Cue usage in financial statement fraud risk assessments: effects of technical knowledge and decision aid use
I am grateful to Hian-Chye Koh, Asheq Rahman and Amit Das for their encouragement and guidance in this project. My special thanks also goes to Hun-Tong Tan for his valuable insights, comments and suggestions. The helpful comments of the anonymous reviewers and associate editor of Accounting and Finance are also appreciated. Financial support from the Singapore Ministry of Education Academic Research Fund Tier 1, RCC12/2000/NBS and support from the Singapore Institute of Directors are also gratefully acknowledged.
Abstract
This paper investigates the effects of technical knowledge and decision aid use on financial statement fraud risk assessments made by directors and students. More extreme fraud risk assessments are made when participants identify and process larger (smaller) numbers of diagnostic (non-diagnostic) factors, with technical knowledge driving diagnostic factor identification. Significant decision aid-technical knowledge effects are also found; decision aid use has a detrimental effect on high-knowledge directors while improving performance in inexperienced, low-knowledge students. These results suggest that although decision aids can afford gains in performance in inexperienced users, they can have unintended and/or paradoxical behavioural effects on experienced users.
1. Introduction
The monitoring role played by directors, particularly by outside (independent) directors and audit committees, has been a focal point in the legal, finance and accounting literature over the past decade. The Blue Ribbon Committee on Improving the Effectiveness of Corporate Audit Committees (Blue Ribbon Report, 1999) and the Sarbanes–Oxley Act (United States Public Laws, 2002) have brought the issues of board and audit committee responsibilities, liabilities and expertise to the forefront. There is increasing concern over the ability of audit committees to discharge their assigned financial reporting, auditing and governance duties (General Accounting Office, 1991; McMullen, 1996), and recognition that some degree of financial sophistication and technical knowledge is required for boards and audit committees to effectively discharge their oversight duties (DeZoort and Salterio, 2001; McDaniel et al., 2002).
Risk management is one of the key oversight responsibilities of boards and audit committees (Turnbull Report, 1999; Organization for Economic Cooperation and Development (OECD), 2004). However, whether directors possess adequate technical knowledge to perform risk assessments is moot. The ability of directors to perform such risk assessments is critical as the failure to adequately assess risk exposure can lead to a lack of appropriate investigative and/or corrective action being taken. Although not all directors are charged with a specific obligation to perform risk assessments relating to the financial statements, this constitutes part of the board's overall responsibility to ensure the integrity of the organization's accounting and financial reporting systems (OECD, 2004). The present study investigates the role of technical knowledge on directors’ ability to perform risk (in particular, financial statement fraud risk) assessments, and their cue usage in performing such risk assessments.
Although little is known about the fraud risk identification and assessments abilities of directors, there is a general belief that providing directors with more explicit guidance can enable them to better identify and assess risk (Turnbull Report, 1999). Presumably, such guidance (operationalized as a decision aid in the present study) should improve the performance of directors, but there is no empirical evidence on this. Nor is there any evidence on the effects of such an aid interacting with technical knowledge in determining directors’ fraud risk identification and assessments. These issues are examined in the present study.
In addition, the impact of these two factors, technical knowledge and decision aid use, on the dilution effect is assessed. Research shows the presence of non-diagnostic (irrelevant) information dilutes the effect of diagnostic (relevant) information in judgement tasks (Zukier, 1982; Tetlock and Boettger, 1989). Prior audit studies on dilution in judgements (Hackenbrack, 1992; Hoffman and Patton, 1997; Shelton, 1999) generally present participants with details of a case scenario that, in one condition, contains only diagnostic information, and in another condition, contains both diagnostic and non-diagnostic information for the purposes of measuring dilution in the mean likelihood judgements made under the two different conditions. The present study refines and extends prior judgement and dilution studies by examining how the number of diagnostic and non-diagnostic factors identified and processed affects risk assessments, and whether the diluting effects of non-diagnostic information can be mitigated by technical knowledge and/or the use of a decision aid.
In the present study, directors and students are provided with a hypothetical case scenario containing both diagnostic and non-diagnostic information for a financial fraud risk assessment task. The decision aid is manipulated by providing some participants with a generic financial fraud risk checklist aid. Prior research finds some evidence of dysfunctional effects of decision aids on the performance of experienced users (Ashton, 1990). Hence, to assess whether the dysfunctional effects of decision aids (if any) apply to both experienced and inexperienced users, the same experimental materials are run using participant groups with vastly different levels of experience as directors: an experienced participant group (directors), and an inexperienced participant group (students).
The findings of the present study contribute to the existing judgement and corporate governance literature through examination of the impact of technical knowledge and decision aid use on directors’ cue usage in judgement tasks. These results have implications for the training of directors and staffing of boards/audit committees to ensure that members possess appropriate technical/task-specific knowledge sufficient to discharge their governance duties. The results also have implications for regulatory bodies that seek to provide operational guidance to directors to enhance director performance on risk assessments by promoting the use of stylized and generic checklist aids.
The remainder of the present paper is organized as follows. Section 2 summarizes the literature and develops the research hypotheses for the experiment conducted. The research design, experimental methods and results are described in Section 3, and a summary of the study's main findings, limitations, and opportunities for future research are presented in Section 4.
2. Literature review and hypothesis development
Information can be characterized as either diagnostic or non-diagnostic in nature. Diagnostic information possesses features or characteristics generally associated with a particular event of interest, while non-diagnostic information possesses features or characteristics not generally associated with a particular event of interest (Hackenbrack, 1992). The psychology literature documents the tendency of individuals to make less extreme (i.e. diluted) judgement decisions when given a mixture of diagnostic and non-diagnostic evidence, compared to when they are given only diagnostic evidence (Nisbett et al., 1981; Zukier, 1982; Tetlock and Boettger, 1989).
Prior audit studies on dilution in judgements (Hackenbrack, 1992; Hoffman and Patton, 1997; Shelton, 1999) present participants with details of a case scenario that, in one condition, contains only diagnostic information, and in another condition, contains both diagnostic and non-diagnostic information for the purpose of making a judgement decision. However, these studies do not investigate how participants process specific pieces of diagnostic versus non-diagnostic evidence, nor do they examine the relation between identification of diagnostic versus non-diagnostic evidence and the eventual judgements made.
Consider the fraud risk identification scenario used in this study. Participants make fraud risk assessments in a case scenario in which both diagnostic and non-diagnostic evidence are provided. Diagnostic evidence in this context includes policies, procedures or other matters (or lack of them) that can lead to financial statement fraud. Non-diagnostic evidence includes routine and/or unexceptional items or situations with no predictive ability for the absence or presence of fraud risk (Hackenbrack, 1992). Consistent with the argument above, if participants mistakenly identify both diagnostic evidence and non-diagnostic evidence (that should not be considered in the first place) as relevant to the fraud risk assessment task, they will be susceptible to dilution effects. In contrast, those that identify only diagnostic evidence as relevant to the fraud risk assessment task should not be susceptible to dilution effects.
What factors influence the identification of diagnostic versus non-diagnostic evidence in fraud risk assessments? Two factors are investigated here: technical knowledge and availability of decision aids.
The psychology literature finds that experienced decision-makers develop more sophisticated knowledge structures (organization of knowledge in memory) that enable them to recognize and discount non-diagnostic evidence (Patel and Groen, 1986; Lesgold et al., 1988). In the accounting literature, Libby and Luft (1993) find that instruction, experience and ability are important elements in the acquisition of requisite task/technical knowledge, which in turn determines task performance. Individuals who possess more technical knowledge have better organization and representation of knowledge (Frederick, 1991), greater knowledge of relevant cues for specific tasks (Bonner, 1990), and more task-specific knowledge (Solomon et al., 1999). Although there is limited empirical research on the intuitively positive effects of technical knowledge on directors’ performance on oversight tasks, DeZoort (1998) finds evidence that audit committee members with task-specific experience make judgements more like experts in an internal control evaluation task, and DeZoort and Salterio (2001) find evidence that greater independent director experience and greater audit-reporting (as opposed to more general financial-reporting) knowledge are associated with audit committee member support for the auditor in auditor–management disputes over financial reporting. McDaniel et al. (2002) also find evidence that financial experts (as opposed to individuals who are merely financially literate) on audit committees have better-developed knowledge frameworks for characteristics related to financial reporting quality that can improve oversight on financial-reporting quality issues. Therefore, individuals who possess higher levels of technical knowledge are a priori expected to achieve better levels of task performance in the identification and processing of diagnostic factors in this study's financial statement fraud risk assessment task. Conversely, they are also less likely to erroneously identify and process non-diagnostic factors as relevant diagnostic cues.
In terms of the effects of a decision aid, Davis (1994) finds no evidence that checklists improve auditors’ going-concern probability judgements, and Pincus (1989) shows that fraud checklists consisting of potential red flags serving as memory aids do not improve auditors’ performance in judging the probability of management fraud. However, Heiman (1990) and Kennedy (1995) find that memory-related biases are reduced, and Butler (1985) and Heiman-Hoffman (1992) find improvements in judgement, in the presence of memory prompts. Although Eining et al. (1997) find that checklist aids do not provide mechanical assistance for weighting and combining red flag cues in an overall likelihood judgement decision, they find that a decision aid that provides a list of potential red flags can heighten a user's awareness of, and sensitivity to, matters or issues flagged in the decision aid. Therefore, although prior research indicates mixed results in the effect of decision aids on judgement, there is evidence that decision aids can improve users’ sensitivity to diagnostic cues mentioned in the aid.
As the provision of a decision aid is intended to enhance performance, users who possess lower levels of technical knowledge are expected to benefit more from the use of a decision aid. Whitecotton (1996) also finds evidence that decision-makers with less expertise rely more on decision aids. This suggests that low-knowledge directors’ identification of diagnostic cues can improve with the presence of a decision aid. Therefore, in the presence of a decision aid, low-knowledge directors’ identification of diagnostic risk cues can improve more than that of high-knowledge directors.
Although high-knowledge directors have the requisite knowledge to identify diagnostic cues, they might perform worse in terms of relying on non-diagnostic cues when a decision aid is present. In the accounting literature, Ashton (1990) finds that experienced auditors aware of the need to justify their decisions perform worse in the presence of an aid because they adopt decision strategies different (and inferior) to that suggested by the aid. In the context of the present study, not all cues mentioned in the decision aid are necessarily associated with the specific fraud setting, and in fact, generic decision aids highlight only some, but not an exhaustive list of, cues specifically related to the decision task at hand.
Directors are likely to be cognizant of the public scrutiny, expectations and legal exposure that their office attracts, and of the need to justify decisions pertaining to important issues such as fraud risk. This can lead them to consider all cues highlighted in the decision aid as well as other cues they can recall, which in turn spurs them to identify events encountered in the course of their work and to place weight on cues that might in reality be unrelated to fraud risk in the specific fraud risk assessment context under consideration. Directors, because of job experience, are likely to form ‘layman theories’ about predictors of fraud, consistent with the phenomenon called ‘illusory correlation’: ‘the report by an observer of a correlation between two classes of events which in reality (a) are not correlated, or (b) are correlated to a lesser extent than reported, or (c) are correlated in the opposite direction than that which is reported’ (Chapman and Chapman, 1967, p. 194). Chapman and Chapman (1967) find evidence of illusory correlation among experienced individuals, where these individuals continue to use hypotheses and signs conditioned on past experience that are not validated by research or use.
In summary, in the presence of a decision aid, high-knowledge directors are more likely (than low-knowledge directors) to consider other ‘lay theories’ about fraud risk factors, besides those cues listed in the aid. As a result, they are also more susceptible to considering more ‘illusorily correlated’ cues when making their fraud risk assessments.
H1: For directors, in terms of the number of diagnostic factors identified as relevant to the task, the difference between directors who possess higher versus lower levels of technical knowledge is smaller in the presence of a decision aid.
H2: For directors, in terms of the number of non-diagnostic factors identified as relevant to the task, the difference between directors who possess higher versus lower levels of technical knowledge is greater in the presence of a decision aid.
