The effects of decision-aid use and reliability on jurors' evaluations of auditor liability.

I. INTRODUCTION

One of the biggest challenges facing the accounting profession is the threat of litigation due to audit failure. Although recent court rulings have brought some relief to the profession by slowing or reversing the trend toward expanded auditor liability (Siliciano 1997; Pacini

In addition to litigation concerns, audit firms face changing market pressures, including increased competition (e.g., the elimination of bans on competitive bidding and advertising in the Code of Professional Conduct), lower audit fees, and the need to offer a dynamic mix of services in response to changing customer demands (Elliott 1998; Melancon 1998). To survive and thrive in this environment, audit firms must continually improve the efficiency and effectiveness of the audit process.

One approach that audit firms have taken to achieve these objectives is to incorporate technology and decision aids (Abdolmohammadi and Usoff 2000). Decision aids have the potential to improve audit quality by increasing the likelihood that auditors will detect and report financial statement misstatements. In addition, decision aids can increase judgment consensus, or the extent to which other auditors would make the same decision under similar circumstances (Ashton and Willingham 1989; Messier 1995). Consensus is particularly important for audit firms defending against claims of negligence, because jurors may evaluate auditors based on the standard of "what other actors (auditors) would do in the same case situation" (Kelley and Michela 1980, 461). Finally, the enhanced documentation that decision aids provide may prove useful in defense of an audit lawsuit (Francis 1994; Lys and Watts 1994).

Despite these potential benefits, some audit firms may be reluctant to use decision aids because of uncertainty about how decision aids will affect their legal liability, and in particular, the fear that decision aids will leave an incriminating audit trail that will come back to haunt the auditor at trial. For example, Anderson et al. (1995, 38) note that decision aids create a veritable "legal minefield" in which "the seeds of subsequent failure will be highlighted." Messier (1995, 214) expressed similar concerns about the legal ramifications of decision aids:

   If an auditing firm makes an expert system available to its auditors and an
   auditor chooses not to use it, will the auditor and firm be held liable if
   some related aspect of the audit is later found defective? Or, suppose an
   auditor overrides a decision aid's recommendation. Will this be viewed as
   evidence of a lack of due professional care? Finally, suppose that an
   expert system makes an incorrect decision. Who is liable and what standard
   should be used to measure the performance of the expert system?

Although these questions are important to the audit profession, surprisingly little research has attempted to address them. Furthermore, the few studies that have investigated the legal implications of decision aids have demonstrated only negative effects from their use (Jennings et al. 1993; Anderson et al. 1995). In other words, the conclusion from existing research would be that decision aids sometimes hurt, but never help, auditors' legal liability.

The current study sheds further light on these issues by investigating audit failures stemming from undetected fraud when (1) the auditor chooses not to use a fraud-prediction decision aid; (2) the auditor uses an aid, then chooses to ignore its prediction of fraud; and (3) the auditor chooses to rely on an aid that incorrectly predicts no fraud. We also include a control condition in which the audit firm has not implemented decision aids, and provide a separate manipulation of the reliability of the decision aid.

Our results suggest that the answers to the questions Messier (1995) posed about the legal liability issues surrounding decision aids not only depend on how the auditor uses the aid during the audit, but also on the reliability of the decision aid. For highly reliable decision aids, jurors attributed more responsibility to the auditor for an audit failure when the auditor ignored a decision aid that predicted fraud, and less when the auditor relied on a decision aid that failed to predict fraud, compared to a control condition in which decision aids were not implemented. Interestingly, jurors' responsibility attributions were unrelated to decision-aid use for the low-reliability conditions, even though this level of reliability reflects the current "state of the art" for fraud prediction decision aids (Bell and Carcello 2000). Our results, in contrast to those from previous research, suggest that decision aids may be advantageous in some situations but disadvantageous in others, depending on how the auditor uses the decision aid during the audit and on the reliability of the decision aid.

In addition to providing more complete evidence about the circumstances that dictate whether decision aids will work to the auditor's benefit or detriment, we also add to the literature by investigating how jurors' responsibility evaluations translate into damage awards against audit firms. Our results confirm Kadous' (2000) conjecture that jurors may consider different factors to determine damage awards than they do to assess responsibility (including the audit firm's financial resources). Understanding how jurors' responsibility attributions translate into damages is particularly important in the context of management fraud, since lawsuits filed against independent auditors frequently involve allegations of management fraud, and these cases are primary contributors to large damage payments by audit firms (Palmrose 1987, 1999).

The remainder of this paper is organized as follows. Section II reviews previous research related to the legal ramifications of decision aids and develops the hypotheses tested in the current study. Section III provides an overview of our research method. In Section IV we present the results of our study. The final section discusses implications, limitations, and directions for future research.

II. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

Previous Research on Fraud Detection and Decision Aids

In 1997, the Auditing Standards Board issued Statement on Auditing Standards 82 (SAS No. 82) to clarify auditors' responsibilities for detecting management fraud (AICPA 1997; Mancino 1997). Although the previous standard (SAS No. 53) required auditors to consider fraud in the overall evaluation of audit risk, it did not distinguish misstatements based on intent. Because audit procedures designed to detect unintentional misstatements (errors) may not be effective at detecting intentional misstatements (fraud), SAS No. 82 requires auditors to evaluate and document the risk of fraud on every audit to provide reasonable assurance that material misstatements are detected (Shibano 1990; Zimbelman 1997).

Even with the new standard in place, it may be difficult for auditors to detect management fraud. Compared to many routine aspects of the audit process, fraud detection is a very complex task. SAS No. 82 identifies more than 30 risk factors that auditors should consider in their evaluations of fraud, but gives little guidance concerning how auditors should combine these "red flags" into an overall assessment of fraud risk. In addition, because intentional errors (fraud) occur much less frequently than unintentional misstatements (errors) (Loebbecke et al. 1989), most auditors will not have had enough experience with fraud to develop elaborate knowledge structures (Bonner 1990). In combination, these findings suggest that fraud detection is a domain in which decision aids can improve the audit process.

