Update: added missing caption to figure
Next year, the New Hampshire House will take up a bill to abolish the death penalty. Several libertarian legislators have signed on as co-sponsors, and observers think the bill has a good chance.
What should libertarians think about the death penalty? In general, hardcore civil libertarians have opposed it, but there does not seem to be anything in the moral principles libertarians have adopted that straightforwardly generate an a priori skepticism of capital punishment. The case for or against capital punishment depends on empirical research in a way that the case, say, for or against certain gun laws does not. (Banning handguns would be wrong even if it reduced the violent crime rate, I would argue.)
Some libertarians say that the state can never be trusted with the death penalty. But this too is an empirical claim. What is the rate of killing of innocents when the government has a death penalty of a certain kind? We also need to think about what rate would be unacceptable. (Every criminal justice system will punish some innocents, because no criminal justice system is perfect.) That latter threshold presumably depends on what the deterrent effect of capital punishment is. If capital punishment deters a significant number of murders, presumably an extremely rare execution of an innocent — while horrible and thoroughly regrettable — does not make the system unacceptable.
So does capital punishment deter murders? Apparently not. The literature on capital punishment in the U.S. has shown mixed results, with some models showing a positive deterrent effect (lives saved) and others showing a negative “deterrent” effect (more murders). The recent research by Durlauf, Fu, and Navarro (here and here) helps us adjudicate among these results.
The most plausible set of models consists of those that assume that a higher probability of execution affects murder rate via a logistic function (which reduces the influence outliers). Here’s how they summarize their results in the Journal of Quantitative Criminology for various sets of models (marginal effects reported for the mean 1996 death penalty state):
The “linear, state coefficients” models are the least plausible: these models assume the deterrent effect of the death penalty (that is, the marginal effect of executions on the decision to commit homicide) varies by state. As they point out, that is a bit like assuming that the treatment effect of a drug is different in Texas than in other states. In general, linear models are less plausible than logistic models, which assume a functional form more appropriate to the data.
All of the logistic models show a net lives lost effect from capital punishment. Varying the other model details seem not to make a big difference to the results. However, the authors also calculate the posterior probability of the model’s being true given its assumptions and the data, and model 16 above comes out best. This model yields one of the highest “negative deterrent” effects of capital punishment.
In summary, the evidence leads me to believe that capital punishment does not, on net, save lives. It may even cost lives through a kind of “brutalization” process. This information is highly relevant to the normative policy implications.