How Test Power Impacts Research Relevance: The Case of Earnings Management Research

Abstract

We argue that the broader applicability of accounting research is often limited by the way accounting researchers typically place far greater weight on the relative cost of type I versus type II errors. To illustrate the extent of this problem, we examine the performance of simple financial ratio-type analysis for detecting earnings overstatements when the total misclassification costs are minimized subject to the relative cost of type I versus type II errors. We then contrast the likelihood of type I versus type II errors from this approach with those arising from several widely used measures of unexpected accruals. The results demonstrate how commonly-used unexpected accruals measures reduce the type I error rate by sacrificing the type II error rate. Given that accounting information users and auditors typically face much higher costs of type II errors, we explicitly identify why unexpected accruals models are likely far less useful in detecting earnings overstatements than a relatively simple approach using financial statement analysis red flags. Our results highlight the fundamentally contrasting incentives facing accounting researchers relative to those who might otherwise use the research in practice, and serve as a warning when the broader relevance of accounting research is increasingly under question.

Keywords: Research Relevance, Earnings Management, Test Power, Unexpected Accruals

JEL Classification: M41