The influence of biased tax research memoranda on supervisors' initial judgments in the review process.
Accounting professionals resolve client issues and recommend reporting positions for financial statements and tax returns. Staff-level accountants do much of the initial work to address these reporting issues, yet prior research indicates that their decision processes are subject to certain judgmental biases such as confirmation bias (Rich et al. 1997; Roberts 1998; Shields et al. 1995). Confirmation bias is a tendency to evaluate evidence that is consistent with the decision-maker's belief as more relevant than evidence contrary to that belief (see e.g., Lord et al. 1979; Snyder and Swann 1978)) Although confirmation bias has been demonstrated in several accounting settings, the question of whether staff-level confirmation bias influences supervisors' assessments made during the review process has received relatively little attention. Studying how supervisors are influenced by staff level confirmation bias is important for understanding the extent to which confirmation bias ultimately affects recommendations to clients. In this study, we address how the confirmation bias that underlies research memoranda written by staff-level accountants affects supervisors' judgments following the first stage of the review process. (2) We also examine whether supervisors take steps to overcome confirmation bias in the memos before reaching final conclusions.
Confirmation bias in professional decision-making contexts is problematic because it can cause professionals to be overly optimistic about the strength of a client-preferred position, which can have adverse consequences to the professional and to the client. For example, Cloyd and Spilker (1999) show that such confirmation bias can cause professionals to recommend an aggressive position that is clearly incorrect, thus exposing both the client and the professional to unanticipated risk. Because tax professionals are client advocates, confirmation bias most naturally favors the client-preferred position (Cloyd and Spilker 1999). We refer to this situation as "positive confirmation bias." In concept, however, confirmation bias could also run in the opposite direction. That is, a tax professional could selectively attend to evidence that refutes the client-preferred position while discounting evidence that supports that position. We refer to this situation as "negative confirmation bias" and, by analogy, suspect that such bias could cause tax professionals to be overly pessimistic about the strength of a client-preferred position, which could also have adverse consequences for the professional and client.
We propose that tax professionals have two primary objectives when resolving tax issues. First, the accuracy objective is to determine whether there is sufficient evidentiary support for the research conclusion, and arises from regulatory standards that penalize tax professionals and their clients when tax positions taken lack such support. Second, the advocacy objective is to marshal evidence to support tax positions that offer the best outcome for the client and arises from professional standards and economic incentives underlying the professional-client relationship. When reviewing tax research memoranda, supervisors seek to accomplish these primary objectives as well as a third objective of providing corrective feedback necessary for novice professionals to learn from their experiences. If reviewers identify and appropriately deal with confirmation bias in staff members' work, then they can (1) fulfill the accuracy objective by ensuring that conclusions are based on an unbiased analysis of pertinent information and (2) provide feedback to the staff member to ensure that the staff member will avoid similar errors in the future. However, it is not possible to fulfill both the accuracy and advocacy objectives unless the client-preferred position can be justified by an objective analysis of the evidence. When the client-preferred position cannot be so justified, there is tension between the accuracy and advocacy objectives, creating an interesting and important setting in which to examine the review process.
Cloyd and Spilker (1999) report experimental evidence of widespread positive confirmation bias by tax staff persons when the client-preferred position cannot be supported by an unbiased analysis of the pertinent facts and tax authorities. We use the same fact pattern in this study, and examine the extent to which supervisors' judgments, including probability assessments, client recommendations, and requests for follow-up work, are influenced by confirmation bias in staff-level memos. We vary the nature of the confirmation bias as (1) contrary to client preferences when the memo cites only negative tax authorities and correctly concludes that the client-preferred position is inappropriate (i.e., negative confirmation bias), (2) consistent with client preferences when the memo cites only positive tax authorities and incorrectly concludes that the client-preferred position is appropriate (i.e., positive confirmation bias), and (3) absent when the memo presents a balanced analysis of both positive and negative tax authorities, and correctly concludes that the client-preferred position is inappropriate (i.e., unbiased). (3)
Our three main results are as follows. First, supervisors find unbiased memos more persuasive than negatively biased memos when both memos conclude that the client's preferred position is not appropriate. This result suggests that when research memos do not achieve the advocacy objective, supervisors view memos that meet the accuracy objective as more persuasive than biased memos that do not. Second, supervisors find positively biased memos that reach an incorrect conclusion consistent with client preferences more persuasive than negatively biased memos that reach a correct conclusion inconsistent with client preferences. This result suggests that when the accuracy objective is not achieved due to apparent bias, supervisors find memos that offer encouragement that the advocacy objective might be met more persuasive than those that do not. Third, supervisors act to correct apparent confirmation search bias in memos by requesting more rework of staff preparing positively and negatively biased memos than of staff preparing unbiased memos.
