Wednesday, May 21, 2008

Does the death penalty save lives? (part 3)

In the previous two death-penalty posts (1, 2), I talked about whether it makes sense to think that the death penalty deters people from killing, given how rarely it's actually applied. John Donohue and Justin Wolfers say this is implausible (PDF). As I explained, I actually think it's pretty consistent with human nature.

But most of their attacks on the recent spate of studies showing the death penalty to be a deterrent are empirical: they say the data just don't support the claim. That gets to the other main problem I have with the Donohue and Wolfers paper:

They compare non-death-penalty states with all death-penalty states, and claim that there are no significant differences between those two types of states in how much they deter homicide. But that seems to be a highly distorted picture of the real situation.

Here's the problem: there are some states that have the death penalty but rarely if ever use it. For instance, several death penalty states have had only one execution each in 40 years. As another example, California executed only 10 people in several decades even though it's the most populous state.

I don't know as much as Donohue and Wolfers know about how to put together an impressive-looking statistical chart, but I know it doesn't make sense to lump together those states with states that regularly execute people.

But is there a way around that? Yes — just break down the death-penalty states into further categories based on how much (and how quickly) they use the death penalty. Joanna Shepherd did just that (PDF), and she found a huge difference among the different kinds of death-penalty states.

In a nutshell, the difference is that only the states that apply the death penalty on a regular basis will achieve the deterrent effect. And those are a small minority of states. The death-penalty states that don't use it much actually have the opposite effect: homicide goes up.

As Shepherd puts it: "On average, an execution in the United States deters crime. [But] these averages are powered by a handful of high-execution, high-deterrence states."

So, if she's right, then that's simultaneously (a) pretty embarrassing for most death-penalty states -- they're actually driving up homicide, but (b) a ringing endorsement of the death penalty itself, as long as it's used right. It basically means that most states would be wise to ramp up their use of the death penalty so that it passes the "threshold" level of death sentences and executions that must be crossed before the death penalty becomes effective.

By the way, Shepherd is no death-penalty cheerleader. She says she's "definitely against the death penalty on lots of different grounds." She's just pursuing the data wherever they go, even when they go against her own personal views.

The New York Times highlighted Shepherd's point about the necessary "threshold" for the death penalty to be an effective deterrent:
According to roughly a dozen recent studies, executions save lives. For each inmate put to death, the studies say, 3 to 18 murders are prevented.

The effect is most pronounced, according to some studies, in Texas and other states that execute condemned inmates relatively often and relatively quickly.
This conclusion strikes me as intuitively plausible. But wait a minute -- wasn't I just speculating in a recent post that people who are deciding whether to commit homicide probably realize only that there is the death penalty in their state but are not thinking about the statistical likelihood of being executed? How can I say that, but then turn around and say that the frequency with which a state uses the death penalty does affect people's incentives?

Well, there are good reasons to think people don't run through the whole calculation to figure out the exact percentage of being executed if you're caught. For instance, a Texan might have no idea if the chances are 5% or 1% or what. But at least they know there's a real chance. In many death-penalty states, there isn't even a real chance -- it's basically 0%. If you were living in New Jersey and you knew anything about the death penalty there (before it was recently abolished), you'd know that your state had the-death-penalty-but-not-really: the last execution was in 1963. I find it very plausible that that would negate any deterrent effect, while a state like Texas or Virginia would exert a strong deterrent effect (even beyond what would be rationally justified based on a sober assessment of the actual risk of being executed).

In other words, it seems plausible that there would be some minimal threshold of executions that must be crossed for the death penalty to even register with people as something their state uses at all. But once the state passes that threshold, human nature will cause people to mentally inflate the risk of getting the death penalty.

Back to the studies: I have to give Donohue and Wolfers some credit: they do acknowledge the finding that strong-death-penalty states are the ones with a deterrent effect. But then they utterly dismiss it! They say that if you take Texas — which has executed far more people than any other state -- out of the equation, the deterrent effect pretty much goes away. Well, gee, what direction does that argue for? That certainly seems to mean that the way to deter homicide is to do what Texas does: vigorously apply the death penalty instead of just keeping it on the books without using it. But Donohue and Wolfers somehow see it as noise that's getting in the way of studying the deterrent effect of the death penalty.

The idea that Texas is noise that might be better left out, rather than one of the strongest indicators of the effects of the death penalty, seems like such a blatant mistake that I kept thinking: "Wait, I can't be right about that -- one of the authors is a Yale law professor, and this article was published in the Stanford Law Review. I must be missing something." Well, Cass Sunstein is a University of Chicago law professor, and his article was also published in the Stanford Law Review. And he says it makes no sense to see Texas as noise that creates a deceptive appearance of a deterrent effect:
States having the largest numbers of executions are most likely to deter, and it does not seem to make sense to exclude those states as “outliers.” By way of comparison, imagine a study attempting to determine what characteristics of baseball teams most increase the chance of winning the World Series. Imagine also a criticism of the study ... which complained that data about the New York Yankees should be thrown out, on the ground that the Yankees have won so many times as to be “outliers.” This would be an odd idea, because empiricists must go here the evidence is; in the case of capital punishment, the outliers provide much of the relevant evidence.
For all the surface complexity and nuance of the Donohue and Wolfers paper, they seem to have drastically oversimplified and distorted the situation.

