Policy critique on death penalty?

853© 2009 American Society of Criminology Criminology & Public Policy • Volume 8 • Issue 4

POLICY ESSAY

D e t e r r e n c e a n D e x e c u t I o n s

Don’t scrap the death penalty

Paul h. rubin E m o r y U n i v e r s i t y

I have been asked to comment on the policy implications of “Does the death penalty save lives? New evidence from state panel data, 1977 To 2006,” (Kovandzic, Vieraitis, and Boots, 2009, this issue). The article does not hesitate to draw its own policy conclusions. Kovandzic et al. conclude that because no evidence is provided for a deterrent effect, “policy makers should refrain from justifying its use by claiming that it is a deterrent to homicide and consider less costly, more effective ways of addressing crime.” I will be equally blunt; I do not think that their evidence is sufficiently persuasive to draw the conclusions they do, and I am not in favor of abolishing the death penalty.

I will consider two questions about the death penalty. First, as execution is currently applied in the United States, does it actually serve as a deterrent? This question will require an exami- nation of some assertions in the Kovandzic et al. (2009) article about the nature of criminals as well as about the predictions of economic theory. Second, if the death penalty is currently a deterrent, is it powerful enough so that existing statistical techniques can measure the deterrent effect? This question will entail an examination of the statistical analysis of the article.

Is capital Punishment a Deterrent theory? In regard to the first question, at some level of certainty and speed, capital punishment could serve as a deterrent. Gordon Tullock has proposed the following thought experiment: Think of a criminal pointing a gun at a potential victim with a policeman standing behind the criminal. The policeman says, “You can shoot him, but if you do then I will shoot you.” In this circum- stance, the threat of death certainly would deter most or all potential murderers. We then can think of actual systems as reducing the probability and speed of execution. With each reduction, the level of deterrence also is reduced because fewer potential murderers will be deterred as the probability and speed of execution is reduced.

Thanks to Hashem Dezhbakhsh, Joanna shepherd, and Griffin edwards for helpful comments. Direct correspondence to Paul H. rubin, Department of economics, emory university, Atlanta, CA 30322 (e-mail: [email protected]).

09008-CrimJournal-Guts.indd 853 10/27/09 7:32:01 PM

Criminology & Public Policy854

Moreover, any probability of execution will deter some potential murderers. In some cir- cumstances, the decision of whether to commit murder is marginal. The criminal sometimes will be on the cusp of murder (“Do I shoot the person I have just robbed or let him go?” “Do I kill the rape victim?” “Will a severe beating punish my enemy enough, or should I kill him?” “Do I kill my spouse or seek a divorce?”), and for at least some of these potential murderers, the extra deterrent will be enough to lead to a decision not to commit murder.

Kovandzic et al. (2009) deny that deterrence is present. But their language allows for some deterrence. Kovandzic et al. downplay the possibility of deterrence but cannot eliminate it. For example:

A substantial body of criminological research on offenders’ decision-making processes and the dynamics of homicide events contradicts the supposition that criminals spend any notable amount of time considering the deleterious consequences of their actions. Instead, ethnographic research suggests these individuals are more likely to focus on the potential gains of their crimes rather than on the costs (2009, emphasis added).

Both of these statements and others throughout the discussion section (e.g., “essentially nullify any deterrent effects” or “the deterrent effect of the DP is unlikely to be substantial”) show that Kovandzic et al. believe that at least some deterrence is associated with more severe punish- ments.

Moreover, Kovandzic et al.’s (2009) argument proves too much. Even if the threat of execution does not deter criminals, it still is the case that a large body of evidence shows that increasing the probability and the severity of punishment in general deters crime. (For a sum- mary, see Eide, Rubin, and Shepherd, 2006.) If it is true as Kovandzic et al. state that “Evidence from interviews with active and captured offenders suggests they do not dwell on the potential consequences of their actions and thus rarely consider the possibility of arrest and imprison- ment,” then no punishment would deter any crime, and that clearly is not the case.

Indeed, deterrence historically has been an important issue in the capital punishment debate. Capital punishment is imposed infrequently in the United States. Evidence of the extent to which it is a deterrent is difficult to obtain, and some studies (which includes the study at issue here) find that it is not a deterrent. But even if capital punishment was not or could not be shown to be a deterrent, that still does not apply to other punishments (which are much more certain and less erratic than capital punishment). It is possible, and generally uncontested, that increased probability and severity of jail sentences for crime reduces the amount of crime, and even those individuals who do not believe the evidence for capital punishment is persuasive generally accept that punishment does deter crime. This fact is an important point to keep in mind.