Recent studies have investigated decision aids designed to facilitate fraud detection. Bell and Carcello (2000), for example, evaluated several logistic regression models that predict the likelihood of fraud based on the presence or absence of various risk factors identified in SAS No. 82. They found that their best model was more accurate in classifying fraud and nonfraud cases than were practicing auditors, suggesting that audit firms consider using such models as decision aids to help them predict the likelihood of management fraud. Eining et al. (1997) found that this type of decision aid improved auditors' ability to discriminate between fraud and nonfraud cases and resulted in higher consensus among auditors. Consensus is important in a legal defense because jurors may judge auditors according to the standard of what other auditors would do in a similar situation.

From these and other studies, we submit the following observations: (1) the detection of management fraud is an important part of auditors' responsibilities; (2) fraud detection is a complex task with which auditors are likely to struggle due to lack of experience and ill-defined professional guidelines; and (3) although decision aids have the potential to improve the audit process, they also raise a number of complex legal issues that are not yet fully understood. The next section explores these legal implications in more detail and develops hypotheses.

Legal Implications of Decision Aids

Kadous (2000, 339) conceptualizes the juror's task as involving three steps: (1) assessing standards of care, (2) assessing audit quality, and (3) comparing the two to determine auditor responsibility for alleged audit failure. Thus, to fully understand the legal implications of decision aids, we must consider both their effect on audit quality and whether they change the standard of care (i.e., whether they raise the bar) against which auditors will be evaluated. As previously discussed, decision aids have the potential to improve audit quality by increasing the likelihood that auditors will detect and report material misstatements. At the same time, decision aids could shift perceived standards of care by providing a ready benchmark for evaluating auditor performance. For example, Jennings et al. (1993) suggest that in the absence of explicit external standards for evaluating auditor performance, jurists may use the "internal guidance" provided by a decision aid as a surrogate standard against which to judge auditor performance.

The major concern for audit firms would be situations in which standards of care escalate without a corresponding increase in jurors' perceptions of audit quality. For example, failure to use an available decision aid might not affect actual audit quality, but could adversely affect jurors' perceptions about audit quality. Furthermore, the availability of a decision aid may make it salient to jurors (ex post) that the standard of care was not met, thus increasing auditor liability. On the upside, an audit firm could argue that using a decision aid provides sufficient evidence of due professional care, thereby reducing auditor liability.

Anderson et al. (1995) tested these propositions by asking judges how likely they would be to support damages against an audit firm under three decision-aid conditions: no decision aid used, full use of a decision aid, and partial use of a decision aid. As expected, judges were more likely to support damages against the auditor when the auditor only partially used a decision aid than when no decision aid was available. However, full use of the decision aid did not decrease judges' support for damages against the auditor.

Taken at face value, the results of Anderson et al. (1995) present a bleak outlook for audit decision aids, illustrating the negative consequences that can occur when decision aids are implemented but not fully used, without a positive benefit when aids are implemented and fully used. However, an important caveat in interpreting these results is that Anderson et al. (1995) simultaneously varied decision-aid use and the reliability, or "face validity," of the decision aid. More specifically, in the partial use condition, the decision aid generated 15 potential explanations for an unusual financial statement fluctuation (and the auditor investigated only 10), but the aid generated only 10 explanations in the full use condition (and the auditor investigated all of them). Anderson et al. (1995) acknowledge that this confound may have contributed to the insignificant results in the full use condition, and conclude that full use of a lower-quality decision aid may have been insufficient to meet judges' due professional care standards. They suggest that "further research should consider alternative methods of conveying decision-aid competence," such as through probabilistic information obtained as part of the decision-aid validation process (Anderson et al. 1995, 52-53).

The current study disentangles these effects by separately manipulating decision-aid use and reliability. Building on Kadous (2000), we expect that jurors will determine whether the auditor should be held responsible for an audit failure by comparing audit quality to some standard of care. The extent to which the auditor uses a decision aid during the audit is likely to affect jurors' perceptions about audit quality. (1) On the other hand, we expect that jurors will only view a decision aid as an implied standard of care if they perceive that the decision aid is reliable. Thus, our general expectation is that decision-aid use will influence jurors' responsibility attributions only if the aid is reliable:

H1: Jurors' evaluations of auditor responsibility for audit failure will be influenced by the extent to which the auditor uses the decision aid, but only if the aid is reliable.

Next, we break this overall interactive hypothesis into three specific hypotheses (H1a, H1b, and H1c) that correspond to the decision aid scenarios described by Messier (1995). In each case, we compare jurors' responsibility evaluations to a control condition in which no decision aid was available to the auditor.

One of the scenarios Messier (1995, 214) described was a situation in which "an auditor overrides a decision aid's recommendation. Will this be viewed as evidence of a lack of due professional care?" We expect the answer to this question will depend on whether jurors believe the decision aid is reliable. Specifically, we expect that overriding the recommendation of a reliable decision aid will make it salient to jurors that the standard of care was not met, thereby increasing jurors' attribution of responsibility to the auditor:

H1a: Jurors will assess higher auditor responsibility for audit failure when the auditor overrides the recommendation of a decision aid than when no decision aid is available, but only if the aid is reliable.

The next situation Messier (1995, 214) described was one in which "an expert system makes an incorrect decision. Who is liable and what standard should be used to measure the performance of the expert system?" In other words, will the fact that the auditor followed the recommendation of a decision aid (even if incorrect) provide sufficient evidence that the standard of care was met? Once again, we believe the answer will depend on whether the aid is reliable:

H1b: Jurors will assess lower auditor responsibility for an audit failure when the auditor relies on an incorrect decision aid than when no decision aid is available, but only if the aid is reliable.

Finally, we examine the situation in which "an auditing firm makes an expert system available to an auditor and an auditor chooses not to use it" (Messier 1995, 214). Our expectation is that jurors will penalize auditors for failing to use a reliable decision aid:

H1c: Jurors will assess higher auditor responsibility for audit failure when the auditor fails to utilize a decision aid than when no decision aid is available, but only if the aid is reliable.

