Firm and Office Level Effects on Audit Quality: A Multilevel Approach

Data in auditing research are typically hierarchical in nature. That is, audit firms consist of multiple audit offices in which multiple audit partners are auditing several client firms. Findings of prior auditing research suggest that characteristics of the different units—the audit firm, the audit office, and the audit partner—matter for audit quality. Statistically, that means that the data are clustered within each unit and that the observations for each unit are dependent. Ignoring the dependency of observations creates potential sta tistical problems like deflated standard errors. We introduce multilevel modeling as a means for auditing researchers to address the potential statistical problems that can arise from analyzing hierarchical data with traditional “single-level approaches” (such as OLS). More importantly, we demonstrate that multilevel modeling enables the examination of new research questions by treating the clustering in the data as an interesting phenomenon per se. Using a large dataset from the U.S. audit market (2000–2010), we re- examine the effects of firm-wide and office-specific industry specialization on audit quality (audit fees; discretionary accruals). Further, we investigate how data are clustered at the audit office level to determine whether it matters to which audit firm an audit office belongs to. Our results show that even despite serious clustering of local offices within audit firms, most OLS estimates are reliable and valid. The results also show that most of the va riance in audit quality is situated at the audit firm level; that is, stems from differences between audit firms rather than from differences within audit firms (i.e., between local offices). Overall, the results demonstrate that firm-wide factors should not be disregarded when studying audit quality.