Is the existing research on the effect of imprisonment on crime biased against finding an effect? Is the research cited with such confidence by the let-em-out crowd wrong?

On SSRN is a new study by Thomas Marvell of Justec Research. Here is the abstract (emphasis added):

This is a critical review of the literature concerning the impact of prison populations on crime. It summarizes 44 time series studies that use prison population in the crime equation, emphasizing problems of simultaneity and disaggregation bias. It briefly reviews studies that estimate the incapacitation impact of prisons by using criminals' individual crime rates, emphasizing problems caused by skewness of the crime rates and their relationship with arrest rates. Almost all the numerous problems with prior research bias results towards finding that prisons have limited impacts, and once the problems are addressed the best estimate of the elasticity of prison populations on crime is about 1.0.

Wow. This could be huge.

The final topic is the policy implications of this body of research. Perhaps the most obvious is that economists and criminologists have not been able to provide policy makers with credible estimates of the impact of prisons on crime and with useable advice about whether further prison expansion is worth the costs.  My conclusion is that the elasticity is roughly one, and that prisons are worth the costs, but other researchers believe that the elasticity is much smaller and that crime prevention money is best spent elsewhere.

In California, we have the administration confidently asserting that tens of thousands of prisoners can be released without danger to the public. The three-judge criminals' dream team rules with even greater confidence that even larger numbers can be safely released. But here we have an expert in the field telling us that the prior estimates by other experts are biased.

Jurors often say after the verdict that when the expert witnesses contradict each other, they just ignore them both and go with common sense. Maybe that's what we should do here.

Thanks to Doug Berman for noting this paper.