Next, we investigate how jurors' responsibility assessments translate into damage awards against the auditor. In audit liability cases, jurors must determine whether the audit firm (defendant) is responsible for losses incurred by the claimant (plaintiff) that result from an audit failure. If jurors attribute the cause of the loss to the auditor's actions, as opposed to other potential causes such as the situation or the actions of the claimant, then they are likely to assess damages against the auditor. Thus, we expect jurors' damage awards to be positively related to their assessment of the auditor's responsibility for the audit failure. In addition, other factors may influence jurors' assessment of damages, independent of their assessment of responsibility. As Kadous (2000, 340) notes, "once a juror has determined that an auditor is financially responsible for a plaintiff's loss, the juror will use a second set of factors to determine the amount of the liability." She suggests that although these factors may overlap with the factors that jurors use to evaluate responsibility, other factors, such as the dollar value of the plaintiff's loss, the auditor's ability to pay, and the availability of additional defendants, may influence damage awards even if they do not influence evaluations of responsibility.

In this study, we investigate whether the auditor's ability to pay (proxied by audit firm size) is an additional determinant of damage awards. Although prior research has suggested that litigation against independent auditors is often motivated by larger firms' perceived "deep pockets," (Carcello and Palmrose 1994; Lennox 1999), previous research on the relation between audit firm size and audit litigation rates has produced mixed results (Palmrose 1988; Stice 1991; Bonner et al. 1998). (2) These prior studies investigated the relation between audit firm size and the incidence of litigation, as opposed to the eventual outcome of the litigation. (3) The incidence of litigation against large audit firms likely depends on more than their perceived deep pockets. For instance, large audit firms arguably perform a higher quality audit and have a greater incentive to avoid litigation than small firms (Dye 1993; Latham and Linville 1998). These factors would tend to mitigate the effect of large audit firms' deep pockets on the incidence of litigation.

To our knowledge, prior research has not investigated the effect of audit firm size on the amount of damages jurors award for audit failures. Once jurors have determined an outcome, they consider the damages that should be assessed against the defendant audit firm to compensate the injured party. At that point, the firm's ability to pay may become an important factor in the award of damages against the audit firm (Kadous 2000). This may be particularly true for jurors (as compared to judges), because jurors may be more prone to bias from extraneous factors, such as the defendant's resources (Palmrose 1991; Cloyd et al. 1998). This reasoning leads to the following hypotheses:

H2: Jurors' damage awards will be positively related to their assessment of the auditor's responsibility for the audit failure.

H3: Jurors will assess higher damage awards against large audit firms than against small audit firms.

III. METHOD

Participants

We conducted an experiment with 149 jurors who were waiting to be called for jury service. (4) Although most audit litigation cases are settled out of court, those that make it to trial are more often decided by jury than by judge, and typically have a lower auditor success rate than judge trials (Palmrose 1991). Jurors completed the case instrument in quiet conference rooms and were subsequently compensated $5.00. Sixty-nine (46.3 percent) of the participants were male and 80 (53.7 percent) were female. The pool of jurors was diverse, with an average age of 40.6 (standard deviation = 13.5), and a wide range of income and education levels.

Task

Participants read and answered questions about a hypothetical audit lawsuit in which an investor was suing a public accounting firm for alleged audit failure. We used separate envelopes for each part of the task to prevent jurors from looking back. The case instrument began with an overview of the audit process including a description of Generally Accepted Auditing Standards. We also explained the auditors' responsibilities for assessing client risk and their overall concern for management fraud. This introductory material educated jurors (as would be done in an actual trial) prior to distribution of information specific to the case.

The case focused on the audit of a publicly traded client in the medical technology industry that had recently experienced a favorable earnings trend. We described the audit firm and the approach the firm and audit manager took in using decision aids in the conduct of the audit (described in the "Design" section below). After performing the audit, the audit firm concluded that the potential for fraud was low and therefore did not perform additional audit procedures. The firm subsequently issued an unqualified audit opinion. Based on the positive trend in earnings in the audited financial statements, an investor purchased 200,000 shares of the firm's stock. It was later revealed that the client had fraudulently overstated net income for the last few years, and that in reality the company was barely breaking even. This discovery caused the client's stock price to drop and the investor (plaintiff) who had relied on the audited financial statements brought a lawsuit against the audit firm. The audit firm claimed that the audit was planned and executed with due professional care and that the court should not hold the firm responsible for failing to detect the fraud.

Design

We used a 4 x 2 x 2 design, with the following between-subjects variables: decision-aid use (four levels), decision-aid reliability (two levels), and audit firm size (two levels). The decision-aid use conditions included a control group in which the firm had not implemented decision aids, plus three conditions designed to represent the scenarios described by Messier (1995) in which an auditor (1) chose to override the recommendation of the decision aid, (2) chose to rely on the recommendation of an incorrect decision aid, or (3) chose not to use an available decision aid during the audit. (5) The decision aid described in the case was similar in structure to the fraud prediction aid proposed by Bell and Carcello (2000).

We described the reliability of the decision aid as either an 81 percent (low) or a 90 percent (high) historical accuracy rate. We chose the 81 percent rate based on previous research on fraud prediction decision aids. (6) For example, Boatsman et al. (1997) reported that their fraud prediction decision aid correctly classified 81 percent of fraud and nonfraud cases. Bell and Carcello (2000) reported that their best model correctly classified 30/37 (81 percent) of fraud cases and 123/143 (86 percent) of nonfraud cases. Once we chose 81 percent as a "realistic" hit rate, our next step was to choose a rate that jurors would view as more/less reliable. After pilot testing with student participants, we chose to go with a higher level of accuracy because we did not believe that jurors would find an 81 percent aid to be highly reliable. Research in psychology (Arkes et al. 1986) and accounting (Ashton 1990; Whitecotton 1996) suggests that decision makers often mistakenly believe they can outperform a statistically superior decision aid. For example, Boatsman et al. (1997) found that auditors were reluctant to rely on an 81 percent accurate fraud prediction aid, even though it outperformed unaided auditors (59 percent) and auditors were told that the aid would "outperform even firm experts." We expected that jurors would be even less likely to perceive an 81 percent aid as reliable because they would be evaluating the aid's performance in hindsight. Based on this reasoning, we chose 90 percent for the high-reliability condition.

