CAV Bounds in Dollar Unit Sampling: Some Simulation Results

This article presents the results of a simulation designed to examine the behavior of five procedures for computing a combined attributes and variables bound from a dollar unit sample. The behavior of five computational methods is simulated for three population error direction conditions, seven error rate conditions, and five sample size conditions. The study population is generated from summary data provided in Auditing Research Monograph No. 2 (ARM #2). Results are analyzed according to reliability, accuracy, and consistency criteria. In addition, results are compared to the classical estimators examined in ARM #2, applied to a very similar population. The results suggest that certain CAV bounds are unreliable, while others are preferable to classical estimators for relatively low error rates in this population. The CAV bound used in ARM #2 is shown to be inferior to two alternatives based upon accuracy and consistency criteria.