This study makes contributions to the literature examining the information search and evaluation processes underlying professional judgment and decision making. Although prior research documents that staff-level accountants' information search and evaluation processes are biased, this study considers whether these biases influence supervisors' initial assessments and recommendations made during the review process. (4) Because accounting firms have a responsibility to ensure that engagements are properly planned and adequately supervised, findings from this study have potential implications for firms' review practices. Supervisors need to be cognizant of potential confirmation bias in memoranda that they review, and they need to have mechanisms in place to counteract the effects of potential bias before making final recommendations to clients.
The remainder of this paper is structured as follows. The next section provides background and develops the hypotheses. The third section describes the research methodology. The final section provides conclusions and suggestions for future research.
BACKGROUND AND HYPOTHESIS DEVELOPMENT
In tax contexts, professionals use their knowledge and research skills to help clients resolve tax issues. The supervising tax professional frequently gathers the relevant facts pertinent to a client's problem and passes them along to a staff tax accountant who generally does the initial information search on the issue in question. After searching a vast database for relevant tax authorities and precedents, the staff member typically writes a memo that summarizes the client's facts and tax issue, cites the relevant authority located, concludes on the client's tax issue, and explains the reasoning used to reach the conclusion. The supervisor then reviews the staff member's memo and may request that the staff person do additional work to help resolve the issue, or the supervisor may conduct her own research before making a final recommendation to the client.
We propose that tax professionals have two primary objectives when resolving tax issues: accuracy and advocacy. The accuracy objective derives from regulatory standards that penalize tax professionals and their clients when they claim overly aggressive tax positions. Taxpayers are subject to an accuracy-related penalty when their tax return positions lack "substantial authority" (IRC [section] 6662). (5) Tax professionals are subject to penalties when a tax position claimed on a return they prepare does not have a "realistic possibility" of success, defined as a "one-in-three, or greater, likelihood of being sustained on its merits" [Treas. Reg. [section] 1.6694-2(b)(1)]. Importantly, Federal tax regulations require that such likelihood judgments be based on a "reasonable and well-informed analysis" [Treas. Reg. [section] 1.6694-2(b)(1)] of all relevant tax authorities, "including the authorities contrary to" the preferred treatment [Treas. Reg. [section] 1.6662-4(d)(3)(i)]. Similarly, professional standards state that CPAs should not recommend tax positions unless they have a "good faith belief that the position has a realistic possibility of being sustained administratively or judicially on its merits if challenged" (AICPA 1988, 4). (6) Thus, the accuracy objective is to reach a well-calibrated assessment of the likelihood that the client's position could be successfully defended if or when disputed by tax authorities, and to avoid recommending positions that lack adequate evidential support.
In order to accomplish this objective, it is important to consider both positive precedents (i.e., precedents with a conclusion consistent with the client's preferred position) and negative precedents (i.e., precedents with a conclusion inconsistent with the client's preferred position). The evidence supports the client's preferred position when the client's facts are similar to the taxpayer's facts in positive precedents and dissimilar to the taxpayer's facts in negative precedents. Conversely, the evidence refutes the client's preferred position when the client's facts are dissimilar to the taxpayer's facts in positive precedents and similar to the taxpayer's facts in negative precedents. Thus, persuasive legal research requires the consideration of both positive and negative authorities (Ashley 1990). Nevertheless, prior studies show that staff-level information search and evaluation is subject to confirmation bias, and that this bias can frustrate attainment of the accuracy objective at the staff-level.
The advocacy objective arises from professional standards that require tax professionals to act as client advocates (AICPA 1988), and economic incentives implicit in the tax professional-client relationship. Prior research demonstrates that tax professionals strongly identify with this objective (Davis and Mason 2003), that greater identification with this objective is associated with more aggressive tax positions (Johnson 1993), that research reports recommending more favorable client outcomes will be given more weight than research reports recommending less favorable client outcomes (Hatfield 2001), and that focusing too strongly on the advocacy objective can increase susceptibility to positive confirmation bias that frustrates attainment of the accuracy objective (Cloyd and Spilker 1999).