The third independent variable was audit firm size, which we described as either "a large international public accounting firm with offices in more than 60 cities, with nearly 20,000 professionals" or "a relatively small auditing firm with offices in four cities and three states, with about 230 professionals."

Two aspects of our design warrant further discussion. Although we made these design choices in an effort to provide a complete and realistic experimental design, they create limitations that should be kept in mind when interpreting our results. First, because the auditor always concluded that the potential for fraud was low when fraud was actually present (otherwise there would be no lawsuit), the decision aid's prediction had to vary across the decision-aid use conditions. For instance, in those conditions in which the auditor did not rely on the decision aid, the case indicated that the decision aid would have predicted fraud. However, when the auditor chose to rely on the decision aid (which proved incorrect), the aid predicted no fraud. Thus, we cannot say for certain whether any differences in juror responses across these scenarios are due to our intended manipulation of decision-aid use, or to the prediction of the decision aid itself (fraud/no fraud). However, manipulation checks included in the exit questionnaire suggest that jurors paid more attention to whether the auditor used a decision aid during the audit than whether the aid predicted fraud or no fraud. (7)

Second, to create a fully crossed design, we included some description of decision-aid reliability in all conditions, including the "no decision aid" control group. To accomplish this, we added a statement from the plaintiff indicating that if the audit firm had used a decision aid that was available in the commercial market or used by other CPA firms ("packages that are correct about 90 percent (or 81 percent) of the time"), then it would have led to the discovery of fraud. Thus, although the audit firm had chosen not to implement decision aids, some form of decision aid was at least potentially available even in the control group. (8)

Dependent Variables

Our first dependent variable captures jurors' beliefs about the extent to which they should hold the audit firm responsible for the losses that the plaintiff claims resulted from an audit failure. We asked three questions to measure these beliefs:

A. Do you feel that the audit manager (and the audit firm) made the right decision in concluding that fraud was not present and that there was no need to perform additional audit procedures?

B. How competent did you perceive the audit manager (and the audit firm) to be in performing its duties in the audit of this client?

C. To what extent do you believe that the plaintiff must assume normal investment risks when purchasing stock, and therefore is largely responsible for their own loss?

Participants responded to these questions on a ten-point Liken scale. The questions were reverse-coded so that higher (lower) values suggest that the auditor was more (less) responsible for the audit failure. We then used factor analysis to combine these three questions into an overall measure of juror attribution of responsibility. (9) Factor analysis attempts to aggregate similar, individual factors into general, underlying constructs. Using principle components analysis, the three questions loaded on a single factor, producing an eigenvalue of 1.8 and a Cronbach's alpha of 0.66. The resulting factor scores serve as the dependent measure for jurors' assessments of auditor responsibility. These scores have a mean of 0 and a variance of 1. Thus, negative scores indicate that jurors attributed "lower than average" responsibility for the audit failure to the auditor, and positive scores imply "above average" responsibility for the audit failure.

The second dependent measure relates to jurors' assessments of damages against the audit firm. We asked jurors to respond to the following three questions using a ten-point Likert scale:

A. To what extent do you believe that the CPA firm's audit was ineffective, and therefore the CPA firm should reimburse the plaintiff?

B. What is the likelihood that you would support the plaintiff's call for some amount of damages by the auditor?

C. What is the likelihood that you would support the plaintiff's call for the total amount of damages by the auditor?

Similar to the procedure described above, we used factor analysis to combine these three questions into an overall measure of jurors' assessments of damages. The three questions loaded on a single factor, producing an eigenvalue of 2.1 and a Cronbach's alpha of 0.79. The resulting factor scores, which have a mean of 0 and a variance of 1, comprised the dependent measure of juror assessment of damages against the auditor. Thus, positive values reflect above-average damage assessments, and vice versa.

Finally, we collected the following demographic variables that could influence jurors' assessments of responsibility and/or damages: gender, age, education, and income level (see Hastie 1993). (10)

IV. RESULTS

Responsibility Evaluations (H1)

The overall prediction of H1 was that auditors' use of decision aids would influence juror responsibility attributions only if the decision aid is reliable. We test this interactive hypothesis using an ANOVA model, with the experimental manipulations of decision-aid use (four levels) and reliability (two levels) serving as the independent variables, and the responsibility factor score designated as the dependent variable. Preliminary analysis revealed that the other model variables--audit firm size and juror demographics--were not significantly related to the dependent variable. Therefore, we excluded these variables from the analyses related to H1. The ANOVA results appear in Panel A of Table 1; the means and standard deviations for the dependent measure are shown in Panel B.

As shown in Panel A of Table 1, the interaction between decision-aid use and decision-aid reliability is statistically significant (F = 2.82; p = 0.041), supporting H1. To provide more evidence on the specific form of this overall interaction, we performed planned comparisons to test the predictions of H1a, H1b, and H1c. Figures 1, 2, and 3 display the relevant cell means for each of these hypotheses in graphical format.

[FIGURES 1-3 OMITTED]

Hypothesis 1a predicted that jurors would attribute more responsibility to an auditor who chose to override the aid than if no decision aid was available, but only if the decision aid is reliable. The cell means displayed in Figure 1 are consistent with the form of this predicted interaction. Planned comparisons confirm that jurors held auditors more responsible for overriding the recommendation of the decision aid (compared to the control condition) when the reliability of the decision aid was high (t = 2.61, p = 0.007), but not when the reliability of the decision aid was low (t = 0.28, p = 0.777). Thus, H1a is supported.

