“There are three kinds of lies: lies, damned lies, and statistics.” (Mark Twain, attributing to Benjamin Disraeli)
Here is how it went:
In 1975, Isaac Ehrlich published in The American Economic Review a paper on “The Deterrent Effect of Capital Punishment: A Question of Life and Death.” He was the first to employ econometric tools in a study of this contentious issue. All previous studies had showed no correlation between capital punishment and deterrence of murders. In fact, Ehrlich himself admitted that raw data showed no deterrent effect at all. This is, however, where econometrics came in to help: Ehrlich created a model.
The Ehrlich model is based on the assumption that murderers are rational people who respond to incentives. In other words, they kill because they think they will derive a benefit or utility (be it material or emotional). Society can alter this criminal behavior by offering countervailing incentives (say, a victim could bribe the murderer) or outright disincentives (the classical case in point would seem to be capital punishment). Ehrlich employed all the august tools of economic prediction: a consumption function based on the probabilities of various outcomes of the consequences of a murder, partial elasticities of the expected utility from crime, a social loss function, marginal cost and revenue from execution, a murder supply function, etc. The model used a range of variables that may seem fairly random (why, for example, choose the population at risk of becoming murderers to be aged 14-24?), especially that not all the data needed was, in fact, available – so the author simply made up some values by ‘estimating’ or interpolating, or substituting them). Pages upon pages of complicated (although still arbitrary) formulae and data manipulations later, the reader is presented with tables of data that now magically purport to show a deterrent effect. Not only that: the author even quantifies this deterrent effect, claiming that a single execution is worth ‘eight saved lives.’
Predictably, this novel approach raised some objections, such as those promptly published by Peter Passell and John B. Taylor in the American Economic Review, “The Deterrent Effect of Capital Punishment: Another view.” This rather short paper resoundingly discredited Ehrlich’s approach for using arbitrary data and variables that could hand the researcher just about any result he desires. More specifically, Passell and Taylor criticized Ehrlich for not using an established theory-based approach (after all, the use of data and of variables needed to be justified theoretically and models needed to reflect behavioral expectations) and instead plugging in whatever made his particular formulae yield the numerical result he happened to be looking for – here, a positive correlation between the number of executions and the number of ‘prevented’ murders. Not only did Ehrlich’s model show precious nothing, but his paper was published at a time when legislatures and courts re-examined their death penalty policy, suggesting that the results might have been produced ‘on demand’ to give support to one policy choice over another.
Fast forward to 2003, when Hashem Dezhbakhsh, Paul H. Rubin, and Joanna M. Shepherd published in the American Law and Economics Review Vol. 5 No. 2 “Does Capital Punishment Have a Deterrent Effect? New Evidence from Postmoratorium Panel Data.” That paper is a continuation of the thread started by Ehrlich, ignoring all scholarship that dismissed the methodology used in “The Deterrent Effect of Capital Punishment: A Question of Life and Death.”
Having at their disposal 28 years of developments in both econometrics and law and economics writing, the authors take the Ehrlich model and improve on it – to the tune of now eighteen saved lives (plus or minus ten, what’s a rounding difference, after all) for each additional execution. As Ehrlich did before, Dezhbakhsh et al also concede that an analysis of raw data comparing the number of murders in executing and non-executing states does not show a deterrent effect, hence they recognize a need to use “more sophisticated empirical techniques” (349) to determine a possible deterrent effect of capital punishment. The superiority of Dezhbakhsh’s approach is stressed by providing a stated rationale for many of their choices of particular function forms and variables (as opposed to “studies [that] often choose the functional form of murder supply rather haphazardly.”(353)) A careful reader will still be puzzled by the authors’ (wholly unsubstantiated) presumptions of what exactly constitutes risk factors for murders: “violent TV programming or movies” (354), National Rifle Association membership rate, population density, per capita income, and demographic variables, such as the percentage of males, of African Americans, and the age of the sample (the population under consideration in their research is aged not even 14-24, as in Ehrlich’s model, but 10-29. Apparently, according to the authors’ implicit logic, ten year olds are much more likely to become murderers – or to respond rationally to the deterrent effect of capital punishment – than do thirty year olds). Population density is, rather oddly, “included to capture any relationship between drug activities in inner cities and murder rate” (358). The higher crime rate in cities is explained as a function of, among other things, “the presence of more female-headed households” (367), and the inclusion of per capita income is explained by “the role of illegal drugs in homicides during this time period. Drug consumption is expensive and may increase with income.” (366) Equally biased are some of the criteria deemed responsible for lowering the incidence of murders: Republican votes and non-African American minorities.