The premise in Hypothesis 2 is that directors’ identification of non-diagnostic factors, by virtue of their job experience, is affected by illusorily correlated cues. If this is true, then in the case of students with no job experience, this effect is not likely to occur. Hence, I expect the benefits of the decision aid to apply to the identification of both diagnostic and non-diagnostic factors in the case of students. Therefore, I hypothesize the following:
H3: For students, in terms of the number of diagnostic factors identified as relevant to the task, the difference between students who possess higher versus lower levels of technical knowledge is smaller in the presence of a decision aid.
H4: For students, in terms of the number of non-diagnostic factors identified as relevant to the task, the difference between students who possess higher versus lower levels of technical knowledge is smaller in the presence of a decision aid.
In the present study, Hypotheses 1 and 2 are tested using experienced directors as participants, while Hypotheses 3 and 4 are tested using inexperienced students as participants.
As it is also anticipated that the strength of fraud risk assessments made is contingent on the number of pieces of diagnostic information and non-diagnostic information (that are erroneously believed to be relevant) identified and processed in arriving at these risk assessments, the effect of the number of identified diagnostic and non-diagnostic factors on fraud risk assessments is also tested.
3. Experiment
3.1. Research design and method
3.1.1. Participants
Director participants in the experiment comprised 32 directors1 recruited at in-house training sessions run by the Singapore Institute of Directors (SID2), and 75 out of 951 directors of Singapore listed companies to whom research materials were mailed (an 8 per cent response rate3). The final sample of 107 comprised 70 directors who sit on boards of listed companies, 31 directors who sit only on boards of non-listed companies, 5 aspiring directors/senior officers assisting directors and 1 recently retired listed company director (who is now no longer a director). Of the 70 listed company directors, 42 were also audit committee members. All Singapore directors, in reality, face a legal responsibility for the preparation and fair presentation of the organization's financial statements (Companies Act, 1994), which is not restricted only to audit committee members. This responsibility includes designing and maintaining internal controls relevant to the preparation and fair presentation of financial statements that are free from material misstatements, whether due to fraud or error.4
The mean technical knowledge score of director participants was 6.82 (standard deviation (SD) 2.16), with technical knowledge scores of audit committee members (mean 7.24, SD 2.06) significantly higher (p= 0.049) than those of non-audit committee members (mean 6.52, SD 2.24). Of the 107 directors in the sample, 51 were randomly assigned to the decision-aid treatment group.
As a joint test of the effects of job experience and of the efficacy of the decision aid, if correctly applied, the same experiment was repeated with final-year accounting major undergraduate students at a business school who had no job knowledge of directorship.5 Student participants comprised 119 final-year accounting major undergraduate students who completed an average of 6.3 accounting courses. The mean technical knowledge score of student participants was 7.52 (SD 1.44). Technical knowledge scores of students are significantly higher (p= 0.005) than those of directors in general, but not significantly higher than those of directors who are audit committee members (p= 0.414).6 Of the 119 students in the sample, 62 were randomly assigned to the decision-aid treatment group.
3.1.2. Method
Participants were presented with summary financial information on a hypothetical listed entity. Other operating information was also provided, into which both diagnostic and non-diagnostic cues were incorporated.
The experimental case consisted of two parts (A and B). Participants were presented with a one page narrative on the firm and summary financial information from the past 3 years’ audited financial statements. Nine (six diagnostic, three non-diagnostic) pretested cues were incorporated into the other operating information given to respondents in Part A of the case. Diagnostic cues were drawn from International Auditing Standard 240: ‘The auditor's responsibility to consider fraud in an audit of financial statements’ (International Auditing and Assurance Standards Board (IAASB), 2004). A brief definition of the term ‘financial statement fraud’ was then provided. In Part B of the case, participants were provided with four additional non-diagnostic cues that had been identified as factors that have no impact on the risk of financial statement fraud during the pretest. All the non-diagnostic cues used generally presented ‘positive’ information about the firm. This was designed to ensure that a directional prediction can be made for the effect of these non-diagnostic cues on the risk assessments made in Part B of the experimental case.
All cues in the experimental case were pretested by five academics with mean external audit experience of 4.9 years from three of the Big Four audit firms, and two audit managers and an audit partner from one of the Big Four firms with mean experience of 9.5 years. Thirty items of information for the risk of financial statement fraud assessments were presented to the eight pretest subjects. Fifteen pieces of information were selected from International Auditing Standard 240 (IAASB, 2004), the London Stock Exchange Combined Code on internal controls (Turnbull Report, 1999) and the Australian/New Zealand risk management standard (Standards Association of Australia, 2004). Fifteen other pieces of information judged to be unrelated to the fraud risk assessment task were also presented.
Pretest subjects were provided with a one page narrative describing the background of the hypothetical firm before being presented with other operating information containing the selected diagnostic and non-diagnostic cues. They were asked to rate each item as diagnostic (the information taken by itself influences the assessment of the company's financial statement fraud risk) or non-diagnostic (the information taken by itself is of no value in deciding if the company faces a financial statement fraud risk). An assessment scale similar to that used in Shelton (1999) was used. Values in the interval (–5, 5) were assigned to each item, with negative (positive) values indicating whether the item made the financial statement fraud risk higher (lower). A zero value for any item indicated that the item was judged to have no impact on the risk assessment decision. The six highest scoring items given unanimous negative scores (i.e. diagnostic items) and the first seven items unanimously rated zero (i.e. non-diagnostic items) by the pretest subjects were used in the experimental materials. Only items unanimously assigned negative scores were included in the experimental materials as the objective is to examine the directional impact of financial statement fraud cues on directors’ risk assessments.
