

I'll try not to butcher the history too much as I am an engineer/designer but I understand that the prevailing view was that  crime in inner cities was seen as a natural behaviour of the poorer masses as a result of their innate psychology. Researchers gradually discovered that its really environment shapes much of make drives a crime to occur (noting the variety of crime that can occur).

So to your point on patrolling large areas, then getting large-scale Urban Planning is absolutely key here. Sprawl plagues North America and many developing countries and results in what you said. Good Urban Planning would facilitate better policing as well as a range of other factors that improve well-being, health, and productivity.

Side story - I recently visited Limerick, Ireland for a project. The city has previously been plagued by drug crime and gang-land family violence. The police there stated that it was a combination of hard-line policing on violence AND community policing (building relationship with the wider community) which changed the situation. So I'm sure there are times when hard-line and tough policing is needed.

So all-in-all, does the solution start with urban planning?

I enjoyed reading this informative and well-researched post, but honestly one of my takeaways was that YIMBY (or "land use reform" as openphil calls it) seems like a marginally better cause area than crime reduction:

  • You say that the total cost of crime to society is similar to the total cost of land-use restrictions.  But the solutions to land-use restrictions seem pretty obvious -- to a first approximation, just let people build more houses by having higher levels of government squash local opposition to projects.  (Then if you want to get fancy, you can address flaws in the first-order solution with patches like congestion pricing, Harberger / Land Value taxation, etc.)  Meanwhile, the solutions to crime seem more nebulous: yes, I agree we're probably underpoliced and overimprisoned, but cutting against this is the "incapacitation" effect you mention of locking up the small percentage of repeat offenders.
  • It's hard to say exactly why, but  to me, these crime-reduction ideas seem like they'd probably run into to diminishing returns.  Even if you managed to encourage optimal policing policy across hundreds of local counties/cities,  it's hard to picture an amazing total victory that turns America into a nordic-style low-crime utopia.  I would expect something more like a 30% reduction in crime, closing some but not all of the gap between the USA and Europe.  To me, total victory in the YIMBY arena seems more plausible (although again, idk why, just a feeling).
  • (For me to believe that a low-crime utopia is possible in the USA -- matching or besting countries like Japan -- I feel like it would have to involve some totally wild mechanism of the sort that Robin Hanson would dream up, or possibly some kind of intensive social-credit system.)
  • YIMBY problems are arguably greater in scale: prohibitively expensive urban real-estate is a plague across the entire developed world -- places like the UK, Canada, and New Zealand have it even worse than the USA.  Meanwhile, most discussion that I see about crime focuses on how America has anomalously high crime rates and should aspire to be more like low-crime Europe.  If crime is dragging down America's GDP by ~10%, but land-use is dragging down the entire developed world's GDP by ~10%, then land-use would be the bigger overall issue.  (On the other hand, you could say this is an argument for the tractability of crime reduction -- just be more like Europe!  Versus the global nature of land-use problems maybe indicates that, as easy as it seems to just let people build more housing, perhaps there is some universal force of creeping vetocracy and stakeholder management that will eventually doom any attempts to go full YIMBY.)

All that said, even here on the EA Forum which is all about prioritizing causes, it feels pretty silly for me to be doing a compare/contrast between YIMBY and crime-reduction policies when I am in favor of both, and the two feel so synergistic!  (You mentioned how one of the major costs of crime is that it causes people to feel unsafe in cities and move to the suburbs.  And in my own life, I personally care about both affording a home and living in a safe, high-trust neighborhood.)  I am not trying to start a debate, rather I suppose I am asking, "As a guy who is excited about institutional experimentation and big potential improvements to civilization's status quo, what are the most interesting/hopeful/exciting ideas about crime-reduction policy?"

Some other questions, if you will forgive the stream-of-consciousness style of this long comment:

  • Some people say that you could reduce crime a lot by looking at "pipeline"-style interventions very far upstream of the crimes themselves.  Do any of these seem to you like they'd plausibly have big effects on crime in the USA?  I've heard people mention things like:
    • investing more in schools (and generally "investing in communities", eg libraries)
    • running a sufficiently hot economy that unemployment is kept low
    • removing pollutants like lead from the environment
    • something about cultural values and fatherhood, idk
  • My impression is that drug use explains some (but not all) of the western hemisphere's high crime rates compared to Europe and East Asia.  So any approaches to reduce drug use (or somehow make drug use less criminogenic / change what drugs are popular / etc) would be very valuable.  But (per this Slow Boring article), this seems like a very hard problem.  Do you see any promising approaches here?

These two pieces are of interest. Lots of similar conclusions.

These are excellent, thank you!

Does anyone have a sense of the costs of inteventions here per say, city?