Further speaking to plausibility, and considering that the authors examine the population of 10-29 year olds, it is only slightly surprising for them to include data on retirement payments, along with income, unemployment, and income maintenance. The authors also aggregate other crimes committed along with murders (even though the ostensible purpose of the article is to show the deterrent effect of executions on murders), and “to address the problem of underreporting” they decide to “use the logarithms of crime rates, which are usually proportional to true crime rates” (emphasis added) (360). Moreover, Dezhbakhsh et al. use “forward-looking and backward-looking expectations” to reflect the conditional execution probability apparently considered by the murderers, and “given the absence of an arrest lag, no lag displacement is used to measure the arrest probability” (361). And apparently, in that model, all murder cases are solved at once.
Obvious contradictions in their obtained results do not deter the authors from stating blithely on page 367 that “expenditure on the judicial-legal system has a positive and significant effect on the conditional probability of receiving a death penalty sentence in all six models of equation (5),” only to appear to reverse themselves just a page later by concluding that “The expenditure on the judicial-legal system has a negative and significant effect on the conditional probability of execution in all six models (equation ). This result implies that more spending on appeals and public defenders results in fewer executions.”
The substitution of data used here is also rather peculiar: “In the absence of conviction data, sentencing is a viable alternative that covers the intervening stage between arrest and execution.” Also, “The estimated coefficients for year and county dummies are not shown.” (362). A problem arises when there happen to be no murders or no death sentences in particular (actually, in several) years in individual counties examined, and Dezhbakhsh et al. deal with it in one of two ways: “Estimates in Table 3 are obtained excluding these observations,” or by substituting “the relevant probability from the most recent year when the probability was not undefined.” In other words, the model excludes the possibility of zero murders and zero death sentences in certain counties, which has, of course, rather dramatic effects on the estimations of the deterrent effects of capital punishment produced by that model. This is how Dezhbakhsh et al. justify it: “The assumption underlying such substitution is that criminals will use the most recent information available in forming their expectations.” (364) It begs the question whether the authors ever tried to imagine, much less to verify empirically, the notion of a murderer planning his crime by researching recent arrest, conviction, and execution statistics for his county, and actually calculating the probability of his execution following conviction.
The entire model is based on specific presumptions of its authors: “Strictly speaking, these measures are not true probabilities. However, they are closer to the probabilities as viewed by potential murderers than would be the “correct” measures. Our formulation is consistent with Sah’s (1991) argument that criminals form perceptions based on observations of friends and acquaintances.” (364) Let us reiterate: the model is based not on facts, but on what the authors think the murderers consider facts based on the experiences of their friends and acquaintances. In other words: the authors try to model the mindset of a murderer, and conclude from it that one additional exercise of capital punishment will dissuade other murderers from killing eighteen (or eight, or twenty eight, or any number in between) innocent people, and all that happens because the prospective murderer of these eighteen victims is presumed to be a friend or acquaintance of an executed person. That is a rather bold statement for scholars who are neither criminologists, nor forensic psychologists, but rather economists.
The purpose of the study is clearly expressed by the authors in their concluding remarks: “our study offers results that are relevant for analyzing current crime levels and useful for policy purposes. Our study is timely because several states are currently considering either a moratorium on executions or new laws allowing execution of criminals.” Given the social divisiveness of capital punishment, the latter would appear to be true at almost any given moment, rendering any such studies ‘timely’ by default. Starting from the assumption that the Ehrlich study was indeed correct in its approach, and then more than doubling Ehrlich’s prediction, the authors clearly took sides in the death penalty debate. In the end, Dezhbakhsh et al.’s specific methodologies are not what matters. Even if they are later dismissed by other scholars as not rigorous enough, as happened with the Ehrlich paper, the mere fact of publication of the research of Dezhbakhsh et al. in a scholarly journal gives its finding, namely, the magical number of eighteen ‘saved lives,’ enough gravitas to be quoted as a “scientific fact” not only, and indeed not so much, by other scholars, but, most importantly, by politicians and death penalty advocates all over. And that is precisely what we see happen in the tendentious paper of Cass R. Sunstein and Adrian Varmeule, “Is Capital Punishment Morally Required?The Relevance of Life-Life Tradeoffs.”
Post a Comment