By the way, there have also been two other academic articles specifically devoted to disproving Donohue and Wolfers's claims: 1, 2. (Those links go to the abstracts, but you can download the full PDF for free by clicking "one-click download.")

Again, I'm not really qualified to judge Donohue and Wolfers's study, but from what I can tell, it seems like a very weak rejoinder to the abundance of new research suggesting that the death penalty is indeed a deterrent.

Now, even if you're convinced by all that, I could understand saying that we should err on the side of not actively killing people if the data are even debatable. But I'm not so sure. We're never going to have definitive proof of the death penalty's deterrent effect or lack thereof. From what I can tell, the stronger argument is that it is a deterrent. You can't make policy with perfect knowledge. You can only make the best possible estimate, and act on that.


Ken Stalter said...

I'm not sure that all this shows that the death penalty is a deterrent. I do believe that it likely is--I'm just not sure that what we have here is a sufficient argument.

Isn't it the case that the states least likely to apply the death penalty are the most liberal states? And we know that the most liberal states are the most urban states. Whereas the conservative states that are likely to apply the death penalty regularly also tend to be more rural.

I wonder if differences in murder rates are better explained by the idea that there is simply more violence in urban areas than in rural ones. Thus, the differences in homicide rates might be explained by demographics alone without any deterrent effect.

John Althouse Cohen said...

Joanna Shepherd says she's controlled for lots of factors -- demographics, different communities' attitudes toward crime, etc. She seems to say that law review articles finding no deterrence (which are typically not peer-reviewed) fail to control for these variables. I hope to post on this in the future.

Anonymous said...

I would guess that there is a threshold level of executions because it is not so difficult for someone to not even be aware that they live in a death penalty state.

Wouldn't most of the people that are likely to end up as murderers not be people who pay close attention to the news?

If I didn't read the paper or watch the local TV news, how else would I know I live in a death penalty state? It is only when an execution takes place that the subject might actually acquire enough buzz for people to be talking about it.

LemmusLemmus said...

"I have to give Donohue & Wolfers some credit: they do acknowledge the finding that strong-death-penalty states are the ones with a deterrent effect. But then they utterly dismiss it! They say that if you take Texas -- which has executed far more people than any other state -- out of the equation, the deterrent effect pretty much goes away."

You then go on to quote Sunstein and Vermeule who say that Texas shouldn't be excluded on the basis of being an outlier (a concept they seem to misunderstand*).

That's an unfair representation of the D&W paper. In fact, they say (p. 815):

"The implication of our Table 5, however, is not that Texas is an outlier (indeed, given the constancy of the coefficient, it probably lies along the regression line), but rather that in its
absence, there is just too little variation in executions to discern an effect with any confidence."

Although they do estimate one specification with Texas excluded (table 5, column 3), they then go on to estimate further specifications that include Texas (columns 4-6) and find no significant deterrent effect. (My preferred specification, and theirs, if I read them correctly, is column 6, which uses the number of executions relative to the number of homicides as the independent variable.)

*Just performing more executions than other states doesn't make Texas an outlier. An outlier is a data point in which the relationship between the two variables is untypical. That's what D&W allude to when they say that given "the constancy of the coefficient, it probably lies along the regression line". As an example, in this graph, the data point in the top right hand corner is an outlier.

John Althouse Cohen said...


I'm hesitant to say anything at all about their charts, since I'm so unqualified to interpret them, and I know that drawing conclusions about causation can be very tricky.

I went back and searched for all references to Texas in the Donohue & Wolfers article -- in the main text, not in the tables or graphs. It seems to me that what they want to do with Texas is remove it to show how "sensitive" the data are. They cite Joanna Shepherd's findings in this context (the same ones I blogged about). They seem to think that her findings actually undermine the deterrent theory. (I detect some subtle snideness in their summary of her position that low-death-penalty states aren't executing "enough" people.)

Even though I've made clear that I'm not qualified to interpret their raw data, I think that if they really do have a good argument, they have the responsibility to state it concisely and persuasively in the text. From what I've seen and understood in the article, they seem to be thinking about the data wrong. The researchers who have found the death penalty to be a deterrent -- Shepherd, Sunstein, Vermeule -- seem to be reasoning about the situation much more lucidly.

John Althouse Cohen said...

I should add that I will have to go look at the data points you mention (no time for that now, though), and I appreciate your pointing them out.

LemmusLemmus said...

John (or do you prefer being called JAC? Mr. Cohen?),

as I said, they find no deterrent effect even when they include Texas. Calling for sensitivity analyses, however, is not unreasonable: This is standard practice in econometrics; you want to make sure your findings are robust to excluding some data points, especially extreme data points - you don't want your results to be driven by just a few numbers of cases.

