The effectiveness of bootstrap methods in evaluating skewed auditing populations: a simulation study

This article describes a comparison among four bootstrap methods: the percentile, reflective, bootstrap-t, and variance stabilized bootstrap-t using a simple new stabilization procedure. The four methods are employed in constructing upper confidence bounds for the mean error in a wide variety of audit populations. The simulation results indicate that the variance stabilized bootstrap-t bound is to be preferred. It exhibits reliable coverage while maintaining reasonable tightness.