Does the Death Penalty Deter Crime?

Many people sentenced to death are never executed. In 2023, five states carried out 24 executions, according to the Bureau of Justice Statistics (BJS). That year, by BJS’s count, 64 people left death row another way. For 34 of them, a court overturned the conviction or the death sentence, and 30 died of other causes.

Whether the death penalty deters murder, and what it costs, both come down to what it adds over the punishment the same murderer would otherwise get. In all 27 states that the Death Penalty Information Center (DPIC) counts as having the death penalty, the alternative sentence can be life without parole.

The most common capital offense is first-degree murder with at least one aggravating factor that the law lists, such as a murder committed during a kidnapping or an armed robbery.

Source: Bureau of Justice Statistics, Capital Punishment, 2023 – Statistical Tables (2025). Four states with a statute (Delaware, New York, Oregon and Wyoming) held no one under sentence of death. New Hampshire, which repealed its statute in 2019, still held one man sentenced before the repeal. DPIC counts 27 states with the death penalty, because it lists Delaware and New York among the states without it. Both states’ top courts struck down part of their death penalty statutes.

In 1975 the economist Isaac Ehrlich, using national figures for 1933 to 1969, estimated that each additional execution a year may have meant 7 or 8 fewer murders on average. In 1978 a National Research Council panel found his results sensitive to minor changes in the model and unstable over time. The panel concluded that the studies then available gave “no useful evidence” on whether the death penalty deters.

In 2003 Hashem Dezhbakhsh, Paul Rubin and Joanna Shepherd reported that each execution meant 18 fewer murders on average, give or take 10. Their county study used data for 3,054 U.S. counties from 1977 to 1996. In 2005 Richard Berk, a statistician at the University of California, Los Angeles, found that several such claims rested on the few state-years with more than five executions. Those state-years were about 1% of the data.

Shepherd later split the same data by state, covering the 27 states that carried out executions in those years. She found that executions deterred murder in 6 of them and made no difference in 8. In 13, she found, executions increased murders, which she called brutalization. Where executions did not deter, she wrote, they “kill convicts uselessly” if deterrence is their purpose.

In 2005 the law professor John Donohue and the economist Justin Wolfers re-ran the recent studies claiming deterrence. The 2003 county study relied on four state figures to tell cause from effect: police payrolls, court spending, prison admissions and the Republican share of the presidential vote. The paper describes the vote share as one figure, but the study’s code used six. Using the authors’ own data and code, Donohue and Wolfers found that entering the vote share as one figure turned 18 lives saved per execution into 18 lost.

Across the county study’s six versions of its model, using only the vote shares gave as many as 429 lives saved. Using only police payrolls, court spending and prison admissions gave as many as 86 lost. Once the margin of error around the published 18 allowed for counties in the same state moving together, it ran from 82 more murders to 119 lives saved per execution.

Executions are so rare, Donohue and Wolfers concluded, that their effect on murder cannot be reliably told apart from the large swings in murder from year to year.

In 2012 a National Research Council committee chaired by Daniel Nagin reviewed the research. It concluded that “research to date on the effect of capital punishment on homicide is not informative about whether capital punishment decreases, increases, or has no effect on homicide rates.”

The committee found that none of the studies accounted for the punishment a murderer would otherwise get, usually a long prison term or life without parole. All of them, it found, relied on unproven assumptions, including about how potential murderers judge their risk of execution. A lack of evidence, it stressed, is not evidence for or against deterrence.

In 2021 Brett Parker compared seven states that had recently abolished the death penalty with blends of the 29 that kept it. He found no evidence that a death penalty law deters murder, even the murder of strangers.

How to read it: each bar is how many lives one execution saved, net, in the version of the county study that gave its 18. Below zero means more deaths. Source: Donohue and Wolfers’s re-run (2005), with the county study’s own data and code.

The National Institute of Justice says the chance of being caught deters far more than even draconian punishment. It draws on Daniel Nagin’s 2013 review of the research. Executions themselves are rare. In 2024 there were 25, while the Centers for Disease Control and Prevention counted 20,162 homicide deaths. By the Death Penalty Information Center’s count, executions nearly doubled in 2025.

Six state studies compared capital murder cases with similar cases where death was not sought, or with a system without the death penalty. Each found the capital cases cost more on average. The studies measure different things, and each row says what it compares.

The difference starts before the trial. Once prosecutors seek death, jurors are questioned about their views on it. Kansas’s auditors counted 230 jurors screened on average, against 89 in other murder trials. In such a case each side usually has two lawyers, and a guilty verdict is followed by a second trial on the sentence, with its own witnesses and experts. In Maryland about 70% of the extra cost of seeking death fell at the trial stage.

A death sentence then brings an automatic review by the state’s highest court, which most other sentences do not get, and years of further challenges that the prisoner can bring.

Seeking death costs more even when no death sentence follows. In Nevada such a case cost $1.2 million, against $775,000 where death was never sought. In Maryland it cost $1.8 million, against $1.1 million. Kansas’s auditors concluded that because the U.S. Supreme Court requires so much of the process, there appeared to be little room to save money by changing it.

How to read it: each bar is the extra cost, in millions of dollars, of a murder case ending in a death sentence. The comparison is a similar case where death was not sought. Nevada’s bar is for a sentence not carried out. The Washington, Kansas, North Carolina and California studies measure other things. Sources: Urban Institute (2008); Nevada Legislative Auditor (2014).

