This paper presents and applies a methodology--"model averaging," which uses weighted averages of a wide set of possible models--that integrates the various statistical studies of the deterrent effects of capital punishment into a single coherent analysis.
The application of this methodology found little evidence of a deterrent effect for capital punishment. Statistical analyses that claim a deterrent effect for capital punishment are apparently based on an inadvertent selection of individual, and highly unlikely, models. Using the methodology of "model averaging," the authors were unable to identify a reasonably sized model space that produced even a single significant negative coefficient on one of the three deterrence variables. This report shows model-averaged coefficients that fail to support the link between deterrence and capital punishment. These are the synthesis of thousands of potential specifications. Existing research on the deterrent effect of capital punishment comes to differing conclusions based upon one or more underlying assumptions that call into question the ability of any single model to explain the impact of capital punishment laws. Such dependence on the specifics of research design--from data cleaning, to aggregation, to model choice--is the basis for using averaging techniques. Since relatively minor variations in model or variable choice can lead to significant changes in conclusions, the inclusion of the information content of all the models lends itself to conclusions upon which policymakers can be more confident. "Model averaging" achieves this. A 50-item bibliography, extensive tables and figures, and appended codebook
Date Published: December 1, 2006