Probabilistic Audits and Misreporting – the Influence of Audit Process Design on Employee Behavior

Abstract

We investigate how the design of audit processes influences employees’ reporting decisions. We focus specifically on detective employee audits for which several employees are randomly selected after a defined period to audit their ex-post behavior. We investigate two design features of the audit process, namely, employee anonymity and process transparency, and analyze their impact on misreporting. Overall, we find that both components influence the extent of individuals’ misreporting. A nonanonymous audit decreases performance misreporting more than an audit in which the employee remains anonymous. Furthermore, the high incidence of performance misreporting in the case of anonymous audits can be decreased when the process transparency is low. Thus, our study informs accountants about how the two design features of employee anonymity and transparency of the audit process can be used to constrain performance misreporting to increase the efficiency of audits.

Keywords:

• Performance misreporting
• Employee audits
• Employee anonymity
• Process transparency

JEL Classifications:

• M41
• M42

Acknowledgement

We thank Victor Maas (editor) and two anonymous reviewers for their insightful suggestions, which have significantly improved our paper. We also thank Markus Arnold, Yin Huaxiang, Thorsten Knauer, Marcel van Rinsum, Laura Wang, Arnt Wöhrmann and Xinyu Zhang for their suggestions. This paper has also benefitted from comments and suggestions received at the Experimental Research in Management Accounting Summer School and Conference 2018 and 2019, the Annual Conference for Management Accounting Research 2019, the European Accounting Association Doctoral Colloquium 2019, the American Accounting Association Annual Meeting 2019, and the European Accounting Review 2020 Annual Conference.

Data Availability

Data are available upon request.

Notes

1 Although anonymity can be considered a construct that relates to transparency, here, employee anonymity describes whether the individual can be identified, whereas process transparency describes whether the individual can gather information about the audit process. Therefore, even if anonymity is considered a transparency-related construct, the direction of the transparency is different, which is why both constructs do not overlap.

2 We assume that a nonanonymous, i.e., onsite audit process is, per se, more transparent and enables employees to gather more information; therefore, we do not distinguish here between low and high process transparency.

3 We focus on random audits with a low detection probability. We assume that our findings will also hold in settings with a higher detection probability as long as the expected value of misreporting is positive. Moreover, we assume that process transparency and anonymity will similarly affect audits where employees are selected due to a suspicion caused by, e.g., an exceptionally high performance or significant jumps in performance. Again, process transparency enables employees to draw conclusions about who will be audited and why. Additionally, reduced anonymity is expected to increase the ethical costs of misreporting when facing the auditor.

4 The experiment was approved by the large European university where the experiment was conducted.

5 We refer to the low detection probability that is more common in practice (Morton, 1993). The Association of Certified Fraud Examiners (2018) states that 15 percent of all noncompliance incidents are detected by internal audits; this finding is similar to the results obtained by our experimental design.

6 Our predictions remain as long as the expected value of misreporting is positive and misreporting is economically beneficial. In our setting, this is the case as long as the detection probability is below 66.67 percent.

7 We would expect even stronger effects if, for instance, the results of the audit were also disclosed.

8 To avoid the distortion of effects, we held the person representing the employee of the head office (i.e., the auditor) and the person representing the experimenter constant across conditions. Moreover, the auditor and the experimenter were two different persons.

9 $ExpectedValueofMisreportingperRound=\left(1-0,2\right)\times 0.5\times 300eurocents=120eurocents$, with 0.2 as the audit probability, 0.5 as the level of profit share and 300 as the maximally reported project earnings per round. Reporting the actual earnings of 200 euro cents leads to a secure compensation of 100 euro cents per round $\left(0.5\times 200eurocents\right)$. Accordingly, the expected value of misreporting is higher than the safe payment.

10 In contrast, misreporting would not be advantageous when the expected value of misreporting is lower than the secure compensation; this is the case under an audit probability higher than 66.67 percent, ceteris paribus. Thus, although the participants did not know the exact audit probability, the information that the audit probability is low should enable them to conclude that misreporting is economically beneficial, specifically, if they assume that a low audit probability means a value of less than 0.5.

11 Because we explicitly investigated employee anonymity and the transparency of the audit process, we never publicized whether misreporting was actually discovered, since this would have different behavioral implications.

12 This time limit was validated in prior experiments and in a pretest.

13 We refer to a sample of 85 students, as explained in the validation of the experimental design.

14 Therefore, in contrast to Lill (2020), there were no significant differences in the participants’ perception of the detection probability at the end of the experiment.

15 Three participants out of 88, who were distributed over the sessions, answered the PEQ without (carefully) reading it and answered the manipulation check questions incorrectly. We excluded these participants’ observations; thus, the sample for our analysis contains the observations of the remaining 85 students. Overall, the results presented are almost identical to the results for the sample including the three participants.

16 Since our dependent variable was already fully surveyed at the time, there are no issues regarding possible distortions or demand effects concerning presenting the participants with the other manipulations.

17 The participants had to assess the statement that they understood in detail how the selection process for the audit worked both ex ante and ex post to the experiment. For the ex-ante assessment, we do not find a significant difference (t = 0.82, p > 0.1, one-tailed). In contrast, when asking the same question again afterward and, therefore, after the actual experience of the audit process and its features, we find statistically significant differences between conditions (t = −1.87, p < 0.05, one-tailed), which supports our reasoning.

18 In addition, we perform the ANOVA with the number of audits in the previous round as a covariate. Since the results are inherently the same and the covariate is not significant (F = 0.05, p > 0.1, two-tailed), we do not include the covariate in our main analysis.

19 In our setting, the cumulative probability of being audited at least once within x rounds is calculated as $p\left(beingaudited\ge 1\right)=1-(1-0.2{)}^x$, with 0.2 as the audit probability and ($1-0.2$) as the probability of not being audited.