At the end of 2023, 41% of the people under sentence of death were Black, the Bureau of Justice Statistics reports. That compares with 13.5% of the U.S. population in the Census Bureau’s 2025 estimate. That gap alone does not show bias, since it could reflect differences in the crimes, the defendants’ records or where the cases were tried. The question is whether gaps remain once those are taken into account.

In 1990 the U.S. General Accounting Office (GAO) reviewed 28 studies of death sentencing since 1972. In 82% of them, the victim’s race influenced the likelihood of a capital charge or a death sentence. In those studies, people who killed white victims were more likely to be sentenced to death than people who killed Black victims.

The GAO’s finding held in studies of high, medium and low quality. The evidence was stronger at early stages such as charging, where larger samples allowed more rigorous analysis. Legally relevant factors such as aggravating circumstances were influential but “did not explain fully” the disparities.

In Georgia, David Baldus and his colleagues studied more than 2,000 murder cases from the 1970s. The Supreme Court summarized one of their models in McCleskey v. Kemp (1987). The model took 39 nonracial variables into account. Even so, it found that defendants charged with killing white victims were 4.3 times as likely to receive a death sentence as those charged with killing Black victims.

The Court assumed the study was statistically valid. But it held that to win under the Equal Protection Clause, McCleskey had to prove that the decisionmakers in his own case acted with discriminatory purpose. The study could not show that purpose.

How to read it: each bar is how many times higher the odds of a death sentence were when a victim was white, after the study’s statistical controls. A bar at 1 would mean no difference. The table shows which cases each study compares. The Supreme Court put Georgia’s figure as “4.3 times as likely”; Phillips and Marceau, who later reworked Baldus’s data, describe it as odds. Sources: Baldus and colleagues, as reported in McCleskey v. Kemp (1987); Paternoster and Brame (2003); Unah and Boger (2001).

In North Carolina, Wesley Jennings and colleagues matched 1,113 capital trials from 1977 to 2009 on about 50 case characteristics. After matching, the white-victim effect on the jury’s sentence was no longer significant. Their study began after prosecutors had chosen to seek death, though. In the Maryland study by Paternoster and Brame, the victim’s race mattered at that choice, not at the sentence.

The evidence on the defendant’s race is less clear. The GAO called it “equivocal”, and the Maryland study found no evidence that the defendant’s race mattered once case characteristics were controlled for. In Washington, though, the state’s Supreme Court relied on a study when it struck down the death penalty in 2018. The study found Black defendants whose cases reached a death penalty hearing 3.5 to 4.6 times as likely as others to be sentenced to death, other factors taken into account. The range depends on the model.

Where a murder happens matters too. In 2013 the Death Penalty Information Center found that 2% of U.S. counties accounted for most of the cases that had led to executions since 1976. The Maryland study compared prosecutors in Baltimore County with those in neighboring Baltimore City. Even after controlling for the facts of each case, the county’s prosecutors were more than 13 times as likely to seek death as the city’s. Where the homicide occurs, the study concluded, “matters and matters a great deal.”

The Death Penalty Information Center counts 203 people exonerated from death row since 1973, in 30 states, the latest in July 2026. The National Registry of Exonerations, which counts all crimes, recorded 147 exonerations in 2024. Of those, 85 were for homicide and four were of people who had been sentenced to death. Official misconduct played a part in at least 104 of the 147, or 71%. It also played a part in 79% of the homicide cases.

Counts miss the innocent who are never cleared, and the chance of being cleared drops sharply once a death sentence is reduced to life. In 2014 Samuel Gross and colleagues modelled that effect and estimated that if everyone sentenced to death stayed under that sentence indefinitely, at least 4.1% would be exonerated. The authors call the estimate conservative.

An execution cannot be undone, and errors can take decades to surface. Derrick Jamison received six death warrants during 20 years of wrongful imprisonment in Ohio. One warrant came within 90 minutes of being carried out. A federal court found that prosecutors had withheld 35 pieces of evidence. The charges were dropped in 2005, and in March 2026 a judge declared him wrongfully imprisoned.

Elwood Jones spent 27 years on Ohio’s death row before a judge found in 2022 that prosecutors had withheld evidence; the charges were dropped in 2025. Tommy Lee Walker, a Black man arrested at 19, was sentenced to death by an all-white Dallas jury and executed in 1956 at 21. In January 2026 Dallas County declared him innocent.

The safeguards meant to catch such errors are also why capital cases cost more. Nevada’s auditor traced the extra spending to procedural safeguards “to ensure the sentence is just and free from error”.

No one has shown that the death penalty deters crime. In 2012 the National Research Council found the research not informative either way, and recommended that the studies not be used to inform policy.

The studies that claimed large effects did not hold up when others re-ran them. The 2003 county study estimated 18 murders prevented per execution. Entering one of its figures as the paper describes gave 18 lives lost instead. Executions are rare, 25 in 2024 against 20,162 homicide deaths, and Donohue and Wolfers concluded that their effect cannot be reliably told apart from the year-to-year swings in murder.

The questions of cost, race and wrongful convictions have firmer answers.

Download the conclusion

The verdict above as one image, with its chart of the 2003 county study’s preferred model as Donohue and Wolfers re-ran it. Its bars are 18.5 lives saved per execution as published, 17.7 lost with one vote-share figure as the paper describes, 52.3 lost with only the police, court and prison figures, and 288.8 saved with only the vote shares. The card is 1,080 by 1,350 pixels; anyone may download and post it.

Download the card (PNG, 1,080 × 1,350)