Supervisors prefer to accomplish both accuracy and advocacy objectives when resolving tax issues. That is, they would like to reach accurate, defensible conclusions that are consistent with the client's preferences. Because the penalty provisions of the Internal Revenue Code allow considerable latitude in meeting the accuracy objective, both objectives can often be met in settings where the proper tax treatment is truly ambiguous (e.g., a 50/50 chance that the client-preferred treatment can be successfully defended). Frequently, however, an objective analysis of the underlying facts and tax authorities indicates that the client's preferred position is too aggressive to recommend (i.e., it has less than a one-in-three chance of being successfully defended). In this situation, the tax professional is unable to fulfill both the accuracy and advocacy objectives because an objective analysis of the client's facts and the relevant tax authority indicates that the client-preferred tax treatment does not meet the "realistic possibility" standard discussed above. The tension between the accuracy and advocacy objectives creates an interesting setting in which to examine supervisors' review assessments, and is the focus of this study. Accordingly, the hypotheses developed below anticipate a situation in which an objective application of tax authorities to the client's facts indicates that the client's preferred position does not have a "realistic possibility" of success if challenged by the IRS.
Within this setting we consider three types of research memoranda to examine how confirmation bias at the staff level influences supervisors' judgments and recommendations. The first is a negatively biased memo that cites only negative tax authorities and correctly concludes against the client-preferred position. The second is a balanced memo that cites both positive and negative tax authorities and, based on an objective analysis of these authorities, correctly concludes against the client-preferred position. The third is a positively biased memo that cites only positive tax authorities and incorrectly concludes in favor of the client-preferred position. The remainder of this section develops hypotheses about how these three types of memoranda will differentially affect supervisors' judgments and assessments during the review process.
Effects of Confirmation Bias on Persuasiveness Negative Confirmation Bias versus No Confirmation Bias (H1)
When a supervisor reviews a negatively biased memo, she cannot be sure whether the accuracy or the advocacy goal is achievable. The advocacy goal is in doubt because all the authority presented in the memo suggests the client-preferred position is inappropriate. Likewise, it is unclear whether the accuracy objective has been achieved because the memo does not cite any positive cases and does not explain why the key facts in these cases are dissimilar to the client's facts. From a legal reasoning point of view, to show that a negative conclusion is accurate, it is necessary to demonstrate how the client's facts are similar to the key facts in negative precedents and dissimilar to the key facts in positive precedents. Without cites to positive precedents, the supervisor is not able to complete the legal reasoning required to determine whether a negative conclusion is appropriate. Thus, from a legal reasoning point of view, a negatively biased memo should not be as persuasive as a memo that presents both positive and negative cases and relates them to the client's facts. Further, from a psychological perspective, because of her role as client advocate, the supervisor is motivated to reject the negative conclusion (Kunda 1990).
Psychology research suggests that when a decision maker is motivated to reject an alternative, such as the conclusion in a negatively biased memo, she will place more weight on reasons to reject the memo (e.g., the apparent negative bias) than on the reasons to he persuaded by the memo (e.g., the memo was completed by a competent staff member) (Shafir et al. 1993). In contrast, because there is no apparent bias in a balanced memo, the supervisor is not able to discount the negative evidence in this situation. Consequently, she is likely to be more persuaded by a balanced memo that cites both negative and positive precedents and reaches a negative conclusion than by a negatively biased memo that reaches a negative conclusion. This suggests our first hypothesis as follows:
H1: When both memos correctly conclude against the client-preferred position, a balanced memo is more persuasive than a negatively biased memo.