Our next hypothesis (H1b) relates to the condition in which the auditor relied on the recommendation of the decision aid, but the aid was incorrect. We expected that jurors would assess lower responsibility for the audit failure when the auditor followed the decision aid's recommendation compared to the control group in which the aid was not available, but only if the aid is reliable. The cell means displayed in Figure 2 are consistent with the form of this predicted interaction. Planned comparisons show that jurors attributed lower responsibility to the auditor who relied on a highly reliable decision aid than the control group (t = 1.80, p = 0.040), but there was no significant difference for the low-reliability decision aid (t = 0.63, p = 0.533). Thus, H1b is supported.

Finally, H1c relates to the condition in which the decision aid was available, but the auditor chose not to use it during the audit. Once again, our expectation was that failure to use an available decision aid would lead jurors to attribute more responsibility to the auditor than in the control condition, but only when the aid was reliable. Figure 3 does not support this predicted interaction. Planned t-tests show no significant differences in jurors' mean responsibility evaluations between the condition in which the decision aid was available (but not used) and the control group, regardless of whether decision-aid reliability was low (t = 0.37, p = 0.713) or high (t = 0.36, p = 0.719). Thus, H1c is not supported.

Although not hypothesized, the results displayed in Figure 3 suggest that jurors attributed more responsibility to the audit firm in the presence of a high-reliability decision aid than a low-reliability aid, regardless of the decision-aid condition (t = 2.09; p=0.039). The only difference between these two conditions was how readily available the decision aid was for the auditor's use. In the control group, the audit firm had not implemented decision aids, but jurors were told that decision aids were commercially available and used by other CPA firms. In the treatment condition, the audit firm had implemented the decision aid, but the auditor chose not to use it during the audit in question. The results shown in Figure 3 suggest that jurors did not differentiate between these two conditions--jurors appear to penalize auditors for failing to use a high-reliability decision aid regardless of whether the decision aid was available within the audit firm or on the commercial market.

Damage Award Assessments (H2 and H3)

Next, we investigate whether jurors' responsibility evaluations (H2) and audit firm size (H3) influence jurors' award of damages against the audit firm. Table 2 reports the results of a regression model to test these hypotheses. The dependent measure was the combined factor score on the three damage-related questions, where positive scores indicate "higher than average" damages against the audit firm, and vice versa. In addition to the hypothesized independent variables, the model includes juror age, gender, education, and income level as control variables. (11)

The overall regression model is highly significant (F = 20.51; p < 0.001), with the six variables combining to explain approximately 44 percent of the variation in jurors' damage assessments (adjusted [R.sup.2] = 0.442). As H2 predicts, the relationship between jurors' responsibility evaluations and damage assessments was significantly positive (t = 9.58; p < 0.001), suggesting that jurors who assessed higher auditor responsibility for the audit failure were more likely to support damages against the auditor. The significantly positive coefficient for audit firm size (t = 2.91; p = 0.004) suggests that jurors were more likely to assess damages against a large audit firm than a small audit firm, supporting H3.

The regression model also shows significant effects for juror age (t = 2.52; p = 0.013), education level (t = -3.76; p < 0.001), and gender (t = -2.04; p = 0.043). These results suggest that older and less educated jurors were more likely to support damages against the auditor, and that male jurors were more likely to support damages against the audit firm than were female jurors. Juror income level was not significantly related to damage assessments (t = 0.66; p = 0.510).

V. DISCUSSION AND IMPLICATIONS

This study adds to the limited evidence on the legal implications of decision aids by examining the effects of auditors' use of decision aids and decision-aid reliability on jurors' evaluations of auditor liability. Our results suggest that the extent to which auditors use decision aids during an audit influences jurors' legal liability judgments only if jurors perceive the decision aid as reliable. For example, jurors assessed higher responsibility for an audit failure when an auditor overrode the recommendation of a highly reliable decision aid than when the audit firm had not implemented decision aids. These results should be of interest to audit firms who are concerned that implementing decision aids may heighten their liability exposure.

On a more positive note, and in contrast to the results of Anderson et al. (1995), we found that jurors attributed lower responsibility to an auditor who followed the recommendation of a highly reliable decision aid, even though the aid turned out to be incorrect. However, regardless of decision-aid reliability, there were no differences in jurors' responsibility evaluations depending on whether (1) the decision aid was available within the firm but the auditor did not use it during that particular audit, and (2) the aid was available in the commercial market but had not been implemented by the audit firm. Apparently, jurors did not view these two scenarios as different enough to affect their legal liability judgments. Thus, our results suggest that highly reliable decision aids can have positive, negative, or neutral effects on auditors' legal liability, depending on how auditors use decision aids during the audit.

In contrast to the high-reliability condition, decision-aid use had virtually no effect on jurors' responsibility evaluations in the low-reliability condition. These insignificant results may be enlightening to audit firms, since this level of accuracy (81 percent) reflects the current "state of the art" for fraud prediction decision aids (Bell and Carcello 2000). Thus, the fear that some audit firms have expressed about the potential adverse legal consequences of decision aids may be unwarranted, at least for currently available fraud prediction decision aids. However, as model developers improve the predictive accuracy of such decision aids (by incorporating additional "red flags," for example), the legal liability stakes may increase.

Thus, audit firms appear to face a dilemma regarding the development and use of decision aids. If firms choose not to implement decision aids, then they may be disadvantaged by the lack of internal guidance and consistency these tools provide. In addition, by not implementing reliable decision aids, audit firms may be at a legal disadvantage if other firms have adopted decision aids, since auditors may be held responsible for failing to use a reliable decision aid, regardless of whether the aid was available internally or on the commercial market. However, if firms develop and use decision aids, then they may be constrained to adhere fully to the decision aids' recommendations, as these aids may serve as implied standards of performance in future litigation, particularly as decision aids gain reliability.

