[Current policy on how auditors should limit uncertainty about misstatements in auditee assertions is based on the audit risk model (AICPA 1992) that decomposes the components of audit risk as inherent risk (IR), control risk (CR), and detection risk (DR). The literature on the audit risk model has focused on a priori analyses of the model's assumptions and implications (see, e.g., Cushing and Loebbecke 1983; Kinney 1983, 1989, 1992; Leslie 1984), and auditors' risk assessments in experimental settings (see, e.g., Colbert 1988; Daniel 1988; Jiambalvo and Waller 1984; Libby et al. 1985). Absent are empirical studies that examine applications of the model in field settings. This article reports empirical evidence on auditors' IR and CR assessments in field settings by analyzing archival data drawn from the audit workpapers of KPMG Peat Marwick. As a part of audit planning, the firm requires its auditors to make and document IR and CR assessments for each assertion of each significant account.1 The assessments are made with respect to tolerable error, an algorithm-based measure of planning materiality at the assertion level.2 The data include approximately 5,000 risk assessments at the assertion level for trade accounts receivable, inventory, and trade accounts payable, on a total of 215 audit engagements. The data also indicate, for each assertion, whether a misstatement exceeding tolerable error was detected by the auditor. The data analysis considers four issues, the first of which is whether there is a statistical association between auditors' IR and CR assessments. A priori researchers (e.g., Cushing and Loebbecke 1983) argue that the audit risk model's multiplicative combination of IR and CR suggests independence between risk components, which contradicts auditors' conventional wisdom of dependence. The analysis in this study concludes that the dependence problem arises because (1) its event structure is ill-defined and (2) it fails to recognize that an auditor's assessments are conditional on his or her knowledge. A knowledge-based dependence may produce a statistical association between IR and CR. Contrary to expectation, the empirical evidence supports the conclusion of an insignificant association between IR and CR; however, in a predominance of cases, CR is assessed at the maximum probably for reasons of efficiency. The remaining issues pertain to the policy requirement of assertion-level risk assessments. Viewed generally over many audit engagements, current policy depends on the following premises: (1) the rate of misstatements varies over assertions, (2) auditors' risk assessments vary over assertions, and (3) the association between the rate of misstatements and auditors' risk assessments is positive (i.e., the assessments are accurate). The data analysis examines the empirical validity of each premise, and supports the first inasmuch as there are significant differences in the rate of detected misstatements over assertions for each account. However, auditors typically assess IR and CR at the same value for all assertions for an account, which is inconsistent with the second premise. When the data pertaining to all assertions for an account are included in the analysis, the association between IR and the rate of detected misstatements (after controlling for CR and DR) tends to be positive but low, which indicates modest support for the third premise. When the analysis includes only the "most important" assertion for each account, the association is considerably stronger. In general, auditors' risk assessments are consistent with a heuristic that deliberately assesses risk for an account's "most important" assertion but does so mechanically for other assertions. Taken as a whole, the results indicate a need either to reconsider the policy of multiple risk assessments for an account or to enhance auditors' ability to assess assertion-specific risk.]

William S. Waller, Self-Selection and the Probability of Quitting: A Contracting Approach to Employee Turnover in Public Accounting, Journal of Accounting Research, Vol. 23, No. 2 (Autumn, 1985), pp. 817-828

[This study reports on two experiments examining the effects of alternative incentive pay schemes for controlling unit manager behavior in intrafirm resource allocation settings. Two control problems are addressed: unit managers' misrepresentation of private information to the central manager prior to allocation, and unit managers' consumption rather than investment of resources subsequent to allocation. The experimental setting was adapted from Groves and Loeb (1979). The firm consists of central management and two units. The role of the (mechanized) central manager is to maximize firm profit by acquiring a common resource and allocating it to the units. The central manager also executes a control mechanism consisting of a performance measure and pay function for the unit managers. The role of each unit manager is to send a message to the central manager before the resource is acquired and to generate profit by investing allocated resources in productive activities. Actual unit profit depends on the unit's profit function and invested resources, which are known only to the unit manager. The linear profit function was in the form of a productivity ratio, i.e., the ratio of outputs to inputs. The central manager learns each unit's actual profit when realized. In both experiments, students served as subjects. Experiment 1 focused on unit managers' misrepresentation, which was measured by the difference between projected and actual p-ratios, under three incentive schemes: (1) unit profit scheme-a unit manager's pay is linear in actual unit profit, (2) unit profit-plus-penalty scheme-a unit manager's pay is linear in actual unit profit except that there is a "large" penalty when an unfavorable profit variance occurs, and (3) Groves scheme-a unit manager's pay is linear in the sum of his or her unit's actual profit and the other unit's budgeted profit. As predicted, misrepresentation was higher under the unit profit scheme than under the other schemes. Contrary to prediction, misrepresentation was higher under the Groves scheme than under the unit profit-plus-penalty scheme. The latter result may have been due to either or both of two factors associated with the Groves scheme. First, some subjects may have tried to gain from tacit collusion. Second, some subjects may have failed to understand the scheme's incentives, given noncooperation. Experiment 2 focused on resource consumption, which was measured by inputs exchanged directly for cash rather than invested, under the Groves and unit profit-plus-penalty schemes. As predicted, resource consumption was higher under the Groves scheme than under the unit profit-plus-penalty scheme. This result was largely due to the latter scheme's penalty, which forces a unit manager to invest enough resources to achieve budgeted unit profit.]

