Disciplines and deterrence
The NY Times has an interesting piece on statistical studies of the deterrent effect (if any) of the death penalty. For those who want to get straight into fact-free debate, the bottom line is that the evidence is too weak to allow a firm conclusion one way or the other. What’s interesting to me, though is the way in which debates within different disciplines proceed, and the lags in transmission between them. Here I think the NYT story, while excellent in many respects, is quite misleading, presenting a story of deterrence-hypothesis economists facing off against legal critics.
That was pretty much the way things stood in the 1970s, after the publication of Isaac Ehrlich’s study in the American Economic Review claiming that one execution deterred 7 or 8 homicides. Ehrlich used multiple regression analysis (quite difficult and computationally demanding in those days, and correspondingly highly regarded) in an attempt to control for other factors affecting homicide rates and isolate the effect of the death penalty.
Over the next decade, economists learned a lot about the limitations of regression analysis. With limited amounts of data, it’s impossible to avoid mining the data for patterns which are then used to fit the model. And if you try enough specifications on weak data, you can get just about any result you want. A classic exposition of this point was Ed Leamer’s 1983 article “Let’s take the con out of econometrics” which pointed out the fragility of regression analysis on time-series data and picked, as an example, the deterrent effect of the death penalty.
As is the way, there was plenty of back and forth after that, on Leamer’s general claims, his proposal of Extreme Bounds Analysis as a remedy, and on the specific question of the deterrent hypothesis. Still, I don’t imagine any economist who’s been paying even moderate attention for the last twenty-five years would have been surprised by the finding of Justin Wolfers and John Donohue, published in the Stanford Law Review in 2006 that “the death penalty – at least as it has been implemented in the United States – is applied so rarely that the number of homicides that it can plausibly have caused or deterred cannot be reliably disentangled from the large year-to-year changes in the homicide rate caused by other factors. … Sampling from the broader universe of plausible approaches suggests not just ‘reasonable doubt’ about whether there is any deterrent effect of the death penalty, but profound uncertainty – even about its sign”
The other point made in the NYT article is that economists tend to believe that people respond to incentives, and no doubt the favorable initial reception of Ehrlich’s work reflected this. But once economists started thinking about actual magnitudes, the story fell apart. The probability of being executed as a result of committing a homicide is tiny, and pales into insignificance compared to, say, the risks of being a street crack dealer as estimated by Levitt. In fact, I recall, but can’t verify a claim that the mortality rate for prisoners on death row is actually lower than that for prisoners in general.
By contrast with this, I was struck by this passage from Cass Sunstein and Adrian Vermeule also in the Stanford Law Review, and quoted by Wolfers and Donohue
More recent evidence, however, has given new life to Ehrlichâ€™s hypothesis. A wave of sophisticated multiple regression studies have exploited a newly available form of data, so-called â€œpanel data,â€? that uses all information from a set of units (states or counties) and follows that data over an extended period of time. A leading study used county-level panel data from 3054 U.S. counties between 1977 and 1996.23 The authors found that the murder rate is significantly reduced by both death sentences and executions. (scare quotes in original)
The reverence with which panel data studies (pretty old hat to economists) are described suggests that not much in the way of critical analysis is going on here. Wolfers and Donohue point out that the study cited here (by Dezhbaksh et al) doesn’t even have fixed year effects, a strategy Wolfers and Donohue describe as “a clear outlier in the literature”. They might have gone on to point out that it’s an outlier because such an omission would lead to immediate rejection at most econ journals.
Like Wolfers and Donohue, I don’t think statistical evidence on the deterrence hypothesis is ever going to get us far. On the other hand, US experience has shown beyond doubt that innocent people are regularly sentenced to death, and it’s virtually certain that some have been executed (the legal system doesn’t provide any real way of establishing this, except in the rare cases when the real killer is convicted subsequently). That’s not the only argument against the death penalty, but it’s sufficient in my view