These results have important implications for audit firms' policies regarding the implementation and use of decision aids. For instance, should the firm's executive office mandate the use of decision aids in their audits? Although the results of our study suggest that such a policy may have advantages in a court of law, it may have some disadvantages in practice. Requiring auditors to use a decision aid may lead auditors to approach their tasks mechanistically (Glover et al. 1997), or, alternatively, may lead to dysfunctional behavior such as intentional circumvention of the aid (Kachelmeier and Messier 1990).

Another objective of this study was to investigate how jurors' responsibility evaluations translate into damage awards against the audit firm. Our results revealed that jurors' responsibility evaluations were significantly related to their damage assessments. This finding provides some evidence that jurors assess damages based upon the degree of corresponding responsibility, in contrast to concern that jurors provide damage assessments based strictly on nonmeritorious (frivolous) claims (Cloyd et al. 1996; Palmrose 1997).

Responsibility assessment was not the only factor affecting jurors' damage assessments. Jurors were also more likely to support damages against large audit firms than small audit firms, which provides further evidence of the perception of large audit firms as "deep pockets." Interestingly, although audit firm size and several juror demographic measures were related to jurors' damage assessments, these factors did not influence jurors' evaluations of auditor's responsibility for the audit failure. These results are consistent with Kadous' (2000) conjecture that jurors may use a different set of factors to determine damages than they do to assess responsibility. Our results suggest that although jurors were able to ignore factors such as the audit firm's "ability to pay" and their own personal beliefs when determining who was responsible for the audit failure, they did allow these factors to influence their damage awards. Additional research is necessary to understand this inconsistency and to determine how and why these factors influenced jurors' damage assessments in an audit liability scenario.

Our results also showed that jurors attributed less responsibility to the auditor when they perceived a decision aid as having low reliability, regardless of how the auditor had used the decision aid during the audit. Defense counsel could therefore emphasize the limitations and incomplete nature of decision aids. The results related to juror demographics may also prove useful to audit firms and their defense counsel in the selection of jurors.

Of course, one must view the results of this study in light of its limitations. In addition to the design limitations previously discussed, many of the items presented in the case may be nonfactual, ambiguous, or a source of contention in a court of law. In a more realistic setting, jurors would receive substantial judicial instructions along with plaintiff and defense counsels' opposing viewpoints. It is not clear how these factors would influence our findings.

Finally, we manipulated the reliability of the decision aid as either 81 percent (low) or 90 percent (high). Although our results suggest that jurors perceived these levels differently, we have no way of identifying the critical threshold for decision-aid reliability. It is also likely that other accuracy rates would produce different results. For example, jurors might penalize auditors for using an aid with a very low level of reliability. Future research should determine what jurors (and judges) consider the minimum accuracy level of decision aids before they are considered reliable and therefore relevant in auditor litigation.

TABLE 1
The Effects of Decision-Aid Use and Decision-Aid Reliability (H1)
On Jurors' Responsibility Assessments (a)

Panel A: Analysis of Variance Model

Source of            Sum off           Mean
Variation            Squares    df    Square      F      p-value

Overall Model         19.06     7      2.72     2.98      0.006
Decision Aid
  Use (b)              5.21     3      1.74     1.90      0.133
Decision Aid
  Reliability (c)      6.59     1      6.59     7.21      0.008
Use x Reliability      7.75     3      2.58     2.82      0.041

Panel B: Mean Juror Attribution Score, (Standard Deviation), Cell
Sizes

                             Low -Reliability    High -Reliability
                               Decision Aid         Decision Aid

Control Group--                  -0.324                0.128
  No Decision Aid                (0.942)              (0.710)
                                  n = 17               n = 22

Auditor Overrides                -0.236                0.816
  the Decision Aid               (0.924)              (0.958)
                                  n = 20               n = 18

Auditor Relied on                -0.093               -0.336
  Incorrect Decision Aid         (1.204)              (0.925)
                                  n = 18               n = 18

Decision Aid Available,          -0.201                0.225
  But Not Used                   (1.005)              (0.970)
                                  n = 18               n = 18

(a) The dependent variable was obtained through principal
component analysis on three individual questions related to jurors'
attributions of auditor responsibility. This method combines the
individual measures into a standardized factor score that, by
construction, has a mean of 0 and a variance of 1. Thus, negative
scores indicate that jurors attributed "lower than average"
responsibility for the audit failure to the auditor, and positive
scores imply "above average" responsibility for the audit failure.

(b) Decision aid use included (1) a control group in which the firm
had not implemented decision aids, plus three conditions in which
an auditor (2) chose to override the recommendation of the decision
aid, (3) chose to rely on the recommendation of an incorrect
decision aid, or (4) chose not to use an available decision aid
during the audit.

(c) The reliability of the decision aid was described as either 81
percent (low) or a 90 percent (high) historical accuracy rate.
TABLE 2
Results of Regressing Jurors' Damage Assessments on Jurors'
Responsibility Evaluations (H2) and Audit Firm Size (H3) (a)

                                    Standard      t-
Variable              Coefficient    Error     statistic   p-value

Intercept                0.275       0.385        0.71      0.476
Responsibility
  Evaluation (b)         0.598       0.063        9.58     <0.001
Audit Firm Size (c)      0.362       0.124        2.91      0.004
Juror Age                0.012       0.005        2.52      0.013
Juror Gender (d)        -0.261       0.128       -2.04      0.043
Juror Education         -0.137       0.037       -3.76     <0.001
Juror Income             0.038       0.058        0.66      0.510
Adj. [R.sup.2] = 0.442

(a) The dependent variable was obtained through principal
component analysis on three individual questions related to jurors'
assessments of damages. Higher (lower) values suggest that jurors
were more (less) likely to award damages against the audit firm.

(b) This variable was obtained through principal component analysis
on three individual questions related to jurors' attributions of
auditor responsibility. Higher (lower) values indicate that jurors
attributed more (less) responsibility for the audit failure to the
auditor.

(c) Audit firm size was coded as either a 0 for "a relatively small
auditing firm with offices in four cities and three states, with
about 230 professionals" or 1 for "a large international public
accounting firm with offices in more than 60 cities, with nearly
20,000 professionals."