Another issue also plagues criminological research. Criminologists study criminals. For

many purposes, this study is useful. But when the issue being studied is deterrence, then clearly criminals are a highly biased sample, because they have been caught; if criminals avoid detection, then they escape the criminologists’ lens. But more importantly, criminologists miss all those

Pol ic y essay Deter rence and execut ions

09008-CrimJournal-Guts.indd 854 10/27/09 7:32:01 PM

855Volume 8 • Issue 4

individuals who actually are deterred. That is, when deterrence does succeed, some potential criminals do not become criminals and so are not examined by criminologists. Criminologists’ sample, therefore, is unsuited for any inference about deterrence. With respect to the issue of deterrence, this point can be a crucial issue.

Kovandzic et al. (2009) realize this issue and try to avoid it. They argue that “most homicide offenders are not otherwise law-abiding citizens who killed during an act of passion or duress.” However, their data belie this argument. Kovandzic et al. cite a study (Wolfgang, 1958), which showed that 64% of homicide offenders had a previous arrest and 42% (66% of the 64% with a previous arrest) had a record of crimes against the person. In other words, 36% of murders had no previous arrest and 58% had no history of crimes against the person. A more recent study, which they cite (Bureau of Justice Statistics, 2007), found that 65.5% of criminals with a death sentence had prior felony convictions, so 34.5% did not. Given these numbers, clearly all murderers are not hardened criminals, and it is plausible that some potential murderers (obvi- ously not in their sample) were deterred by the threat of execution. These potential murderers would not show up in any study of convicted murders.

Finally, Kovandzic et al. (2009) misunderstand the economic arguments in regard to crime. It is true that Becker’s (1967) model is a formal cost–benefit analysis of committing a crime. This model is what Kovandzic et al. (2009) criticize: “Again, although some offenders might follow Becker’s model of rational decision making and factor in the potential legal sanctions of their actions, it is likely a small portion of offenders.” But for deterrence not to be present, it is not necessary that offenders undertake accurate cost-benefit analysis any more than it is neces- sary for consumers to undertake complex calculations of marginal use for the law of demand to hold. Rather, behavior must respond to changes in prices or costs. If one jurisdiction executes more criminals than another or if a jurisdiction becomes more likely to execute a criminal, and criminals perceive the direction of change in probability, then deterrence can be increased. Caricatures of economics that require criminals to be calculators of pleasure and pain are no more accurate than similar caricatures in other areas of behavior.

They also overstate the arguments of economists. Economists generally believe that punish- ment deters crime; we do not believe that feasible levels of punishment can lead to a “solution to America’s crime problem.” Rather, economists who study the issue believe that deterrence is one tool that can be used to reduce crime. We have studied this tool more than others, because deterrence is where we have a comparative advantage, not because we believe that it is the only solution. Economists believe in the division of labor, and clearly, criminologists have a role in studying methods of reducing crime. But, as I show in the next section, criminologists might do better to use the tools in which they have a comparative advantage rather than attempting to use econometric tools in their research.

rubin

09008-CrimJournal-Guts.indd 855 10/27/09 7:32:01 PM

Criminology & Public Policy856

econometric Issues First, I must state that Kovandzic et al.’s (2009) conclusions go beyond their evidence. They implicitly conclude that they have written the definitive study on capital punishment and de- terrence and have shown deterrence is not present so that, with respect to capital punishment, “policy makers should refrain from justifying its use by claiming that it is a deterrent to homicide.” Yet, they themselves cite a meta-analysis of empirical studies of capital punishment (Yang and Lester, 2008). This meta-analysis found 104 refereed studies, of which 95 had data adequate for analysis. Of these 95, 66 found a deterrent effect. Thus, Kovandzic et al’s study is only one more data point. As such, it is somewhat hubristic to claim that it is the definitive study and should itself be the basis for policy. Indeed, to make such a claim would require examining all 66 studies that found a deterrent (or at least the 15 panel data studies that found a deterrent) and showing why this particular study is preferable. In fact, only 10 studies are cited (Table 1), and Kovandzic et al. actually do not compare many of these findings with their results.

Kovandzic et al. (2009) assert that they have accounted for differences:

The most likely explanation for the divergence between our largely null findings and studies reporting robust deterrent effects that result from increases in execu- tion risk is the failure of the latter to (1) address adequately omitted variable bias by failing to include year dummies and or state-specific trends in the regression model, (2) adjust standard errors to correct for serial correlation, and (3) rely on IV estimation using invalid instruments to address phantom simultaneity effects of homicide on execution risk.

I examine each point. I mainly will rely on the article I know best, Dezhbakhsh, Rubin, and Shepherd (2003). This article is cited in Kovandzic et al. (2009) and so is presumably one they believe to be surpassed by their research.