Positive Confirmation Bias versus Negative Confirmation Bias (H2)
When a supervisor reviews a memo that cites only positive cases and concludes that the client-preferred position is appropriate, she is receiving encouraging signals about the potential for accomplishing the advocacy objective. Given this signal, according to motivated reasoning theory (Kunda 1990), the supervisor is likely to search her memory for instances that may bolster the pro-client case and, despite the apparent positive bias, she may hesitate to discount the positive cases because they help her accomplish the advocacy goal. Further, because the supervisor desires to choose the pro-client alternative, she will likely place more weight on the reasons to accept the memo's conclusion (e.g., positive precedents cited in the memo) than on the reasons to reject it (e.g., the memo is positively biased) (Shafir et al. 1993). Consequently, the supervisor is likely to be persuaded by the memo in the direction of the client-preferred position, even though the memo appears to be biased because it cites only positive cases. On the other hand, as discussed above, when the memo is apparently negatively biased and reaches a negative conclusion, the supervisor is likely to focus on the reasons to reject the negative conclusion (e.g., the memo is negatively biased, possibly excluding positive authority) and is thus unlikely to be swayed much by the negative memo because it does not allow her to accomplish either the advocacy or accuracy objectives. This leads to our second hypothesis as follows:
H2: When both memos exhibit confirmation bias that frustrates achievement of the accuracy objective, a positively biased memo that incorrectly concludes in favor of the client's preferred position is more persuasive than a negatively biased memo that correctly concludes against the client's preferred position.
To summarize, H1 predicts that balanced (or unbiased) memos will be more persuasive to supervisors than negatively biased memos. Hypothesis 2 predicts that positively biased memos will be more persuasive to supervisors than will negatively biased memos. Both hypotheses deal with supervisors' judgments based solely on the evidence provided in the memos. However, during the review process, supervisors have the opportunity to correct perceived deficiencies in the information provided in the memo by requiring the staff member to perform additional work before ultimately resolving the issue. The following section considers the effects of apparent confirmation bias on supervisors' feedback to staff persons, as reflected by the amount of additional work they require the staff person to perform in order to complete the research task.
Effects of Confirmation Bias on Supervisors' Rework Expectations Negative Confirmation Bias versus No Confirmation Bias (H3)
When the supervisor reviews a negatively biased memo reaching a conclusion inconsistent with the client's preference, she is unsure whether the memo achieves the accuracy goal because it ignores positive precedents. In this situation, the supervisor is likely to require the staff member to do additional research to locate positive precedents so she can gain better insight into whether the accuracy and/or advocacy objectives can be accomplished. In contrast, supervisors reviewing an unbiased memo that reaches a conclusion inconsistent with the client's preference should not require as much rework because the staff member has provided persuasive evidence that the accuracy objective has been met and that the advocacy objective cannot be met. This leads to our third hypothesis as follows:
H3: When both memos reach conclusions inconsistent with client preferences, supervisors will require more rework when reviewing negatively biased memos than when reviewing unbiased memos.
Positive Confirmation Bias versus No Confirmation Bias (H4)
Although tax professionals attempt to accomplish the advocacy goal during the research process, they are not free to abandon accuracy goals due to professional standards and penalty provisions in the tax code. Thus, even if supervisors are persuaded by positively biased memoranda (H2), they may require additional research by the staff member to search for negative precedents that might have been overlooked in the initial information search process. By requesting this additional information, the supervisor is able to cover her bases on the accuracy goal. In contrast, as discussed previously, when the supervisor reviews a balanced memo, this additional rework is not necessary because she will have greater confidence in the accuracy of the conclusion (H1) even when the conclusion is inconsistent with the client's preferences. This leads to our final hypothesis as follows:
H4: Supervisors reviewing positively biased memos reaching conclusions consistent with client preferences will require more rework than will supervisors reviewing unbiased memos reaching conclusions inconsistent with client preferences.
Participants and Procedures
Participants in this study were recruited from among tax professionals attending either the AICPA's National Tax Education Program or a continuing education program of an international accounting firm. Although 75 tax professionals participated in this study, 13 are excluded from the analyses because they reported no experience in supervising research projects. Due to the nature of our research questions, it is important that participants have at least some supervisory experience. Five additional participants are excluded because their responses to one or more of the measures underlying the dependent variables were either missing or unintelligible. Finally, two participants are excluded due to their failure to correctly answer a manipulation check question (discussed below). Therefore, the final sample on which all analyses are based consists of 55 tax professionals with supervisory experience. (7) All professionals were compensated for their participation in the study. (8) All phases of the experiment were administered in written form. The sequence of experimental procedures is shown in Figure 1 and discussed below.
