What Economists Found When They Counted Police Officers and Crimes

Economists cannot agree on whether the economy will improve next quarter, whether the minimum wage costs jobs, or whether the coffee at the faculty club is a fair price. When a result survives three different research designs, run by different scholars in different decades on different continents, it deserves a seat at the table. It has earned that seat by getting past people who spend their careers looking for the flaw.

The result in question is that putting more police on the street reduces crime. That sounds like something your grandmother could have told you, and in fact she could have. But for decades it was surprisingly hard to prove, and the difficulty is instructive. It is also worth hearing for conservatives who have watched a political fashion treat the proposition as naive. The evidence says it is not.

Why it was hard to prove

State the problem plainly. Suppose you collect data on every American city and plot police per capita against crime. You will find that cities with more police have more crime. A naive reader concludes that police cause crime. That conclusion is absurd, and the reason it is absurd is the heart of the problem.

The data are tangled by reverse causation: cities with high crime hire more police, so crime drives staffing, not the other way around. Meanwhile any benefit of police on crime is hiding behind that relationship, canceled out by it. Economists call this the identification problem. To measure whether police reduce crime, you need a source of variation in police numbers that has nothing to do with the crime rate. You need, in other words, a natural experiment: something that put more police in one place than another for reasons unrelated to how much crime was there.

The three studies below found three such experiments. Each is clever. None is perfect. Together, they are much stronger than any one alone.

Study one: election-year hiring (Levitt, 1997)

The economist Steven Levitt published a paper in 1997 in the American Economic Review titled "Using Electoral Cycles in Police Hiring to Estimate the Effect of Police on Crime." His observation was about politics. Mayors and governors like to be seen adding police before an election. If hiring rises in election years in a way that is driven by the political calendar and not by the crime rate, then the election cycle supplies the variation economists need. Hiring that comes from a campaign promise has no relationship to the crime rate that might follow.

Levitt used data on large American cities and found that additional police reduced violent crime and, to a lesser degree, property crime.

The paper's later history is a useful lesson in how science works. The economist Justin McCrary later found an error in the original analysis, a programming mistake in how the data were handled, and published a comment in the same journal in 2002 showing that it changed the results. Levitt acknowledged the error and published a reply that used a different instrument, the number of firefighters in a city, which, like police, tends to rise in election years but has no direct effect on crime. With that approach he again argued that police reduce crime. Later researchers have questioned whether election timing and firefighter hiring are strong enough instruments to settle the matter. The record is a mixed one, and I would be misleading you if I presented the 1997 numbers as the last word. But the correction itself is a reason for confidence in the field: the error was found by a rival, admitted, and addressed with a new design. That is what an honest scientific community does.

Study two: after the bombing in Buenos Aires (Di Tella and Schargrodsky, 2004)

The second study is the one I find most persuasive, because the experiment is the cleanest.

On July 18, 1994, a bomb destroyed the headquarters of the AMIA, the Argentine Israelite Mutual Association, in Buenos Aires. It was the deadliest terrorist attack in Argentine history, and 85 people were killed. In its aftermath, the government ordered police protection for Jewish institutions across the city, and officers were posted on the blocks where synagogues, community centers and schools stood.

Now notice what that does to the economists' problem. The decision to station police on a particular block was driven by the presence of a Jewish institution on that block, not by the amount of ordinary crime in the neighborhood. A synagogue is not located on a street because the street has a car theft problem. The police went to the blocks for a reason that had nothing to do with local crime, which is exactly the variation needed.

Rafael Di Tella and Ernesto Schargrodsky published "Do Police Reduce Crime? Estimates Using the Allocation of Police Forces after a Terrorist Attack" in the American Economic Review in 2004. They compared car theft on blocks that got a police presence with car theft on other blocks, before and after the attack. They found that car theft fell sharply on the protected blocks, by roughly three quarters, and that the effect was concentrated on the protected blocks themselves. They looked for crime that might have been displaced to nearby blocks and found no appreciable effect outside the narrow area where police were posted. A later re-examination of the same natural experiment has argued that displacement matters more than the original paper found, so this part of the result is contested.

Think about what that means. The usual objection to the police-reduce-crime claim is that police merely push crime around the corner. On the original paper's reading, crime fell on the blocks with police, and the thieves did not simply find the next street. They were deterred, at least to a significant degree. The authors were careful in noting what they measured: car theft, in one city, in a specific period. But the design is hard to argue with.

Study three: terror alerts in Washington (Klick and Tabarrok, 2005)

The third design is the funniest to describe, because it relies on a colored chart.

After September 11, 2001, the federal government introduced a color-coded Homeland Security Advisory System. When the threat level was raised to orange, the city of Washington, D.C., put additional police officers on the streets, especially in areas near the National Mall and federal buildings. The decision to raise the alert level had nothing to do with the daily rate of ordinary street crime in the city. It was driven by intelligence about terrorism. So, as in Buenos Aires, the economists had a source of police presence that was unrelated to crime.

Jonathan Klick and Alexander Tabarrok published "Using Terror Alert Levels to Estimate the Effect of Police on Crime" in the Journal of Law and Economics in 2005. They compared daily crime in the District on high-alert days with crime on other days. They found that total crime fell by about 6.6 percent on high-alert days, roughly seven crimes a day, and commentary on the study describes the decline as concentrated around the National Mall, where the added police were posted.

An obvious worry arose: perhaps on orange-alert days, tourists and commuters stayed home, leaving fewer potential victims on the street. If that were the explanation, crime would fall without the police doing anything. The authors addressed that worry in the paper, and a reader who wants to judge the study should look at how they handled it. The point of the design is that a decline concentrated where the police were posted is hard to explain by empty streets alone.