Can just be an upper bound

Note: You might think you don't know  this number, but everyone does. We all know that such interventions are going to 95% cost more than 1$ and 95% cost less than World GDP. But I think some of you will have a sense that the 5% lower bound is quite a lot higher and the higher bound is quite a lot lower.  It's these intuitions I'm interested in.

My starting 30s  5 min 30m  1h[1] calculation  squiggle model, you can run here. It gives an invervention 6x as effective as GiveDirectyly with a 90% CI [.8, 260] for the first city.

 Happy to hear: "your top number seems very high" or "you are missing this important detail in your model, it is X".

If you want to know more about this model, if something isn't clear or your interested in this process, DM me[2]. If this gets 30 karma, I'll write/record a longer explanation.

Multiple of GiveDirectly efficiency


uSGDPPerCapita = 64k
totalCrimeGDPImprovmement = .07 to .11
easiestCityInUSToImproveCrimePop = 500k to 10M
gDPperCapitaInThatCityComparedToAverage = .3 to 2
costToFindCityAndIntervention = 100k to 10M
costToImplementIntervention = 200k to 1M + (1 to 100)*easiestCityInUSToImproveCrimePop
interventionAnyEffect = .2
inverventionCrimeReduction = .8 to 1.05 //lower numbers better, it might make it worse
valueOfCrimeReducedInThatCityPerCapita = interventionAnyEffect * inverventionCrimeReduction * totalCrimeGDPImprovmement * uSGDPPerCapita * gDPperCapitaInThatCityComparedToAverage
valueOfCrimeReduction = valueOfCrimeReducedInThatCityPerCapita * easiestCityInUSToImproveCrimePop
costOfInvervention = costToFindCityAndIntervention + costToImplementIntervention
uSGDPComparedToGiveDirectlyMultiplier = .01 to 1 //US consumption is worse than developing nations, right?
benefitMultipleOfGiveDirectly = (valueOfCrimeReduction-costOfInvervention)*uSGDPComparedToGiveDirectlyMultiplier/costOfInvervention
model = benefitMultipleOfGiveDirectly


  1. ^

    I've heard of nerdsniping, this once nuked me from orbit

  2. ^

    I'm trying to crowdsources some models for both this and the criticicm contest. I aim to crowdsource models of intervenions, average them, then rank all interventions by them, then write a listicle. 

    I have built the tech, but it still very eary stage and user-unfriendly. If actively want to help (happy to pay a bit if that makes the difference), have time in the next two weeks and are competent at google docs, DM me on here.  Even better if you like modelling, squiggle, or google apps script. 

Thanks very much for writing this great article, I think it's the best EA treatment of the subject I've seen.

 One solution area worth adding to this potentially is Crime Prevention Through Environmental Design (CPTED) (I'm biased because its part of my day-job). 

CPTED is a theory of discouraging crime through architecture and urban planning. Think consideration for circulation routes, lighting, 'eyes on the street', and encouraging positive use of space. 

Embedding crime prevention on a project-by-project basis would have a highly scalable impact.  In the UK, Aus, and in Europe, some local and city authorities have local urban plans which usually  require designs to have a positive impact in reducing crime and increasing community safety through consideration for security. 

So I wonder if expanding this through planning policy and connecting it further into social treatments for crime could have a large impact. 

While will not treat the direct economic and social causes of crime, it would:

  1. Provide the environment to help stop crime propagating.
  2. Support well-designed cities, indirectly improving mental, physical, and social health.
  3. Supporting urban environments more functional and more economically productive.

Further to this, one other important area is Fear of Crime.  Which designing environments in a specific way can directly reduce.

This is really interesting, thank you Darren! I mentioned it briefly in the context of lighting and cameras but this is really promising. Is there a set of sources you would recommend reading more in?

My apologies for missing this! Its worth noting that lighting alone doesn't make a space safer (in feeling or reality), it is a mixture of factors with other people (bystanders or allies) being around.  My lighting designer colleagues did some interesting work on this here with Plan International and Monash University

The CPTED page on wiki gives a good starter but if you want to read more, you can have a look at this academic paper.

This is a fascinating piece. I would be keen to read more on various aspects and see what research could be carried out. If the GDP impact estimates are correct, crime might cause even more loss of welfare in the US than is caused by housing shortages.

If you think this is a great entry you can bet that it will win, using fake money, here. These exist for all entries: 

You write:

(iv) Roodman’s review of incarceration notes that multiple studies find strong evidence that incapacitation works to prevent crime.

But that's tautological or close to tautological given Roodman's definition of incapacitation (p. 10, Sec. 2.2.2):

Because prison crime is so rarely studied, and because crime outside of prison has particular political salience, “incapacitation” in this review will refer to crime outside of prison.