In the background phase of the study, participants reviewed a set of facts describing a transaction entered into by the client and disclosing the client's preferred tax position. The client fact pattern was the same as that used in Cloyd and Spilker's (1999) study 2. Specifically, the client had initially purchased undeveloped land for investment purposes, but later developed and subdivided the land into residential lots that he held out for sale. Moreover, consistent with the status of a dealer in real estate, the client made a second purchase of adjacent property that he also developed and subdivided into residential lots prior to the sale in question. Although the client preferred investor treatment (i.e., long-term capital gain) in all three experimental conditions, such tax treatment was insupportable given his fact pattern. Therefore, according to a panel of four experts, the client fact pattern used in this study unambiguously required dealer treatment, resulting in ordinary income. (9) Consequently, the fact pattern presents a conflict between the tax supervisor's goals of accuracy and advocacy.
After reading the client facts, participants reviewed a summary of the factors courts have used to distinguish between investor and dealer treatment in similar transactions. This authoritative background was intended to refresh participants' memories and ensure that they understood the relevant factors in the experimental case. Based on the client facts and authoritative background, we asked participants to provide a preliminary assessment of the probability that the courts would conclude the client is an investor with respect to his real estate sales. The response scale for this assessment ranged from 0.00 = Certain to be considered a Dealer to 1.00 = Certain to be considered an Investor. Participants were also asked for a preliminary recommendation regarding the tax treatment of the client's transaction. The response scale for this question ranged from 1 = Strongly recommend Dealer (ordinary income) to 7 = Strongly recommend Investor (long-term capital gain). Finally, participants provided demographic and work experience information before beginning the review phase of the experiment.
In the review phase, participants reviewed a research memorandum prepared by a hypothetical tax staff person. The research memorandum contained the experimental manipulation. Each participant received only one of three versions of the research memorandum that differed with regard to the staff's recommendation (investor or dealer) and the set of court cases cited. This manipulation is described in detail below.
After participants reviewed the research memorandum, we asked them to complete the initial round of the review process by making judgments and decisions about the staff-level work exhibited therein. First, they provided a confidence assessment on the correctness of the conclusion reached in the memorandum. Second, they were asked how much additional work they would require the staff person to conduct, and to explain the reasons for this rework, listing specific points they would have the staff person address. Finally, they provided a final probability assessment and final recommendation regarding the tax treatment of the client's transaction. The scales used for these latter two responses were the same as for the preliminary probability assessment and preliminary recommendation provided in the background phase of the experiment.
On concluding the review phase, participants completed the post-experimental questionnaire. These questions included two manipulation checks, discussed in the next section, and an assessment of the perceived realism of the experimental case. The average rating of task realism is 5.50 on a scale ranging from 1 = Very Unrealistic to 7 = Very Realistic. Perceived task realism did not vary significantly across experimental conditions ([F.sub.(2,52)] = 1.49, p = .234). The final set of questions assessed participants' client advocacy attitudes using a measure created by Davis and Mason (2003) and further tested by Mason and Levy (2001).
We randomly assigned participants to one of the three experimental conditions in which they reviewed one of three versions of a staff-prepared research memorandum. Immediately prior to the review task, we told participants that the staff person conducted the research task by searching an electronic tax service using a specific string of keywords that produced a hit list of 37 court cases addressing investor versus dealer issues related to real estate sales. Of the 37 cases, the courts concluded that the taxpayer was a dealer in 21 cases and an investor in 16 cases. (10) We also told participants that the staff person drew upon cases from this list to generate the memorandum they would review. The three memorandums are described below. (10)
* Negatively Biased (NEG_BIAS). The memo correctly concludes in favor of dealer treatment, cites six court cases in which taxpayers were treated as dealers, but cites no investor-outcome cases.
* Unbiased (UNBIASED). The memo correctly concludes in favor of dealer treatment, but cites three investor-outcome and three dealer-outcome court cases. The facts of investor-outcome cases are distinguished from those of the client, while similarities between the facts of dealer-outcome cases and the client's facts are highlighted.
* Positively Biased (POS_BIAS). The memo erroneously concludes in favor of investor treatment, cites six court cases in which taxpayers were treated as investors, but cites no dealer-outcome cases. (12)
Although the cases cited and the language linking these cases to the client facts necessarily differed across conditions, the memorandums were very similar in terms of length. Each contained between 650 and 665 words.
As noted earlier, the post-experimental phase included two questions to ensure that participants attended to the experimental manipulation. First, we asked participants to indicate the conclusion reached in the staff person's memorandum. All but three participants answered this question correctly. However, other responses provided by one of these participants (e.g., review points written about the staff person's memo) indicated an understanding of the memo's conclusion. Therefore, we excluded only two participants due to their failure to correctly identify the conclusion reached by the staff person.
