The Death Penalty: Deterrence, Cost & Racial Disparities

The death penalty debate conflates several distinct things. A death sentence is a judicial pronouncement; an execution is the carrying out of that sentence. Many sentenced individuals are never executed — they die in custody, receive commutations, or win appeals. At year-end 2023, BJS reported 2,192 people under sentence of death across 26 states plus the federal system, with only 24 executions carried out that year. The average time from sentence to execution for those executed in 2023 was 279 months — about 23.3 years.

Life without parole (LWOP) is the real comparator for both deterrence and cost questions. Parole-eligible life sentences are less relevant because most capital-eligible states now use LWOP as the alternative. A capital-eligible murder is a first-degree murder plus statutory aggravating circumstances (e.g., multiple victims, felony murder, prior record). This distinction matters because bad cost or deterrence comparisons often compare death cases to average prison cases instead of comparable aggravated murder cases.

Source: Bureau of Justice Statistics, Capital Punishment 2023 — Statistical Tables.

The evidence here focuses on the modern era of U.S. capital punishment — post-Furman v. Georgia (1972 moratorium) and especially post-Gregg v. Georgia (1976 reinstatement) — with emphasis on the strongest research from the last 25 years. The three questions — deterrence beyond LWOP, comparative cost, and racial disparities in application — are addressed using named studies with disclosed methodologies.

Isaac Ehrlich (1975), an economist at the University of Chicago, published the seminal pro-deterrence paper in the American Economic Review. Using national time-series data from 1933–1969, he found an elasticity suggesting each execution prevented roughly 7–8 murders. This paper revived the econometric deterrence debate. However, the National Academy of Sciences (1978) found Ehrlich’s results highly sensitive to model specification, with key findings disappearing under alternative assumptions.

Dezhbakhsh, Rubin & Shepherd (2003) used county-level panel data (3,054 counties, 1977–1996) with instrumental-variables regression and reported each execution saves approximately 18 murders (±10). This is the most-cited pro-deterrence finding. However, Donohue & Wolfers (2005) demonstrated that the DRS model produces the opposite sign with minor instrument changes. Berk (2005) showed that roughly 1% of the data (high-execution years in a few states) drove the entire result.

Joanna Shepherd (2004–2005) found heterogeneous effects: among 27 states with at least one execution, deterrence appeared in only 6 states, brutalization (increased murders) appeared in 13, and no effect in 8. Even on the pro-deterrence side, this is not a simple “the death penalty clearly deters” result.

Donohue & Wolfers (2005/2006), a Stanford law professor and an economist respectively, conducted a comprehensive re-analysis of all major panel and time-series studies. Their finding was stark: estimated deterrent effects range from +429 lives saved per execution to −86 lives lost with trivial specification changes. The data are too noisy and executions too rare (<2% of capital murders result in execution) for any robust causal inference.

The National Research Council (2012) panel, chaired by criminologist Daniel Nagin, conducted the definitive review. Their conclusion: “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.” This was not an abolitionist slogan — it was the official conclusion of the National Academies. The reasons: studies fail to model the full sanction regime (LWOP vs. parole), use implausible perception models, and face massive confounders including low execution certainty, demographic shifts, drug markets, urbanization, and prosecutorial variation.

Parker (2021) used synthetic control methods comparing seven states that abolished the death penalty against 29 that retained it and found no evidence that death penalty statutes deter murders, including stranger homicides. This is methodologically stronger than older panel studies.

Sources: Ehrlich (AER 1975), Dezhbakhsh/Rubin/Shepherd (ALER 2003), Donohue & Wolfers (Stanford Law Review 2005), National Research Council (2012). Chart shows the low end of the D&W reanalysis (−86 lives); the high end (+429 lives saved) is omitted because it makes all other bars invisible — that instability is itself the point. NRC (2012) produced no numeric estimate, concluding the field is “not informative.”

The broader criminological literature repeatedly finds that certainty of punishment — the likelihood of being caught and convicted — deters crime far more than severity of punishment. The death penalty, being rarely imposed and even more rarely executed, is unlikely to provide meaningful additional deterrence over LWOP. With only ~24 executions per year in a country with ~21,000 homicides, the probability of any given murderer actually being executed is vanishingly small.

The cost question has one of the clearest evidence bases of any issue in this article. Every modern state audit — across Republican and Democratic states, across different methodologies — finds the same thing: capital cases cost substantially more than comparable non-capital cases. The extra costs come from bifurcated trials, specialized counsel, longer proceedings, mandatory multi-level appeals, and separate death-row housing.

Sources: Urban Institute (Roman et al. 2008), Seattle University (Collins 2015), Kansas Judicial Council (2014), Cook (Duke 2009), Nevada LA (2014), California CRPC (2021).

The popular argument that the death penalty would be cheaper “if appeals were shortened” is not a description of how the current U.S. system actually operates. It is a counterfactual that assumes away safeguards central to American capital jurisprudence. In practice, the cost drivers are structural:

Trial phase: Capital trials require a separate penalty phase with its own jury instructions, witnesses, and expert testimony. Kansas found death penalty jury trials averaged ~40 court days versus ~17 for non-death cases. Defense costs averaged $395,762 in capital trials versus $98,963 in non-capital trials.

Appeals: Automatic direct appeal, state post-conviction review, federal habeas corpus, and sometimes additional rounds of litigation are mandatory. Kansas Supreme Court justices spent 20 times more hours on capital appeals. Washington found average state and federal review time of 11.2 years.

Incarceration: Death-row housing costs more due to higher security and isolation. Kansas estimated death-row incarceration at ~$49,380/year versus $24,690 for general population — roughly double.

Sources: Urban Institute (MD), Collins et al. (WA), Nevada LA. Figures represent the extra cost above a comparable non-capital aggravated murder case, not the total cost. California and Kansas figures omitted from this chart because they use different cost metrics (system-wide annual or trial-only ratios).

The raw numbers show stark overrepresentation. BJS (2023) reports that 41.6% of people under sentence of death are Black, compared to roughly 13–14% of the U.S. population. About 42.2% are white. However, raw disparities alone do not prove bias — they could reflect differences in offense severity, prior records, aggravating circumstances, or geographic concentration. The question is whether disparities persist after controlling for legally relevant factors.

The most consistent and best-documented racial pattern is the race-of-victim effect: cases with white victims are far more likely to result in death sentences than cases with Black victims, even after statistical controls. About 75–80% of executions involve white-victim cases, even though only 40–50% of murder victims nationally are white.

The U.S. GAO (1990) synthesized 28 post-Furman studies and found that in 82% of them, race of victim influenced the likelihood of capital charging or a death sentence. This pattern held in both high-quality and medium-quality studies, at all stages of the process, and was strongest at the earlier discretionary stages (charging and sentencing). Legally relevant variables were influential but “did not fully explain” the victim-race disparities.

The Baldus study in Georgia, covering ~2,484 cases from 1973–1979, found that after controlling for over 100 case characteristics (crime severity, prior record, multiple victims, etc.), defendants accused of killing white victims were 4.3 times more likely to receive a death sentence than those accused of killing Black victims. This study was central to McCleskey v. Kemp (1987), where the Supreme Court assumed the study’s validity but held that statistical disparities alone, without proof of discriminatory intent in the specific case, do not constitute an Equal Protection violation.

The national evidence on defendant race alone is less consistent. The GAO found it “equivocal” across 28 studies. Some studies show Black defendants more likely to be sentenced to death; others find null effects or even reverse effects after controls for case severity. Jennings et al. (2014) used North Carolina data and propensity-score matching on over 50 factors, finding that after matching, both victim-race and defendant-race effects disappeared — suggesting the raw gap reflected case severity more than pure racial bias.

However, “equivocal nationally” does not mean “no defendant-race problem anywhere.” In Washington, Black defendants were more than four times as likely as non-Black defendants to be sentenced to death. In Louisiana, Black male defendant/white female victim cases were more than five times as likely to produce a final capital charge. These jurisdiction-specific findings cannot be dismissed.

Sources: Baldus et al. (Georgia, 1983), Paternoster et al. (Maryland), Robinson (North Carolina). All figures are adjusted odds ratios after multivariate controls. North Carolina figure compares Black defendant/white victim vs. Black defendant/Black victim combinations.

Capital punishment is geographically concentrated. About 2% of U.S. counties account for a majority of death-sentencing outcomes leading to executions since 1976. Prosecutorial discretion in charging death-eligible cases is a major driver of disparity. A similarly situated crime in one county may draw the death penalty while the identical crime in a neighboring county does not. Maryland’s study found extreme county variation: some counties virtually never sought death, others often did. As Paternoster summarized: “Whom you kill and where in Maryland you commit the crime make a difference.”

The Death Penalty Information Center documents 202 death-row exonerations since 1973 across 29 states. The National Registry of Exonerations reported 147 exonerations in 2024 alone, including 85 homicide exonerations and 4 involving former death sentences. Leading causes include official misconduct (present in ~77% of 2024 cases), perjury, false accusations, inadequate defense, and mistaken eyewitness identification.

Gross, O’Brien, Hu & Kennedy (PNAS 2014) used statistical survival analysis to estimate that if all death-sentenced defendants remained under sentence indefinitely, at least 4.1% would eventually be exonerated. This is a model-based estimate — a conservative lower bound — not a direct observed innocence rate, but it remains one of the most important quantitative attempts to measure hidden innocence in capital cases.

Sources: DPIC innocence database; National Registry of Exonerations 2024 Annual Report; Gross et al. (PNAS 2014).

The unique moral weight of wrongful execution is that unlike wrongful imprisonment, it cannot be undone. Recent exonerations underscore the ongoing reality of this risk: Derrick Jamison (Ohio) was formally vindicated in 2026, 41 years after conviction; Elwood Jones (Ohio) had his case dismissed in 2025 after 27 years on death row; Tommy Lee Walker (Texas) was posthumously exonerated 70 years after execution.

This risk also worsens the cost argument. The safeguards needed to minimize wrongful execution — mandatory appeals, post-conviction review, specialized defense — are themselves a major reason capital cases cost so much more than LWOP cases. The system must either accept greater error risk or bear very high process costs to reduce it.