A decision aid was provided together with the case materials in the decision aid manipulation group. The decision aid lists recognized red-flag cues based on International Auditing Standard 240 (IAASB, 2004). The decision aid (Panel B of Table 1) comprises three red flag cues that match only a subset (three out of six) of the total number of pieces of information present in the experimental case that are relevant for the fraud risk judgement task. It also lists three red flag cues that are not present in the experimental case. This decision aid was designed with a mix of both relevant and non-relevant cues for the chosen financial statement fraud risk assessment context because any operational guide/aid available to users is unlikely to include all relevant information cues, while also excluding all information cues not relevant to a particular judgement decision. Participants were told that this list is by no means exhaustive, and merely provides illustrative examples.7
|
Panel A: Factors in the financial statement fraud risk assessment task |
|
|---|---|
| Pretest impact ratingb | |
| Diagnostic factors used in Part A of the experimental casea | |
| D1: The bonus allocation for top executives is based on increasing percentages of audited profit before taxationd. | −3.6 |
| D3: The group faces pressure to improve on its revenue and profit trends to support stock market prices of its sharesd. | −3.4 |
| D5: Revenues and profits have declined in the past 2 years because of poor economic conditions and the group expects negative cash flows from operating activities in the current yeard. | −3.2 |
| D6: The group is planning to issue more ordinary shares to raise funds for capital investmentd. | −2.8 |
| D4: The Managing Director exerts a dominant presence across all phases of the group's operationsd. | −2.6 |
| D2: Approximately 30 per cent of group revenues are derived from related companies in foreign countriesd. | −2.4 |
| Non-diagnostic factors used in Part A of the experimental casec | |
| N1: The group prides itself on being environmentally responsible, and has voluntarily performed an environmental audit in the current year. | 0 |
| N2: The group has introduced a more stringent quality control procedure for production. | 0 |
| N3: The group is committed to meeting consumer needs and actively develops its ability to better serve their needs. | 0 |
| Non-diagnostic factors used in Part B of the experimental casec | |
| N4: In accordance with the Singapore Exchange listing rules, external audit partner rotation has taken place in the current year. | 0 |
| N5: The same external audit firm also prepares the corporate taxation returns and performs routine, formal administrative company secretarial services for the group. | 0 |
| N6: The group has introduced a middle management job rotation scheme targeted at skills development and team-building. | 0 |
| N7: The group is known for its regular financial support for charitable and social welfare organizations. | 0 |
|
Panel B: Decision aid red flag cues |
|---|
| • The quality of earnings is deteriorating, and/or increased risk taking with respect to credit sales, changes in business practice or selection of accounting policy alternatives that improve income. |
| • Inability to generate cash flows from operations while reporting earnings and earnings growth. |
| • Significant related-party transactions that are not in the ordinary course of business or with related entities not audited or audited by another audit firm. |
| • High turnover rate for key accounting and financial personnel. |
| • Management continuing to employ an ineffective accounting, information technology or internal auditing staff. |
| • Domineering management behaviour in dealing with the auditor, especially involving attempts to influence the scope of the auditor's work. |
- a Diagnostic factors received a non-zero (negative) impact rating by all subjects in the pretest.
- b Means from combined pretest impact ratings. Subjects were asked to rate the impact of factors on a scale of –5 (negative impact) to +5 (positive impact) on the assessment of the financial statement fraud risk.
- c c Non-diagnostic factors received a zero impact rating by all subjects in the pretest.
- d Based on/derived from International Standard on Auditing 240 (IAASB, 2004).
The technical knowledge task comprised 10 questions selected from the US certified public accountant examinations covering knowledge required to perform the task of assessing financial statement fraud risk. These questions are a recognized and well-tested source and have been used in prior studies (e.g. Tan and Kao, 1999). The selected questions deal with internal control issues with implications for financial statement misstatement/fraud (6), related-party transactions (2), basic knowledge of double-entry accounting (1), and financial (gross margin) ratios (1). These questions were specifically chosen on the basis that they do not require complex arithmetical calculations of financial ratios, or detailed knowledge of internal or external audit procedures. All questions were pretested by seven academics with audit and accounting experience, and pilot tested by two council members of the SID; the Cronbach's alpha coefficient was 0.62 for director participants and 0.39 for student participants.8
3.1.3. Procedure
The research instrument was administered in two envelopes (I and II) and participants were given instructions on the sequence in which to perform the tasks: the contents of Envelope I (the experimental case with/without decision aid) was to be completed and re-inserted into the envelope before participants proceeded to the contents of Envelope II (a distractor task, followed by the technical knowledge task).
In Part A of the experimental task, participants were asked to rate their assessment of the likelihood of fraud occurring in the current year's financial statements on an 11-point scale (‘0’: extremely unlikely that fraud will occur; ‘10’: extremely likely that a fraud will occur), and to indicate how confident they were of their likelihood assessment. Participants were then asked to identify the factor(s) from the case that they believed either increased or decreased the risk of financial statement fraud occurring in the current year's financial statements. They were also asked to indicate the perceived impact of each factor they identified by circling a number on a (–5, 5) scale. The experiment requested participants to list factors that they believed either increased or decreased the risk of financial statement fraud occurring; as the scale in the experimental case indicates that a rating of ‘0’ means ‘no impact at all’, both factors selected by participants and given a ‘0’ rating and items that were not selected were coded ‘0’.
In Part B, participants were asked to re-assess the likelihood of financial statement fraud occurring on a (0 to 10) scale (‘0’: extremely unlikely that fraud will occur, ‘10’: extremely likely that a fraud will occur), and to indicate how confident they were of their revised risk assessments. Participants were then asked to identify the factor(s) from the additional case information provided that they believed increased or decreased the risk of financial statement fraud occurring in the current year's financial statements. They were asked to assess the factors chosen in relation to their impact, on a (–5, 5) scale, on the risk of fraud occurring in the financial statements. Again, this (–5, 5) rating was used as the measure of the perceived strength of the identified factors. Participants were also asked to indicate whether they would initiate action on the part of the audit committee or the company on the basis of their revised fraud risk assessments.
After completing both Parts A and B of the experimental case in Envelope I, participants were presented with a separate envelope (Envelope II) and instructed not to refer to the previous materials, questions or answers in completing the questions in the envelope. A distractor task comprising a short questionnaire eliciting information on participants’ demographics was first administered to clear working memory before the technical knowledge task was administered. Dilution was measured as the difference between participants’ initial risk assessments (Part A of the experiment), and the risk assessments made after being presented with the additional non-diagnostic factors presented in Part B of the experiment.
