I made the mistake of trying to rip off a quick post in a hurry and ended up not even satisfying me with the post below on the death penalty and the problems, not just with the studies cited (hopefully well by the reporters) but also with having economists once again sticking their noses into places they don't understand (which in my long experience is most of the world). The whole premise behind the death penalty being a deterrent is that people will see someone executed and rethink their own criminal behavior. So, the first thing any design should do is not run some aggregate numbers (and please don't think using county instead of state data relieves that problem). The first thing is . . . to find out if anyone even knows there's an execution going on. Maybe in states with few executions, it would be a big media event and affect people's thinking, but I can tell you that here in OK and I suspect in TX (Grits, help me out) that news gets wedged in local media between the latest car chase and the bulldog that can climb a tree. I would frankly be surprised if one person in ten could tell you much of anything at all about an execution in OK.
So, just do a friggin' survey to find out if and how much people know about the execution before you start crunching numbers on the assumption that they know enough to do detailed cost-benefit analysis, or even undetailed cost-benefit. Establish that as your baseline. Do some focus groups, interview guys after the execution who did go on to commit crimes to find out why the execution didn't affect them, interview other guys a year later to find out why they held off for a year or more before deciding to commit the next crime, whatever. Find out what the state of knowledge is before you start basing conclusions on people having a hypothesized state of knowledge. But that would require a commitment to understanding the reality of a situation rather than the models that economics as a discipline prefers to gambol in. It does no good to use "real world" data if you insert it into a model of human behavior that is still unfounded.
It wouldn't be hard to do a real analysis along the lines I just described. Lord knows there are states you could do it in and you might even be able to find distinct states in the same time period that do few, some, and many executions in a year. (The current "moratorium" may put a crimp on it for a little while.) But the studies reported sound way too much like a case of "give a kid a hammer" and applying statistical methods blindly. I can use the same approach to show you that increased eating of ice cream led to increased mental illness.
Really, I can.