First, consider omitted variables. Dezhbakhsh et al. (2003) use panel data and includes year and county (or state) dummies; Dezhbakhsh, et al. does not include state-specific time trends. However, Dezhbakhsh et al. also include measures of probability of arrest as well as the conditional probability of a death sentence; both of these variables are omitted in the Ko- vandzic et al. (2009) specification. Dezhbakhsh et al. include aggravated assault and robbery rates as well as NRA membership and racial composition; Kovandzic et al. does not include any of these variables, which generally are significant in our specification (In our robustness checks, we dropped the measures of assault and robbery, and some of our deterrence measures became insignificant.) Thus, their specification includes some variables we had omitted but omits some significant variables we had included. Moreover, no strong reason or evidence is given for including state-specific time trends. As shown in Table 4 of Kovandzic et al., when this trend is dropped, the results become negative and significant for executions and homicide. More generally, the specification generally suffers from overparametrization. This specificity reduces the estimation efficiency, which leads to insignificant coefficient estimates. Examples include using state fixed effects and year fixed effects along with state-specific time trends. The

Pol ic y essay Deter rence and execut ions

09008-CrimJournal-Guts.indd 856 10/27/09 7:32:01 PM

857Volume 8 • Issue 4

overparametrization is reflected in Table 3 in which few variables are significant in regression equations with R2 of 94%.

Kovandzic et al. (2009) claim that the second improvement is an adjustment for serial correlation. They assert that the deterrent finding in some other studies has been the product of serial correlation. However, they have neither tested nor established the presence of serial correlation in their data set, let alone in the other data sets critiqued. Thus, this assertion is without foundation.

Third, Kovandzic et al. (2009) do not use simultaneous equations or instrumental vari- ables; they estimate only ordinary least-squares specifications. But their discussion implies that simultaneity likely is to be an issue. At the end, however, they neither apply the methods used by other authors to address this simultaneity issue, nor do they use any valid alternative ap- proach to treat this problem. Instead, they use the inadequate and simplistic method of looking at the bivariate correlation between homicide rate and the number of death-row inmates. This method is only a (weak) test for one possible source of simultaneity, and Kovandzic et al. do not test for any others. In our debate with Donohue and Wolfers (2005) and Dezhbakhsh and Rubin (2009), a discussion takes place of the proper instruments for controlling simultaneity, but Kovandzic et al. completely ignore this discussion and simply disregard the possibility of simultaneity.

Finally, it is improper to claim that the Kovandzic et al. (2009) article is stronger than the Dezhbakhsh et al. (2003) article because they ultimately are noncomparable. In Dezhbakhsh et al., deterrence is measured six times. However, all measurements are conditional on arrest and on sentencing to death. That is not the case in Kovandzic et al. Moreover, only one mea- sure used in Kovandzic et al. is the same as any of the measures in Dezhbakhsh et al. (2003). Dezhbakhsh et al. uses executions lagged by 6 years, executions leading by 6 years, and a 6-year moving average. (The 6-year period is based on average lag between sentencing and execution.) In three specifications, jurisdictions with no murders or death sentences were excluded. In the other three specifications, the most recent available values were used to fill in cases in which the value is zero. Although Kovandzic et al. use eight measures of execution, only one (the 6-year lagged execution measure) is the same as Dezhbakhsh et al., and this specification had the smallest measured effect. Indeed, in the model in which state (rather than county) data were used, this measure is insignificant, although the other five measures were significant. Because Kovandzic et al. uses state data, the fact that it uses our weakest measure (which we also found to be insignificant in state data) indicates that the results are noncomparable. Kovandzic et al. (2009) do not consider or test the specifications in which significant deterrence was found.

Other econometric issues are present in the Kovandzic et al. (2009) article as well. Many selections and remedies are ad hoc and at odds with sound econometric practice. For example, introduction of lag of the dependent variable is used as a method to correct for serial correlation. Despite using heteroskedastic robust standard errors, Kovandzic et al. argue that weighted least square needs to be used to correct for heteroskedasticity. This assertion is not evidence-based, and the solution also is adopted without testing to see whether this remedy deals with the issue

rubin

09008-CrimJournal-Guts.indd 857 10/27/09 7:32:02 PM

Criminology & Public Policy858

of assumed heteroskedasticity. The specification does not address the choice of the functional form adequately, which ignores double log, Poisson, and negative binomial regressions that some other death penalty studies have used to show robustness. Kovandzic et al.’s statement that “Any deterrent impact due to the presence of these laws is captured by the state fixed- effects variables” is incorrect and misleading. Such effects only can be captured by extending the sampling period to those years.

In sum, Kovandzic et al (2009) change the model specification, estimation method, as well as both the dependent and independent variables used by earlier death penalty studies that report deterrence, and they find no deterrence. This conclusion is not surprising to those scholars well versed in econometrics. Kovandzic et al., however, interpret such a divergence of results as validation of their findings. Their logic reads as follows: If our results are different from others, then ours must be the correct one. To prove their assertions, Kovandzic et al. in- stead should have established, with rigor, that their results are derived from more appropriate statistical models and must, therefore, be the correct one. Moreover, their statistical methods are unjustified and, at times, inappropriate. Their assertion about the lack of a deterrent effect is, therefore, unwarranted given their evidence.