(d) Female jurors were coded 1 and male jurors were coded 0.

The authors acknowledge the helpful comments and suggestions of two anonymous reviewers, Steve Kachelmeier (associate editor), Steve Kaplan, Paul Carlson, Ed O'Donnell, workshop participants at the University of Arizona and Arizona State University, and participants of the 2000 Accounting, Behavior, and Organizations Research Conference.

Submitted September 1999

Accepted May 2001

(1) This is consistent with Kadous' (2000) definition of audit quality, which she operationalized as whether the auditor consulted a specialist regarding inventory measurement.

(2) The Private Securities Litigation Reform Act of 1995 replaced the long standing "joint and several" liability rule with a modified "proportionate" liability rule for most federal cases (King and Schwartz 1997; Hillegeist 1999). Although this change should reduce the likelihood that auditors and other "deep pocket" defendants will bear a disproportionate share of the damages from audit failure, it is not likely to eliminate the issue entirely. For example, because "joint and several" liability still applies in many state jurisdictions, plaintiffs may simply shift their litigation from federal courts to more favorable state venues (Palmrose 1997). Furthermore, Cloyd et al. (1998) note that jurors may circumvent the "proportionate" liability rule by apportioning a greater share of the liability to "deep pocket" defendants to compensate for the limited recoveries expected from "shallow pocket" defendants.

(3) These studies assume that plaintiff attorneys anticipate how juries will attribute responsibility and assess damages in audit trials, and factor these considerations into their decision to initiate litigation against audit firms.

(4) Prior arrangements had been made with the court administrator. Jurors are called from a general jury list to serve in a juror pool. The jury list includes names drawn from county driver's license registration files and from a master file of registered voters from the Board of Elections. Individuals are deleted from the jury list if they (1) are less than 18 years old, (2) are not current residents of the county, or (3) are presently exempt from jury duty. The court administrator sorts the random file into numerical order and each jury pool is selected sequentially. Plaintiff or defense counsel can then dismiss jurors due to perceived conflicts or biases (i.e., occupation, attitudes, etc.).

(5) Based on current practice, we believed that auditors who would not use a computer-based decision aid would at least rely on some sort of checklist. To make the case as realistic as possible, we included a checklist when the auditor did not use a computer-based decision aid. We also acknowledge that checklists are a type of decision aid.

(6) Most prior research has relied on the sample of 77 fraud and 305 nonfraud cases from Loebbecke et al. (1989) and Bell et al. (1991), respectively. Although the accuracy rates of the models vary depending on the red flags included in the model, the actual outcome (fraud or no fraud), and the cutoff criterion for predicting fraud/no fraud (Bell and Carcello 2000), 81 percent emerges as a "realistic" hit rate for fraud prediction decision aids.

(7) When asked whether the audit firm used a decision aid during the audit, 85 percent of participants responded correctly, whereas only 56 percent responded correctly to a similar question asking whether the decision aid predicted fraud or no fraud.

(8) Although we recognize that this has the potential of weakening our results, we felt it was important to provide a control condition in our design. Furthermore, Sutton et al. (1995) note that audit firms are increasingly called upon to provide this type of expert testimony on behalf of plaintiffs who have filed lawsuits against other audit firms.

(9) We also conducted two alternative analyses. First, we performed a multivariate analysis (MANOVA) with each of the individual questions treated as dependent variables. We also combined the individual questions into an aggregate measure using an equally weighted scoring rule (Kadous 2000). The results from each analysis were qualitatively similar to the factor analysis results.

(10) We also included these demographic variables to check on the randomization of the various levels of the independent variable. Statistical tests performed on the demographic variables indicate randomization was successful across treatment levels.

(11) Since there were no significant interactions among the independent variables, only main effects were included in the model.

REFERENCES

Abdolmohammadi, M., and C. Usoff. 2000. A longitudinal study of applicable decision aids for detailed tasks in a financial audit. Working paper, Bentley College, Waltham, MA.

American Institute of Certified Public Accountants (AICPA). 1997. Consideration of Fraud in a Financial Statement Audit. Statement on Auditing Standards No. 82. New York, NY: AICPA.

Anderson, J. C., M. M. Jennings, S. E. Kaplan, and P. M. J. Reckers. 1995. The effects of using diagnostic decision aids for analytical procedures on judges' liability judgments. Journal of Accounting and Public Policy 14: 33-62.

Arkes, H. R., R. M. Dawes, and C. Christensen. 1986. Factors influencing the use of a decision rule in a probabilistic task. Organizational Behavior and Human Decision Performance 37: 93-110.

Ashton, R. H. 1990. Pressure and performance in accounting decision settings: Paradoxical effects of incentives, feedback and justification. Journal of Accounting Research 28 (Supplement): 148-180.

--, and J. J. Willingham. 1989. Using and evaluating audit decision aids. In Auditing Symposium IX: Proceedings of the 1988 Touche and Ross/University of Kansas Symposium on Auditing Problems, edited by R. P. Srivastava, and J. E. Rebele, 1-25. Lawrence, KS: University of Kansas.

Bell, T. B., and J. V. Carcello. 2000. A decision aid for assessing the likelihood of fraudulent financial reporting. AUDITING: A Journal of Practice & Theory 19 (Spring): 169-184.

--, S. Szykowny, and J. J. Willingham. 1991. Assessing the likelihood of fraudulent financial reporting: A cascaded logit approach. Working paper, KPMG LLP, Montvale, NJ.

Boatsman, J. R., C. Moeckel, and B. Pei. 1997. The effect of decision consequences on auditors' reliance on decision aids in audit planning. Organizational Behavior and Human Decision Processes 71 (2): 211-247.

Bonner, S. E. 1990. Experience effects in auditing: The role of task-specific knowledge. The Accounting Review 65 (January): 72-92.

--, Z-V. Palmrose, and S. M. Young. 1998. Fraud type and auditor litigation: An analysis of SEC accounting and auditing enforcement releases. The Accounting Review 73 (October): 503-532.

Carcello, J. V., and Z-V. Palmrose. 1994. Auditor litigation and modified reporting on bankrupt clients. Journal of Accounting Research 32 (Supplement): 1-30.

Cloyd, C. B., J. R. Frederickson, and J. W. Hill. 1996. Motivating factors in lawsuits against independent auditors: Experimental evidence on the importance of causality. Journal of Accounting and Public Policy 15 (Fall): 185-218.

--, --, and --. 1998. Independent auditor litigation: Recent events and related research. Journal of Accounting and Public Policy 17 (Summer): 121-142.

Dye, R. A. 1993. Auditing standards, legal liability and auditor wealth. Journal of Political Economy 101 (October): 887-914.

Eining, M. M., D. R. Jones, and J. K. Loebbecke. 1997. Reliance on decision aids: An examination of auditors' assessments of management fraud. AUDITING: A Journal of Practice & Theory 16 (Fall): 1-19.

Elliott, R. K. 1998. Assurance services and the audit heritage. The CPA Journal 68 (June): 40-47.

Francis, J. R. 1994. Auditing, hermeneutics, and subjectivity. Accounting, Organizations and Society 19 (April): 235-269.

Glover, S. M., D. F. Prawitt, and B. C. Spilker. 1997. The influence of decision aids on user behavior: Implications for knowledge acquisition and inappropriate reliance. Organizational Behavior and Human Decision Processes 72 (2): 232-255.

Hastie, R. 1993. Inside The Juror: The Psychology of Decision Making. New York, NY: Cambridge University Press.

Hillegeist, S. A. 1999. Financial reporting and auditing under alternative damage apportionment rules. The Accounting Review 74 (July): 347-369.

Jennings, M., D. Kneer, and P. Reckers. 1993. The significance of audit decision aids on pre-case jurists' attitudes on perceptions of audit firm culpability and liability. Contemporary Accounting Research 9 (Spring): 489-507.

Kachelmeier, S. J., and W. Messier. 1990. An investigation of the influence of a nonstatistical decision aid on auditor sample size decisions. The Accounting Review 65 (January): 209-226.

Kadous, K. 2000. The effects of audit quality and consequence severity on juror evaluations of auditor responsibility for plaintiff losses. The Accounting Review 75 (July): 327-341.

Kelley, H. H., and J. L. Michela. 1980. Attribution theory and research. In Annual Review of Psychology, edited by M. R. Rosenzweig, 457-501. Palo Alto, CA: Annual Reviews.

King, R., and R. Schwartz. 1997. The Private Securities Litigation Reform Act of 1995: A discussion of three provisions. Accounting Horizons 11 (March): 92-106.

Latham, C. K., and M. Linville. 1998. A review of the literature in audit litigation. Journal of Accounting Literature 17: 175-213.

Lennox, C. S. 1999. Audit quality and auditor size: An evaluation of reputation and deep pockets hypotheses. Journal of Business, Finance, and Accounting 26 (Sept/Oct): 779-806.

Loebbecke, J., M. Eining, and J. Willingham. 1989. Auditors' experience with material irregularities: Frequency, nature and detectability. AUDITING: A Journal of Practice & Theory 9 (Fall): 1-28.

Lys, T., and R. L. Watts. 1994. Lawsuits against auditors. Journal of Accounting Research 32 (Supplement): 65-95.

Mancino, J. 1997. The auditor and fraud. Journal of Accountancy 183 (April): 32-36.

Melancon, B. C. 1998. The changing strategy for the profession, the CPA and the AICPA: What this means for the education community. Accounting Horizons 12 (December): 397-406.

Messier, W. E 1995. Research in and development of audit decision aid. In Judgment and Decision-Making Research in Accounting and Auditing, edited by R. H. Ashton, and A. H. Ashton, 207-228. New York, NY: Cambridge Press.

Pacini, C., W. Hillison, and D. Sinason. 2000. Three's a crowd: An examination of state statutes and court decisions that narrow accountant liability to third parties for negligence. Advances in Accounting 17: 151-186.

Palmrose, Z-V. 1987. Litigation and independent auditors: The role of business failures and management fraud. AUDITING: A Journal of Practice & Theory 6 (Spring): 90-103.

--. 1988. An analysis of auditor litigation and audit service quality. The Accounting Review 63 (January): 55-73.

--. 1991. Trials of legal disputes involving independent auditors: Some empirical evidence. Journal of Accounting Research 29 (Supplement): 149-185.

--. 1997. Audit litigation research: Do the merits matter? An assessment and directions for future research. Journal of Accounting and Public Policy 16 (Winter): 355-378.

--. 1999. Empirical Research in Auditor Litigation: Considerations and Data. Sarasota, FL: American Accounting Association.

Shibano, T. 1990. Assessing audit risk from errors and irregularities. Journal of Accounting Research 28 (Supplement): 110-140.

Siliciano, J. A. 1997. Trends in independent auditor liability: The emergence of a sane consensus? Journal of Accounting and Public Policy 16 (Winter): 339-353.

Stice, J. D. 1991. Using financial and market information to identify pre-engagement factors associated with lawsuits against auditors. The Accounting Review 66 (July): 516-533.

Sutton, S. G., R. Young, and P. McKenzie. 1995. An analysis of potential legal liability incurred through audit expert systems. Intelligent Systems in Accounting, Finance, and Management 4: 191-204.

Whitecotton, S. M. 1996. The effects of experience and confidence on decision aid reliance: A causal model. Behavioral Research in Accounting 8: 194-216.

Zimbelman, M. F. 1997. The effects of SAS No. 82 on auditors' attention to fraud risk factors and audit planning decisions. Journal of Accounting Research 35 (Supplement): 75-97.

D. Jordan Lowe
University of Nevada, Las Vegas

Philip M. J. Reckers
Stacey M. Whitecotton
Arizona State University