A sample of 200 studies empirically analyzing deterrence in some way is evaluated. Various methods of data mining (stepwise regressions, Extreme Bounds Analysis, Bayesian Model Averaging, manual and naive selections) are used to explore different influences of various variables on the results of each study. The preliminary results of these methods are tested against each other in a competition of methodology to evaluate their performance in forecasting and fitting the data and to conclude which methods should be favored in an upcoming extensive meta-analysis. It seems to be the case that restrictive methods (which select fewer variables) are to be preferred when predicting data ex ante, and less parsimonious methods (which select more variables) when data has to be fitted (ex post). In the former case forward stepwise regression or Bayesian Model Selection perform very well, whereas backward stepwise regression and Extreme Bounds Analysis are to be preferred in the latter case.
meta analysis data mining deterrence criminometrics