Finally, Kovandzic et al. (2009) overinterpret their results. They do not find evidence of deterrence, but this does not mean no deterrence is present. As a recent op-ed by two participants (on opposing sides) in the debate says: “But the absence of evidence of deterrence should not be confused with evidence of absence.” (Sunstein and Wolfers, 2008:A11). Statistical tests fail unless criminologists can show some effect with greater (90% or 95%) levels of confidence. But the failure to show effects does not indicate that they do not exist.

conclusion The Kovandzic et al. (2009) article is one of more than 100 refereed articles that examine the deterrent effect of capital punishment. Most of these articles find a statistically significant deter- rent effect (Yang and Lester, 2008). Moreover, it would be incredible and a violation of the law of demand if the chance of execution did not deter at least some murders. The data are murky, and it is difficult to measure empirically the effects under consideration, but the weight of the evidence as well as the theoretical predictions both argue for deterrence, and econometrically flawed studies such as this article are insufficient to overthrow this presumption.

One final note: The overwhelming weight of academic opinion is opposed to capital punishment. Even articles that find a deterrent are hesitant about drawing policy conclusions. For example, Yang and Lester (2008: 459) explained: “Even if executions are shown to deter potential murderers, alternative strategies to reduce the murder rate may be more effective and more ethically acceptable.” But professors live in safe neighborhoods, have relatively safe jobs and are not likely to be drug dealers. Most murders occur in poor neighborhoods and among relatively uneducated persons, often with risky lifestyles. An element of elitism may be present in academic recommendations for abolishing the death penalty, because others will bear the costs.

Pol ic y essay Deter rence and execut ions

09008-CrimJournal-Guts.indd 858 10/27/09 7:32:02 PM

859Volume 8 • Issue 4

references Becker, Gary S. 1967. Crime and punishment: An economic approach. Journal of Political

Economy, 76: 169–217.

Bureau of Justice Statistics. 2007. Capital punishment, 2006—Statistical tables. Retrieved September 22, 2009 from ojp.usdoj.gov/bjs/pub/html/cp/2006/tables/cp06st08.htm.

Dezhbakhsh, Hashem and Paul H. Rubin. 2009. From the “econometrics of capital punishment” to the “capital punishment” of econometrics: On the use and abuse of sensitivity analysis. Emory University working paper.

Dezhbakhsh, Hashem, Paul H. Rubin, and Joanna Shepherd. 2003. Does capital punishment have a deterrent effect? New evidence from postmoratorium panel data. American Law and Economics Review, 5: 344–376.

Donohue, John J. and Justin Wolfers. 2005. Uses and abuses of empirical evidence in the death penalty debate. Stanford Law Review, 58: 791–846.

Eide, Erling, Paul H. Rubin, and Joanna Shepherd. 2006. The economics of crime. Foundations and Trends in Microeconomics, 2: 291–363.

Kovandzic, Tomislav V., Lynne M. Vieraitis, and Denise Paquette Boots. 2009. Does the death penalty save lives? New evidence from state panel data, 1977 to 2006. Criminology & Public Policy. This issue.

Sunstein, Cass R. and Justin Wolfers. 2008. A death penalty puzzle. Washington Post, June 30: A11.

Wolfgang, Marvin E. 1958. Patterns in criminal homicide. Philadelphia: University of Pennsylvania.

Yang, Bijou and David Lester. 2008. The deterrent effect of executions: A meta-analysis thirty years after Ehrlich. Journal of Criminal Justice, 36: 453–460.

Paul h. rubin is Samuel Candler Dobbs Professor of Economics at Emory University in Atlanta and editor-in-chief of Managerial and Decision Economics. He is a Fellow of the Public Choice Society and former Vice President of the Southern Economics Association, and is associated with the American Enterprise Institute, Independent Institute, and the Technology Policy Institute. Rubin has been Senior Staff Economist at President Reagan’s Council of Economic Advisers, Chief Economist at the U.S. Consumer Product Safety Commission, Director of Advertising Economics at the Federal Trade Commission, and Vice President of Glassman-Oliver Economic Consultants, Inc. He has taught economics at the University of Georgia, City University of New York, and George Washington University Law School. Rubin has written or edited 11 books and has published more than 150 articles and chapters on economics, law, regulation, and evolution in leading journals, and he also frequently contributes to the Wall Street Journal and other national newspapers.

rubin

09008-CrimJournal-Guts.indd 859 10/27/09 7:32:02 PM