Having said that, it may be exactly wrong to exclude Texas because only there is the death penalty used at some notable frequency. It is entirely possible that there is some threshold. Searching for threshold effects, however, is a tricky business - it reeks of data massaging. If only you try out enough things, sooner or later you're going to find a statistically significant effect.

What I'd really like to see is a dataset from a country in which some regions execute one percent of convicted murderers, others fifty percent, and yet others are in between. I don't think such a dataset exists.

Simon Kenton said...

The question as phrased is whether the death penalty saves lives. The question as discussed seems to have been whether it has a deterrent effect. I don't speak to the deterrence question - everyone else has, somewhat diffusely. I do want to bring a perspective on the life-saving, original question. There are a couple of implicit assumptions that seem to be made here - that prisons are non-porous, and that they are homicide-safe. A little reflection and research will point out that murderers escaping is not uncommon even in relatively recent times. Their record while fleeing has been spotty. A bunch of them escaped in New Mexico 20+ years ago, and in responding the governor covered himself with a fatuity so rich and deep that it was unusual even for New Mexico. He'd commuted their sentences, thus putting them in the position to get away. When they did, he called plaintively and publicly upon them to give themselves up and return to custody, because as long as they stayed at large, it would change the whole tenor of the debate on the death penalty. I award him 7 Dave Barry "you-can't-make-this-shit-up points" with an Oak-leaf cluster. With regard to the imperviousness of prisons and the trustworthiness of their escapees, I'd suggest you not bet much on it.

More to the point, google Terrible Tommy Silverstein. Here's a result:

A friend of mine was at one point responsible for keeping Silverstein in a federal prison, and knew some of the guards he killed. If I remember right, he got to take the death notice to their families. Tommy killed first, as I recall, at 12 - grandparents. He's sworn to kill anyone in his presence who raises their hands above their shoulders, and has done so. Check out how many serial life sentences Terrible Tommy has amassed. Last I heard, law students at the University of Denver were contemplating bringing an action to get him released from 23-hour lockdown. I don't suppose that if they succeed they will be planning to take any necessary death sentences to families around Florence, CO.

You might also look at the Aryan Brotherhood, 1/10 of 1% of the prison population and responsible for 18 - 25% of the homicides in the federal system. Do these lives - other inmates particularly of other races, and guards - matter? I don't see much evidence in 'the studies' that they do. We just seem to figure that if we stick these unfortunate criminal persons in the slammer, that'll ... do it. They'll be OK in there. We'll be OK out here. In an acridly cynical sense the assumption is true. Most of these scrotes don't get out so lives on the outside are safe from them. But it's not true at all for lives inside the walls. Would it have saved these all these intramural lives to have snuffed the leaders of the AB? I'd say that these are far clearer instances where a death penalty would have saved lives. But the discussion after your posts would suggest that the actual lives taken in these prisons are valued less - at least for purposes of argument - than hypothetical extramural lives. I don't see that they are even thought of, when it is their lives that are more likely to be saved than ours.

Reading all this is giving me a bit of the Whitman reaction in "When I heard the learn'd astronomer...."

John Althouse Cohen said...

Simon: Excellent points, thanks. I certainly agree that we should take prison escapes and prison homicide into account, and that these are too often absent from the debate.

Of course, it would be blatantly inconsistent for opponents of the death penalty to ignore the prison-homicide factor, since their premise (and one I agree with) is that we should value the lives of even the worst offenders.

Simon Kenton said...

"we should value the lives of even the worst offenders"

If you define life as the biologists do - carrying on 8 or 10 functions, such as respiration, reproduction, etc - you are logically forced to value prions, AIDS virii, Yersinia pestis, etc. You should find yourself first eschewing antibiotics; and second, refusing to operate an immune system: suicide. Take a shit into a modern sewer and you are condemning 10s of trillions of lives to death by chlorine intoxication, their version of mustard gas. Which is absurd.

It has always seemed to me that the intrinsic value of lives comes not from their biologic functionality, but from their concatenation of choices through time. By age you are closer to it than I, so you may recall, and when you breed will discover, that respecting the choices of children turns them into valuable adults. Obviate their consequences, rob them of their development.

So it seems to me that genuinely valuing the lives (the actual, not the merely biological lives) of even the worst offenders means you should be willing to serve on their firing squads. I am.

John Althouse Cohen said...

If you define life as the biologists do - carrying on 8 or 10 functions, such as respiration, reproduction, etc - you are logically forced to value prions, AIDS virii, Yersinia pestis, etc. You should find yourself first eschewing antibiotics; and second, refusing to operate an immune system: suicide. Take a shit into a modern sewer and you are condemning 10s of trillions of lives to death by chlorine intoxication, their version of mustard gas. Which is absurd.
Fallacy of equivocation. When I say "lives," I'm using it as shorthand for "the lives of those who have moral standing," which in my opinion includes all humans. We could debate about whether it includes non-human animals or what kinds of animals or if it even includes life forms other than animals -- those might be interesting debates for another blog post, but it's not what I was referring to.