4. Results
4.1. Demographics and manipulation checks
Table 2 reports the descriptive statistics for the dependent and independent variables used in the testing of hypotheses. A manipulation check shows that the mean impact ratings assigned to all diagnostic (non-diagnostic) factors are positive (negative), indicating that both director and student participants interpreted each of the factors included in the experimental case as anticipated. A review of the level of reliance on/self-assessed usefulness of the decision aid by director and student participants in the decision aid treatment group (directors: n= 51; students: n = 62) shows a mean rating of 4.89 (SD 2.55) for directors and 5.16 (SD 1.69) for students out of a scale of ‘0’ (‘not useful – did not refer to it at all’) to ‘10’ (‘made extensive use of it’). These findings indicate that both director and student participants in the decision aid treatment group did read and perceive the decision aid to be of use when performing the assigned task.
| Directors (n = 107) | Students (n = 119) | |||
|---|---|---|---|---|
| Mean | SD | Mean | SD | |
| ASSESS_A | 5.50 | 2.03 | 6.80 | 1.17 |
| ASSESS_B | 3.64 | 1.93 | 4.57 | 1.38 |
| DILUTION | 1.86 | 1.29 | 2.22 | 1.70 |
| TK | 6.82 | 2.16 | 7.52 | 1.44 |
| NUM_DA | 2.89 | 1.37 | 2.97 | 0.97 |
| NUM_NDA | 0.31 | 0.57 | 0.71 | 0.77 |
| NUM_NDB | 2.16 | 1.03 | 2.13 | 0.79 |
| NUM_NDAB | 2.47 | 1.32 | 2.84 | 1.26 |
| D1 (bonus plan) | 2.10 | 1.73 | 3.24 | 1.36 |
| D2 (related party transactions) | 1.18 | 1.71 | 0.89 | 1.59 |
| D3 (pressure to improve revenues) | 1.08 | 1.59 | 1.49 | 1.88 |
| D4 (dominant Managing Director) | 1.16 | 1.70 | 0.62 | 1.28 |
| D5 (poor operating revenue/cash flow) | 1.83 | 1.78 | 1.39 | 1.56 |
| D6 (plan to issue shares) | 1.27 | 1.63 | 1.91 | 1.86 |
| N1 (environmental audit) | −0.22 | 0.79 | −0.79 | 1.21 |
| N2 (customer needs) | −0.26 | 0.84 | −0.21 | 0.70 |
| N3 (product quality control) | −0.30 | 0.89 | −0.57 | 1.04 |
| N4 (compliance with listing rules) | −2.04 | 1.57 | −2.41 | 1.10 |
| N5 (auditor provides non-audit services) | −0.82 | 1.57 | −0.29 | 0.80 |
| N6 (middle management rotation) | −1.97 | 1.48 | −1.27 | 1.32 |
| N7 (supports welfare organizations) | −0.57 | 1.27 | −0.89 | 1.18 |
- ASSESS_A is the risk assessment in Part A of the experiment; ASSESS_B is the risk assessment in Part B of the experiment; DILUTION is the risk assessment for Part A minus risk assessment for Part B of the experiment; TK is the mean-centred technical knowledge score; NUM_DA is the number diagnostic factors identified in Part A of the experiment; NUM_NDA is the number of non-diagnostic factors identified in Part A of the experiment; NUM_NDB is the number of non-diagnostic factors identified in Part B of the experiment; NUM_NDAB is the number of non-diagnostic factors identified in Parts A and B of the experiment; D1–D6: 11-point scale impact ratings of diagnostic factors identified in Part A of the experiment; N1–N3: 11-point scale impact ratings of non-diagnostic factors identified in Part A of the experiment; N4–N7: 11-point scale impact ratings of non-diagnostic factors identified in Part B of the experiment. SD, standard deviation.
4.2. Tests of hypotheses
To test the predicted effects of the presence of a decision aid on the number of diagnostic (Hypotheses 1 and 3) and non-diagnostic (Hypotheses 2 and 4) factors identified by directors and students who possess higher versus lower levels of technical knowledge, multiple regression analyses are performed, with the number of diagnostic factors identified in Part A of the experiment (NUM_DA), along with non-diagnostic factors identified in Part B (NUM_NDB) and both Parts A and B of the experiment (NUM_NDAB) as dependent variables, and technical knowledge (TK) and decision aid (AID) as independent variables.
Regression results in Part A of Table 3 show that in terms of the number of diagnostic factors identified by directors, there is a highly significant TK main effect for dependent variable NUM_DA (p= 0.000), no statistically significant main effect of AID, and contrary to the prediction in Hypothesis 1, no statistically significant TK × AID interaction. These results provide evidence that directors who possess higher levels of technical knowledge identified and processed a significantly larger number of diagnostic factors. However, AID does not moderate the effect of technical knowledge in terms of the identification of diagnostic factors.
|
Panel A: NUM_DA = b0 + b1 (TK) + b2 (AID) + b3 (TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 2.938 (0.169) | 11.408 | 0.000*** |
| TK | 0.633 (0.166) | 3.823 | 0.000*** |
| AID | −0.117 (0.245) | −0.476 | 0.635 |
| TK × AID | −0.103 (0.247) | −0.415 | 0.340 |
| Students | |||
| Intercept | 2.440 (0.130) | 18.742 | 0.000*** |
| TK | 0.713 (0.130) | 5.329 | 0.000*** |
| AID | 0.616 (0.189) | 3.249 | 0.001*** |
| TK × AID | −0.533 (0.215) | −2.479 | 0.008*** |
|
Panel B: NUM_NDAB = b0 + b1(TK) + b2(AID) + b3(TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 2.423 (0.178) | 13.624 | 0.000*** |
| TK | −0.122 (0.174) | −0.699 | 0.243 |
| AID | 0.132 (0.258) | 0.513 | 0.305 |
| TK × AID | 0.360 (0.260) | 1.386 | 0.084* |
| Students | |||
| Intercept | 2.930 (0.178) | 16.443 | 0.000*** |
| TK | −0.641 (0.183) | −3.500 | 0.000*** |
| AID | 0.330 (0.259) | 1.274 | 0.205 |
| TK × AID | 0.181 (0.294) | 0.615 | 0.270 |
|
Panel C: NUM_NDB = b0 + b1(TK) + b2(AID) + b3(TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 2.117 (0.138) | 15.287 | 0.000*** |
| TK | −0.096 (0.136) | −0.705 | 0.241 |
| AID | 0.121 (0.201) | 0.601 | 0.274 |
| TK × AID | 0.296 (0.202) | 1.464 | 0.073* |
| Students | |||
| Intercept | 2.234 (0.117) | 19.124 | 0.000*** |
| TK | −0.278 (0.120) | −2.311 | 0.012** |
| AID | −0.001 (0.170) | −0.006 | 0.498 |
| TK × AID | 0.100 (0.193) | 0.516 | 0.304 |
- *** Significant at p < 0.01;
- ** significant at p < 0.05;
- * significant at p < 0.1. n = 107 (directors); n = 119 (students).
- a One-tailed p-values are reported where the expected sign is unidirectional. All other p-values are two-tailed. NUM_DA is the number of diagnostic factors identified in Part A of the experiment; NUM_NDB is the number of non-diagnostic factors identified in Part B of the experiment; NUM_NDAB is the number of non-diagnostic factors identified in Parts A and B of the experiment; TK is the mean-centred technical knowledge score; AID is equal to 1 when a decision aid is provided, and 0 otherwise.
In terms of the number of non-diagnostic factors identified by directors, Panels B and C of Table 3 show that, consistent with Hypothesis 2, the TK × AID interaction is marginally significant for both NUM_NDB (p= 0.084) and NUM_NDAB (p= 0.073). None of the main effects is statistically significant. Further analysis of the significant TK × AID interaction is performed by comparing the slope of the simple regression of AID on the dependent variable at values of TK one standard deviation above, and one standard deviation below, the centred mean (Cohen and Cohen, 1983; Aiken and West, 1991). The regression slopes of AID on the dependent variables at values of TK one standard deviation above, and one standard deviation below, the centred mean show AID is marginally positively significant for NUM_NDB (p= 0.060) and NUM_NDAB (p= 0.073) at TK levels one standard deviation above the centred mean but not significant at TK levels one standard deviation below the mean. These results provide support for Hypothesis 2: in the presence of the decision aid, directors who possess higher levels of technical knowledge identify and process a larger number of non-diagnostic factors.
These results show that although high-knowledge directors perform significantly better at identifying diagnostic factors, the decision aid does not significantly narrow the difference in the number of diagnostic factors identified by directors who possess higher versus lower levels of technical knowledge. However, as predicted in the case of identification of non-diagnostic factors, high-knowledge directors perform worse (i.e. identify a larger number of non-diagnostic factors) in the presence of a decision aid.
In terms of the number of diagnostic factors (NUM_DA) identified by students, Panel A of Table 3 shows that the main effects of TK and AID are statistically significant (p= 0.000 and p= 0.001, respectively), along with their interaction effect (p= 0.008). Further analysis reveals that AID is highly positively significant (p= 0.002) for NUM_DA at TK levels one standard deviation below the centred mean but is not significant at TK levels one standard deviation above the mean (p= 0.669). Therefore, the decision aid improves performance in students who possess lower levels of technical knowledge in the task of identifying diagnostic factors, a result consistent with predictions made in Hypothesis 3.
In terms of the number of non-diagnostic factors (NUM_NDAB, NUM_NDB) identified by students participants, results in Panels B and C of Table 3 reveal a statistically significant negative effect of TK (p= 0.000 and p= 0.012, respectively), but neither the main effect of AID nor the interaction variable is statistically significant. Students who possess higher levels of technical knowledge identify a smaller number of non-diagnostic factors than those with lower levels of technical knowledge. Although the decision aid does not significantly improve student performance in the identification of non-diagnostic factors, and Hypothesis 4 is not supported, the absence of a significant TK × AID interaction is in contrast to director participants, where Hypothesis 2 is supported.
In summary, the positive effect of the decision aid in terms of the number of diagnostic factors identified by student participants, when juxtaposed with the evidence of a negative decision aid effect in terms of the number of non-diagnostic factors identified by high-knowledge directors, suggests that it is the job knowledge or experience of directors that leads to dysfunctional decision-aid effects among high-knowledge directors.
4.3. Effects of diagnostic and non-diagnostic factors identified on risk assessments
To test the effects of the number of identified diagnostic and non-diagnostic factors on risk assessments made by director and student participants, multiple regression analyses are run on the risk assessments made in Parts A and B of the experimental case (ASSESS_A, ASSESS_B), with the number of diagnostic factors (NUM_DA) and the number of non-diagnostic factors identified in both Parts A and B of the experimental case (NUM_NDAB) as independent variables, and with technical knowledge (TK), decision aid (AID) and the interaction of technical knowledge and decision aid (TK × AID) as control variables. An additional regression is run on dilution (DILUTION) in risk assessments (ASSESS_B–ASSESS_A), with the number of non-diagnostic factors identified in Part B of the experimental case (NUM_NDB) as the independent variable.9
In the case of directors, regression results in Panels A and B of Table 4 indicate that NUM_DA has a significant (p= 0.002) positive effect on the risk assessment made in Part A of the experiment (ASSESS_A), while NUM_NDA has a significant (p= 0.009) negative effect on ASSESS_A. Results of the regression on the risk assessment made in Part B of the experimental case (ASSESS_B), where directors are asked to make a revised fraud risk assessment after being presented with four additional non-diagnostic pieces of information, show that NUM_DA has a marginally significant (p= 0.067) positive effect on the risk assessment made. The total number of non-diagnostic factors identified in Parts A and B of the experiment (NUM_NDAB) has a highly significant (p= 0.001) negative effect on the risk assessment made in Part B of the experiment (ASSESS_B). Results in Panel C of Table 4 show that the number of non-diagnostic factors identified in Part B of the experiment (NUM_NDB) has a highly significant (p= 0.000) positive effect on dilution. Neither TK, AID, nor their interaction has a direct effect on all three dependent variables (ASSESS_A, ASSESS_B, and DILUTION), suggesting that it is the number of diagnostic and non-diagnostic factors that directly influenced these variables.
|
Panel A: ASSESS_A = b0 + b1(NUM_DA) + b2(NUM_NDA) + b3(TK) + b4(AID) + b5(TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 4.163 (0.520) | 8.011 | 0.000*** |
| NUM_DA | 0.465 (0.159) | 2.918 | 0.002*** |
| NUM_NDA | −0.824 (0.345) | −2.388 | 0.009*** |
| TKb | 0.140 (0.276) | 0.508 | 0.311 |
| AIDb | 0.487 (0.379) | 1.284 | 0.101 |
| TK × AID | −0.319 (0.383) | −0.832 | 0.204 |
| Students | |||
| Intercept | 5.537 (0.323) | 17.121 | 0.000*** |
| NUM_DA | 0.329 (0.122) | 2.704 | 0.004*** |
| NUM_NDA | −0.023 (0.147) | −0.156 | 0.438 |
| TKb | 0.495 (0.202) | 2.457 | 0.008*** |
| AIDb | 0.349 (0.245) | 1.420 | 0.079* |
| TK × AIDb | −0.471 (0.274) | −1.719 | 0.044** |
|
Panel B: ASSESS_B = b0 + b1(NUM_DA) + b2(NUM_NDAB) + b3(TK) + b4(AID) + b5(TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 3.851 (0.539) | 7.150 | 0.000*** |
| NUM_DA | 0.231 (0.153) | 1.508 | 0.067* |
| NUM_NDAB | −0.451 (0.145) | −3.106 | 0.001*** |
| TKb | 0.175 (0.265) | 0.660 | 0.256 |
| AIDb | 0.450 (0.362) | 1.242 | 0.108 |
| TK × AIDb | −0.317 (0.369) | −0.858 | 0.196 |
| Students | |||
| Intercept | 4.112 (0.538) | 9.911 | 0.000*** |
| NUM_DA | 0.385 (0.142) | 2.699 | 0.004*** |
| NUM_NDAB | −0.318 (0.104) | −3.055 | 0.001*** |
| TKb | 0.436 (0.237) | 1.841 | 0.034** |
| AIDb | 0.344 (0.282) | 1.220 | 0.113 |
| TK × AIDb | −0.648 (0.318) | −2.039 | 0.044** |
|
Panel C: DILUTION = b0 + b1(NUM_NDB) + b2(TK) + b3(AID) + b4(TK × AID) |
|||
|---|---|---|---|
| b (SE) | t-statistic | p-valuea | |
| Directors | |||
| Intercept | 0.534 (0.278) | 1.922 | 0.057* |
| NUM_NDB | 0.617 (0.109) | 5.637 | 0.000*** |
| TKb | 0.139 (0.151) | 0.922 | 0.179 |
| AIDb | −0.015 (0.223) | −0.067 | 0.474 |
| TK × AIDb | −0.100 (0.227) | −0.438 | 0.331 |
| Students | |||
| Intercept | 0.350 (0.305) | 1.149 | 0.253 |
| NUM_NDB | 0.830 (0.119) | 6.984 | 0.000*** |
| TKb | 0.054 (0.157) | 0.346 | 0.365 |
| AIDb | 0.068 (0.217) | 0.314 | 0.754 |
| TK × AIDb | 0.180 (0.246) | 0.730 | 0.467 |
- *** Significant at p < 0.01;
- ** significant at p < 0.05;
- * significant at p < 0.10. n = 107 (directors); n = 119 (students).
- a One-tailed p-values are reported where the expected sign is unidirectional. All other p-values are two-tailed.
- b Control variable. ASSESS_A is the risk assessment in Part A of the experiment; ASSESS_B is the risk assessment in Part B of the experiment; DILUTION is the risk assessment for Part A minus risk assessment for Part B of the experiment; NUM_DA is the number of diagnostic factors identified in Part A of the experiment; NUM_NDA is the number of non-diagnostic factors identified in Part A of the experiment; NUM_NDB is the number of non-diagnostic factors identified in Part B of the experiment; NUM_NDAB is the number of non-diagnostic factors identified in Parts A and B of the experiment; TK is the mean-centred technical knowledge score; AID is equal to 1 when a decision aid is provided, and 0 otherwise.
Results reported in Table 4 also show that directors’ risk assessments are influenced by the number of identified diagnostic and non-diagnostic factors; risk assessments are more (less) extreme when a larger (smaller) number of diagnostic factors is identified and processed, and less (more) extreme when a larger (smaller) number of non-diagnostic factors is identified and processed. Dilution in risk assessments is also greater when a larger number of non-diagnostic factors is identified and processed.
The combined results of the regression analyses show that directors’ fraud risk assessments are more extreme when a larger (smaller) number of diagnostic (non-diagnostic) factors is identified and processed. High-knowledge directors perform significantly better at identifying diagnostic factors, and there is evidence that the provision of a decision aid to experienced and knowledgeable users (e.g. directors) has associated costs in the form of unintended behavioural effects that impair task performance.10
Results for student participants are similar to those obtained for directors. However, a notable difference can be seen in the statistically significant TK × AID interaction variable on ASSESS_A and ASSESS_B (p= 0.044 in both cases).
Further analysis reveals a significant positive main AID effect at values of TK one standard deviation below the centred mean for ASSESS_A and ASSESS_B (p= 0.039 and p= 0.033, respectively), but no significant main AID effect at values of TK one standard deviation above the centred mean. These results suggest that with respect to students with no job experience (and in contrast to directors), TK and AID jointly have a direct effect on risk assessments ASSESS_A and ASSESS_B.
4.4. Tests of mediation
To test whether the effects of TK and AID (along with their interaction) on risk assessments is mediated by the number of diagnostic and non-diagnostic cues identified, tests of mediation are performed. Kenny et al. (1998, p. 260; see also David Kenny's website: http://davidakenny.net/cm/mediate.htm) state that mediation can be established if: (1) the independent variable is related to the dependent variable; (2) the independent variable is related to the mediator; and (3) the mediator is in turn related to the dependent variable.11 In the context of this study, the second requirement is met by the earlier regression analysis results reported in Table 3, which show that TK and the interaction variable TK × AID have significant effects on the mediator variables NUM_DA, NUM_NDAB and NUM_NDB for both director and student participants (AID also has significant effects on the mediator variables NUM_DA, NUM_NDAB and NUM_NDB for student participants). Step three of the mediation analysis is also fulfilled for director participants as regression results reported in Table 4 show that the mediator variable NUM_DA (NUM_NDAB) has a significant positive (negative) influence on ASSESS_A and ASSESS_B while controlling for the independent variables TK and AID. In addition, NUM_NDB has a significant positive influence on DILUTION while controlling for the independent variables TK and AID. In the case of directors, the independent variables TK and AID and the interaction variable TK × AID are also insignificant when the mediator variables NUM_DA, NUM_NDAB and NUM_NDB are included in the regression equations as reported in Table 4, which suggests that NUM_DA, NUM_NDAB and NUM_NDB fully mediate the relationship between the independent variables and the dependent variables ASSESS_A, ASSESS_B and DILUTION. In the case of student participants, the independent variables TK and AID, and the interaction variable TK × AID are generally either insignificant or less significant when the mediator variables NUM_DA, NUM_NDAB and NUM_NDB are included in the regression equations as reported in Table 4, suggesting that, in relation to student risk assessments, NUM_DA, NUM_NDAB and NUM_NDB partially mediate the relationship between the independent variables and the dependent variables ASSESS_A, ASSESS_B and DILUTION.
5. Conclusion
In the present study, I show that directors who possess higher levels of technical knowledge perform better at identifying diagnostic (risk) factors associated with financial statement fraud. Although not all directors might assume the responsibility, authority and liability relating to oversight tasks relating specifically to financial matters in their companies, these results provide evidence to support recent moves by regulators to require greater expertise among directors, and, in particular, greater financial expertise among audit committee members (Blue Ribbon Report, 1999; United States Public Laws, 2002).
However, in the presence of a generic decision aid, I find that directors with higher technical knowledge perform worse in that they consider more non-diagnostic factors as relevant compared to directors with lower technical knowledge. These results suggest that caution should be exercised in the use of generic decision aids as a supplemental aid to directors’ decision-making, as unintended dysfunctional effects can occur.
In contrast to the results for the director participants, I find that with student participants, the presence of a decision aid facilitates students with lower technical knowledge in their identification of diagnostic factors. Furthermore, the adverse effect of a decision aid on high-knowledge directors in terms of the erroneous identification of non-diagnostic factors as relevant does not occur for students. These results suggest that generic decision aids can be useful, but only for relatively naïve decision-makers with little business experience. The student results also imply that the dysfunctional effects of the decision aid for director participants (but not for students) can be attributable to the business knowledge that directors bring to bear on the judgement task.
In additional analyses, I find that higher fraud risk assessments are made when more (fewer) diagnostic (non-diagnostic) factors are identified and processed, and that the identification and processing of a larger number of non-diagnostic factors is associated with less extreme (diluted) risk assessments. There is also support for a mediation model where technical knowledge and decision aid influence the number of diagnostic and non-diagnostic factors identified, which in turn influence risk assessments made.
These findings contribute to the accounting and psychology literature as they provide the first evidence that technical knowledge and decision aid use interactively determine decision-makers’ identification of diagnostic and non-diagnostic factors, and that the specific effects depend on the participant group (directors versus students). The findings reported in the present paper also have practical implications as they provide the first evidence on how directors’ identification of fraud risk factors is jointly influenced by their knowledge and the presence of generic decision aids. The susceptibility of directors to the diluting effect of non-diagnostic information is an area of concern and has implications for board and audit committee staffing and composition. Specifically, boards and audit committees comprising members who lack relevant technical knowledge are less likely to be able to effectively perform the important tasks of fraud risk identification and evaluation. In addition, my finding that the number of diagnostic factors identified mediate the effects of knowledge and decision aids on risk assessments suggest that directors’ performance on risk assessments can be also improved by better equipping or assisting them to distinguish between diagnostic and non-diagnostic information. However, this study's results find that the provision of generic decision aids to assist directors to make fraud risk judgements is not necessarily the answer. Although a generic decision aid can serve as a valuable tool for supplementing lack of knowledge on an unstructured judgement task to the novice user (e.g. students), the provision of a relatively simple decision aid might not improve task performance in experienced but low-knowledge users (e.g. low-knowledge directors), and can instead spur experienced and high-knowledge users (e.g. high-knowledge directors) to want to outperform it (Ashton, 1990), and to invoke ‘illusorily correlated’ cues that cause them to perform more poorly (Chapman and Chapman, 1967).
These findings suggest that directors’ performance on risk assessment tasks might be enhanced more through task-specific training and on-the-job experience than through providing them with simple and generic decision aids. Furthermore, any decision aids designed for use by directors are likely to require the incorporation of more specific, detailed and contextual guidance (as opposed to simple generic decision aids that provide only limited assistance for unstructured tasks) in order for the desired objective of enhancing task performance to be achieved. Organizations should also ensure that sufficient expert advice or assistance is available on hand for directors who do not possess the requisite levels of technical knowledge for the risk management/assessment tasks they need to perform.
The present study has several limitations that should be borne in mind when interpreting the results. First, the diagnostic and non-diagnostic factors incorporated into the experimental case and the decision aid cues were limited in number to keep the time needed to complete the experimental tasks at a manageable length. Second, student participants used in this study might not be good surrogates for inexperienced directors. Furthermore, although one of the objectives of the present research is to investigate the ability of directors to perform risk management oversight tasks, the present study deals only with the preliminary processes of risk identification and risk assessment. The actions and measures that directors take in response to the fraud risk assessments made present opportunities for future research.
Additional research can investigate the effects of decision aids on director performance in the context of performance pressure (DeZoort et al., 2006) to ascertain if the dysfunctional effects of a generic decision aid on high-knowledge directors found by Ashton (1990) and in the present study can be mitigated by explicit accountability pressure imposed on directors. Further research can also be conducted to investigate the intervening effects of certain firm-specific variables that influence the risk assessment procedures of directors; these include block ownership, shareholder dispersion, profitability and leverage, among others. An investigation into firm characteristics will allow for a better appreciation by policy-makers of the many facets to the processes of risk identification and evaluation within the broader context of promoting good corporate governance.
Notwithstanding these limitations, the present study contributes to the extant literature by providing evidence on how technical knowledge and decision aid use influence directors’ identification and processing of diagnostic and non-diagnostic evidence, which in turn influence the fraud risk assessments they make.