(I say "close to tautological"  because some crimes committed by someone in prison, e.g. mail fraud, might nonetheless be considered crimes outside of prison.)

If we consider total crime, not just reported crime or just crime outside of prison, then the quoted incapacitation figures will be overstated.  Indeed, it's quite plausible that the full incapacitation effect is negative on some margins.  So for instance consider someone in jail for a relatively short period of a week or a month.  I would guess their risk of being a victim of crime (theft, assault, rape)  is typically higher than the risk of them committing or being a victim of a crime outside of jail over the same period.

Given the line you quote it's totally reasonable for you to think this, but I think Roodman's summary is actually very misleading. If you look at the tables where the actual calculations take place, you can see negative coefficients for incapacitation for both Rape and Aggravated Assault: he assumes imprisonment increases these crimes during the period of 'incapacitation'! If we ignored crimes among felons, these figures would be positive, and increased incarceration would actually look more desirable on the current margin than his report concludes:

Leaving Roodman's report to one side for the moment, there is an important empirical (non-tautological) insight in the incapacitation benefits of prison:  crimes are very concentrated among a small segment of the population, so you can imprison 1% of people and catch much more than 1% of all crimes. 

Thanks for the reply.  I agree that's a natural tentative interpretation of Table 26, taken in isolation.  But note that table doesn't give any indication of confidence intervals for the relevant column.   

Have a look at Table 11 (below).   We see the same numbers (up to rounding) in the final row,  with 90% confidence intervals.  Note the  negative point estimates for rape and assault are very close to the center of their confidence intervals and thus not distinguishable from zero.  Basically, there was a natural experiment in which California reduced the prisoner population and for those two categories, on a relative basis to (most of) the rest of the country, crime happened to decrease very slightly, but only to an extent well within the statistical noise.  (In fact of the seven categories, only motor vehicle theft is significant -- and barely so -- at the p=0.1 level and none is significant at the conventional p=0.05 level.)

Note the crime numbers being used here are inferred from official crimes reported, rescaled using national estimates of the reporting rate (which helps to put e.g. murder and larceny on the same footing, despite their very different reporting rates).  Since only a tiny fraction of crimes in prison are reported (at least that's my sense), that means that crimes in prisons and jails are essentially being ignored (as Roodman states in his definition of incapacitation I initially cited).

The bottom line as I see it:  If someone wants to do a CEA analysis of an intervention in this space, they should think carefully about the incapacitation term as sources (Roodman and I suspect the underlying literature) will tend to exclude crimes in prisons and jails.  In fairness, it's likely hard to get good estimates, but things don't fail to be real because they are hard to estimate.


I don't think that's correct. There's a recent study showing a reduction in mortality for prisoners that are incarcerated relative to a control group of people who were not incarcerated. It appears that most of the reduction in mortality is from reduction in overdose death risk but also a reduction in homicide. The implication is that people in prisons are safer than they would be outside. Probably most of the data showing otherwise comes from poorly done studies that don't use appropriate control groups, EG they may use the general population as control group relative to prisoners.


What about ankle bracelet surveillance as an alternative to incarceration? More humane, lower cost and appears to reduce recidivism as well (vs. prison). I'm bullish, but eager to hear opposing points of view. Including possible effects on e.g. risk of totalitarianism. 

1. Can Electronic Monitoring Reduce Reoffending? by Williams and Weatherburn

Abstract: We evaluate electronic monitoring as an alternative to prison for nonviolent offenses. Leveraging plausibly exogenous variation in sentencing outcomes generated by quasi-random assignment of judges, we find electronic monitoring reduces reoffending at both extensive and intensive margins. Compared with prison, electronic monitoring is estimated to reduce the probability of reoffending by 22 percentage points five years after sentencing and by 11 percentage points ten years after sentencing, with the cumulative number of offenses reduced by 40% ten years after sentencing. These results demonstrate that electronic monitoring has sustained crime-reducing effects.

2. Criminal Recidivism after Prison and Electronic Monitoring by Di Tella & Schargrodsky


Abstract: We study criminal recidivism in Argentina by focusing on the rearrest rates of two groups: individuals released from prison and individuals released from electronic monitoring. Detainees are randomly assigned to judges, and ideological differences across judges translate into large differences in the allocation of electronic monitoring to an otherwise similar population. Using these peculiarities of the Argentine setting, we argue that there is a large, negative causal effect on criminal recidivism of treating individuals with electronic monitoring relative to prison.

Rob Wiblin inspired twitter discussion.

Great work! Would be even better if it also touched demographics and immigration

Curated and popular this week
Relevant opportunities