On recombinant estimation for experimental data

The recombinant estimation technique of Mullin and Reiley (2006) can be a useful tool for analyzing data from normal-form games. The recombinant estimator falls within a general category of statistics known as U-statistics. This classification has both theoretical and practical implications: (1) the recombinant estimator is optimal (minimum variance) among unbiased estimators, (2) there is a computationally simple method for computing its asymptotic standard error, and (3) the estimation technique can be extended to multiple outcomes and to other types of inferential procedures commonly used for experimental data, such as the sign test. Simulation evidence suggests that researchers should use the asymptotic standard error rather than the standard error of Mullin and Reiley (2006) since the latter exhibits a downward bias. Copyright Economic Science Association 2008