The arithmetic of a policy decision

A study's coefficient is not a budget. Here is a hypothetical calculation, labeled as one, to show how a city council might use the findings. None of the numbers below come from any study.

Suppose a city has 500 sworn officers and 2,000 violent crimes in a year. Suppose, for illustration, that a 10 percent increase in police reduces violent crime by 4 percent. That is a ratio of 0.4, in the range that economists in this literature often report for violent crime, though I am choosing it for easy arithmetic, not because it is a result for your city.

• A 10 percent increase is 50 more officers, because 500 times 0.10 is 50.
• A 4 percent reduction in 2,000 violent crimes is 80 fewer crimes, because 2,000 times 0.04 is 80.
• If each added officer costs the city $100,000 a year in salary and benefits, the 50 officers cost $5 million, and the cost per violent crime prevented is $5,000,000 divided by 80, or $62,500.

Is that a good deal? It depends on what a violent crime costs the victim and the community, in medical care, lost work, fear and, in the worst cases, a life. A defender of the policy will say that the true cost of a single serious violent crime is many times $62,500. A skeptic will say that the number should be compared with other uses of the same money. Both are legitimate arguments, and both start from the same arithmetic. That is the value of the studies: they let the debate be about numbers instead of about slogans.

The limits critics raise

Now the other side, because the evidence has limits and an honest reader should hear them.

1. External validity. The studies measure the effect of extra police in particular places and times: Buenos Aires after a bombing, Washington on orange alert, American cities in election years. Critics point out that the results may not carry to a very different city, or to a very different kind of policing. A hot-spot patrol in a crowded downtown is not the same as an extra officer in a quiet suburb.
2. Marginal versus average. The studies estimate the effect of adding police, at the margin. They do not tell us whether a city that already has a great many police would see the same benefit from adding more, or whether it would lose the same amount if it cut. Returns may diminish.
3. Which police, doing what. The studies measure presence. They do not say which tactics work best, which ones create friction with the community, or which ones carry costs that offset the benefit. Visible patrol, investigation, targeted enforcement against repeat offenders and community engagement are different things.
4. The costs of policing. Police have the power to stop, search and arrest. More police contact with citizens means more of those interactions, some of which go badly. A fair account of policy has to count those costs too. That is a major reason constitutional limits on police power matter, and we take up some in Graham v. Connor and the Officer's View in a Split Second.
5. Reported crime versus actual crime. More police can mean more crime reported, which would make police look like they raise crime. Careful researchers address that problem, and the studies above use crimes, like car theft, that are reported consistently. But it is one more reason crime data need careful handling; see Seven Things Crime Statistics Cannot Tell You, and Two Things They Can.
6. Substitutes. The studies compare police with no police. They do not compare police with alternative ways to spend money, such as treatment programs, lighting, or the courts. A skeptic who says the money would do more good elsewhere has raised a question the studies cannot answer by themselves.

None of these limits overturns the finding. They shape what it can bear. The honest summary is that the question "do police reduce crime?" has been answered, repeatedly and by different methods, with a yes. The question "how many police, doing what, at what cost?" is still a live one.

What the studies mean for the "defund" argument

This is where the research meets politics. In 2020 a slogan that called for cutting police budgets spread across American cities, and several cities cut funding and then restored it. We tell that story in Defund the Police Was a Slogan Cities Tried and Then Quietly Walked Back. I would put the research this way: the economists did not set out to rebut a slogan, and they published their work years before it appeared. But the conclusions are directly relevant. If more police reduce crime, then fewer police increase it. A city that cuts its force is running the same experiments in reverse, and it should not be surprised at the results.

That is not an argument that police budgets should be inviolable. It is an argument that they should be debated on evidence. A city that wants to spend less on enforcement should say what it expects to happen to crime and why. A city that wants to spend more should say how it will measure results. A hiring shortage in many departments is itself a constraint, and we look at that in Why Cities Cannot Hire Enough Police Officers, and What Would Fix It.

The Founding view of this question

The oldest justification of government in the American tradition is protection. The Declaration of Independence holds that to secure the rights to life, liberty and the pursuit of happiness, "Governments are instituted among Men." The first right in that list is life, and a government that cannot keep a citizen safe on his own street has failed at the one job it was formed to do. John Locke, whom the Founders read closely, argued that people enter civil society to have an impartial authority that protects their persons and property, instead of leaving each man to defend himself.

Police are the everyday face of that protection. The finding that they work is not just a statistic. It is a statement that the Founders' premise can be tested and has passed. A conservative who believes in ordered liberty and in the duty of government to secure the basic rights of its citizens should welcome the evidence and insist that policy reflect it.

The case in brief

1. Naively comparing police numbers with crime shows more police where there is more crime, because cities hire police in response to crime, so economists needed natural experiments to isolate the effect.
2. Levitt's 1997 study using election-year hiring (corrected and re-tested after McCrary's 2002 critique), Di Tella and Schargrodsky's 2004 study of police posted after the 1994 AMIA bombing in Buenos Aires, and Klick and Tabarrok's 2005 study of Washington terror alerts all found that more police reduced crime.
3. The Buenos Aires study found car theft fell by about three quarters on protected blocks without displacement to nearby blocks on the original reading, and the Washington study found about a 6.6 percent drop in crime on high-alert days.
4. The limits are real: results come from particular places, measure the margin, do not say which tactics are best and do not count the costs of police contact, so cost-benefit decisions still need local judgment.
5. The evidence supports funding police on the basis of results and measuring what the money buys, which is consistent with the Founders' view that government exists first to secure the right to life.

What you can do: get your city's or county's police budget and staffing figures from its public budget documents, and ask a council member at a public meeting what crime outcomes the department is expected to deliver for that money.