[When making audit judgments and decisions, an auditor often relies on knowledge or information about how task variables covary. Typically, an auditor generates this information without the aid of a formal covariation model. Thus, the quality of audit judgments and decisions which rely on covariation information may depend on an auditor's ability to judge covariation in a manner that is paramorphic to a formal model. This paper reports two experiments examining the rules by which auditors integrate joint frequency data when making covariation judgments and whether their judgments are affected by context, prior expectations, and amount of auditing experience. A main result was that the subjects generally used data-integration rules that were sensitive to the objective covariation level, but often overstated or understated this level. The effects of context, prior expectations, and amount of auditing experience were generally small.]

[Auditors' self-perceptions of their judgment ability may affect how well they learn from experience and how much they rely on decision aids. This study tests the hypothesis that incomplete outcome feedback causes auditors' self-perceived judgment ability to be affected by factors that are not necessarily related to actual judgment ability. The results of an experiment that used professional auditors as subjects making internal control judgments were, for the most part, consistent with this hypothesis.]

Current policy on how auditors should limit uncertainty about misstatements in auditee assertions is based on the audit risk model (AICPA 1992) that decomposes the components of audit risk as inherent risk (IR), control risk (CR), and detection risk (DR). The literature on the audit risk model has focused on a priori analyses of the model's assumptions and implications (see, e.g., Cushing and Loebbecke 1983; Kinney 1983, 1989, 1992; Leslie 1984), and auditors' risk assessments in experimental settings (see, e.g., Colbert 1988; Daniel 1988; Jiambalvo and Waller 1984; Libby et al. 1985). Absent are empirical studies that examine applications of the model in field settings. This article reports empirical evidence on auditors' IR and CR assessments in field settings by analyzing archival data drawn from the audit workpapers of KPMG Peat Marwick. As a part of audit planning, the firm requires its auditors to make and document IR and CR assessments for each assertion of each significant account. The assessments are made with respect to tolerable error, an algorithm-based measure of planning materiality at the assertion level. The data include approximately 5,000 risk assessments at the assertion level for trade accounts receivable, inventory, and trade accounts payable, on a total of 215 audit engagements. The data also indicate, for each assertion, whether a misstatement exceeding tolerable error was detected by the auditor. The data analysis considers four issues, the first of which is whether there is a statistical association between auditors' IR and CR assessments. A priori researchers (e.g., Cushing and Loebbecke 1983) argue that the audit risk model's multiplicative combination of IR and CR suggests independence between risk components, which contradicts auditors' conventional wisdom of dependence. The analysis in this study concludes that the dependence problem arises because (1) its event structure is ill-defined and (2) it fails to recognize that an auditor's assessments are conditional on his or her knowledge. A knowledge-based dependence may produce a statistical association between IR and CR. Contrary to expectation, the empirical evidence supports the conclusion of an insignificant association between IR and CR however, in a predominance of cases, CR is assessed at the maximum probably for reasons of efficiency. The remaining issues pertain to the policy requirement of assertion-level risk assessments. Viewed generally over many audit engagements, current policy depends on the following premises: (1) the rate of misstatements varies over assertions, (2) auditors' risk assessments vary over assertions, and (3) the association between the rate of misstatements and auditors' risk assessments is positive (i.e., the assessments are accurate). The data analysis examines the empirical validity of each premise, and supports the first inasmuch as there are significant differences in the rate of detected misstatements over assertions for each account. However, auditors typically assess IR and CR at the same value for all assertions for an account, which is inconsistent with the second premise. When the data pertaining to all assertions for an account are included in the analysis, the association between IR and the rate of detected misstatements (after controlling for CR and DR) tends to be positive but low, which indicates modest support for the third premise. When the analysis includes only the "most important" assertion for each account, the association is considerably stronger. In general, auditors' risk assessments are consistent with a heuristic that deliberately assesses risk for an account's "most important" assertion but does so mechanically for other assertions. Taken as a whole, the results indicate a need either to reconsider the policy of multiple risk assessments for an account or to enhance auditors' ability to assess assertion-specific risk.

[This paper provides empirical evidence on a truth-inducing pay scheme widely discussed and analyzed in the incentive contracting literature. An experiment was conducted in which subjects acted as subordinates who performed a production task. Budgets were participatively set under either a truth-inducing or slack-inducing pay scheme and either the presence or absence of a superior-subordinate information asymmetry about subordinate performance capability. Slack was defined as expected performance minus the participatively set budget. The results showed that, when the information asymmetry was absent, slack did not differ significantly between the pay schemes. However, when the information asymmetry was present, slack was significantly lower under the truth-inducing scheme. Similarly, the pay scheme and information asymmetry variables interacted to affect performance.]