Contact FutureLearn for Support
Skip main navigation
We use cookies to give you a better experience, if that’s ok you can close this message and carry on browsing. For more info read our cookies policy.
We use cookies to give you a better experience. Carry on browsing if you're happy with this, or read our cookies policy for more information.

Evidence on the crime-reducing effect of prison

Prison Population and Crime Rates

Did the recent prison population increase in the U.S. reduce crime? If so, by how much? Based on the U.S. data from the 1970s and 80s, Steven Levitt (1996) and William Spelman (2000) find that the elasticity of crime with respect to prison population is somewhere between -0.2 and -0.4. In other words, doubling the prison population is associated with a 20-40 percent reduction in crime.

However, the crime-reduction effect of having additional prisoners is likely to fall as the level of prison population increases. Prisons will be increasingly populated by less dangerous criminals (lower incapacitation effect) and those who have not entered prison will be even less likely to commit crime (lower deterrent effect). Thus, the crime-reducing effect of more prison population is probably smaller now (with more than 1.5 million inmates) than 30 years ago (with about 500,000 inmates).

However, there is an important limitation in the empirical studies on the relationship between prison population and crime rates. . Policymakers can choose which criminals to send to prison and for how long, but they cannot directly choose the level of prison population. (In other words, prison population is not a policy choice.) The level of prison population is an outcome jointly determined by the certainty and severity of incarceration, and how potential criminals respond to these policy choices by committing fewer or more crimes. Even if the best research evidence indicates that “a 50 percent increase in prison population leads to a 15 percent decrease in crime rates”, how exactly should policymakers achieve a 50 percent increase in prison population? What should be more relevant in a policy discussion is the extent to which crime rates are influenced by the probability and length of incarceration.

Incapacitation Effect of Incarceration

While the incapacitation effect is easy to understand conceptually, empirically measuring its magnitude is a lot more difficult. One naïve approach would be to compare two inmates, one serving a short prison term and another serving a longer term. After the first inmate is released from prison, we can observe how many new crimes he commits while the second inmate is still in prison. Then we may interpret this number as the magnitude of incapacitation effect: the number of crimes prevented by keeping the second inmate in prison.

There is an obvious problem with this approach, however. This approach works only if the two inmates are similar in their criminal risks. But if they are really similar, why did one get a longer prison sentence than the other in the first place? Usually, the length of prison sentence in determined by the severity of the current offense and the prior criminal history of the offender. In this example, the inmate with a longer prison sentence is probably a more dangerous criminal than the other, and we would be comparing apples with oranges.

Ideally, this problem can be eliminated by running the following experiment. We would recruit a large number of inmates who entered prison at the same time with similar criminal background, and randomly divide them into two groups. Inmates in the first group would be released a lot sooner than the second group, and we observe how many crimes are committed by those in the first group while the second group is still in prison. This number can be plausibly viewed as the extent of incapacitation effect.

Clearly, it would not be feasible for a researcher to run such an experiment. But sometimes researchers find a real-life situation that closely resembles this hypothetical experiment (“quasi-experimental variation”). For example, economist Emily Owens noted that a change in the sentencing guideline in Maryland created a variation in the length of sentence among criminals with very similar criminal backgrounds.

Maryland uses a pre-determined formula to compute the recommended length of incarceration for each criminal, based on the type of his current offense and his prior criminal history. Until 2001, the sentencing guideline included juvenile criminal records as part of the prior criminal history for offenders below age 26. Juvenile criminal records for offenders at or above age 26 were disregarded in the sentencing guideline computation.

In 2001, the state lowered this cutoff age from 26 to 23. Following this policy change, offenders between age 23 and 25 were suddenly subject to more lenient prison terms, because their juvenile criminal records no longer counted in their sentencing formula.

Owens computed the additional days of incarceration they would have served had they been sentenced before the change, and then counted the number of new arrests they received during this additional period. She found that the offenders were re-arrested about 3 times and committed 1.5 index crimes on average per year. In other words, an additional year of incarceration for a 23-26 year offender appears to prevent 3 arrests and 1.5 index crimes per year.

