Power Characteristics of Test Statistics in the Auditing Environment: An Empirical Study

Abstract

The increasing use of statistical sampling in auditing has led to a number of studies to evaluate alternative estimators (e.g., Kaplan [1973] and Neter and Loebbecke [1975]) and to consider alternative statistical testing procedures which might be used in auditing (e.g., Elliott and Rogers [1972] and Roberts [1974; 1978]). The latter studies focus on the two types of decision errors which the auditor might commit. The power curve for a statistical decision rule describes the magnitude of the risks of making these decision errors, given different total error amounts in the population. It is important for auditors to know whether, and under what conditions, the risks of making the two types of incorrect decisions are excessively high for different statistical test statistics, and whether the power characteristics differ for various types of accounting populations and statistical test statistics. A major difficulty in studying the power curve of a statistical decision rule in auditing is that a given total error amount in a population can arise in several ways-for example, as a result of a few large errors or as a result of many small errors. A model is therefore required for generating errors in a systematic and consistent fashion to study the power at different total error amounts. Except for a recent study by Beck [1980] of the regression estimator, the power characteristics of the different statistical procedures commonly