xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001795461
007 cr mnu|||uuuuu
008 070710s2006 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001572
Taylor, Eileen Zalkin.
The effects of in-group bias and decision aids on auditors' evidence evaluation
h [electronic resource] /
by Eileen Zalkin Taylor.
[Tampa, Fla] :
b University of South Florida,
ABSTRACT: This study examines the effect of in-group bias and decision aid use on auditor judgments, confidence, and decisions in an analytical procedures task. In-group bias, a product of Social Identity Theory, may impair auditor independence by influencing auditor judgments. Auditors rely on client representations to support their opinion of the financial statements; however, clients are sometimes former auditors of the external audit firm. This prior relationship could lead the auditor to exhibit unwarranted trust of client representations. In an online mixed design experiment using staff and senior auditors, I test whether auditor judgments, confidence in those judgments, and decisions to extend testing differ based on a client's prior affiliation. I find that there is insufficient evidence of in-group bias in auditor judgments, confidence, or decisions. Lack of support could be due to the small sample size. In the same experiment, I give auditors access to a decision aid. Practice and prior literature suggest using decision aids should improve audit judgment. I find that a structured decision aid improves audit judgments and decisions for all auditors, and improves confidence for auditors who initially made good judgments. Audit managers can benefit from noting the usefulness of decision aids in improving judgment.
Dissertation (Ph.D.)--University of South Florida, 2006.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 129 pages.
Adviser: Uday S. Murthy, Ph.D.
x Business Administration
t USF Electronic Theses and Dissertations.
The Effects of In-Group Bias and Decision Aids on AuditorsÂ’ Evidence Evaluation by Eileen Zalkin Taylor A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy School of Accountancy College of Business Administration University of South Florida Major Professor: Uday S. Murthy, Ph.D. Stephanie Bryant, Ph.D. Terry Engle, Ph.D. Joseph Vandello, Ph.D. Date of Approval April 6, 2006 Keywords: Experimental, judgment, decision-making, auditing, social identity Copyright 2006, Eileen Zalkin Taylor
DEDICATIONS I dedicate this dissertation to my family. Glenn, my husband, you have supported me every step of the way, taking on all of the daily responsibilities of raising our children and keeping our home, as well as being a gr eat husband and dad. You managed to do all of these things, and keep an eye on me at th e same time. You recognized when I needed a break from work, and made sure I rested. For over twenty years, you have been my best friend. I look forward to twenty more. I also dedicate this work to my ch ildren: Adam, Jordan, and Isabella. Adam, you have grown up to be a wonderful, happy, person. You are dedicated to your goals, kind to everyone you meet, and able to put people at ease. Hold fast to your dreams and you will reach your goals. Someday, IÂ’ll have seats on the glass right behind you, telling everyone, Â“ThatÂ’s my son and I love him!Â” Jordan, you are my Renaissance man. A second child, like me, you are independent and fearless. I admire you for your willingness to try anything and everything, and to do it all with heart. School, sports, spirituality, music, art and politics Â– you know something about them all. You make me proud and I love you. Isabella, you and I started school the same year: for you it was the beginning of your education, for me, it was the beginning of a second career. You are assertive, yet perceptive. You do best when challenged and I encourage you to strive for more and aim for higher heights. You have a zest for life and I know will accomplis h so much. I love you. I thank my mom, Susan Zalkin, for belie ving in me always, and for knowing just the right thing to say; I hope to make you proud. I thank my dad, Max M. Zalkin, for
teaching me that I could do anything, and fo r introducing me to the world of business from a very young age. I thank my in-laws Jerry Taylor, for careful ed iting of my initial drafts, and Paloma Sparrowhawk for reminding me to take time to play.
ACKNOWLEDGMENTS To Dr. Uday Murthy, my dissertation ch air, thank you for expecting more and more Â– your high expectations encouraged me to keep working, even when I was tired. I am grateful to have learned about resear ch design and the written word from you. I thank Dr. Stephanie Bryant, my mentor at the University of South Florida and dissertation committee member. Ov er the last four years, yo u have been available to answer my questions, listen to me talk, and lead me in the right direction. Whenever I sent an e-mail, day or night, I would get an instant response. Your accomplishments stand as a reminder that I, too, can succeed. Thank you, Dr. Terry Engle, for servi ng on my dissertation committee. You took this responsibility so seriously, and I appr eciate all of your pe rceptive comments and suggestions. You also expect the best a nd I am thankful for your time and effort. Dr. Joseph Vandello, thank you for serv ing on my committee. You taught me what I needed to know about social psychol ogy. You always welcomed me in your office and took the time to make sure I really understood the theory behind the behavior. To Dr. Jacqueline Reck, thank you for supporting me and for acting as an advocate for the students in our program. You make sure we are on track and give us the best chance to succeed. To all of my professors at the University of South Florida, thank you for preparing me to be the best, not only in research, but in te aching and service as well. Thank you, Melanie Kelly, for always reminding me to keep things in perspective. Finally, thanks to Ann Dzuranin, an ever-present source of encouragement.
i Table of Contents List of Tables iii List of Figures iv Abstract v Chapter 1: Introduction 1 Chapter 2: Literature Review and Hypotheses 6 2.1 Introduction 6 2.2 Analytical Procedures and Source Reliability Judgments 6 2.2.1 Analytical Procedures 6 2.2.2 The Judgment Process 8 2.2.3 Source Reliability Judgments 11 2.2.4 ClientsÂ’ Insider Knowledge of Audit Process 12 2.3 Auditor Affiliation and Related Studies 13 2.4 Auditing Judgment: Biases and Social Identity 16 2.4.1 Biases in Auditing Judgments 16 2.4.2 Social Identity Theory 17 2.4.3 Inter-group Bias in Social Psychology 18 2.4.4 In-group Bias in Auditing 21 2.4.5 A Normative View 22 2.5 Statement of Hypotheses Â– In-group Bias 23 2.6 Discussion of Potential Covariates 26 2.7 Debiasing in Auditing 27 2.8 Debiasing with Decision Aids 28 2.9 Decision Aid Reliance in Auditing 30 2.10 Statement of Hypotheses Â– Performance Improvement 31 Chapter 3: Method 35 3.1 Introduction 35 3.2 Sample 35 3.3 Experimental Task 36 3.4 Research Design 37 3.4.1 Procedure 37 3.4.2 Characterization of Client Explanation 39 3.5 Independent Variables 40
ii 3.5.1 Between-subjects treatment: Group Affiliation 40 3.5.2 Within-subjects treatment: Decision Aid 41 3.6 Dependent Variables 42 3.6.1 Plausibility 42 3.6.2 Confidence 42 3.6.3 Extent of Testing 43 3.7 Internal and External Validity 43 3.7.1 Internal Validity 43 3.7.2 External Validity 47 3.8 Manipulation Checks 47 3.9 Planned Statistical Analyses 48 3.10 Pilot Study 50 3.10.1 Pilot Background and Descriptive Statistics 50 3.10.2 Pilot Study Results 52 3.10.3 Discussion of Design Changes 53 Chapter 4: Results 55 4.1 Background and Descriptive Statistics 55 4.2 Correlation Matrices 58 4.3 Statistical Analysis 60 4.3.1 In-Group Bias and Its Effect on Initial Audit Judgment 61 4.3.2 In-Group Bias and Its E ffect on Auditor Confidence 63 4.3.3 In-Group Bias and Its Eff ect on Auditor Decisions to Extend Testing 66 4.3.4 Discussion of Analysis of Decision Aid Hypotheses 67 4.3.5 Decision Aid Use and its Effect on Auditor Plausibility Judgments 68 4.3.6 Effect of Decision Aid on Confidence 70 4.3.7 Effect of Decision Aid on Extent of Testing 71 4.4 Post Hoc Analysis 73 Chapter 5: Conclusion 78 5.1 Discussion of Results 78 5.2 Summary 83 5.3 Limitations 85 5.3.1 Small Sample Size 85 5.3.2 Alternative Explanations 86 5.3.3 Experimental Context 87 5.4 Future Research 88 References 91 Appendix A: Client Background and DecisionSERVE Report 101 Appendix B: Survey Instrument 103 About the Author end page
iii List of Tables Table 1 Pilot Study 51 Table 2 Participant Demographics for Group Hypotheses Tests 57 Table 3 Correlation Matrix 59 Table 4 Tests of Normality 60 Table 5 Test of In-Group Bias on Initial Plausibility Judgment 62 Table 6 Test of In-Group Bias on Initial Confidence 64 Table 7 Test of In-Group Bias on Decision to Extend Testing 67 Table 8 Participant Demographics for Decision Aid Hypotheses Tests 68 Table 9 Effect of Decision Aid on Auditor Plausibility Judgments 69 Table 10 Change in Confiden ce Post-Decision Aid 71 Table 11 Test of Effect of Decisi on Aid on Extent of Testing 72 Table 12 Post Hoc Analysis of Extent of Testing 74 Table 13 Post Hoc Analysis of Confidence by Initial Judgment 75 Table 14 Post Hoc Analysis of Extent of Testing by Init ial Judgment 76 Table 15 Summary of Findings 78
iv List of Figures Figure 1. Model of the Evaluation of Causes in Analytical Procedures 9 Figure 2. Auditor Affiliations 14 Figure 3. Primary Co ncerns and Motives of the Social Self: a Taxonomy 20 Figure 4. Diagram of the Experimental Design 38 Figure 5. Frequency Di stribution of Time to Complete for All Participants 56
v The Effects of In-Group Bias and Decision Aids on AuditorsÂ’ Evidence Evaluation Eileen Zalkin Taylor ABSTRACT This study examines the effect of in-g roup bias and decision aid use on auditor judgments, confidence, and decisions in an an alytical procedures task. In-group bias, a product of Social Identity Theory, may im pair auditor independence by influencing auditor judgments. Auditors rely on client representations to support their opinion of the financial statements; however, clients are sometim es former auditors of the external audit firm. This prior relationship c ould lead the auditor to exhibi t unwarranted trust of client representations. In an online mixed design expe riment using staff and senior auditors, I test whether auditor judgment s, confidence in those judgments, and decisions to extend testing differ based on a clientÂ’s prior affiliati on. I find that there is insufficient evidence of in-group bias in auditor judgments, confiden ce, or decisions. Lack of support could be due to the small sample size. In the same e xperiment, I give auditors access to a decision aid. Practice and prior literature suggest using decision aids should improve audit judgment. I find that a structured decision ai d improves audit judgments and decisions for all auditors, and improves c onfidence for auditors who initially made good judgments. Audit managers can benefit from noting th e usefulness of decisi on aids in improving judgment.
1 CHAPTER 1: INTRODUCTION This study examines whether in-group bias an inclination to trust oneÂ’s own group members, affects auditorsÂ’ judgment s, confidence in those judgments, and decisions in an analytical procedures tas k. It also explores whether a decision aid successfully mitigates in-group bias and impr oves auditorsÂ’ decisions. Auditors perform analytical procedures in which they gather information from multiple sources to justify and explain changes in account balances; they often rely on client representations for supporting evidence (Biggs et al. 1995). When evaluating client representations, auditors must consider the clientÂ’s source reliability, which includes both competence and objectivity (Hirst 1994).1 In group bias, which occurs when group members extend unjustified trust to other group members (Hew stone et al. 2002), could impact this evaluation.2 As companies hire members of their external audit firm to work in key financial positions, former auditors become clie nts, yet current auditors may still consider them group members. The resulting in-group bi as could lead auditors to overrate a clientÂ’s objectivity, which would lead to an inappropriately high source reliability judgment. Auditors could conclude that evid ence is sufficient when it is insufficient, prematurely end the search for additional or corroborating eviden ce, or exhibit an unjustified confidence in the fi nal audit opinion. All of these outcomes could result in an ineffective audit. 1 Objectivity, in this context, simply refers to the client Â’s willingness to be truthful to the auditor. In other auditing literature, objectivity is also a measure of bias in an individualÂ’s judgment. In this study, I am exploring the effect of a bias on the auditorÂ’s judgment of the clientÂ’s objectivity. 2 Biases in auditor judgments, specifically those rela ted to the auditorÂ’s evaluation of source reliability, have been the topic of several auditing studies (Anderson et al. 2003, Anderson et al. 1994, Bamber 1983, Hirst 1994).
2 Companies often hire employees from thei r external audit firm, (Beasley et al. 2000; Bleed 2002; Lennox 2005). In several of the most recent audit failures, highranking accounting personnel were also alumn ae of the companyÂ’s external audit firm (Barrionuevo 2002). Congress has recognized the potential for in-group bias to influence auditor judgments and has restricted public co mpanies from hiring their external auditors in positions of financial au thority for a one-year period. 3 AICPA Ethics Interpretation 101-2 cautions that client hiring of their ex ternal auditors might impair independence (AICPA 2005). The Independence Standard s Board also warned that auditor independence could be threatened by their fa miliarity or prior longstanding relationships with attest client (Independe nce Standards Board 2000). While these bodies recognize ingroup bias as a threat, empirical studies of the phenomenon in auditing are scarce. King (2002) demonstrated in-group bias among aud itors in a behavioral experiment. Lennox (2005) found that companies with affiliated executives (employees who were former members of the current external audit firm) were more likely to receive a clean opinion than companies without affiliated executives, and Menon and Williams (2004) found evidence of abnormal accruals in firms with affiliated executives. These studies suggest that auditors may e xhibit in-group bias. Even though the above studies suggest in -group bias exists among auditors, there are some reasons why individual auditors may be immune to this bias. Auditors have strict professional standards, training in professional skepticism and independence, are 3 Sarbanes-Oxley mandates that the CEO, Controller, CFO, Chief Accounting Officer or person in an equivalent position cannot have been employed by the company's audit firm during the 1-year period preceding the audit (Sarbanes-Oxley Act (SOX) 2002).
3 subject to public accountabil ity, and must meet stringent exam requirements. These factors suggest an auditor could accurately assess client objectivit y, or lack thereof, regardless of past associations. Indeed, Bamb er et al. (1995) notes several studies which suggest that auditors are less sus ceptible to psychological biases. Whether and to what extent auditors demonstrate in-group bias toward former audit team members is an empirical question. It is important to answer this question since this bias could threaten the auditorÂ’s prof essional judgment, resulting in an unacceptably high risk of audit failure. Given the widespread practi ce of companies hiring their external auditors, and considering the recent C ongressional laws, I firs t test the extent of in-group bias on auditor judgments and decisi ons. I then test whet her a decision aid can improve audit judgments. Practitioners use decision aids to im prove auditing (Bedard and Graham 2002). These aids support decision-making by ove rcoming human information processing limitations (Rose 2002), automating structur ed decisions (Abdolmohammadi 1991), and providing models and data to assist the audito r in choosing between al ternatives in semistructured tasks (Abdolmohammadi 1991). I us e analytical procedures in this study because they are semi-structured tasks: they have a reasonably well-defined problem, with limited alternatives, re quiring some judgment (Abdol mohammadi and Usoff 2001; Abdolmohammadi 1991). I supply auditors with a decision aid that lists plausible explanations for a given account fluctuation. If in-group bias causes auditors to exhibit unjustified trust, which results in an incorrect audit judgment, a decision aid could provide the auditor with guidance regarding the correct judgment, thus mitigating the
4 negative effect of in-g roup bias. In practice, this deci sion aid could be something as simple as a listing of expected account rela tionships or something more complex, such as an interactive computer model that provides a probability report. A decision aid can also improve audit judgment in the absence of gr oup bias by directing attention to relevant indicators. I use a mixed design experiment with one between-participants variable, group affiliation (in and out) and one within-participants variable, decision aid (pre and post). Senior and staff auditors evaluate a client -provided explanation for the results of an analytical procedure; the clie nt is either a former audit team member or a longstanding client employee. Dependent variables, measur ed preand post-decision aid, include the auditorÂ’s plausibility judgme nt, his confidence in that j udgment, and his decision about how much to extend audit test ing. After evaluating demographi cs (task experience, level, and affiliation) as possible covariates, I analyze the data using the appropriate statistical methods. To summarize, first, I test for in-group bi as in an audit context, exploring how this bias affects auditor judgment, confidence in that judgment, and decisions to extend audit testing. Second, I evaluate whether a deci sion aid is effective in improving audit judgments, confidence and decisions. Based on data collected, th ere is insufficient evidence to conclude that in-group bias exis ts among auditors perf orming an analytical procedure. Further, there is no indication that in-group bias affects either confidence or decisions to extend testing. Findings indicat e that decision aid use improves auditor
5 judgments and auditor decisions for all au ditors, but only impr oves confidence for auditors who initially provi ded a correct judgment. One cannot conclude that insignificant findings indicate an absence of in-group bias. The small sample size and use of nonparametric tests reduce the likelihood of finding an effect, should one exist. Future re search with a larger sample could yield results that are more conclusive. Audit firm s should be especially interested in the findings related to the effec tiveness of the decision aid in improving both judgments and decisions. The simple decision aid used for this task offers a feasible and cost-effective tool for practice improvement. The dissertation continues as follows: Ch apter 2 includes the literature review and development of the hypotheses, Chapte r 3 describes the method, discusses the research design and provides results of the pilot test, Chapter 4 includes the statistical analysis, and Chapter 5 concludes with a di scussion of results, limitations, and future research..
6 CHAPTER 2: LITERATURE RE VIEW AND HYPOTHESES 2.1 Introduction This section provides an overview of the audit and psychology literature pertaining to this study. The firs t part of this section focu ses on a review of the audit literature related to analytical procedures a nd the auditorÂ’s judgment process during those procedures. I use the Anderson and Koonce (1998) model, which is a two-step approach including plausibility and sufficiency checks. I then integrate source reliability literature from the audit field, focusing on the auditorÂ’s judgment of client objectivity. I proceed to review the recent studies on auditor affiliati on, discussing how affiliation can impact the audit process. I follow this di scussion by a review of the seminal literature on social identity theory (Tajfel 1981), moving toward a definition of inter-group bias, and finally discussing how group biases can impact audito rs. Hypotheses related to in-group bias are then stated. The second part of this section provides a discussion of the aud it-related debiasing literature, followed by a review of how decision aids can be used to improve auditor judgment and decisions. 2.2 Analytical Procedures and Source Reliability Judgments 2.2.1 Analytical Procedures Generally Accepted Auditing Standards require auditors to obtain sufficient, competent, evidential matter to reasonably s upport their opinion of a clientÂ’s financial
7 statement presentation and disclosures (AICPA 2005). AU Section 329 Â“Analytical ProceduresÂ” defines these procedures as Â“Â…evalua tions of financial information made by a study of plausible rela tionships among both financia l and nonfinancial dataÂ” (AICPA 2005, 465). This standard requires auditors to perfor m analytical procedures in both the planning and review stages of th e audit (AICPA 2005). When used during planning, analytical procedures help aud itors identify accounts that need further investigation, allowing them to budget more time and testing to these areas. AU Section 329.09 also suggests that auditors use analytical procedures as a substantive test during fieldwork to obtain evidence about financ ial statement assertions (AICPA 2005). Analytical review procedures require aud itors to develop exp ectations about account balances based on their knowledge of internal and external factors. These factors might relate to industry averages, current econom ic indicators, changes in accounting or operations policies, or firm-specific growt h. After developing expectations, auditors compare their expectations to financial stat ement assertions, investigating differences. This investigation requires auditors to gather information from multiple sources. Auditors must also have a complete understanding of how accounts are related, in order to assess the reasonableness of account balance changes. Practicing auditors commonly place gr eat reliance on analytical procedures (Anderson et al. 1994; Biggs et al. 1995). Hirst and Koonce (1996) conducted a field study in which they interviewed 36 audit prof essionals (seniors, managers, and partners) about the use of analytical procedures in prac tice. They found that au ditors use analytical procedures as a substantive test to dete rmine account balance validity. Further, during
8 substantive testing, auditors emphasize Â“Â…th e explanation developm ent /evaluation and information search aspectsÂ…Â” of analytic al procedures (Hirst and Koonce 1996, 476). The study also notes that although auditors report improvements in judgment over time (as they gain experience) they admit that ev en with additional experience their confidence in evaluating client explanations remains low. Auditors obtain explanations for unexpect ed fluctuations from multiple sources (Anderson et al. 2003). Changes in balanc es may relate to external events ( e.g., a change in general economic indicators or competition), or may result from internal client decisions ( e.g., discontinuation of a product line or re placement of depreciated assets). When changes result from external causes, a uditors can probably gather evidence from objective external sources. However, when changes result from internal management decisions, auditors often rely on client expl anations. In practice, Hirst and Koonce (1996) find that while experienced auditors are more likely to self-generate possible explanations before turning to the client, less experienced auditors are more likely to ask the client first. Both experienced and novice auditors turn to the client e ither to confirm or to seek explanations regarding the cau ses of observed fluctuations. The degree of reliance the auditor places on these explanat ions depends on his assessment of the clientÂ’s objectivity. 2.2.2 The Judgment Process According to the Anderson and Koonce (1998) model in Figure 1, auditors proceed through a two stage proc ess when evaluating evidence. Auditors start with a twostep plausibility check. In the first step, they assess whether an explanation is consistent with the observed fluctuation. Fo r example, if net income in creases, an increase in sales
9 would be a plausible reason; an increase in common stock would not be a plausible reason. Once auditors judge a cause consistent with the fluctuation, they then consider whether the cause is consistent with the available information. Did sales, in fact, increase? If the evidence shows that sales decreased, th is explanation would not be consistent with the facts and auditors would judge th is explanation implausible (Anderson and Koonce 1998). After compiling a list of plausible hypotheses, auditors perform the sufficiency evaluation task. This evaluation requires an assessment of how much of the variation in the account is explained by the explanation overall. Figure 1 Model of the Evaluation of Causes in Analytical Procedures (Anderson and Koonce 1998, 3) Although prior research has found that audi tors often fail to adequately assess sufficiency (Anderson et al. 2003; Anders on and Koonce 1998; Hirst and Koonce 1996), little research has examined auditorsÂ’ abi lity to assess plausib ility. Since auditors Is cause consistent with unexpected fluctuation? Is cause consistent with the available information? Is cause of a sufficient magnitude to account for substantially all of the fluctuation? Plausibility Check Sufficiency Check
10 recognize basic accounting relati onships, it is likely that they would accurately complete step one of the plausibility judgment (explanation is consistent with change in account balance). However, because step two of the plausibility check requires auditors to search for confirming evidence, they could fail to iden tify explanations that are inconsistent with actual circumstances. Given a seemingly plausi ble client-provided explanation, auditors could fail to complete step two accurately, independent confirmation that the hypothesis fits the circumstances. The current study uses an explanation that is consistent with the change in account balance, yet inconsistent with the actual facts (a s evidenced by changes in other account balances). To identify the explanation as implausible, auditors must search beyond the account of interest. Prior research suggests that experien ced auditors can detect implausible explanations when source objectivity is manipul ated at two levels: cl ient and some other outside source (e.g., decision aid, external th ird party, audit team member) (Anderson et al. 2003; Bamber 1983; Hirst 1994; Joyce a nd Biddle 1981). None considers the case where the client is a former audit team member. In the current study, the auditorÂ’s judgment of the clientÂ’s objectivity depends on both the clientÂ’s former position as a fellow audit team member and the clientÂ’s cu rrent position within his or her firm. The clientÂ’s former position with the audit fi rm should increase the clientÂ’s objectivity because the client (former auditor) has an understanding of proper financial statement preparation and of the importance of providing a high-quality audit. The clientÂ’s current position with the firm could decrease objectivit y because the client could be motivated by
11 bonuses and promotions related to strong fina ncial results. In addi tion, the client could also be motivated to falsify financial statements in order to cover up fraud. 2.2.3 Source Reliability Judgments Auditors weigh client explanations ba sed on their assessment of the clientÂ’s source reliability. The source reliability judgment includes an evaluation of both competence and objectivity (Bamber 1983; Hirst 1994). According to Hirst (1994), Â“Â…competence means an individualÂ’s ability to measure or interpret an item or event accurately. Objectivity means the likelihood an individual will report his measurement or interpretation truthfully, re gardless of its accuracyÂ” (p.114). Ceteris paribus the level of source reliability increases with the level of competence, as well as with the degree of objectivity. Source reliability judgments include an evaluation of both competence and objectivity. Therefore, any bi as that impairs auditor judgment about either the competence or the objectivity of a source could reduce the auditÂ’s effectiveness and increase audit risk. When evaluating the competence of a client who was once an audit team member, the auditorÂ’s past interaction with that indivi dual on audit engagements, as well as the auditorÂ’s knowledge of firm trai ning and promotion policies, should result in an accurate competence judgment. Further, an individualÂ’s competence is unlikely to decrease when he or she goes to work for a client firm. Unlike competence, objectivity is subject to situati onal pressures. Clients, although knowledgeable, might not be objec tive (Hirst 1994). Compensation plans, promotion opportunities, and stock options provide motivation for clients to report
12 untruthfully. AU 316, Â“Consideration of Fraud in a Financial Statement AuditÂ” requires auditors to evaluate client assertions with professional sk epticism, directing them to inquire about management incentives, pr essures and motivations (AICPA 2005). Auditors must consider how these motivations can influence clients to provide untruthful explanations. When evaluating the objectivity of a former audit team member, the current auditor must consider how the ex-auditorÂ’s objectivity may have changed, and how that change may affect the client Â’s overall source reliability. 2.2.4 ClientsÂ’ Insider Knowledge of Audit Process The potential for client deception is especia lly relevant when the client is a former member of the current audit firm. A significan t threat to financia l reporting involves the ex-auditorÂ’s specialized know ledge of the continuing audit firmÂ’s processes and operations (Beasley et al. 2000). As a former audit team member, the client knows which tasks lower-level auditors complete. He or she can apply this knowle dge strategically to hide his or her misdeeds, using certain accounts assigned to novice team members. Further, the client knows the audit firmÂ’s inte rnal procedures for determining materiality, evaluating evidence, and conducting substantive testing. While AICPA Ethics Interpretation 101-2 contains a requirement that th e ongoing engagement team consider the necessity to modify engagement proce dures, insider information does increase the risk that the client can anticipate and s ubvert those procedures. Admittedly, although a client might attempt to deceive the auditor, successful deception depends on the auditorÂ’s inability to detect the deception. The focu s of this paper remains on the auditorÂ’s judgment of the plausibility of client explanations; the above discussion merely
13 highlights the increased potential for a client to plan his dece ption, as well as the need for auditors to effectively detect deception when it occurs. 2.3 Auditor Affiliation and Related Studies Lack of independence is an often-cited cause of audit failure. In some salient audit failures, the top executives at the client corporations were also past employees of the firms that audited them. For example, in the Enron case, both Richard Causey, Chief Accounting Officer, and Sherron Watkins, Vice President, were Andersen alumni (Barrionuevo 2002). Being past employees of the audit firm, the concern is that these key client personnel are able to exercise undue influence on th e auditor, thereby impairing auditor independence. The federal governme nt has responded to the auditor affiliation threat to independence by restricting the employment options of audit team members (Sarbanes-Oxley Act (SOX) 2002). In addition, AICPA Ethics Interpretation 101-2 has identified the hiring of an external audito r by the client firm to be a threat to independence and suggests several mitigation techniques (AICPA 2005). As displayed in Figure 2, the timing of an auditor affiliation can occur in one of three ways (Lennox 2005). This study focuse s on employment affiliations (Panel B), which arise when the client company hires a me mber of the recurring external audit team. There are two reasons I focus on employment affiliations. First, they are the most common (Lennox 2005). Second, they are particul arly susceptible to bias because the auditor goes directly from being a member of th e audit team to being an audit client. This change in circumstance could alter the ex-a uditorÂ’s motivations, a nd potentially, his or her objectivity.
14 Figure 2 Auditor Affiliations (Lennox 2005, 212) Panel A: The timing of chance affiliations Panel B: The timing of employment affiliations Panel C: The timing of alma mater affiliations It is common for clients to hire employ ees from their current audit firm. In fact, the relationship between audit firm and client has been referred to as a Â“revolving doorÂ” (Bleed 2002, 1). Three benefits accrue from hi ring former external auditors (Beasley et al. 2000). First, auditors are often highly trained by their firms. Second, auditors commonly have had exposure to varied cl ients, businesses, and complex financial transactions. Third, a client companyÂ’s former auditors have an insiderÂ’s knowledge of the clientÂ’s current strategi es and corporate environmen t and therefore can quickly acclimate themselves to client practices. Individual leaves audit firm Company selects audit firm Individual joins client company Company selects audit firm Individual leaves audit firm Individual joins client company Individual leaves audit firm Individual joins client company Company selects audit firm
15 Beasley et al. (2000) also identify three threats to the financial reporting process associated with such hirings. While one of these threats relates to the potential for auditor shirking before hiring, two relate to the time period after the auditor is hired. The first threat, detailed previously in Section 2.2.4 relates to the clientÂ’s advantage over the auditor. The ex-auditorÂ’s intimate knowledge of the audit firmÂ’s plans and procedures logically makes it easier for him or her to suc cessfully hide improprieties in the financial statements from the current auditors. The sec ond threat and the focus of this study, stems from the effect of an in-g roup bias, explained later in Section 2.4.2 which causes the auditor to overestimate the clientÂ’s object ivity, leading to undera uditing. This bias can cause a reluctance of the current auditors to question the assertions of clients who were once their co-workers. Although auditor affiliation threats have at tracted the interest of regulators, researchers have published little on the s ubject. Lennox (2004, 202) observes that Â“Â…no published archival evidence exists on the t ypes of affiliations or whether affiliations impair audit quality.Â” Using an estimation model to identify companies whose unfavorable opinion probabilities are gr eater than 10%, Lennox partitions these companies based on the presence or absence of an affiliated executive. Findings suggest that firms with affiliated executives were sta tistically more likely to have a clean audit opinion. Menon and Williams (2004) examined cas es where the affiliated client was a former audit firm partner. Using an archival approach, after contro lling for performance characteristics, they found evidence of an a ffiliation effect. Firms employing former audit partners were more likely to have larger abnormal accruals. In addition, they noted an
16 affiliation effect on earnings such that firms with former affiliated partners were more likely to just meet analystsÂ’ earnings forecas ts than were firms without former affiliated partners. While the above studies examine correlations between affiliation and external measured variables, the current study uses an experimental approach to explore the effect of staff and senior auditor affiliati on on individual audit judgments. 2.4 Auditing Judgment: Biases and Social Identity Theory 2.4.1 Biases in Auditing Judgments Much has been written rega rding the process, and part icularly the weakness of human judgment and decision-making (Bambe r et al. 1995; Hogarth 1980; Kahneman et al. 1982; Libby 1991). One such weakness is bias, defined as Â“a preference or an inclination, especially one th at inhibits impartial judgmen tÂ” (American Heritage 2000). Biases can be strategic (individuals are consci ous of their bias) or implicit (individuals are unaware of their bias). AuditorsÂ’ profe ssional skepticism likely prevents them from exhibiting strategic biases; howev er, implicit biases may persist.4 Bias identification is particularly important in a uditing since auditors are requ ired to make many judgments, the results of which can significantly impact multiple stakeholders. For example, if an auditor incorrectly believes there is suffi cient evidence to support an account balance (overweighting), he could wrongly curtail furt her testing on that acc ount. While a single judgment error is not likely to increase risk considerably, the final audit opinion is the 4 For a discussion of the two types of biases, see Kunda (1990).
17 sum of multiple judgments; th erefore, the cumulative eff ect of these errors could significantly increase the risk of an a udit failure (Moeckel and Plumlee 1989). Some frequently researched biases in the audit liter ature include: anchoring and adjustment (Hogarth and Einhorn 1992), prim acy/recency effects (Kahneman et al. 1982), base rate frequency (T uttle 1996), common information-sampling bias (O'Donnell et al. 2000), and information search strategy (Kida 1984). The roots of these biases reside in the psychology field. However, Bamber et al. (1995) suggest that auditing has unique attributes that prevent the bl anket application of psychology findings to auditors. Indeed, research results are mixed. While auditors performed better than non-auditors in a representativeness judgment (Joyce and Bi ddle 1981), and demonstrated a better understanding of subpopulation error rates (Tut tle 1996), in anchoring and adjustment studies they exhibited a recency effect consistent with general psychology findings (Bamber et al. 1995). Although auditors are prof essionals, trained to detect errors and misstatements, they are still human, and as such, demonstrate many of the biases long established through years of psychology research. 2.4.2 Social Identity Theory In-group bias, based on Social Identit y Theory, influences human decisionmaking in social contexts (Tajfel 1981). This theory proposes that group members are an extension of the self, and as such, each group member has a Â“Â…systematic tendency to evaluateÂ…( his )Â… own membership group (the in-gr oup) or its members more favorably than a nonmembership group (the out-group) or its membersÂ” (Hewstone et al. 2002, 576). In-group bias, characterized by oneÂ’s unque stioning belief in the assertions of a
18 fellow group member, provides an individual wi th a positive social identity, thereby satisfying his need for self-esteem (Hewstone et al. 2002). This bias is quite robust. Oakes et al. (1994) note that discriminatory behavior and attitudes can be brought about by a mere cognitive division of people into groups. Towry (2003) successfully manipulated team identity simply through the use of colored props and seating assignments. Auditors become part of an audit firmÂ’s in-group when they are hired. As they work together on the same audit team, th ey develop familiarity through repeated interactions, increasing the level of in-gr oup bonding. Although audito rs who eventually leave the firm to go work for a client are technically no longer members of the audit team, this change in employment does not n ecessarily exclude them from the audit ingroup. Levine et al. (1998) not es that individuals may si multaneously be members of multiple groups. When an auditor becomes a client, the remaining audit team members may view the ex-auditor as part of both the client and the audit groups. Therefore, even after auditors go to work for a client, rema ining audit team members could continue to identify them as group members; they are, in fact, still working together on the same audit, albeit on opposite sides. 2.4.3 Inter-group Bias in Social Psychology According to Apfelbaum and Lubek (1979) three characteristics define intergroup bias. First, in-group members view th emselves as a homogeneous group; second, in-group members view out-group members as a homogeneous group; and third, in-group members view themselves as different from out-group members. Inter-group bias causes
19 people to draw distinctions based on group me mbership, rather than on individual traits. Inter-group bias can take the form of in-g roup trust or out-group derogation. In-group favoritism results in the Â“exte nsion of trust, positive rega rd, cooperation, and empathy to in-group, but not out-group me mbersÂ” (Hewstone et al. 2002, 578). Out-group derogation is the underlying source of stereotyping and discrimination. Interestingly, much of the psychology research seeks to reduce out-group derogation, and, in turn, reduce the intergroup conflict (Hewstone et al 2002; Tajfel 1981). In this study, the focus lies not with unwarranted out-group skepticism, but with unjustified in-group trust. The danger comes from overweighting assertions made by an in-group member, not from underweighting assertions made by an out-group member. Self and social identity theories are ofte n used to explain an individualÂ’s behavior in groups (Ellemers et al. 2002; Oakes et al. 1994; Tajfel 1981). Ellemers et al. (2002) presents a taxonomy of the primary concerns a nd motives of the social self. The two axes are level of group commitment (high and low) and level of perceived threat (none, individual, and group). The taxonomy in Figure 3 details concerns and motives for each response. For the purposes of this study, I cl assify auditors with a rank of senior and below auditors as belonging to Cell #4: high commitment to the group and exposure to individual-directed threats. While there is no prior research to directly support this classification, these auditors are likely to be highly committed to their firm; seniors have chosen to stay with their firms by accepting promotions and have been given additional responsibilities within the firm, and staff members have ju st completed years of training and study, as well as a competitive interview pr ocess. The individual threat is one of
20 exclusion from the group (e.g., being fired). These auditors are more likely to make decisions that further their acceptance as part of the group than decisions that might lead to their rejection by the group. Figure 3 Primary concerns and motives of the social self: a taxonomy (Ellemers et al. 2002, 167) Group Commitment Low High No threat 1. 2. Concern: Accuracy/efficiency Social meaning Motive: Noninvolvement Identity expression Individual-directed threat 3. 4. Concern: Categorization Exclusion Motive: Self-affirmation Acceptance Group-directed threat 5. 6. Concern: Value Distinctiveness, value Motive: Individual mobility Group-affirmation Ellemers et al. (2002, 173) points out th at new group members Â“Â…tend to be more anxious and lack confidence reflecting acceptance concernsÂ….Â” I surmise that this lack of confidence could negatively affect an aud itorÂ’s professional skepticism, causing him to be reluctant to question aff iliated clients, an idea echoe d by Beasley et al. (2000). Within a single organization, there ar e both in-group and out-group members. Napier and Ferris (1993) not e that, among other factors, the higher the perceived similarity between supervisors and subordinates, the lower th e psychological distance. In turn, Â“Â…less Psychological Distance is asso ciated with greater attraction and liking, greater subordinate satisfac tion, and higher supervisor evaluations of subordinate performanceÂ” (Napier and Ferris 1993, 333). Given these benefits, it is likely that staff
21 and senior auditors would seek to nurture perceived similarity between themselves and their superiors, including the form er auditor who is now a client. 2.4.4 In-group Bias in Auditing An extensive literature sear ch revealed only one behavi oral study on the effects of in-group bias among auditors. King (2002) ch allenged the idea th at auditors are subservient to self-serving biases, and that they are unable to objectively audit a client upon whose business they depend. In an experi ment, he created a strong group identity among the auditors by having them meet frequen tly with each other. This strong identity resulted in the auditorsÂ’ increased ability to de tect client deceptions. Auditors in the weak group treatment interacted primarily with clie nts and were less likely to detect client deception. The team identity in the strong group Â“Â…motivates auditors to focus more on the collective goal of conduc ting appropriate auditsÂ” (K ing 2002, 267). The result was that this motivation overcame the auditorÂ’s self-serving biases.5 In the above study, auditorsÂ’ in-group bias toward other auditors resulted in better audits yet individuals belong to many gr oups simultaneously, resulting in differing degrees of group identity (Ellemers et al. 2002). I might identify myself as a graduate of a particular university, an account ant, an auditor, an employ ee of a large audit firm, and specifically, an employee of a particular fi rm. Depending on how strongly I identify with each group, I will exhibit a concomitant level of in-group bias. In an auditor/client relationship, auditors may view clients as fellow group members based on their common 5 A self-serving bias in this case is defined as the auditorÂ’s need to please the client so that the client will continue to contract with the auditor for services.
22 socio-economic class, college alma mater, re ligious affiliation, or, where the client was once an auditor. In the case of employment affiliation, the auditor c ould still view the exauditor, now the client, as an audit firm group member. In-group bias is particularly relevant in auditing because it can affect the auditorÂ’s professional skepticism. For example, analytic al procedures often require auditors to gather and evaluate explanations from clients. A key part of this evaluation involves the auditorÂ’s ability to judge the clientÂ’s objectivity correctly, an d the effect that objectivity has on the clientÂ’s truthfulness. In the contex t of the current study, clients were also once fellow auditors, thus confounding group identity A likely outcome is that auditors will continue to identify affiliated clients with their former audit group and thus will fail to adjust their assessment of the clientÂ’s objectivity appropriately. The resulting unwarranted trust could cause the auditor to accept the clientÂ’s implausible explanation, resulting in an incorr ect audit judgment. 2.4.5 A Normative View Generally Accepted Auditing Standards require auditors to approach an audit engagement with professional skepticism; th e notion that a seemi ngly irrelevant past association could result in unjus tified bias is a cause for concern. It is important to investigate whether this past relationship is truly irrelevant. Hirst (1994) suggests that both competence and objectivity should be considered in a source reliability judg ment. Based on an insiderÂ’s knowledge of hiring criteria, professional certifications, firm training and evaluation procedures, along with the direct experience of working together, auditors shoul d correctly assess thei r former co-workerÂ’s
23 competence. The validity of this assessment should not change regardless of the fellow co-workerÂ’s employment. Auditors must al so assess a clientÂ’s objectivity -an individualÂ’s motivation to communicate his be liefs honestly. In fulfilling their obligation to reduce the risk that acco mpanies the principal-agent relationship characteristic of owners and managers, auditors must maintain objectivity. Auditor objectivity arises from the motivation to provide a quality audit. Cont rary to this, client bias arises from the motivation to present the financial statements in the best possible light. Because of this difference in motivations, it is likely that a c lientÂ’s representations are more biased (less objective) than those of an auditor. 2.5 Statement of Hypotheses Â– In-group Bias The first part of this study tests whether staff and senior audi tors demonstrate ingroup bias when assessing a client-provide d explanation. As noted earlier, senior auditorsÂ’ experience should enable them to detect an implausible explanation. However, senior auditorsÂ’ tenure with the firm should lead to a strong in-group association. Staff auditors, although less experienced, have a need for acceptance by the group and are likely to align themselves with establishe d group members. Staff members could either seek to impress the audit team by demonstrat ing skepticism of client explanations or could view the client as part of the audit fi rm in-group and thus be reluctant to question the assertions. In sum, both levels have th e potential to exhibit in -group bias. Given that arguments exist for and agains t in-group bias at each leve l, I make no formal hypotheses about level. Rather I make a general proposal that in-group bi as persists from the original
24 association between the client and auditor, making the auditor more likely to overrate the plausibility of a client explanation. The following hypothesis tests for a si mple effect of in-group bias. H1: Given an implausible explanation, auditors will judge that explanation as more plausible when it comes from an in-group client, than when it comes from an out-group client. Auditors also must expre ss confidence in thei r judgments. Rose (2002, 114) notes that individuals may exhibit either overconfidence (Â“Â…inc reases in confidence without the associated improvements in decision quality Â…Â”), or underconfiden ce (failure of the individual to recognize when the decision is accurate). General psychology research finds overwhelmingly that individuals are overconfid ent (Fischoff 1982). In the audit literature, findings on confidence are mixed (Ahlawat 1999; Bamber and Ramsay 2000; Einhorn and Hogarth 1978; Moeckel and Plumlee 1989). To massini et al. (1982) find that auditors demonstrate less overconfidence than suggested by the general psychology literature for an audit-related task. Solomon et al. (1982) find that auditors were underconfident in an audit task; however, similar to general ps ychology findings, were overconfident in a general knowledge task. In an audit evid ence recall task, Moeckel and Plumlee (1989) find that participants are e qually confident in their inaccurate memories as in their accurate memories. Bamber (1995) suggest s that there is some underlying, unknown reason for underconfidence in an audit context. Given that auditors would not expect an in-group client to present an implausible explanation, they may question their own j udgment, causing their confidence to be lower
25 than it would be if the implausible explan ation came from an out-group client. This discussion leads to the following hypothesis: H2: Given an implausible explanation, auditors will be less confident in their initial plau sibility judgment when the explanation comes from an in-gro up client than when it comes from an out-group client. Auditors rely on their judgments to adju st future audit plans (Cohen and Kida 1989). It is important to eval uate whether in-group bias has an effect on auditorsÂ’ decisions to extend or curta il further testing. Auditors who correctly identify an explanation as implausible could still suspe nd testing on that item because a fellow group member supplied the explanation. To explain further, an auditor could believe that a client explanation is implausible, but not belie ve that the client is intentionally lying. An auditor who has an in-group re lationship with the client could still choose to extend testing; however, this extension of testing could be less than if the auditor did not have an in-group relationship with the c lient. In-group bias could result in an auditor deciding to give a fellow group member Â“the benefit of th e doubt.Â” On the other hand, an auditor who receives an implausible explanation from an in-group client could believe that the client is intentionally lying and compensate for this discovered decepti on by increasing testing. I propose that consistent with in-group bias, an auditor wi ll extend testing by less when the client is an in-group member than when the client is an out-group member.6 H3: Given an implausible explan ation, auditors who correctly identify an explanation as implausible will extend testing less when the client is an ingroup member than when the client is an out-group member. 6 This hypothesis refers to decisions without benefit of a decision aid; however, auditors may reassess their decision after using a decision aid. I test this hypothesis pre and post-decision aid.
26 2.6 Discussion of Potential Covariates I consider the following potential covariat es for inclusion in the model: perception of client competence, prior task experience, and prior experience with affiliated clients.7 Hirst (1994) finds that competence and objectivity interact in an a uditorÂ’s determination of source reliability. To contro l for this possible interaction, I measure each participantÂ’s perceived client competence rating. I plan to compare these ratings across groups to rule out a competence effect on auditor judg ments, confidence, and decisions. Prior research finds that task experience is positivel y related to performance on audit tasks in general, as well as on anal ytical procedures (H irst and Koonce 1996; Kaplan et al. 1992; Libby and Frederick 1990) Thus, auditors who are experienced in analytical procedures are likel y to give lower plausibility judgments than are auditors with less experience. Gi ven that the sample includes a uditors from staff through senior levels, it is reasonable to assume that participants have vary ing levels of experience with procedures. Therefore, I include a measure of analytical procedures experience as a potential covariate in the model. An auditorÂ’s prior experience with affiliate d clients could impact their attention to the group manipulation, and as a result, affect their judgment. For example, an auditor who has no prior experience working with an a ffiliated client could interpret th e in-group manipulation as unusual. This interpretation co uld lead him or her to weight the in-group factor more than an auditor who had prior ex perience with affiliated clients. Accordingly, 7All covariates are measured post-task to avoi d confounds within the research design.
27 I ask for participantsÂ’ prior experience working with affiliated clients for inclusion in the model as a potential covariate. If auditors exhibit in -group bias, and this bias potentially increases audit risk, it is valuable to examine whether there is a t ool to mitigate this bias effectively and efficiently. In sections 2.7 Â– 2.10 I develop an argument that a valid, objective decision aid will be successful in mitigating in-group bias. 2.7 Debiasing In Auditing Multiple techniques exist for debiasing in an audit environment. Justification (Peecher 1996), counterexplanation (Ke nnedy 1995), accountability (Kennedy 1993; Tetlock 1983), documentation (Ballou 2001) and the review process (B razel et al. 2004; Trotman 1985) all influence the auditorÂ’s judgment and performance on audit tasks. Although research has shown the prior met hods to be effective, there are three noteworthy drawbacks to using them. First, since the cost of an audit depends on the number of hours worked, efficiency is of key importance. The review process, while effective, takes both the a uditorÂ’s and the reviewerÂ’s time. Second, review and documentation procedures are detective or corr ective controls -they do not prevent staff members from making initial errors in judgm ent. Third, because individuals implement these methods, execution could be incons istent, resulting in more audit risk. Decision aids are not subj ect to the above drawbacks. Abdolmohammadi and Usoff (2001) find that practitioners identify a mu ltitude of audit tasks that are well-suited to the use of decision aids. Rose (2002) notes that decision aids can mitigate systematic
28 information-processing biases.8 By their nature, they offer a consistent, objective recommendation to the audito r (Ashton 1992). This consiste ncy reduces variability in both an individual auditorÂ’s judgments, as well as auditorsÂ’ judgments firm-wide. Although decision aids do not completely prev ent incorrect judgments, they can provide auditors with suggestions and direction. 2.8 Debiasing and Improving Judgment a nd Decisions with Decision Aids Several studies establish the effectivene ss of decision aids in mitigating audit judgment biases and improving audit judgments overall (Butler 1985; Eining et al. 1997; Emby and Finley 1997; Rose and Rose 2003). Bu tler (1985) developed a decision aid that focused the userÂ’s attention away from specifi cs and to a broader vi ew of the situation. Since analytical procedures require auditors to consider the in terrelationships among accounts, a decision aid that informs the us er about these interrelationships should improve judgments. Eining et al (1997) find that for a comple x task (fraud detection) an expert system with a constructive dialogue feature is effective in improving judgments. The authors considered a combination of char acteristics from the ps ychology literature to design a constructive dialogue feature that woul d increase decision ai d reliance. Increased reliance led not only to improved assessments but also to improved decisions. Emby and Finley (1997) successfully used an evidence rating technique to mitigate framing effects for internal control asse ssments and decisions. 8 Decision aids can create new judgment biases, especi ally in the presence of other debiasing strategies such as accountability and incentives (Ashton 1990). For a discussion of the literature on decision aids, see Rose (2002).
29 Decision aids can effectively mitigate bi ases in auditing. KennedyÂ’s (1993; 1995) framework classifies biases as either Â“dat a-relatedÂ” or Â“effort-related.Â” Effort-related biases occur when the decision-maker has ei ther insufficient capacity or insufficient motivation to complete the task. Suggested solutions include increasing internal capacity, providing incentives, or introduci ng accountability. Data-related biases occur when either internal or external information (or both) are imprecise. Internal data (individual memory) is the source for individual biases such as framing (Emby and Finley 1997), first impression bias (Lim et al. 2000), and anchoring and adjustment (George et al. 2000). External data biases arise when the information provided to the individual is unclear, irrelevant, or presen ted in complex format. In th e current study, I classify ingroup bias as an internal data bias becau se the in-group influence arises from the individualÂ’s biased perception of the affiliated clientÂ’s trustworthiness. Both Kennedy (1995) and Roy and Lerch (1996) suggest the following solutions to minimize data-related bias. First, firm s can modify information presentation. This approach is used successfully by Lim et al. (2000) to reduce re interpretation of secondary data (a framing bias) and present the secondary data in such a way that it could not be ignored. Second, firms can train individuals to use appropriate information processing strategies. Firms can provide feedback during a task so that individua ls can adjust their decision processes and subseque ntly apply the improved proc ess to similar situations. Eining et al. (1997) uses this approach in designing a decision aid that includes constructive dialogue. Third, firms can replace decision-makers with a model that suggests a normative answer. Libby and Libby (1989) find less variability and better
30 performance when auditors used a decision aid to combine multiple judgments into a global answer. Rose and Rose (2003) also find that decision aids mitigated recency bias in an audit evidence evaluation task. In this study, in-group bias involves a s ubconscious leaning toward believing an in-group member. The debiasing agent will display information in a structured format, as well as provide cues to guide the auditor in his search for support. I discuss the decision aid design in Section 3.5.2 2.9 Decision Aid Reliance in Auditing Technology use is increasing in todayÂ’s a udit process. A long itudinal survey of auditors indicates an increase in the number of audit tasks that are amenable to the application of a decision aid (Abdolmohammadi and Usoff 2001). Audit firms use a variety of decision aids, decisi on support systems, and expert systems in the audit process (Abdolmohammadi and Usoff 2001; Bedard a nd Graham 2002). Rele vant decision aid studies find that reliance is influenced by face validity (Ashton 1990), and source objectivity (Anderson et al. 2003; Lim et al. 2000). A decision ai dÂ’s face validity refers to the usersÂ’ assessment of Â“the extent to wh ich it appears sensible and reasonableÂ” (Ashton 1990, 170). Source objectivity refers to the trustw orthiness of the decision aidÂ’s source. Ye and Johnson (1995) find that auditors are more likely to accept expert system advice if the advice is reasonabl e. In addition, the study finds that just ification, described as Â“Â…an explicit description of the causal argument or rationale behind each inferential step taken by the ESÂ” is most effective in user acceptance of e xpert systems (Ye and Johnson 1995, 158). Justification requires aud itors to have a d eep understanding of
31 accounting; in this study, auditors must be familiar with the relationships among accounts, in order to judge the decision aid predictions as reasona ble. Ye and Johnson (1995) posit that decision aid reasonableness increas es the auditorÂ’s c onfidence in the aid, and thus increases the probability that the auditor will rely on the aid. As noted above, Eining et al. (1997) successf ully increase decision aid reliance by incorporating a constructive dialogue feature in their decision aid. However, there is ample evidence in the literature that decision makers do not always rely on decision aids (Rose 2002). Individuals may work around th e decision aid (Kachelmeier a nd Messier 1990) or try to outperform the decision aid (Arkes et al. 1986). Thus, there is a possibility that auditors will not rely on a decision aid. Anderson et al. (2003) find that auditors judged decision aid expl anations as more sufficient than client-provided explanations when, in fact, such explanations were insufficient. Overreliance on the decision ai d resulted from the a uditorÂ’s assessment of the decision aidÂ’s objectivity. Since validity and objectivity are both important to decision aid reliance, I will confirm that par ticipants judged the decision aid in the study to be both valid and objective. 2.10 Statement of Hypotheses Â– Decision Improvement Auditor judgments can be influenced by in-group bias, as posited above. However, factors other than group biases can also negatively impact auditor judgments. For example, even auditors who receive an implausible explanation from a non-affiliated client can incorrectly accept the explanation as plausible. This error in judgment can arise from the auditorÂ’s reliance on perceived clie nt competence. In other words, an auditor
32 can successfully complete step one of the plausibility check (plausibility of hypothesis given the change in account ba lance), yet fail to complete step two (plausibility of the hypothesis given other, external information) successfully. A decision aid that redirects the auditorÂ’s attention to ot her possible hypotheses (similar to the approaches of Lim et al. (2000) and Butler (1985)) should improve auditor judgments. Further, a decision aid that provides reasonable justification (a s found by Ye and Johnson (1995)) should result in auditor reliance, which is necessary for audit judgment improvement. I propose that a decision aid will improve auditor judgment by directing auditorsÂ’ attention to the implausibility of the client-provided explan ation. The decision aid will provide more information, lowering cognitive effort, as suggested by Kennedy (1993). Finally, auditors should judge a firm-developed decision aid as more valid and objective, causing them to weight the decision aidÂ’s recommendation mo re than the clientÂ’s explanation, as evidenced in Anderson et al. (2003). Hypothesis 4 tests the effec tiveness of decision aid use on auditors who initially incorrectly judge plausibility to be high. H4: Given an implausible explan ation, auditors who make an initial incorrect judgment will decrease their plausibility judgment after using a decision aid. Ahlawat (1999) finds that confidence incr eases with an increase in the amount of information provided. The decision aid report provides additional information to the auditor by directing his or he r attention to alternative explanations and expected relationships among relevant accounts. Hogart h and Einhorn (1992) provide a model for belief adjustment that addresses how beliefs change when new information is received.
33 Srivastava and Mock (2004) suggest that an auditorÂ’s belief assessment regarding audit evidence includes three components: first, th e belief that the evidence supports the conclusion, second, the belief that it suppor ts an opposing conclusion, and third, the ambiguity related to unknown information. As auditors gather new information the amount of ambiguity decreases and they can classify information as confirming or disconfirming. As ambiguity about the judgment decreases, auditors should feel more certain about their decisions. Ye and Johnson (1995) suggest that the use and acceptance of decision aid recommendations will im prove user confidence. Chung and Monroe (2000) find that judgment confidence decrease s as perceived task difficulty increases. Use of a decision aid should reduce cognitive effort and therefore reduce task difficulty. As task difficulty decreases, I expect confid ence to increase. In this study, the decision aid offers feedback by providing expected relationships between relevant accounts. Auditors who have the requisite accounti ng knowledge and rely on the decision aid should recognize whether their prior judgmen t was correct. If they were initially incorrect, this realization shoul d lead them to the correct answer, about which they should be confident. If they were initially correct reliance on the decision aid reinforces their original answer and should also increase confidence. I propose the following hypothesis. H5: Auditors will be more confident in their post-decision aid plausibility judgment than in th eir pre-decision aid plausibility judgment. Finally, I explore the effect of decision aid reliance on auditor decisions to extend testing. Based on the reasoning used for Hypothe sis 4, a logical result of the change in plausibility judgment is a change in extent of testing. Auditors who rely on the decision
34 aid, and subsequently change their judgment of the client explanation from plausible to implausible, should logically adjust their extent of testing to reflect their revised belief. Eining et al. (1997) noted that not only did auditors improve their judgments after using a decision aid, but they also improved their subs equent decisions. Bukszar (2003) finds that individuals treat decision s with more consideratio n than they do judgments.9 In a forecast and investment task, he finds that indivi duals perform an additional evaluation step between making a judgment and making a deci sion, which results in individuals being likely to act on their acc urate judgments. In an audit contex t, it is important to explore the effect of a decision aid not only on judg ments, but on subsequent decisions. Auditors who initially judge a client explanation plausible, will likely extend testing little, if at all. Post decision-ai d, auditors who reevaluate their decision and conclude that the client explanation is implau sible, will likely increase testing. Further, an auditor who changes his or he r plausibility judgment to implausible will also likely reassess the clientÂ’s ob jectivity, also leading to a decisi on to increase testing. Therefore, I propose the following hypothesis. H6: Given an implausible explanat ion, auditors who make an initial incorrect judgment will in crease their extent of testing after using a decision aid. 9 A judgment is an individualÂ’s inference about an external event or phenomenon. A decision is an individualÂ’s choice of action (Hastie 2001).
35 CHAPTER 3: METHOD 3.1 Introduction This section details the experimental me thod. I first justify the sample selection, noting that seniors and staff me mbers both have the potential to exhibit in-group bias for different reasons. Then I discuss the choice a nd design of analytical procedures as the experimental task. Analytical review of expenses is an a ppropriate task for staff and senior auditors and it is amenable to decision aid development (Abdolmohammadi 1999). I proceed to discuss the research desi gn, detailing the procedure, instrument development, and measurement of dependent variables. I include a discussion of the client explanation, noting that the explanation is reasonable given the related change in account balance, but implausible given the ch ange in related acc ounts. I describe the establishment of the between-subjects mani pulation group affilia tion, and the creation of the within-subjects treatment Â– decision ai d. I describe the dependent variable scales, noting their use by prior research ers in the audit literature. Finally, I include a discussi on of threats to internal and external validity, noting how this study addresses those threats. I fo llow with a discussion of manipulation checks. I also describe the pilot study. Finally, I detail planne d statistical analyses. 3.2 Sample Participants are staff and senior auditors. I chose staff auditors because, as noted, clients will likely use their inside knowl edge to deceive less -experienced (novice) auditors. Further, analytical procedures are often comp leted by assistant auditors
36 (Abdolmohammadi 1999). I chose senior audito rs to explore whether in-group bias affects multiple levels within the firm. Se niors, because of their experience, should accurately detect implausible explanations. Ho wever, their longer affiliation with the firm may increase their in-group bias The use of auditors (as opposed to audit students) is necessary to establish the in-group treatment. Auditors ha ve had time to develop in-group feelings toward their co-workers, and shoul d also have sufficient task experience. Online access to the experimental materials simplified data collection from various locations. I recruite d participants from several national CPA firms. All participation was voluntary; I contacted firms an d asked them to distribute the web link to their staff through senior auditors, along w ith a letter endorsing th e study. I provided no incentives for performance; however, partic ipants were asked to voluntarily provide contact information if they wanted indi vidual feedback. To encourage completion, I allowed participants to direct a $5.00 donation to their choice of ch arity (from a select list). 3.3 Experimental Task The experimental task required an audito r to perform an analytical procedure on the repair and maintenance e xpense account during the substa ntive testing phase of the audit. There are three reasons for this choi ce of task: it is appropr iate for staff through senior auditors, expense accounts have been used to hide fraud (high inherent risk), and the analytical procedure related to expenses is amenable to decision aid use. First, auditors identify this task as appropriate for a staff auditor to conduct (Abdolmohammadi 1999). Second, asset misappropriation often oc curs in expense accounts (Hall 2004).
37 Third, analysis of repair and maintenance expe nse is a substantive testing task that is amenable to the development of decision support systems (Abdolmohammadi 1999). Analytical procedures are a semi-structured task; they include a reasonably well-defined problem, with limited alternatives, requiring some judgment (Abdolmohammadi 1991). Although a decision aid can list plausible r easons for an account balance fluctuation, auditors must also consider many intangible, non-financial factors th at cannot or typically are not covered by a decision aid. Auditors mu st use their judgment to make a final determination regarding the likelihood that a given explanation is plausible. Task materials included a narrative descri ption of the firm, a copy of the current and prior yearÂ’s financial statements (with the unexpected increase in the repair and maintenance expense account highlighted), a description of the cl ientÂ’s background (to establish the varying group tr eatments), and the clientÂ’s explanation for the unexpected fluctuation.10 After the first measurements, partic ipants had access to a decision aid.11 3.4 Research Design 3.4.1 Procedure Figure 4 details the mixed research design with one between-subjects factor (group affiliation) and one within-subjects f actor (decision aid). Prior to completing the task, participants filled out an online in formed consent, as well as a demographic questionnaire to elicit the identity of their current employer, as well as their level in the 10 Overstatement of expenses is often an indication of asset misappropriation; a fraud which is more likely to be committed by mid to lower management (Hall 2004), such as a controller or assistant controller. 11 The decision aid was labeled as Â“firm developedÂ”, but was developed by the author and was the same decision aid for all participants.
38 firm (staff or senior).12 It was necessary to gather this information before the experiment to operationalize the group mani pulation. I randomly assigned part icipants to either an ingroup or an out-group treatment (the difference between in-group and out-group was the client representativeÂ’s history). For in-g roup participants, the client was a former employee of the participantsÂ’ audit firm, for out-group member s, the client was a longtime employee of the client firm. Figure 4 Diagram of the Experimental Design Read Current and Prior Year Financial Statements (Treatment) Implausible Client Explanation (Treatment) Plausibility, confidence, extension of testing (Observation) Decision Aid Report (Treatment) PostDecision Aid Plausibility, confidence, extension of testing (Observation) Ingroup X X O X O Experimental Condition Outgroup X X O X O Time All participants then had access to the above-referenced task materials. They made the following judgments: plausibility of the clientÂ’s explanation (scale of 0 Â– 100), their confidence in that judgment (scale of 0 Â– 100), and whether and how much to extend testing on that item (num ber of hours). After making those judgments, participants were shown a decision aid report (attributed to their firmÂ’s national office research department), as well as to the materials provided earlier. They then answered the same 12 Demographics also include age, gender, certifications held, and highest education level.
39 questions regarding plausibility, confiden ce, and extent of testing. A post-test questionnaire included manipul ation checks and further measur es that may be significant, including the participantsÂ’ perc eptions of client competence, prior analytical procedures experience, and experience with affiliated clients. At the conclusion of the online experiment, participants were thanked for thei r participation and were allowed to direct a contribution to a charity of their choosing. 3.4.2 Characterization of Client Explanation There are two steps to the plausibility ch eck (see Figure 1). Step one relates to how well the explanation fits w ith the unexpected fluctuation. This judgment is a test of accounting knowledge (Libby 1985). The auditor n eed only do a search of his internal knowledge base to judge the expl anationÂ’s plausibility. It is also unlikely that a client, especially a competent client, would present an explanation that violates the accounting relationships. Therefore, I use a client-provided explanation th at is plausible, given an increase in the repair and maintenance account.13 The second step requires th e auditor to confirm that the explanation fits the circumstances. The auditor must search for information to confirm or disconfirm the clientÂ’s explanation. To judge implausibility, the auditor cond ucts an external information search, rather than an internal accounti ng knowledge search. This search requires additional effort. In a situation where in-group bias exists the auditor could 13 A manipulation check confirmed that the participan t has sufficient accounting knowledge to identify the explanation as plausible, given the fluctuation.
40 subconsciously choose to forego the additiona l work and rely instead on his positive assessment of the clientÂ’s source objectivity. In the experimental task, the client explan ation is consistent w ith the direction of the unexpected fluctuation, yet inconsistent with certain financial statement information (fixed assets have increased). The client explanation provided to participants follows: The unexpected increase in re pair and maintenance expense comes from an internal decisi on to forego replacing certain capital equipment until next year. We were planning to replace our fleet of trucks with a new fleet, but due to the increase in interest rates, we decided to repair, rather than replace them. 3.5 Independent Variables 3.5.1 Between-subjects treatment: Group Affiliation I manipulate group affiliation at tw o levels between subjects. Although individuals concurrently claim various group affiliations, in this study, I vary only the former employment of the client. In-group clie nts are either former managers or seniors from the recurring audit team.14 Out-group clients have worked only for the client firm. I expect the manipulation to affect the audito rÂ’s judgment of the clientÂ’s objectivity. However, since source reliability incl udes both competence and objectivity, I hold competence constant between treatment groups. Manipulations occur after this brief introduction. 14 Senior auditors received an explanation from a fo rmer audit team manager who became a controller for the client. Staff auditors received an explanation from a form er audit team senior who became an assistant controller for the client. This design maintains one level between the auditor and his or her superior.
41 As part of the current audit fi eldwork, your assignment is to evaluate the changes in expe nse accounts. Noticing that the current year's repair and maintenance expense account balance is unexpectedly hi gh, you have asked Chris, the controller, to provide an explanation. In group Out group Chris's Background Chris's Background Chris worked for ( your firm) for the last several years, where he was a manager (senior) on the Continental Transport audit. Chris has worked for Continental Transport for the last several years He recently took a job at Continental as the controller (assistant controller). He was recently promoted to Controller (Assistant Controller) at Continental. Chris is technically proficient in accounting. Chris is technically proficient in accounting. 3.5.2 Within-subjects trea tment: Decision Aid I propose that the decision aid will m itigate in-group bias by modifying the presentation of information and providing the auditor search cues. Based on the clientÂ’s financial statements, the decision aid report li sts possible explanations for the unexpected account fluctuation. I establish decision ai d validity and objectivity as follows. The following report was generated by Â“DecisionSERVEÂ” audit software, developed by the (your firmÂ’s) national office research department. Aud itors should use it to assist them in evaluating client expl anations. The process uses the clientÂ’s current and past yearÂ’s financial data to generate possible explanations for ch anges in account balances. Past experience indicates th at DecisionSERVE provides valid explanations.
42 See Appendix A for an example of the decisi on aid output and the financial statements. 3.6 Dependent Variables Three dependent variables are measur ed both preand post-decision aidÂ— plausibility, confidence, a nd extension of testing. 3.6.1 Plausibility Prior source reliability studies (Bambe r 1983; Hirst 1994), used a 100-point scale to evaluate participantsÂ’ judgments. Bamber (1983) asked participants to evaluate the sufficiency of an internal control system. E nd points were Â“No LikelihoodÂ” and Â“Certain Likelihood.Â” Hirst (1994) asked participants to provide a probab ility estimate that inventory was materially misstated. On a 100-point scale, endpoints were Â“there is absolutely no chance that Inventory is materi ally misstatedÂ” and Â“I am absolutely certain that Inventory is materially misstatedÂ” ( p.119). This study uses a 0-100 point scale: end points are Â“not at all plausi bleÂ” and Â“highly plausibleÂ”. 3.6.2 Confidence Final audit opinions are the result of combining multiple audit judgments. Confidence in each judgment should be sufficien t to prevent an audit failure. Bamber et al. (1995) reviews research on auditor confidence finding that auditors are overconfident in their general knowledge, but underconfident in their performance of financial and audit tasks. This underconfidence could be a result of conservatism. I measure confidence on a 0-100 point scale; end points are Â“not at all confidentÂ” and Â“completely confident.Â”
43 3.6.3 Extension of Testing The extent of testing variable measures the effect of in-group bias on auditor decisions. I inform participants that a norma l budget for expense testing for this type of client and risk level is 40 hours. The measure allows for the participant to answer Â“0Â” if they choose not to extend testing. Alt hough staff auditors generally do not make decisions to increase testing, they have leew ay to investigate items further and/or make recommendations to their superiors. Seni or auditors do make decisions regarding extension of testing therefore this measur e mirrors practice. Fo llowing prior research (Cohen and Kida 1989), I use number of hours budgeted to measure planned increases in testing. The scale has end points of 0 hours and 10 hours. 3.7 Internal and External Validity 3.7.1 Internal Validity Pedhazur and Schmelkin (1991, 224) define in ternal validity as Â“Â…the validity of assertions regarding the e ffects of the independent variable(s) on the dependent variable(s).Â” In this study, I establish internal validity through a car eful research design. My goal is to eliminate alternative explanati ons so that any signifi cant findings related to judgments, confidence in those judgments and decisions to extend testing, are, in fact, due to either the group manipulation or the use of a decision aid. Common threats to internal validity include hi story, maturation, testing, instru mentation, regression to the mean, selection bias, and mortalit y (Pedhazur and Schmelkin 1991).
44 History threats relate to events that occur during or immedi ately preceding the study, which could influence par ticipantsÂ’ responses. In this study, legislation restricting auditor hiring by clients was passed one year prior data collection. Since auditors would likely be aware of this legisl ation, they could have been mo re attuned to the concept of in-group bias toward affiliated clients. This awareness could have led participants to guess the group hypothesis and as a result, overco mpensate for the bias by reducing their initial plausibility judgment. Maturation occurs during studies that occur over time Â– allowing individual personal changes to affect outcomes. Particip ants completed this study in a single sitting over less than one hourÂ’s time; therefore, maturation is not a significant threat. The testing threat is applicable when individuals are measured multiple times using the same variable (Pedhazur and Schmelkin 1991). In this study, all three dependent variables are measured twice (pre and post-decision aid). To increase the likelihood that individuals would give true re sponses to the post-de cision aid questions, I designed the survey so that prior answers were unavailable for viewing (participants could not access prior survey pages). This desi gn prevents auditors from merely repeating their original answers. Inst ead, they should have been mo re likely to incorporate new information (from the decision aid) into their second responses. Instrumentation threats arise from differen ces in the instrument or differences in the administration of the instrument. I used the same instrument for all participants, however, since data collection occurred online, individual di fferences in browsers or computing speed could have influenced part icipant responses. For example, although two
45 individuals could have spent the same amount of time comple ting the survey, the individual with a faster online connection sp eed could have spent more time reading the background information and thinking about his answers befo re responding. The participant with a slower online connection woul d have had to wait longer for the page to load, thus shortening the time used to consider responses. Other than an analysis of Â“time to completeÂ” (I verified that no participan t took less than 9 minutes or more than 60 minutes), the only other way to control this threat would have been to administer the survey in a computer lab. Given the geogra phical disparity of part icipants, this option was not feasible. Regression to the mean occurs when ever two variables are not perfectly correlated with each other (P edhazur and Schmelkin 1991). In this study, the threat of regression to the mean is relevant to the s econd measurements of the dependent variables. For example, regression to the mean predicts th at auditors who initia lly rate plausibility low (the correct answer), will increase their second plausibility rating toward the mean. Likewise, auditors who initially rate pl ausibility high will decrease their second plausibility ratings toward the mean. To test for this effect, I eval uate the direction of change for both high and low initial plausibility ratings to assure th at they do not assume this pattern. Random assignment to treatment groups mi nimizes the threat of selection bias. However, the method of partic ipant recruitment could lead to sample selection bias. I recruited auditors by co ntacting each firmÂ’s national or lo cal office. Partners distributed the survey site link through an internal email. Bias could occur from the partnersÂ’
46 selection of employees to send the e-mail t o, or could occur from the employees who chose to respond to the e-mail. A larger sample size would allow me to confirm that there is no difference between early or late res ponders with respect to demographics and dependent variable measures. However, th e small sample size prevents a thorough analysis of non-response bias. Mortality occurs when individuals do not complete the entire survey. Given that this experiment was voluntary, individuals were free to drop at any time. In addition, since the survey was online, there was little cost to dropping out (par ticipants would just close their browser). Further, mortality in an online context could be unintentional (e.g., technology breakdowns, lost Inte rnet connections). During th e data collection period, two audit firms distributed links to the surv ey immediately preceding the Thanksgiving holiday. The following Monday is referred to as Â“Black MondayÂ” because of the increased online shopping tra ffic (Kopytoff 2005). Auditors who attempted to logon to the Internet could have expe rienced slower connections due to this phenomenon. This could have influenced participants to drop out of the study prior to completion. When I became aware of this threat, I contacted the two firms and re quested that they send an email encouraging auditors to return to the survey if they had experienced Internet slowdowns. I also designed the survey to al low participants to logon multiple times. This allowed participants a chance to complete su rveys that were unintentionally interrupted. Using demographic analysis and IP address data, I confirmed all responses were from different individuals.
47 3.7.2 External Validity External validity is a measure of how we ll findings can be generalized to or across target populations, settings, or time (Pedh azur and Schmelkin 1991). Participants, task, and time are all limitations to external valid ity. I designed this study and collected data from only staff and senior auditors at large, national audit firms. Application of findings is limited to this population group. Because ma nagers and partners have longer tenure with their firms, as well as more advanced audit skills, generaliza tion to levels above senior auditor are inappropriate. In addition, this study makes use of a single audit task, analytical procedures. Audit research indi cates that task stru cture and complexity influence outcomes and should be adequate ly considered (Abdolmohammadi and Usoff 2001). Care should be taken in extending findings to expectat ions of auditor behavior on other audit tasks, especially tasks of differe nt complexity. Finally, as noted before, data collection occurred during a tim e of heightened awareness of possible biases related to auditor affiliation. On a larg er scale, given se veral large re cent audit failures, audit quality was also a concern during data collec tion. Generalization to future time periods may be unsupported and should be approached with caution. 3.8 Manipulation Checks The post-task questionnaire includes a se ries of manipulation checks to evaluate the strength of the manipulation and rule out alternative explanations It also contains several questions related to prior audit experi ence and prior experien ce with clients who are former audit firm employees.
48 The first series of questions measures the participantÂ’s judgment of client competence and client and decision aid objectiv ity. Consistency of competence ratings between group treatments rules out the possibi lity that perceived differences in client competence influenced plausibility judgments. Following Hirst (1994), I had the participants rate the clie ntÂ’s objectivity (defined as Â…the likelihood that the clie nt would give you, the auditor, a fictitious reason for an account fluctuation, when, in fact, he knew that the real reason was different). End points are Â“Extremely lowÂ” and Â“Extremely HighÂ”. Finally, as in Anderson et al. (2003), participants rate d the objectivity of the decision aid. To rule out part icipant non-reliance on the de cision aid due to a perception of low validity, I also measured the participan tÂ’s perception of the d ecision aidÂ’s validity. The second set of questions elicits info rmation about the participantÂ’s past experience and general opinions. Participan ts answered questions about their audit experience, experience with clie nts who were former audit te am members, and analytical procedures experience. They also rated whether their firm alumnae are more or less competent and/or objective than are alumnae of other audit firms, or non-firm accountants. 3.9 Planned Statistical Analyses The first step in data analysis, before hypothesis testing, is to evaluate responses for adherence to the manipulation checks. In this study, I am testing for group bias on the basis of auditor affiliation. I cannot assume that partic ipants who fail the betweensubjects group manipulation adequately attende d to the group affiliati on factor; therefore,
49 I plan to exclude those participants from the analysis. I will also evaluate how much time each participant spent logged onto the survey. Based on pilot study findings ( section 3.10 ), participants who spend less than eight minutes likely have not put forth the minimum effort to complete the task, therefor e, I will eliminate those responses from the dataset. I will then analyze the remaining data for violations of the statistical assumptions of normality of the dependent variables and constant and equal variance of the residuals using visual analysis of the stem and leaf pl ots and histograms and fo rmal statistical tests including the Kolmogorov-Smirnov and Shapiro-Wilk tests for normality. I will use LeveneÂ’s test for equality of variances. If the data adhere to the required assumptions, I will analyze the dependent variables and potential continuous covariat es for significant correlations using the Pearson correlation coefficient ( r ). Significant correlations between the dependent variables suggest that they must be evaluated simultaneously using multivariate statistics. If the dependent variables are significantly co rrelated, as I expect them to be for the repeated measures variables, I will use MANOVA or MANCOVA, as indicated. If the dependent variables are not significantly correlated, I will use univariate analysis, ANOVA or ANCOVA to test hypotheses. If the data does not adhere to the required assumptions for parametric tests, I will analyze the dependent variables and covariat es for significant correlations using the Spearman rank correlations (rho). I will then test the hypotheses using nonparametric tests. I will use the Mann-Whitney, two inde pendent samples test in place of ANOVA. I
50 will use the Friedman Test for K related samples in place of repeated measures MANOVA. I will also complete a post hoc analysis to explore interesting or unusual findings not formally specified by the hypotheses. 3.10 Pilot Study 3.10.1 Pilot Background and Descriptive Statistics I conducted a pilot study to ga ther preliminary data and as sess the validity of the instrument.15 Participants were undergraduate a udit students at a large metropolitan university. Since students do not have an in-group affiliation with a particular audit firm, I manipulated the group variable by characterizing the client as either a graduate of the participantÂ’s university (in-group) or as a graduate of an unnamed university (out-group). I also collected additional demographic data relating to grade point average and courses taken. Twenty-three participants took the survey ; I eliminated four because they failed the group manipulation check. Of the remaining 19, four took under 8 minutes to complete the instrument. Given the length of th e instrument, it is unreasonable to believe that those participants supplie d the requisite effort and I excluded them from the final analysis. Of the remaining 15 participants seven received the in-group treatment and eight received the out-group tr eatment. The mean time to complete was approximately 24 15 Prior to the pilot data collection, two expert auditors previewed the instrument to determine face validity, realism, and clarity. Experts indicated that the task and background information was both believable and appropriate for novice auditors. I made several small changes to the question text to improve clarity.
51 minutes, average age was 25 years, and m ean GPA was 3.3. Gender was fairly even within each group. Table 1, Panel A includes de scriptive statistics. Panel B of Table 1 includes dependent variable data by group. TABLE 1 Â– PILOT STUDY (All participants n = 15) Panel A Descriptive Statistics Panel B Dependent Variables by Group (m ean, standard deviation, range) Initial Plausibility Initial Confidence Initial Extent of testing Post-decision Aid Plausibility Post-decision Aid Confidence Postdecision Aid Extent of testing Ingroup n = 7 56.43 (20.56) 30-90 62.14 (17.29) 40-90 11.43 (7.84) 0-20 47.14 (19.12) 25-80 67.14 (20.38) 40-90 14.43 (8.81) 3-90 Outgroup n = 8 49.50 (27.73) 10-90 78.75 (13.29) 60-100 10.38 (6.84) 0-20 55.63 (24.12) 10-85 80.63 (14.25) 60-100 8.88 (5.64) 0-20 Overall n= 15 52.73 (24.05) 71.00 (17.03) 10.87 (6.91) 51.67 (21.60) 74.33 (18.11) 11.47 (7.58) Time to Complete Mean 24.08 Standard Deviation 8.08 Minimum 16 Maximum 36 Gender Male 9 Female 6 Age Mean 24.08 Standard Deviation 4.82 Minimum 21 Maximum 36 Grade Point Average Mean 3.31 Standard Deviation .39 Minimum 2.30 Maximum 3.80
52 3.10.2 Pilot Study Results The small sample size made analysis of the data for assumptions of normality and equal variance problematic. To address this issue, I used nonparametric methods to analyze the data. Conover (1999) suggests using the Mann-Whitney, two independent samples test, when analyzing data which are not normally distributed. Prior to analysis, I analyzed the data for outliers. Noting none, all data were retained within the analysis. Although the raw mean for init ial plausibility indicates th at in-group auditors rate plausibility higher than do out-group auditors statistical tests show insufficient support (p=.310) for hypothesis one. Hypothesis two predic ts that auditors wi ll be less confident in their judgment when an implausible expl anation comes from an in-group member than when it comes from an out-group member. M ean confidence measurements support this hypothesis as mean in-group confidence (62.14) is lower than mean out-group confidence (78.75). Using a Mann-Whitney test, there is a significant difference in confidence between groups (p=.035) demonstr ating support for hypothesis two. Hypothesis 3 suggests that in-group bias could affect an auditorÂ’s decision to extend testing, even though the auditor has correctly identified the explanation as implausible. Hypothesis 3 is tested both preand post-decision aid. To test this hypothesis, responses were split into high ( 50) and low (<50) plausibility groups, resulting in 9 high responses and 6 low res ponses pre-decision ai d and 6 high responses and 9 low responses post-decision aid. Using the Mann-Whitney test, I found no significant support either predecision aid (p=.251) or post-decision aid (p=.400) to indicate that auditors who correctly identif y an explanation as implausible will extend
53 testing less when the explan ation is given by an in-gr oup member. Although I found no support during the pilot test, recal l that the pilot s ubjects are not auditors, and therefore, have little experien ce in making decisions about ex tending testing during fieldwork. Hypothesis 4 predicts that decision ai d use will reduce effectively reduce plausibility judgments for auditors who have initial incorrect judg ments. Conover (1999) suggests using the nonparametric Friedman test as a substitute for parametric repeated measures analysis when comparing several related samples. Using the Friedman test, results indicate that the d ecision aid did not significantly change auditor judgments (p=.353). Hypothesis 5 suggests that confiden ce will increase post-decision aid. Using the Friedman test, there is insufficien t evidence to support Hypothesis 5 (p=.125). Finally, Hypothesis 6 predicts that aud itors who make a correct post-decision aid judgment will also increase their decision to extend testing. Although raw mean hours increased post-decision aid, there is insu fficient evidence to support a significant difference between pre-and post-deci sion aid extent of testing. 3.10.3 Discussion of Design Changes The pilot study was undertaken to provide preliminary data as well as identify potential weaknesses in the research design. One caveat is that the group manipulation in the pilot study did not exactly replicate the planned group ma nipulation in the main study (university rather than audit firm affiliation). Based on the pilot study results, I made several changes to the instrument. Since four of the 23 participants (17 percent) could not recall the group manipulation, I made the manipulation more salient. I significantl y reduced the amount of information given
54 about the client background and presented it in bullet point format. I added an accounting knowledge question to confirm the audito rsÂ’ internal knowle dge about account relationships. I included a question about th e participantÂ’s experience with affiliated clients, and modified a ques tion pertaining to skepticism tr aining. Finally, I simplified the decision aid in order to make its content more salient. Expert auditors indicated that staff audito rs often complete tasks similar to the one in the study. However, they also note d that based on the amount of background information, several alternative explanati ons exist for the change in repair and maintenance expense. Peecher and Solomon (20 01) suggest that intern al validity is more important than mundane realism. Therefore, I reduced the amount of information in the financial statements and firm background to make the task more manageable for the participants. This change reduces noise and eliminates alternative explanations for the change in account balances. The final in strument can be found in Appendix B.
55 CHAPTER 4: RESULTS 4.1 Background and Descriptive Statistics I collected data over a three-month pe riod using an online survey software application. Participants were staff and senior auditors from five large national firms; all were located in the Southeastern United Stat es. Fifty-five auditors answered the survey; fifty-three completed all ques tions. Eleven failed the group manipulation check and three failed the accounting knowledge check, leaving forty-one usable responses for the group bias hypotheses.16 Table 1 includes the descriptive statistics. Participants included twenty-two seniors and nineteen staff audito rs. Mean time to complete the survey was nineteen minutes, with a minimum of 9.25 and a maximum of 58.18. A frequency distribution is included in Fi gure 5. Average age was about th irty for seniors and twentyfive for staff members. Mean experience with analytical procedures was twenty-seven times (about thirty-nine times for seniors a nd only six times for staff members). On average, senior auditors worked with affiliate d clients thirteen percent of the time, staff members worked with affiliated clients only about six percent of the time. Table 2 shows the number of participants pe r treatment group. Twenty-four were in-group (client was a 16 The 25 percent failure rate is high and indicates that these individuals did not attend to the manipulation. While none of the participants who failed the group treatment manipulation check indicated an incorrect client affiliation, all eleven answered that they were unable to tell the clientÂ’s prior affiliation given the information provided. Given that these participants did not attend to the manipulation, they were dropped from the analysis of hypotheses related to group. Inclusion of these eleven does not qualitatively change results. The three auditors who failed the accountin g knowledge check did not demonstrate sufficient knowledge to accurately complete the initial step in plausibility determination. They were also dropped form the analysis.
56 former audit firm employee) and seventeen were out-group (client was a long-time employee of the client firm). FIGURE 5 0:56:00 0:48:00 0:40:00 0:32:00 0:24:00 0:16:00 0:08:00 Time To Complete 25 20 15 10 5 0 Frequency Frequency Distribution of Time to Complete For All Participants
57 TABLE 2 Â– PARTICIPANT DEMOGRAPHICS FOR GROUP BIAS HYPOTHESES TESTS (All participants n = 41) Panel ADescriptive Statistics Seniors (n=22) Staff (n=19) Overall (n=41) Time to Complete Mean 17.51 20.28 19.04 Standard Deviation 5.31 14.43 10.44 Minimum 9.53 9.25 9.25 Maximum 30.28 58.18 58.18 Gender Male 17 10 27 Female 5 9 14 Age Mean 30.59 25.63 28.39 Standard Deviation 8.18 4.30 7.00 Minimum 24 23 23 Maximum 55 41 55 Analytical Procedures Experience (# of times) Mean 39.41 6.43 26.37 Standard Deviation 44.73 10.56 37.12 Minimum 2 0 0 Maximum 200 40 200 Experience with Affiliated C lients (% of total clients) Mean 13.41 6.16 9.22 Standard Deviation 20.01 13.01 16.87 Minimum 0 0 0 Maximum 80 50 80 Certified Public Accountant 18 7 25 Highest Education Level B.S. Accounting 6 4 10 Master of Accounting 14 10 24 Master of Business Administration 2 3 5 Master Other 0 2 2 Panel B Â– NUMBER OF PARTICIPANT S IN EACH TREATMENT CONDITION Participants per Treatment N Senior Staff In-group 22 11 11 Out-group 19 13 6 Total 41 24 17
58 4.2 Correlation Matrices A comprehensive statistical analysis re quires the evaluation of correlations among the dependent variables. If the dependen t variables are significantly correlated, a multivariate approach is appropriate. Furt hermore, significant correlation of potential continuous covariates with the dependent va riables justifies thei r inclusion in the statistical analysis. I also evaluate correla tions between the demogr aphic variables (age, gender, level, analytical procedures experience, and affiliated percentage) and the dependent variables (plausibilit y, confidence, and extent of testing). Noting no significant correlations, I do not plan to include thes e demographics variables in the model. Table 3 includes the Spearman rank co rrelation coefficient matrices (chosen because of the non-normality of the data) fo r the fifty participants who passed the accounting knowledge check17. All three post-decision ai d dependent variables, plausibility, confidence, and extent of tes ting are significantly positively correlated with their respective pre-decision aid variables. This correlation supports the use of the Friedman test for repeated measures. Pre-decisi on aid extension of testing is significantly negatively correlated with initial plausibil ity (rho = -.481). Likewi se, post-decision aid extension of testing is significantly nega tively correlated with post-decision aid plausibility. These findings suggest support th at lower plausibility judgments lead to increased testing, as predicted by Hypothesis 6. 17 Results for the sample excluding the group manipula tion check failures are not qualitatively different.
60 4.3 Statistical Analysis Prior to hypothesis testing, I evaluated the data for compliance with the required statistical assumptions. Random assignment to groups implies that observations are independent for the between-subjects variab le, group, but are not independent for the within-subjects factor, decision aid. Univar iate analysis relies on an assumption of normality of the dependent variable (Me ndenhall and Sincich 1996). To test for univariate normality, I visually analyzed st em and leaf plots and histograms for each dependent variable (plausibil ity, confidence, and extent of testing) across groups. Given the small sample size, it was difficult to j udge normality from the graphs alone. I also analyzed the dependent variables for normality using formal statistical tests: ShapiroWilk and Kolmogorov-Smirnov with Lilliefors significance correc tion. Table 4 includes results. Based on these tests, the dependent variables are not normally distributed. However, these tests are highly sensitive to even small departures from normality and are therefore of limited use (Me ndenhall and Sincich 1996). TABLE 4 Â– TESTS OF NORMALITY FOR DEPENDENT VARIABLES Group Kolmogorov-Smirnov Shapiro-Wilk Statistic df Sig. Statistic df Sig. Pre-Decision Aid Plausibility Out .185 19 .088 .870 19 .015 In .172 20 .124 .880 20 .018 Confidence Out .286 19 .000 .816 19 .002 In .204 20 .028 .820 20 .002 Extent of Testing Out .223 19 .014 .829 19 .003 In .187 20 .066 .860 20 .008 Post-decision Aid Plausibility Out .223 19 .014 .852 19 .007 In .232 20 .006 .826 20 .002 Confidence Out .258 19 .002 .856 19 .008 In .222 20 .011 .822 20 .002 Extent of Testing Out .203 19 .039 .775 19 .001 In .207 20 .024 .832 20 .003
61 To address these concerns, I used the mo re appropriate nonparametric tests, which do not rely on the assumption of normality. For hypotheses that test variables with independent observations, I used the MannWhitney tests. For hypotheses that test repeated measures variables, I used the Friedman test. I evaluated the data for outliers and note d observations with values greater or less than two standard deviations from the mean. Three observations for initial confidence fit this criterion (all rated confidence at 10), wh ile two observations for extent of testing fit this criterion, (rated at 40.00 and 60.00) Removing these observations did not qualitatively change results. Given the small sample size and limited justification for removal, I retained these observations within the dataset. 4.3.1 In-Group Bias and its Effect on Initial Audit Judgment The purpose of this section is to report the findings as to whether in-group bias affects auditor plausibility judgments. Prior to testing this hypot hesis, I used MannWhitney to confirm that there was no statis tically significant diffe rence between groups in auditorsÂ’ ratings of client competence (p =.200). Thus, I am reasonably assured that each group perceived the client equally compet ent and that differences in plausibility judgments are unaffected by differences in client competence judgments. Table 5, Panel A displays the plausibility judgment mean, standard deviation, range and number of participants by level and trea tment group. Participants rated plausibility on a 101-point scale where 0 indicates Â“not at all plausibleÂ” and 100 indicate s Â“highly plausible.Â” I first reviewed the plausibility raw means for the total sample, noting that means were in the expected direction (in-group 54.50 and out -group 47.84). Using the Mann-Whitney nonparametric statistic, I found insufficient support to conclude that groups were
62 significantly different (p=.572 two-tailed). U pon further analysis, I noted that senior means were in the expected direction (in-group 56.64 and out-group 35.36), while staff means were in the opposite direction. Using Mann-Whitney, I tested the significance for seniors only, finding insufficient support (p =.097 one-tailed) to c onclude a significant difference between groups. TABLE 5 Â– TEST OF IN-GROUP BIAS ON INITIAL PLAUSIBILITY JUDGMENT a (mean, standard deviation, range, n) Panel ADescriptives Senior Staff In-Group 56.64 (36.34) 10-95 11 52.36 (36.35) 1-100 11 54.50 (35.54) 22 Out-Group 35.36 (31.96) 10-89 11 65.00 (23.76) 25-95 8 47.84 (31.83) 19 46.00 (35.12) 22 57.68 (31.54) 19 aPlausibility is measured on a 101-point sc ale where 0 is Â“not at all plausibleÂ” and 100 is Â“highly plausibleÂ” (1) Using Mann-Whitney, one-tailed p-value is insignificant at the .05 level. (2) No test is performed as means are in the opposite direction of prediction. Panel B Â– Total Sample Mann-Whitney Ranks for Total Sample Group N Mean Rank Sum of Ranks Plausibility Out-group 19 19.87 377.50 In-group 22 21.98 483.50 Total 41
63 Test Statistics Plausibility by Group for Total Sample Plausibility Mann-Whitney U 187.500 Wilcoxon W 377.500 Z -.565 Asymp. Sig. (2-tailed) .572 Panel C Â– Seniors Only Mann-Whitney Ranks for Seniors Ingroup 1 N Mean Rank Sum of Ranks 0 11 9.68 106.50 1 11 13.32 146.50 Plausibility Total 22 Test Statistics by Group for Seniors Plausibility Mann-Whitney U 40.500 Wilcoxon W 106.500 Z -1.338 Asymp. Sig. (2tailed) .181 Exact Sig. [2*(1tailed Sig.)] .193 One tailed Sig. .097 4.3.2 In-group Bias and it Effect on Auditor Confidence Hypothesis 2 predicts th at participants who r eceived an implausible explanation from an in-group client would be less confident in their initial judgment than auditors who received an implausible explanation from an outgroup client. Table 6, Panel A displays th e mean, standard deviation, range and number of participants by level and tr eatment group for the initial confidence variable. Participants indicated their c onfidence level on a 101-point scale where 0 indicates Â“not at all co nfidentÂ” and 100 indicates Â“completely confidentÂ”. Means
64 are in the predicted direction for the total sample (in-group 62.50 and out-group 71.95), for the seniors (in-group 64.09 and out-group 74.27) and for the staff (ingroup 60.91 and out-group 68.75). Using Mann-Whitney tests, there is insufficient evidence to support Hypothesis 2 for the total sample (p=.654), for senior auditors (p=.562), or for staff auditors (p=.968). TABLE 6 Â– TEST OF IN-GROUP BI AS ON INITIAL CONFIDENCE a (mean, standard deviation, range, n) Panel A Descriptives Senior Staff In-Group 64.09 (32.24) 10-95 11 60.91 (32.47) 15-90 11 62.50 (31.61) 22 Out-Group 74.27 (30.06) 10-100 11 68.75 (15.30) 45-85 8 71.95 (24.51) 19 69.18 (30.86) 22 64.21 (26.31) 19 a Â– confidence is measured on a 101-point scale where 0 is Â“not at all confidentÂ” and 100 is Â“completely confidentÂ”. Panel B Â– Total Sample Mann-Whitney Ranks for Total Sample Group N Mean Rank Sum of Ranks Out-group 19 21.89 416.00 In-group 22 20.23 445.00 Initial Confidence Total 41
65 Test Statistics by Group for Total Sample Initial Confidence Mann-Whitney U 192.000 Wilcoxon W 445.000 Z -.449 Asymp. Sig. (2tailed) .654 Panel C Seniors Mann-Whitney Ranks for Seniors Group N Mean Rank Sum of Ranks Out-group 11 12.36 136.00 In-group 11 10.64 117.00 Initial Confidence Total 22 Test Statistics by Group for Seniors Initial Confidence Mann-Whitney U 51.000 Wilcoxon W 117.000 Z -.632 Asymp. Sig. (2tailed) .527 Exact Sig. [2*(1tailed Sig.)] .562 Panel D Â– Staff Mann-Whitney Ranks for Staff Group N Mean Rank Sum of Ranks Out-group 8 9.88 79.00 In-group 11 10.09 111.00 Initial Confidence Total 19
66 Test Statistics by Group for Staff Confidence 1 Mann-Whitney U 43.000 Wilcoxon W 79.000 Z -.083 Asymp. Sig. (2tailed) .934 Exact Sig. [2*(1tailed Sig.)] .968 4.3.3 In-group Bias and it Effect on Aud itor Decisions to Extend Testing Hypothesis 3 predicts that in -group bias persists in an auditorÂ’s decision to extend testing, even when the auditor correctly identifies th e explanation as implausible. For this test, I used a subset of Â“low plausibilityÂ” aud itors to represent audito rs who are correct. In lieu of a normative answer, I considered auditors who rated plausibility less than 50 % to be correct. Table 7, Panel A displays the m ean, standard deviation, and range for the dependent variable extent of testing. Panel A also includes the number of pa rticipants by level and treatment group. The extent of test ing measurement represents the number of hours participants chose to exte nd analytical procedures testin g. Participants selected this amount after making plausibility and confid ence judgments. Using a Mann-Whitney test, I evaluated this hypothesis both pre and post-decision aid. As noted in section 4.2 plausibility rating is signifi cantly correlated with the depe ndent variable of interest Â“decision to extend testingÂ” bot h preand post-decision aid. Partitioning the sample to include only Â“correctÂ” re sponses should sufficiently address this correlation. Initial extent of testing is significantly co rrelated (rho = .881) with post-de cision aid extent of testing, but cannot be accommodated by nonparametric procedures. An analysis of the raw means for each group indicates that extent of testi ng is in the opposite direction from predicted
67 for both pre-decision aid (in-group 14.89 a nd out-group 11.11) and pos t-decision aid (ingroup 17.44 and out-group 12.44). Given the raw m eans for auditors who correctly assess an explanation, a statistical te st is unjustified. There is in sufficient evidence that group affiliation impacts the auditorÂ’s decision to extend testing either preor post-decision aid. A post hoc analysis in section 4.4 further investigates this finding. TABLE 7Â– TEST OF IN-GROUP BIAS ON DECISION TO EXTEND TESTING: CORRECT JUDGMENTS ONLY (mean, standard deviation, range, n) Pre-decision Aid Post -decision Aid In-Group Extent of Testing 14.89 (11.67) 2-40 9 17.44 (13.99) 4-40 9 Out-Group Extent of Testing 11.11 (9.33) 0-20 9 12.44 (8.05) 0-20 9 4.3.4 Discussion of Analysis of Decision Aid Hypotheses As noted in section 4.1 eleven participants failed to answer the group manipulation check correctly. Given that I found insufficient s upport to indicate the presence of in-group bias (and there is no reason to believe that group bias affects decision aid effectiveness), there is no justif ication to exclude the participants who failed the manipulation check from the analysis. I test the decision aid hypotheses using the complete sample18. Table 8 includes descriptive st atistics for the sample including manipulation check failures. 18 Note that two of the additional participants failed to indicate post-decision plausibility, confidence, and extent of testing, resulting in a final sample of 50.
68 TABLE 8Â– PARTICIPANT DEMOGRAPHICS FOR DECISION AID HYPOTHESES TESTS (All participants n = 50) Seniors (n=26) Staff (n=24) Overall (n=50) Time to Complete Mean 19.27 18.17 18.53 Standard Deviation 12.49 12.49 11.14 Minimum 9.53 9.25 9.25 Maximum 45.36 58.18 58.18 Gender Male 21 14 35 Female 5 10 15 Age Mean 31.04 26.08 28.66 Standard Deviation 8.59 4.39 7.28 Minimum 24 23 23 Maximum 55 41 55 Analytical Procedures Experience (# of times) Mean 38.12 5.72 25.87 Standard Deviation 44.67 10.75 39.91 Minimum 2 0 0 Maximum 200 40 200 Experience with Affiliated C lients (% of total clients) Mean 13.48 7.71 10.65 Standard Deviation 18.56 16.77 17.92 Minimum 0 0 0 Maximum 80 75 80 Certified Public Accountant 22 8 30 Highest Education Level B.S. Accounting 8 4 12 Master of Accounting 15 14 29 Master of Business Administration 3 4 7 Master Other 0 2 2 4.3.5 Decision Aid Use and its Effect on Auditor Plausibility Judgments Hypothesis 4 tests for the e ffectiveness of a decision ai d on improving plausibility judgments for auditors who were initially incorrect. In this study, a reduction in plausibility rating represents an improvement in judgment (since the explanation given is
69 implausible). I partition the total sample by initial plausibility judgments considering judgments greater than or equal to 50% as incorrect. As shown in Table 9, Panel A, overall, plausibility judgments decreased after use of the decision aid (initial plausibility mean 75.77 and post-decision aid plausi bility mean 49.76). The appropriate nonparametric test for related samples is th e Friedman test (Conover 1999). Results for the Friedman test for the total sample (Table 9, Panel B) find support for the effectiveness of the decision aid to improve plausibility judgments (p=.000). Hypothesis 4 is supported. Additional analysis finds that hypothesis 4 is also supporte d for seniors (Table 7, Panel C) (p=.008) and for staff (Table 9, Panel D) (p=.001). TABLE 9Â– EFFECT OF DECISION AID ON AUDITOR PLAUSIBILITY JUDGMENTS (Incorrect auditors only) (mean, standard deviation, range, n) Panel A Â– Descriptives Seniors Staff Pre-decision Aid Plausibility 80.33 (14.00) 50-95 12 72.89 (16.36) 50-100 19 75.77 (15.68) 31 Post-decision Aid Plausibility 55.73 (34.65) 0-95 11 46.11 (32.34) 0-100 18 49.76 (32.96) 29 Panel B Â– Total Sample Friedman Ranks for Total Sample Mean Rank Pre-decision Aid Plausibility 1.84 Post-decision Aid Plausibility 1.16
70 Test Statistics for Change in Plausibility for Total Sample N 29 Chi-Square 18.182 df 1 Asymp. Sig. .000 Panel C Â–Seniors Friedman Ranks for Seniors Mean Rank Pre-decision Aid Plausibility 1.82 Post-decision Aid Plausibility 1.18 Test Statistics for Change in Plausibility for Seniors N 11 Chi-Square 7.000 df 1 Asymp. Sig. .008 Panel D Â– Staff Friedman Ranks for Staff Mean Rank Pre-decision Aid Plausibility 1.86 Post-decision Aid Plausibility 1.14 Test Statistics for Change in Plausibility for Staff N 18 Chi-Square 11.267 df 1 Asymp. Sig. .001 Section 4.3.6 Effect of Decision Aid on Confidence Hypothesis 5 predicts that auditor confidence will increase post decision aid. I measured confidence on a 101-point scale where 0 is Â“not at all c onfidentÂ” and 100 is Â“completely confident.Â” Table 10 includes the mean, standard deviation, range and
71 number of observations. Predecision aid confidence ratings ranged from 10 to 100, with a mean of 68.79 (standard deviation 27.96). Po st-decision aid confidence ratings ranged from 0-100 with a mean of 65.74 (standard de viation 32.72). Contrary to expectations, raw mean confidence scores decreased fo r both seniors and st aff members; thus Hypothesis 5 is not supported. Section 4.4 contains a post hoc anal ysis that explores the changes in confidence. TABLE 10Â–CHANGE IN CONFID ENCE POST-DECISION AID (mean, standard deviation, range, n) Descriptives Seniors Staff Pre-decision Aid Confidence 69.70 (30.99) 10-100 27 67.80 (24.88) 15-95 25 68.79 (27.96) 52 Post-decision Aid Confidence 66.92 (35.28) 0-100 26 64.46 (24.88) 0-100 24 65.74 (32.72) 50 Section 4.3.7 Effect of Decisi on Aid on Extent of Testing As noted, it is insufficient to examine auditor judgments alone, as auditor decisions ultimately impact audit effectivene ss. This test explores whether auditors will improve their decisions after using a decision ai d. Hypothesis 6 predicts that auditors who initially make an incorrect judgment, will incr ease their extent of testing after using a decision aid. In keeping with prior procedures, I restrict my analysis to auditors who initially provided an incorrect judgment (pla usibility judgment greater than or equal to 50%). Table 11, Panel A includes the mean, standard deviation, and range for the dependent variable extent of testing both preand post-de cision aid. After having access to the decision aid, participants increa sed testing by 61 %, from 6.34 hours to 10.21
72 hours. Using the Friedman test, which is th e appropriate nonparametric statistic for a repeated measures analysis, this increase in th e extent of testing is significantly greater post-decision aid for the total sample (p=.000). Additional analysis shows that Hypothesis 6 is also supporte d for staff auditors (p=.001), but not for senior auditors (p=.083). Section 4.4 includes a post hoc anal ysis of changes in extent of testing for auditors who are initially correct. TABLE 11 Â– TEST OF EFFECT OF DECI SION AID ON EXTENT OF TESTING (INITIALLY INCORREC T AUDITORS ONLY) (mean, standard deviation, range, n) Panel A Â– Descriptives Seniors Staff Pre-decision aid Extent of Testing 5.00 (5.79) 0-16 11 7.17 (8.38) 0-30 18 6.34 (7.28) 29 Post-decision aid Extent of Testing 6.27 (5.06) 0-16 11 12.61 (14.86) 0-60 18 10.21 (12.37) 29 Panel B Â– Total Sample Friedman Ranks for Total Sample Mean Rank Pre-decision Aid Ex tent of Testing 1.26 Post-decision Aid Extent of Testing 1.74 Test Statistics for Change in Extent of Testing for Total Sample N 29 Chi-Square 14.000 df 1 Asymp. Sig. .000
73 Panel C Â– Seniors Friedman Ranks for Seniors Mean Rank Pre-decision Aid Ex tent of Testing 1.36 Post-decision Aid Extent of Testing 1.64 Test Statistics for Change in Extent of Testing for Seniors N 11 Chi-Square 3.000 df 1 Asymp. Sig. .083 Panel D Â– Staff Friedman Ranks for Staff Mean Rank Pre-decision Aid Ex tent of Testing 1.19 Post-decision Aid Extent of Testing 1.81 Test Statistics for Change in Extent of Testing for Staff N 18 Chi-Square 11.000 df 1 Asymp. Sig. .001 4.4 Post Hoc Analysis I perform the following post hoc analyses to investigate prior nonsignificant findings and to explore re levant relationships. Hypothesis 3 suggested that in-group bi as could cause aud itors who correctly identified a client explanation as implausi ble to curtail additi onal testing. I found no support to indicate that group bias affects de cisions to extend testing. However, it is important to confirm that auditors who corre ctly identified a client explanation as implausible did, in fact, increase testi ng (independent of group). I conducted the following Mann-Whitney nonparametric test to confirm that auditors who judged the client explanation as implausible increased testing more than a uditors who judged the
74 client explanation as plausible. Table 12 incl udes the ranks and test statistics. Findings suggest that auditors acted as expected and that deci sions to extend testing logically followed judgments (p=.008). TABLE 12 POST HOC ANALYSIS OF EXTENT OF TESTING Mann-Whitney Ranks for Total Sample Initial Plausibility N Mean Rank Sum of Ranks Correct 21 33.21 697.50 Incorrect 31 21.95 680.50 Initial Extent of Testing Total 52 Test Statistics by Group for Total Sample Initial Extent of Testing Mann-Whitney U 184.500 Wilcoxon W 680.500 Z -2.653 Asymp. Sig. (2tailed) .008 Hypothesis 5 proposed that confidence would increase post-decision aid for all auditors. A post hoc analysis analyzed the changes in confidence by initial plausibility, separating auditor into groups of initially correct and ini tially incorrect. Table 13, Panel A includes descriptive st atistics. Table 13, Panel B reports th e Friedman test statistics for preand post-decision aid confidence for audito rs who were initially correct. Confidence significantly increased post-dec ision aid (p=.005). This re sult is logical because the decision aid provided confirming evidence. It also partially supports Hypothesis 5. Table 13, Panel C reports Friedman test statistics for preand post-decision aid confidence for auditors who were initially incorrect. While raw means indicate that confidence decreased post-decision aid, this decrease was not significant (p=.127).
75 TABLE 13 POST HOC ANALYSIS OF CONFIDENCE BY INITIAL JUDGMENT (mean, standard deviation, range, n) Panel A Initially Correct Initially Incorrect Pre-decision aid Confidence 64.52 (35.03) 10-100 21 72.66 (21.33) 10-95 29 Post-decision aid Confidence 71.57 (33.36) 10-100 21 61.52 (32.16) 0-100 29 Panel B Friedman Ranks for Initially Correct Auditors Mean Rank Pre-decision aid Confidence 1.31 Post-decision aid Confidence 1.69 Test Statistics for Initially Correct Auditors N 21 Chi-Square 8.000 df 1 Asymp. Sig. .005 Panel C Friedman Ranks for Initially Incorrect Auditors Mean Rank Pre-decision aid Confidence 1.62 Post-decision aid Confidence 1.38
76 Test Statistics for Initially Incorrect Auditors N 29 Chi-Square 2.333 df 1 Asymp. Sig. .127 Hypothesis 6 examines whether decision aid use improves decisions for auditors who are initially incorrect. I performed additi onal testing to confirm that auditors who are initially correct also increase testing pos t-decision aid. Tabl e 14, Panel A provides descriptive statistics of preand post-decision aid extent of testing for initially correct auditors. Table 14, Panel B show s test results. There is sign ificant support (p=.014) that initially correct auditors also increased testing post-decision aid. TABLE 14 POST HOC ANALYSIS OF EXTENT OF TESTING FOR INITIALLY CORRECT AUDITORS (mean, standard deviation, range, n) Panel A Pre-decision aid Extent of Testing 13.29 (10.05) 0-40 21 Post-decision aid Extent of Testing 15.57 (11.01) 0-40 21 Panel B Friedman Ranks for Initially Correct Auditors Mean Rank Pre-decision aid Extent of Testing 1.36 Post-decision aid Extent of Testing 1.64
77 Test Statistics for Initially Correct Auditors N 21 Chi-Square 6.000 df 1 Asymp. Sig. .014
78 CHAPTER 5: CONCLUSION 5.1 Discussion of Results Table 15 includes a summary of findings. I found no support for an in-group bias effect on auditor plausibility judgments, conf idence in those judgments or decisions to extend testing. I found strong support for the e ffect of a decision aid on improvements in auditor plausibility judgments and decisions to extend testing. I found no support for an increase in confidence post-decision aid. I follow with a discussion of findings and possible reasons for lack of significant findings. TABLE 15 SUMMARY OF FINDINGS Hypothesis IV DV Supported p-value 1 Group Plausibility No .572 2 Group Confidence No .654 3 (pre-decision aid) Group Extent of Testing No --3 (post-decision aid) Group Extent of Testing No --4 Decision Aid Plausibility Yes .000 5 Decision Aid Confidence No --6 Decision Aid Extent of Testing Yes .000 Legislation that restricts client hiring of former ex ternal auditors provides evidence that there is a belief that in-gr oup bias exists and that it affects auditor independence. Although theory s uggests that individuals dem onstrate in-group bias in the form of extending unjustified trust to thei r group members, auditors may or may not exhibit this bias in an audit context. I em ploy an experiment to investigate potential differences in auditor judgments based on the clientÂ’s former employment with the audit
79 firm. Hypothesis 1 predicted th at auditors would judge an explanation from an in-group client as more plausible than an ex planation from an out-group client. I partitioned the sample into seniors and st aff to analyze the data in more detail. An analysis of raw means for seniors indicate d that plausibility judgments were, in fact, higher for the in-group treatment, (56.64 vers us 35.36); however, the difference was not statistically significant. Raw m eans for staff auditorsÂ’ plausi bility judgments are in the opposite direction with out-group plausibility ju dgments higher than in-group plausibility judgments (65.00 versus 52.36); again, the differe nce is not statistically significant. The raw means do suggest that seniors are more lik ely to exhibit in-group bias than are staff auditors. Seniors could be more likely to ex hibit in-group bias because they likely have been a part of the audit firm group for a longe r period of time than have staff auditors. In addition, staff auditors are likely recent graduates of accounting programs. These programs typically cover professional st andards, which emphasize professional skepticism. The emphasis on skepticism could cau se staff auditors to pay close attention to client source reliability, thus mitiga ting in-group bias. Although the current study found insufficient evidence to support an eff ect of in-group bias on auditor plausibility judgments in an analytical procedures task, increasing the sample size of senior auditors only might shed light on the pr evalence of in-group bias. While theory supports finding a difference, there are several possible reasons why I did not find a significant difference. Th ese include lack of power, experimental weaknesses, or absence of a di fference in fact. First, I had access to a limited sample of auditors, which resulted in a sm all pool of participants. To rec tify this situation, I plan to collect additional data. Second, several par ticipants failed the ma nipulation check and were removed from the analysis. A failed mani pulation check is often the result of an
80 experimental weakness. I plan to improve the study by making the client affiliation manipulation more salient, perhaps by in cluding detailed information about the controllerÂ’s background, particul arly his or her experience at the audit (client) firm. Finally, it may be that auditors do not e xhibit in-group bias when performing audit procedures. Both extensive training and atten tion to professional skepticism act against an individualÂ’s inclination to exhibit in -group bias and could mitigate this bias. Hypothesis 2 predicted an eff ect of in-group bias on confid ence such that auditors who received an implausible explanation from an in-group client would be less confident in their plausibility judgments than an auditor who received the same explanation from an out-group client. Although there is insufficien t evidence to support a statistical difference between groups, the raw means are in a dire ction consistent with Hypothesis 2. In-group seniors have a mean confidence level of 64.09, while out-group seniors demonstrate a higher mean confidence of 74.27. The same rela tionship holds for staff membersÂ’ mean confidence: in-group, 60.91 and out-group, 68.75. Prior research has indicated that factors such as experience and gender coul d moderate confidence. Although I collected data regarding participantsÂ’ analytical proc edures experience and gender demographics, the use of nonparametric statistics prevented thei r inclusion in the analysis, since there is no nonparametric procedure that allows for the inclusion of covariat es. A larger sample size could allow the use of parametric statis tics, which, in turn, accommodate models that are more powerful and allow for the inclusion of covariates. Hypothesis 3 predicted that in-group bias would affect an auditorÂ’s decision to extend testing such that even though the audito r had made a correct plausibility judgment, he or she would extend testing less if the client was a former audit team member. In other words, even though an auditor Â“knowsÂ” that a client is providing an implausible
81 explanation, he or she coul d still choose to Â“overlookÂ” the inconsistency of the explanation because the client is a former aud itor from his or her firm. I tested for this effect both preand post-decision aid. I exam ined the means and found that pre-decision aid, in-group auditors extended testing by 14.89 hours, while out-group auditors extended testing an average of 11.11 hours. Likewise, post-decision aid measures show that ingroup auditors extended testing by 17.44 hours, while out-group members extended testing by only 12.44 hours. Although there is no evidence of a group effect, in a post hoc analysis, I analyzed extent of testing to confirm that an auditor who makes a correct initial plausibility judgment extends testing mo re than an auditor who makes an incorrect initial plausibility judgment. I find signifi cant support that audito rs do, in fact, extend testing more when they are correct than when they are incorrect. This finding indicates that participants expended the requisite cognitive effort to the task. Thus, results likely indicate that there is no in-g roup bias in auditorsÂ’ decisi ons to extend testing in an analytical procedures task. In addition to testing for group biases, I also examined whether a simple decision aid could improve auditorsÂ’ pl ausibility judgments, confidence in those judgments, and decisions. Hypothesis 4 predicte d that auditors who provide d initially in correct (high plausibility) judgments, would decrease t hose judgments after using a decision aid. Nonparametric statistical analyses provide d evidence that decision aids improved auditorsÂ’ plausibility judgments in an anal ytical procedures task. This effect was supported for the total sample and for staff and senior auditors independently. Seniors significantly reduced their plausibility judgments from 80.33 to 55.73, while staff auditors significantly reduced their pl ausibility judgments from 72.89to 46.11. These
82 findings justify the effectiveness of a si mple decision aid in improving auditorsÂ’ performance during analytical procedures. Prior literature indicates confidence improves as individuals gather more information. Hypothesis 5 predicted that audi tors would increase their confidence after using a decision aid. A review of the raw m eans indicated that confidence decreased overall for both seniors (69.70 to 66.92) a nd staff members (67.80 to 64.46). I explored the change in confidence further in a post hoc analysis. Confirmation bias suggests that individuals tend to disregard disconfir ming evidence and overweight confirming evidence. Although I found no prior l iterature indicating that this effect is associated with changes in confidence, I chose to partition the sample by ini tial plausibility judgment to explore this variable further. I found that for auditors who were initially correct (low plausibility), confidence significantly increa sed after using a decision aid. The decision aidÂ’s confirmation of their or iginal judgment likely is re sponsible for their increased confidence. However, confidence for auditors who were initially incorrect showed a marginally significant decrease. This decrea sed confidence is possi bly a result of the disconfirming evidence provided to those audi tors by the decision aid. Although initially incorrect auditors improved their plausibi lity judgment post-d ecision aid (indicating reliance on the decision aid), th ey would logically have fe lt less confident about their own ability to audit. It is possible that wh en they answered the confidence question, they were indicating confidence in their ability, rather than c onfidence in that particular judgment. Hypothesis 6 predicted that the use of a decision aid would improve auditorsÂ’ decisions to extend testing. As I did before in the tests for Hypothesis 4 (effect of decision aid on plausibility judgm ents); I partitioned the sa mple, choosing only auditors
83 who were initially incorrect (h igh plausibility). I found that those auditors significantly increased the extent of testing after us ing a decision aid. This finding supports the effectiveness of a decision aid on auditor deci sions to extend testing in an analytical procedures task. I also analyzed the participants by level. While senior auditors increased their extent of testing from a mean of 5.00 to 6.27, (the correct directio n but a statistically insignificant difference), staff auditors incr eased their extent of testing from 7.17 to 12.61, (a statistically significant difference ). Although only staff auditors increased testing significantly, seniors also increased testing. These findings support the hypothesis that decision aids improve decisions to extend testing. A post hoc analysis explores the effect of a decision aid on extent of testing for initially correct (low plausibility) auditors Raw means for extent of testing increased from 13.29 pre-decision aid to 15.57 post-deci sion aid. This increase was statistically significant, demonstrating that decision aids are effective in improving auditor decisions for both initially correct and in itially incorrect auditors. 5.2 Summary This study had two objectives: first, to investigate whether in-group bias was evident in auditorsÂ’ judgment s, confidence in those judgments, and decisions and second, to examine whether a decision aid was eff ective in improving auditorsÂ’ judgments, confidence in those judgments, and decisions Auditors completed an online task in which they evaluated client explanations fo r changes in an account balance. The client sourceÂ’s affiliation differed betw een participants Â– in-group c lients were former members of the participantÂ’s audit firm, out-group members were long-time client employees. Based on Social Identity Theory, I predicted th at auditors would exhi bit in-group bias in their judgments and decisions, assigning a highe r level of plausibil ity to explanations
84 obtained from a former group member, and redu cing testing for in-group client audits. I found no effect for in-group bias on judgment, confidence in judgment, or extent of testing. After collecting auditorsÂ’ initial plausibi lity judgments, confidence ratings, and decisions, I presented them with a decision aid report. I expected the structured design of the report to improve audit pl ausibility judgments, confiden ce in those judgments, and decisions to extend testing. The decision ai d improved plausibility judgments for both staff and senior auditors, and for both initiall y incorrect and initially correct auditors. The decision aid also increased c onfidence for auditors who made initially correct judgments, but not for auditors who were initially incorrect. For auditors who were initially incorrect, there was a marginally significant reduction in confidence. Although decision aid use did not result in increased confidence for all aud itors, the decision aid resulted in improved plausibility judgments and decisions to extend testing. Practitioners should note the positive effects of providing a decision aid during analytical review. Professional skepticism is necessary to audit effectively; however, auditors are subject to human biases. An auditorÂ’s failure to adjust appropriately his or her assessment of client objectivity may compromise indepe ndence and audit effectiveness. Audit firms should be aware of the potential for this bias so that they can reduce the risk of audit failure. Congress and the AICPA already have noted that the hiring of former audit team members could lead to an impairment of i ndependence and objectivity. This study sought to improve the understanding of both the existe nce and extent of this claim. However, due to the small sample size, results about in -group bias are inconclusive. Additional data collection could provide result s that are more conclusive.
85 A simple decision aid was effective in improving judgments overall. Both seniors and staff members improved their judgments, as well as their decisi ons post-decision aid. The decision aid also improved judgments a nd decisions not only fo r auditors who were initially incorrect, but also for auditors who initially rated plausibility low. An added benefit is that the decision aid increased confidence for auditors who were initially correct. This increase in confidence possibl y stems from the positive feedback offered by the decision aid. The decision aid used in this study was a simp le listing of account relationships and expectations related to t hose relationships. The de cision aid provided valid, reasonable advice to auditors during th e task. Audit firms could find the use of simple decision aids a low-cost wa y to improve auditor performance. 5.3 Limitations 5.3.1 Small Sample Size Pedhazur and Schmelkin (1991) list four elements to consider when using a decision-based strategy for hypothesis testing.19 These elements are effect size, Type I error, Type II error and sample size. In this study, effect size refers to the magnitude of the difference between groups (and between preand post-decision aid) for the dependent variables plausibility, confiden ce, and extent of testing. A Type I error (d esignated by ) is the error of reject ing the null, when it s hould not have been re jected (Pedhazur and Schmelkin 1991). In this case, a Type I error would be to c onclude that there is an ingroup bias, when there is not actually an in -group bias. A Type II error (designated by ) is the error of failing to reject the null hypot hesis, when, in fact, it should be rejected. 19 A decision-based strategy refers to using a pre-determined value for hypothesis testing. For example, when comparing two groups, setting an (alpha) value to determine rejection of the null.
86 This is also known as the power to detect a difference, should one exist. An example in this study would be finding no significant group bi as, when, in fact, there is a significant group bias. Sample size, the fourth element, in ter-relates with effect size, and both Type I and Type II errors, such that increases in sa mple size, increase power, while decreases in sample size decrease power (holding effect si ze constant). In this study, the sample size was small, which made determination of nor mality of the data problematic. Without the ability to confirm that the data was normal I chose to use nonparametric statistical methods (which do not rely on normality). N onparametric methods are more likely to result in a Type II error (le ss likely to detect differences). Given that I designed the study with careful attention to internal validity, I estimate that my failure to detect group bias is a result of either small sample size or sma ll (no) bias effects in fact. While there is insufficient evidence to reje ct the null hypothesis (of no group bias), based on the data collected, I likewise cannot c onclude that in-group bias doe s not exist for auditors. By increasing sample size in the future, I hope to arrive at results that are more conclusive. 5.3.2 Alternative Explanations There are a number of limitations to cons ider in interpreting the results of the current study. Given the hei ghtened awareness of threats to independence resulting from auditor affiliation, particip ants could have engaged in hypothesis guessing. Demand effects from hypothesis guessing typically re sult in participants trying to Â“give the researcher what he or she wants.Â” In this st udy, participants could ha ve wanted to appear in the best light possible, answering in such a way as to obscure their inclination toward in-group bias.
87 Limitations to the findings of a reduction in plausibility post decision-aid could be due to a recency effect, ra ther than a mitigation of in-g roup bias. Recency argues that auditors overweight information received later in a sequence.20 In this study, since auditors receive the decision aid report last they could have placed more weight on its recommendation. Both Hogarth and Einhorn (1992) and Ashton and Ashton (1988) find recency effects for a series of conflicting evidence. However, the tasks used in those studies were not analytical pr ocedures tasks. Asare and Messier (1991) note that in an unpublished study, Bonner and Butler (1989) did not find recency effects in an analytical procedures task. Confirmation bias (Church 1990) could also mitigate the effectiveness of the decision aid. Confirmation bias exists wh en individuals tend to overweight evidence that supports their initi al beliefs. Auditors who initially believe the client could be more likely to disregard the deci sion aid report, resulting in a non-significant finding. However, Smith and Kida (1991) find that au ditorÂ’s conservatism precludes the use of confirmatory strategies. Since participants are practicing auditors, confirmation bias is unlikely. 5.3.3 Experimental Context The experimental context is also a limita tion. The sterility of an online experiment cannot capture the face-to-face in teractions present in an actual audit. When faced with individuals that they know personally and w ith whom they have a working relationship and history, auditors may subconsciously make different judgments than they would in an experimental setting. In-group bi as in an audit context might be more subtle and difficult to recreate in an experimental setting. Th is study is also limited to positive prior 20 Asare and Messier (1991) provide an in-depth summary of belief adjustment audit research.
88 relationships between the parties. Circumst ances in which the past relationship is negative could result in different findings. The online method of data collection has limitations as well. The researcher cannot observe participants as they proceed th rough the survey; this lack of supervision reduces experimental control. Participants can engage in multiple tasks (e.g., surfing the web, talking on the phone, answering e-ma ils) while completing the online survey. Participants can also leave the computer and return later leaving th e researcher to guess whether the extra time spent online was, in fact, representative of added effort or lack of effort. In this particular study, an additional limitation arose from the recruiting method. The researcher had no control over which audi tors at a firm took the survey. Therefore, selection bias could have been a factor in th e results. Auditors who took the survey could have been the Â“less capableÂ” auditors with more free time. On the other hand, partners could have selected the Â“more capableÂ” auditors to answer the survey in order to present their firm in the best light. 5.4 Future Research There are several avenues for future research including addressing research design weaknesses, usin g alternative research methods, extending research parameters, and altering the decision aid. As noted above, the research design was limited. Recency provides an alternative explanation to findi ngs of decision aid effectiveness. Prior research suggests a recency effect for mixe d evidence in a conten t-rich audit setting (Tubbs et al. 1990). This issue could be addressed by includ ing a group that receives the decision aid concurrently with the client e xplanation and comparing that group with predecision aid judgments. A large number of part icipants (11 out of 55 or 20%) failed the
89 between-subjects group affiliation manipula tion check. While the removal of these observations is justified, the smaller sample size reduced the power of the study. Future trials can be modified to make the gr oup manipulation more salient. Alternatively, researchers can require particip ants to respond to a set of que stions that ensures they are aware of the manipulation before proceeding with the experiment. Another avenue for future research is to use an alternative research method. Given that in-group bias appears sensitive to face-to -face cues, an experiment that uses actual firm auditors interacting with participants could improve results. An archival approach using working papers for completed audits woul d provide a richer data set. By analyzing auditorsÂ’ work, I could explor e whether auditor judgments and decisions differ based on the presence/absence of an affiliated client. A natural and relevant extension of this re search is to vary the participants of interest. Archival studies including Le nnox (2005) and Menon and Williams (2004) find evidence of affiliation bias at the partner level. Based on the current study, there is some evidence that seniors exhibit bias, while staff members do not. Using managers and partners in an experimental study could reveal stronger bi ases. Another extension would explore affiliation at various levels; for example, does affiliation (in-group bias) occur between members of the same office, the sa me firm, or even between Big Four group members? In addition, does in-group bias depend on the audit task? This study used a single task, analytical procedures related to expense accounts, ofte n completed by a lower level employee. Given the multitude of task s completed during an audit, it would be worthwhile to explore tasks that have a larger impact on the final audit opinion ( e.g., evaluation of a going concern). As a final point given the effectiveness of the decision aid report, future research should investig ate the development and effectiveness of
90 decision aids in audit practice. Eining et al. ( 1997) find that constructive dialogue a form of interaction between participants and the decision aid, auditor performance. The decision aid in this study c ould be modified to include an interactive component. The current study found that staff auditors relied heavily on their plausi bility judgments in making the decision to extend testing. Given the link between judgments and decisionmaking, it is worthwhile to study how deci sion aids can improve audit practice.
91 References Abdolmohammadi, M. and C. Usoff. 2001. A Lo ngitudinal Study of Applicable Decision Aids for Detailed Tasks in a Financial Audit. International Journal of Intelligent Systems in Accounting, Finance and Management 10(3): 139-154. Abdolmohammadi, M. J. 1991. A Test of the Relationship Between Task Structure and Decision Aids Type in Auditing. In L. A. Ponemon, & D. R. L. Gabhart (Eds.), Auditing Advances in Behavioral Research New York: Springer Verlag. Abdolmohammadi, M. J. 1999. A Comprehensive Taxonomy of Audit Task Structure, Professional Rank, and Decision Ai ds for Behavioral Research. Behavioral Research in Accounting 1999(11): 51-79. Ahlawat, S. S. 1999. Order Effects and Memory for Evidence in Individual versus Group Decision Making in Auditing. Journal of Behavioral Decision Making 12(1): 7188. AICPA. 2005. AICPA Professional Standards New York: American Institute of Certified Public Accountants, Inc. Anderson, J. C., K. K. Moreno, and J. M. Mueller. 2003. The Effect of Client vs. Decision Aid as a Source of Explanati ons upon Auditor's Sufficiency Judgments: A Research Note. Behavioral Research in Accounting 15: 1-11. Anderson, U. and L. Koonce. 1998. Evaluating the Sufficiency of Causes in Audit Analytical Procedures. Auditing: A Journal of Practice and Theory 17(1): 1-12. Anderson, U., L. Koonce, and G. Marchant 1994. The effects of source-competence information and its timing on auditors' pe rformance of analytical procedures. Auditing: A Journal of Practice and Theory 13(Spring): 137-148.
92 Apfelbaum, E. and I. Lubek. 1979. Relati ons of Domination and Movements for Liberation: An Analysis of Power be tween Groups. In W. G. Austin, & S. Worchel (Eds.), The Social Psychology of Intergroup Relations : 188-204. Monterey: Brooks/Cole Publishing Company. Arkes, H. R., R. M. Dawes, and C. Christ ensen. 1986. Factors influencing the use of a decision rule in a probabilistic task. Organizational Behavior and Human Decision Processes 37(1): 93-110. Asare, S. K. and W. F. Messier Jr. 1991. A Review of Audit Research Using the BeliefAdjustment Model. In L. A. Ponemon, & D. R. L. Gabhart (Eds.), Auditing Advances in Behavioral Research : 75-92. New York: Springer Verlag. Ashton, A. and R. H. Ashton. 1988. Sequential belief revision in auditing. The Accounting Review 63: 623-641. 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. Ashton, R. H. 1992. Effects of justificat ion and a mechanical aid on judgment performance. Organizational Behavior and Human Decision Processes 52: 292306. Ballou, B. 2001. The relationship between auditor characteristics and the nature of review notes for analytical procedure working papers. Behavioral Research in Accounting 13: 25-49. Bamber, E. M. 1983. Expert Judgment in the Audit Team: A Source Reliability Approach. Journal of Accounting Research 21(2): 396-413.
93 Bamber, E. M., P. R. Gillett, T. J. Mock, and K. T. Trotman. 1995. Audit Judgment. In T. B. Bell, & A. M. Wright (Eds.), Auditing Practice, Research, and Education : 210. New York: American Institute of Certified Public Accountants, Inc. Bamber, E. M. and R. J. Ramsay. 2000. The e ffects of specializati on in audit workpaper review on review efficiency and reviewers' confidence. Auditing 19(2): 147-158. Barrionuevo, A. 2002. Questioning the Books: Court Documents Show Andersen's Ties With Enron Were Growing in Early '90s, Wall Street Journal Eastern ed.: A.6. New York. Beasley, M. S., J. V. Carcello, and D. R. Hermanson. 2000. Should You Offer a Job to Your External Auditor? Journal of Corporate Accounting & Finance 11(4): 3542. Bedard, J. C. and L. E. Graham. 2002. The Effects of Decision Aid Orientation on Risk Factor Identification an d Audit Test Planning. Auditing 21(2): 39-56. Biggs, S. F., W. R. Knechel, N. R. Walker W. A. Wallace, and J. J. Willingham. 1995. Analytical Procedures. In T. B. Bell, & A. M. Wright (Eds.), Auditing Practice, Research, and Education : 110-143. New York: American Institute of Certified Public Accountants, Inc. Bleed, J. 2002. Industry Official Says Accounting Firms Face Unprecedented Challenge, Knight Ridder Tribune Business News : 1. Washington. Brazel, J. F., C. P. Agoglia, and R. C. Hatfield. 2004. Electronic versus Face-to-Face Review: The Effects of Alte rnative Forms of Review on Auditors' Performance. The Accounting Review 79(4): 949-966.
94 Bukszar, E. 2003. Does Overconfidence Lead to Poor Decisions? A Comparison of Decision Making and Judgment Under Uncertainty. Journal of Business and Management 9(1): 33-43. Butler, S. A. 1985. Application of a deci sion aid in the judgmental evaluation of substantive tests of details samples. Journal of Accounting Research 23: 513-526. Chung, J. and G. S. Monroe. 2000. The effect s of expereince and task difficulty on accuracy and confidence assessments of auditors. Accounting and Finance 40: 135-152. Church, B. 1990. Auditors' use of confirmatory processes. Journal of Accounting Literature 9: 81-112. Cohen, J. and T. Kida. 1989. The Impact of Analytical Review Results, Internal Control Reliability, and Experience on Audito rs' Use of Analytical Review. Journal of Accounting Research 27(2): 263-276. Conover, W. J. 1999. Practical Nonparametric Statistics (Third ed.). New York: John Wiley & Sons, Inc. Einhorn, H. J. and R. M. Hogarth. 1978. Conf idence in judgment: Persistence of the illusion of validity. Psychological Review 85(5): 395-416. Eining, M. M., D. R. Jones, and J. K. Loebbecke. 1997. Reliance on Decision Aids: An Examination of Auditors' Assessment of Management Fraud. Auditing: A Journal of Practice and Theory 16(2): 1-19. Ellemers, N., R. Spears, and B. Doosje. 2002. Self and Social Identity. Annual Review of Psychology 53: 161-186.
95 Emby, C. and D. Finley. 1997. Debiasing Frami ong Effects in Auditors' Internal Control Judgments and Testing Decisions. Contemporary Accounting Research 14(2): 55-77. Fischoff, B. 1982. Debiasing. In D. Kahne man, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and Biases : 422-444. Cambridge: Cambridge University Press. George, J. F., K. Duffy, and M. Ahuja. 2000. Countering the anchoring and adjustment bias with decision support systems. Decision Support Systems 29: 195-206. Hall, J. A. 2004. Accounting Information Systems (Fourth ed.): Thomson Southwestern. Hastie, R. 2001. Problems for Judgment and Decision Making. Annual Review of Psychology 52: 653-683. Heritage, A. 2000. The American Heritage Dictio nary of the English Language (Fourth ed.): Houghton Mifflin Company. Hewstone, M., M. Rubin, and H. Willis. 2002. Intergroup Bias. Annual Review of Psychology 53: 575-604. Hirst, D. E. 1994. Auditor's sensitivity to source reliability. Journal of Accounting Research 32(Spring): 113-126. Hirst, E. and L. Koonce. 1996. Audit analyt ical procedures: A field investigation. Contemporary Accounting Research 13(Fall): 457-486. Hogarth, R. M. 1980. Judgment and Choice Chichester: John Wiley & Sons. Hogarth, R. M. and H. J. Einhorn. 1992. Or der effects in belief updating:The belief adjustment model. Cognitive Psychology 24(1): 1-55.
96 Independence Standards Board. 2000. A Conceptual Framework for Auditor Independence. Exposure Draft on St atement of Independence Concepts New York: ISB. Joyce, E. J. and G. C. Biddle. 1981. Are A uditors' Judgments Sufficiently Regressive? Journal of Accounting Research 19(2): 323-349. Kachelmeier, S. J. and W. F. Messier, Jr 1990. An Investigation of a Nonstatistical Decision Aid on Auditor Sample Size Decisions. The Accounting Review 65(1): 209-226. Kahneman, D., P. Slovic, and A. Tversky. 1982. Judgment under uncertainty:Heuristics and biases Cambridge, MA: Cambridge University Press. Kaplan, S. E., C. Moeckel, and J. D. Willia ms. 1992. Auditors' Hypothesis Plausibility Assessments in Analytical Review Setting. Auditing 11(2): 50-66. Kennedy, J. 1993. Debiasing Audit Judgment with Accountability: A Framework and Experimental Results. Journal of Accounting Research 31(2): 231-245. Kennedy, J. 1995. Debiasing the Curse of Knowledge in Audit Judgment. The Accounting Review 70(2): 249-273. Kida, T. 1984. The impact of hypothesis testing strategies on auditors' use of judgment data. Journal of Accounting Research 22(1): 332-340. King, R. R. 2002. An Experimental Investigatio n of Self-Serving Biases in an Auditing Trust Game: The Effect of Group Affiliation. The Accounting Review 77(2): 265284. Kopytoff, V. 2005. Online shoppers set to st orm the Internet Monday: Consumers use workplace's faster computers for e-commerce, San Francisco Chronicle San Francisco.
97 Kunda, Z. 1990. The Case for Motivated Reasoning. Psychological Bulletin 108(3): 480498. Lennox, C. 2004. Audit Quality and executive officers' affiliations with CPA firms. Journal of Accounting and Economics In Press, Corrected Proof. Lennox, C. 2005. Audit quality and executive o fficers' affiliations with CPA firms. Journal of Accounting and Economics 39(2): 201-231. Levine, J. M., R. L. Moreland, and C. S. Ryan. 1998. Group Socialization and Intergroup Relations. In C. Sedikides, J. Schopler, & C. A. Insko (Eds.), Intergroup Cognition and Intergroup Behavior : 283-308. Mahwah: Lawrence Erlbaum Associates, Inc. Libby, R. 1985. Availability and the Generation of Hypotheses in Analytical Review. Journal of Accounting Research 23(2): 648-667. Libby, R. 1991. Auditing: Advances in Behavioral Research New York: SpringerVerlag. Libby, R. and D. M. Frederick. 1990. Experience and the Ability to Explain Audit Findings. Journal of Accounting Research 28(2): 348-367. Libby, R. and P. A. Libby. 1989. Expert Measurement and Mechanical Combination in Control Reliance Decisions. The Accounting Review 64(4): 729-747. Lim, K. H., I. Benbasat, and L. M. War d. 2000. The Role of Multimedia in Changing First Impression Bias. Information Systems Research 11(2): 115-136. Mendenhall, W. and T. Sincich. 1996. A Second Course in Statistics Regression Analysis (FIfth ed.). Upper Saddl e River: Prentice Hall. Menon, K. and D. D. Williams. 2004. Former Audit Partners a nd Abnormal Accruals. The Accounting Review 79(4): 1095-1118.
98 Moeckel, C. L. and R. D. Plumlee. 1989. A uditors' Confidence in Recognition of Audit Evidence. The Accounting Review 64(4): 653-666. Napier, B. J. and G. R. Ferri s. 1993. Distance in Organizations. Human Resources Management Review 3(4): 321-357. O'Donnell, E., V. Arnold, and S. G. Sutton. 2000. An Analysis of the Group Dynamics Surrounding Internal Contro l Assessment in Inform ation Systems Audit and Assurance Domains. Journal of Information Systems 14(Supplement): 97-116. Oakes, P. J., S. A. Haslam, and J. C. Turner. 1994. Stereotyping and Social Reality Oxford: Blackwell Publishers. Pedhazur, E. J. and L. P. Schmelkin. 1991. Measurement Design and Analysis: An Integrated Approach (819 ed.). Hillsdale: Lawerence Erlbaum Associates Inc. Peecher, M. 1996. The Influence of Auditors' Justification Processe s on Their Decisions: A Cognitive Model and Experimental Evidence. Journal of Accounting Research 34(1): 125-140. Peecher, M. E. and I. Solomon. 2001. Theory and Experimentation in Studies of Audit Judgments and Decisions: Avoi ding Common Research Traps. International Journal of Auditing 5: 193-203. Rose, A. M. and J. M. Rose. 2003. The Effect s of Fraud Risk Assessments and a Risk Analysis Decision Aid on Auditors' Evaluation of Evidence and Judgment. Accounting Forum 27(September): 312-338. Rose, J. M. 2002. Behavioral Decision Aid Re search: Decision Aid Use and Effects. In V. Arnold, & S. Sutton (Eds.), Researching Accounting as an Information Systems Discipline : 111-133. Sarasota: American Accounting Association.
99 Roy, M. C. and F. J. Lerch. 1996. Overcoming Ineffective Mental Representations in Base-rate Problems. Information Systems Research 7(2): 233-247. Sarbanes-Oxley Act (SOX). 2002. Public Law No. 107-204 Washington D.C.: Government Printing Office. Smith, J. F. and T. Kida. 1991. Heuristics a nd Biases: Expertise and task realism in auditing. Psychological Bulletin : 472-489. Solomon, I., J. L. Krogstad, and M. B. Romney. 1982. Auditors' Prior Probability Distributions for Account Balances. Accounting, Organizations and Society 7(1): 27-41. Srivastava, R. P. and T. J. Mock. 2004. Belief Functions in A ccounting Behavioral Research. Advances in Accounting Behavioral Research 3: 225-242. Tajfel, H. 1981. Human groups and social categories : Studies in social psychology Cambridge: Cambridge University Press. Tetlock, P. E. 1983. Accountability and th e Perseverance of First Impressions. Social Psychology Quarterly December: 285-292. Tomassini, L. A., I. Solom on, M. B. Romney, and J. L. Krogstad. 1982. Calibration of Auditors' Probabilistic Judgments: Some Empirical Evidence. Organizational Behavior and Human Performance 30(3): 391-407. Towry, K. L. 2003. Control in a Teamwork E nvironment-The Impact of Social Ties on the Effectiveness of Mutu al Monitoring Contracts. The Accounting Review 78(4): 1069-1095. Trotman, K. 1985. The Review Process and the Accuracy of A uditor Judgments. Journal of Accounting Research 23: 740-752.
100 Tubbs, R. M., J. William F. Messier, and W. R. Knechel. 1990. Recency Effects in the Auditor's Belief-Revision Process. The Accounting Review 65(2): 452-460. Tuttle, B. M. 1996. Using base rate frequency perceptions to diagnose financial statement error causes. Auditing 15(1): 104-121. Ye, L. R. and P. E. Johnson. 1995. The Impact of Explanation Facilities on User Acceptance of Expert Systems Advice. MIS Quartely 19(2): 157-172.
101 Appendix A Background Material21 and DecisionSERVE Report Client Background Continental Transport Inc. is one of North America's larges t logistics companies, with operations in the United States, Canada, Mexi co, South America, Eu rope, and Asia. Most of their revenue comes from providing tr uck, rail, ocean, and air transportation throughout the world. Continental Transport Inc. works with Fort une 500/Blue Chip companies and familyowned and start-up businesses. They devel op logistics plans and provide the people, transportation, and execution to make the plans work. Their 2,000+ motor carriers provide flatbed, temperature c ontrolled, expedited, and speci al handling services. They are publicly owned and traded on the NASDAQ. They have 27 offices and 750 employees. Continental Transport, Inc. Income Statement FYE 12/31/05, 12/31/04 (unaudited) (in thousands) 12/31/2004 12/31/2003 Actual Change Percent Change Revenue: Transportation Revenue 284,593 251,721 32,872 13.06% Cost of Transportation: Fuel and Depreciation 238,123 210,590 27,533 13.07% Repair and Maintenance 6,532 3,862 2,670 69.14% Total Cost: 244,655 214,452 30,203 14.08% Gross Profit 39,938 37,269 2,669 7.06% Total selling, general, and administrative expenses 24,470 24,203 267 1.10% Income from operations 15,468 13,066 2,402 18.38% Net interest expense 87 64 23 35.94% Income before taxes 15,555 13,130 2,425 18.47% Provision for income tax (7,196) (6,158) (1,038) 16.86% Net Income 8,359 6,972 1,387 19.89% 21 Information adapted from CH Robinson Worldwide Inc. website and Financial Statements.
102 Continental Transport, Inc. Balance Sheet FYE 12/31/05, 12/31/04 (unaudited) (In thousands) 12/31/2004 12/31/2003% Change 12/31/2004 12/31/2003 % Change Current Assets 91,393 85,333 7.10% Current Liabilities 35,850 31,468 13.93% Property, Plant and Equipment Land 15,000 15,000 0% Buildings 26,000 26,000 0% Vehicles 52,844 29,749 77.63% Total Long-term Liabilities 57,580 43,542 32.24% (Less accumulated depreciation) (46,719) (44,273) 5.52% StockholdersÂ’ Equity Net Property, Plant and Equipment 47,125 26,476 77.99% Common Stock 8,400 8,400 0% Goodwill, net of accumulated amortization 15,297 15,297 0% Additional Paid in Capital 9,668 9,668 0% Other Assets 550 480 14.58% Retained Earnings 42,867 34,508 24.22% Total StockholdersÂ’ Equit y 60,935 52,576 15.9% Total Assets 154,365 127,586 20.99% Total Liabilities and StockholdersÂ’ Equit y 154,365 127,586 20.99% DecisionSERVE Report Possible Explanations for Unexpected Increases in Repair and Maintenance Client: Continental Tr ansport, FYE 2005 Reason Information Source Related Accounts Expected Direction Increase in volume Income Statement Sales Increase Increase in labor rates Income Statement Salary Increase Repair rather than replace fixed assets Balance Sheet PP&E Either No Change or Decrease Fictitious Payments/Billings Evidence may be found through additional substantive testing.
129 About the Author Dr. Eileen Zalkin Taylor was born in Liberty, New York and raised in Tampa, Florida. She is married to Glenn Taylor and has three children, Adam, Jordan, and Isabella. She earned her Bachelor of Scie nce in Business Administration, Master of Accountancy, and Doctor of Philosophy degrees at the University of South Florida (USF) in Tampa. Before enrolling in the Ph.D. program Dr. Taylor worked for Deloitte and Touche, was a controller for the Tampa Orla ndo Pinellas Jewish Foundation, and taught as an adjunct for the USF School of Accountancy. She has received several awards and sc holarships during her time at USF. She was awarded best research paper at the Am erican Accounting Asso ciationÂ’s Accounting Information SystemÂ’s mid-year meeting, 2004. Dr. Taylor is interested in behavioral research that aids accounting and audit practice. She plans to inve stigate accounting information systems and knowledge sharing in organizations, as well as ethics in auditing firms.