Deterrent Effect of Incarceration

While incapacitation effects reduce crime by preventing inmates from committing new crimes, deterrent effects reduce crime by deterring would-be criminals from offending. To estimate the extent of deterrent effects, economists usually look for quasi-experimental variations in sentencing severity among seemingly-comparable individuals.

Consider the famous three-strikes law in California. It requires a minimum sentence of 25 years when an offender with two serious felony convictions commits a new crime. Until its revision in 2012, the third “strike” could be any felony conviction, including relatively minor violations of law.

Economists Eric Helland and Alexander Tabarrok noted that, under the three-strikes law, the costs of committing a new crime are very much different between offenders with one strike and two strikes. Offenders with one strike would probably end up with a short jail term when they are caught for a relative minor crime. On the other hand, offenders with two strikes who commit the same minor crime can be sentenced to 25 years or more in prison. Comparing recidivism patterns between 1) offenders with two prior strikes and 2) offenders who were tried for two strikeable crimes but only received one conviction, Helland and Tabarrok concluded that the threat of a 25-years-to-life prison sentence reduces the likelihood of a new felony offense by 20 percent.

Helland and Tabarrok’s study examines one extreme example of deterrent effects. A felony offender who commits a relatively minor crime may receive a life sentence (if he had two strikes already) or be sentenced to a short jail term (if not). But what about deterrent effects in less extreme cases? Can government expect strong deterrent effects when they increase the length of prison sentence for a given crime from, say, two years in prison to four years in prison?

A recent study by economists Francesco Drago, Roberto Galbiati, and Pietro Vertova attempts to answer this question by exploiting the variation in punishment severity caused by a large-scale pardon in Italy. A government pardon in 2006 reduced the length of incarceration for all inmates in Italy by 3 years, and as a result, about 22,000 inmates who had less than 3 years of prison time remaining were released. On a single day (August 1, 2006), the number of prison population in Italy fell by 40 percent.

The pardon came with an important caveat. If a former inmate released early because of the pardon commits a new crime within 5 years, he will have to serve the remaining portion of the original sentence in addition to the sentence for the new crime. For example, if an inmate was released six months earlier than the scheduled time of release because of the pardon, when he receives a new conviction in the future, his sentence will automatically increase by six months.

This created a variation in the severity of future punishment among seemingly-comparable offenders. Suppose the pardon led to the early release of two inmates who committed the same crime and received the same sentence. However, if the first inmate entered prison six months earlier than the second inmate, the first inmate must have served six more months at the time of release. If they are re-convicted of the same offense in the future, the second inmate will have to serve six months longer than the first inmate. Would this extra time in prison deter the second inmate from committing a new offense?

Analyzing the recidivism patterns of Italian inmates and how it varied across the expected length of future incarceration, authors found that the increase in the expected length of incarceration significantly reduced the likelihood of recidivism. (They estimate that the elasticity of crime with respect to the expected length of incarceration is equal to -0.74.) These findings suggest that both small and large increases in the severity of punishment can successfully deter would-be criminals.


  • Drago, Francesco, Roberto Galbiati, and Pietro Vertova. “The Deterrent Effects of Prison: Evidence from a Natural Experiment.” Journal of Political Economy 117.2 (2009): 257-280.
  • Helland, Eric, and Alexander Tabarrok. “Does Three Strikes Deter? A Nonparametric Estimation.” Journal of Human Resources 42.2 (2007): 309-330.
  • Levitt, Steven D. “The Effect of Prison Population Size on Crime Rates: Evidence from Prison Overcrowding Litigation.” Quarterly Journal of Economics 111.2 (1996): 319-351.
  • Owens, Emily G. “More Time, Less Crime? Estimating the Incapacitative Effect of Sentence Enhancements.” Journal of Law and Economics 52.3 (2009): 551-579.

Share this article:

This article is from the free online course:

Economics of Crime

Hanyang University

Course highlights Get a taste of this course before you join: