Disclosure Checklists and Auditors’ Judgments of Aggressive Accounting

Abstract

This study investigates if auditors who feel accountable to management (as opposed to the audit committee) are more susceptible to pro-client bias after using a disclosure checklist. We theorize that the use of a disclosure checklist, even though it is uninformative about the aggressiveness of the accounting methods used, can influence auditors’ judgments of the acceptability of aggressive reporting by inducing a less critical state of mind. We propose that this less critical state of mind is reflected in higher levels of pro-client bias, particularly when management is the appointing party. Our experimental data for two cases, obtained from experienced auditors working at a Big Four audit firm, support this prediction. Our findings imply that threats to auditor independence are more subtle than has often been assumed.

Acknowledgements

This paper has benefitted from comments and suggestions of seminar participants at the 2014 ISAR conference in Maastricht, the 2013 AAA Auditing Section Midyear Meeting in New Orleans, the 2013 European Accounting Association conference in Paris and seminar presentations at Pablo de Olavide University, the University of Amsterdam and Tilburg University. We would like to thank Laurence van Lent (editor) and an anonymous reviewer for their help and valuable comments. We would also like to thank Sanjay Bissessur, Jan Bouwens, Willem Buijink, Bart Dierynck, Marcus Doxey, Anna Gold, Hans Gortemaker, Kris Hoang, Kathryn Kadous, Christopher Koch, Stephan Kramer, Steve Salterio and Philip Wallage for help and comments on previous versions of this manuscript.

Notes

1 Following Moore et al. (2010) and Koch et al. (2012) we designed our experiment to test whether auditors who are provided with the same financial statements reach different conclusions when they feel accountable to different parties. We exploit the opportunity provided by the corporate governance regime in the Netherlands where external auditors can be hired by either a company’s management board or an audit committee consisting of members of the supervisory board. However, it is important to emphasize that our theoretical expectations extend to situations in which the influence on the hiring and firing of external auditors is not a dichotomous variable. As Cohen et al. (2010) and Dhaliwal et al. (2015) show, for example, even in a one-tier context in the post-Sarbanes-Oxley era, management influence on auditor selection decisions varies cross-sectionally and is often substantial.

2 We predict this will hold in a situation in which an accounting treatment complies with minimal disclosure requirements.

3 More information about the corporate governance system in the Netherlands can be found on the website of the Dutch Corporate Governance Code Monitoring Committee: http://www.mccg.nl/.

4 This approach implies that the average time taken by participants with a checklist exceeds that of those without a checklist, possibly also increasing processing depth. From a theoretical perspective, if anything, deeper processing potentially works against finding less skeptical judgments. Moreover, time spent or processing depth cannot explain an interaction effect, detailing that auditors respond more leniently to checklist use when management is the appointing party. From an experimental design perspective, keeping the average time allocated to the entire task constant would only be possible by introducing a pause or an unrelated in-between task of similar duration to the no checklist conditions. Doing so, however, would give rise to various potentially problematic effects, including a distraction or emotional effect of the unrelated task, and boredom or annoyance in case of a pause period. Finally, and most importantly, when a checklist is introduced in practice, this extends the audit process and thus also automatically implies more time is needed. Therefore, our manipulation mimics auditing firms' practical audit process design choice whether to use a checklist or not.

5 The checklist for the first case contained one item (IAS 36.130e) that was irrelevant and the checklist for the second case contained three irrelevant items (IAS 18.14a, b and c). For these items the box N/A should be checked instead of the box Yes. Including a few irrelevant items should reduce participants' tendency to mechanically tick the Yes boxes and thus work against us finding support for our hypothesis.

6 Nelson further concludes that: ‘(…) studies provide evidence that auditors are more likely to make judgments that exhibit more professional skepticism (i.e., are less willing to allow aggressive reporting) as the balance of incentives tips (…) away from concerns about client importance (…)’. (Nelson, 2009, p. 13, italics added). Moreover, Kim and Trotman (2015) verify empirically that audit judgments provide one method of measuring skepticism.

7 The judged acceptability of an accounting method (i.e. our dependent variable) is somewhat broader than the judged probability of a material misstatement. Using this broader measure allows us to capture the participants' assessment of the accounting method without adding an additional layer of complexity by introducing specific materiality levels. There can be reasons why an auditor would find an accounting treatment, though not materially misstated, less acceptable. For instance, managers could manipulate certain figures only to increase their bonus, which may not necessarily directly affect misstatement likelihood. Thus, our variable is very suitable to studying pro-client bias.

8 The data furthermore show that all participants in the LIST conditions actually did use the checklists and indicated whether the specified items were disclosed. Of the 14 items in the checklist for the first case (impairment of patent) participants on average indicated that 2.15 (SD = 2.30) items were not disclosed. Of the 11 items in the checklist for the second case (construction contract) participants on average found 2.78 (SD = 2.82) items were not disclosed. If we exclude the items for which ‘N/A’ should have been ticked (items that indeed were not disclosed, but also did not need to be because they were not applicable), the mean numbers decline to 1.67 and 1.81, respectively, or 12% and 16% of the items in each checklist. Thus, on average at least 84% of the items were checked off as disclosed, providing the participants with a relatively positive view on disclosure status as intended. The number of undisclosed items according to the participants’ completed checklists is not correlated with the acceptability ratings of the accounting methods (i.e. the DVs), nor is the number of times a No box was ticked correlated with age, experience, gender or rank (all p > .1). Importantly, the number of checklist items flagged as (not) disclosed is also not significantly different in the management and the audit committee conditions.

9 Despite the general effectiveness of the manipulations, seven participants answered one of the two manipulation check questions incorrectly. Of these seven participants, four indicated that they at least somewhat agreed with the statement that they had used a checklist, while they were actually in a NOLIST condition. One possibility is that these participants were well aware that they did not use a provided IFRS disclosure checklist but that in deciding about the acceptability of the described accounting methods they used a self-developed (mental) checklist. We retain the data from these participants for our hypothesis tests. However, if we test the hypothesis excluding these seven cases the results are inferentially identical. We also ran full-factorial ANOVAs for both manipulation check questions. The results (not tabulated) show that the interaction between the appointing party and checklist use is not significant and that only the corresponding main effect is significant, which further validates our manipulations.

10 The results are inferentially identical, and our conclusions regarding the hypothesis do not differ, if we exclude either or both of these covariates from our analyses. Auditor rank has also been identified as a possible antecedent of skepticism in prior research (Hurtt et al., 2013, p. 68). If we include this variable as a covariate, we again find inferentially identical results, regardless of whether we omit experience.

11 As a robustness check, we re-analyzed our ANCOVAs using a newly defined variable, based on checklist use and how the checklist was filled out. We assigned this variable a value of 0 when no checklist was used, a value of 1 in case a checklist had at least one ‘No’ checkbox ticked, and a value of 2 for checklists filled out without any ‘Nos’. These values were chosen to ensure that a higher number corresponds to more ‘Yes’ ticking, as we theorized that this may lead to less skeptical judgments. We ran ANCOVAs for both cases with this variable replacing the original checklist dummy variable. We find that our results are inferentially identical. The interaction effects between the new checklist variable and appointing party are still significant for both cases (DV1 and DV2, both p < .10), and the pattern of results is similar to the ones displayed in Figures 1 and 2, with only small differences between the cases without any ‘Nos’ and those with at least one ‘No’.