Steven D. Levitt believes “that Wolfers” (who criticized the study I wrote about yesterday) “is on the right side of this debate.” Steven goes on to write:

There are recent studies of the death penalty — most bad, but some reasonable — that find it has a deterrent effect on crime. Wolfers and John Donohue published an article in the Stanford Law Review two years ago that decimated most of the research on the subject…

Analyses of data stretching farther back in time, when there were many more executions and thus more opportunities to test the hypothesis, are far less charitable to death penalty advocates. On top of that, as we wrote in Freakonomics, if you do back-of-the-envelope calculations, it becomes clear that no rational criminal should be deterred by the death penalty, since the punishment is too distant and too unlikely to merit much attention. As such, economists who argue that the death penalty works are put in the uncomfortable position of having to argue that criminals are irrationally overreacting when they are deterred by it.

In this regard, it might also be interesting for you all to take a look at these stats (comparing homicide rates in death penalty and non-death penalty states):

A new survey by the New York Times found that states without the death penalty have lower homicide rates than states with the death penalty. The Times reports that ten of the twelve states without the death penalty have homicide rates below the national average, whereas half of the states with the death penalty have homicide rates above…

When comparisons are made between states with the death penalty and states without, the majority of death penalty states show murder rates higher than non-death penalty states. The average of murder rates per 100,000 population in 1999 among death penalty states was 5.5, whereas the average of murder rates among non-death penalty states was only 3.6.

H/t Holly for forwarding me the links to these (and other) articles.

Also: reader Mikkel left a comment in yesterday’s thread, which might be of interest:

So I downloaded the Emory paper. To say that I don’t believe it is an understatement. Not only are there so few states that have enough executions that I think you could have any chance of showing any correlation (even if you had a perfect idea of what to look at) but near the top of the paper is a figure that in my eyes tends to discount the whole idea.

It’s a graph of executing vs. non-executing states from 77-99. The non-executing goes (per 100k, 5 year increments) 6-7-6-9-8-6. The executing goes 9-9-8-9-8-6. They have TONS of variables in the model, but I can tell you right now that the reason why they got the result they did was because executions increased in the 90s and that’s when the executing states started tracking the non-executing ones.

The paper was pretty dense and I couldn’t figure out how they got the conclusion they did. I think all they did was make a model of all the variables and then did a regression analysis and saw how much the “death penalty” variable mattered.

This is the chief difference between modelling economics and say…biological systems. Doing regression (even if the equations are properly formed) says almost nothing predictively speaking. The difference between this and global warming models is that the global warming ones are created and they use regression to get a feel for the values to set for the variables…but then they do a “free-run” which means they set the model to an initial state and let it run its course and see how well it matches data they already know.

As far as I could tell, they didn’t do this last step so there is no way of telling if their model could predict the murder rate of the 90s based on the laws of the 80s for instance.

They also assume all variables are linear. This means that all variables changes in variables are proportional (for instance, doubling the death penalty doubles the amount of negative deterrence while increasing economic conditions by two decreases its effect). This assumption was made for statistical reasons, not because it necessarily reflects reality.

So in general, in order for the paper’s conclusion to be “right” (as in real, I’m not saying that the paper is wrong in any math or anything) a) crime would have to work under an economic model b) they would have to properly disassociate their variables (of which there were a ton) c) the variables must be linear d) the regression would be proper for looking in the past and future. [I also couldn’t tell how much was done to account for general trends. I saw it in there but again it was too dense to tell.]

Even if all that was true, only ten states had more than 20 executions so I would be loathe to even think about having it apply in general to the other forty.

And the debate carries on…