Sampling error in first order stochastic dominance

This study analyzes the standard method of testing for first order stochastic dominance from a statistical viewpoint and applies a boundary crossing algorithm to approximate the resulting error probabilities. Error probabilities can be estimated even when the two distributions are not equal. This approach, which is useful when large sample simulations are not feasible, helps clarify some of the unusual results obtained by Kroll and Levy in an earlier study.