Selection bias in audit firm tenure research

• Published in: Review of accounting studies Vol. 29; no. 4; pp. 3085 - 3129
• Main Authors: Zhou, Ying, Weber, David P., Wen, Ce
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2024; Springer Nature B.V
• Subjects: Accounting firms; Accounting/Auditing; Audit quality; Business and Management; Corporate Finance; Public Finance; Tenure; M48; Auditor tenure; Audit quality; Selection bias; Fixed effects; M42; Auditor turnover; C18
• ISSN: 1380-6653; 1573-7136
• DOI: 10.1007/s11142-023-09787-4

Whether audit firm tenure affects audit quality is a question of longstanding regulatory and scholarly interest. While this question centers on how audit quality changes over time within client-audit firm relationships (longitudinal variation), prior studies tend to instead compare audit quality across relationships of different lengths (cross-sectional variation). We show that conventional pooled cross-sectional tests are subject to selection bias: client-auditor pairs that select into longer versus shorter relationships differ systematically in a variety of ways that also relate to measures of audit quality. These differences exist at the outset of relationships and thus are not attributable to tenure but rather predict the ultimate length of relationships. We then re-examine models of audit quality and tenure after including client-audit firm relationship fixed effects, moving the level of analysis from between clients (cross-sectional) to within client-auditor pairs (longitudinal). We find no evidence that audit quality changes with tenure over time within client-audit firm relationships. The combined evidence suggests that the previously documented association between audit firm tenure and audit quality is likely attributable to selection bias and cannot be taken as evidence of a causal link. More broadly, our results highlight the importance of a design perspective that isolates variation tightly linked to the research question being examined.