The second manipulation check asked participants to rate the extent to which the staff person's analysis of the existing judicial authority was biased against or in favor of the client. The response scale ranged from 1 = Biased Against the Client to 7 = Biased in Favor of the Client. The average response to this question was 5.95 in the POS_BIAS condition, 1.89 in the NEG_BIAS condition, and 2.67 in the UNBIASED condition. This pattern of cell means is consistent with the manipulation, and each cell mean is significantly different (p < .01) from the other two means. We conclude, therefore, that our manipulation of the extent of confirmation bias underlying the staff-level memos in each condition was successful. Although we present this result primarily as a manipulation check, it is worth noting that this is the only evidence we are aware of that supervisors recognize confirmation bias underlying the research memoranda they review. (13)
Information on Participants
Table 1 summarizes five demographic variables for the final sample of 55 participants by experimental condition. Participants had an overall average of 51.16 months of tax experience, although the range of their experience was quite broad (i.e., from 13 to 144 months). Participants rated their own experience in supervising research projects on a scale ranging from 1 = No Experience to 7 = Extensive Experience. The average rating of research supervision experience is 3.58. In terms of industry-specific experience, participants indicated that, on average, they devoted 8.93 percent of their chargeable hours to real estate clients or real estate transactions. Participants' average self-evaluation of their issue-specific experience is 2.24, on a scale ranging from 1 = No Experience to 7 = Extensive Experience. Our final demographic variable is participants' aggregate scores on the nine-item measure of taxpayer advocacy attitudes. The average advocacy score is 46.67 compared to a theoretical maximum score of 63.
We performed a one-way ANOVA on each demographic variable using the experimental condition as a single three-level variable. F-statistics from these ANOVAs, shown in the far fight column of Table 1, indicate that none of the demographic measures is significantly different across the three experimental conditions. These results are consistent with random assignment of participants to the experimental conditions. (14)
We use two dependent measures to test hypotheses relating to persuasiveness (HI and H2). SHIFTPROB is the difference between participants' final probability assessment of judicial success for the client's preferred position, after exposure to the experimental manipulation, and their preliminary assessment of this same probability. A positive (negative) value of this variable indicates an increase (decrease) in the belief that the investor position would be judicially supported. Similarly, SHIFTREC is the difference between participants' final and preliminary recommendation regarding the tax treatment of the client's transaction. As with SHIFTPROB, a positive (negative) value of SHIFTREC indicates a shift toward investor (dealer) treatment. Table 2, Panel A summarizes participants' responses to the persuasiveness dependent measures by experimental condition. Because the directional predictions for the persuasiveness of POS_BIAS and NEG_BIAS memos are in opposite directions, we use the absolute value of the dependent variables SHIFTPROB and SHIFTREC for testing H2.
The dependent measure for the rework hypotheses (H3 and H4) is participants' response to the question "How much additional work would you require the staff person to conduct before completing this assignment?" Responses were given on a seven-point scale with a 1 = No rework and 7 = Extensive rework. Table 2, Panel B, reports participants' responses to the REWORK dependent measure by experimental condition. In addition to indicating the amount of rework necessary to complete the assignment, participants explained the reasons for their responses and indicated the major points they would have the staff person work on to complete the research.
Hypothesis 1 predicts that a negatively biased memo correctly concluding that the client-preferred position is inappropriate is less persuasive than an unbiased memo reaching the same conclusion. Support for this hypothesis is provided if the SHIFTPROB and SHIFTREC scores are more negative (i.e., greater shifts toward dealer treatment) in the UNBIASED condition than in the NEG_BIAS condition. Consistent with this prediction, Table 3 indicates that the mean value of SHIFTPROB is significantly more negative (t = 1.66, p < .05 one-tailed) in the UNBIASED condition (-.0772) than in the NEG_BIAS condition (-.0228). The mean value of SHIFTREC is also significantly more negative (t = 2.14, p < .02 one-tailed) in the UNBIASED condition (-0.83) than in the NEGA_BIAS condition (0.00). These results support H1.
Hypothesis 2 predicts that a positively biased memo incorrectly reaching a conclusion consistent with client preferences is more persuasive than a negatively biased memo correctly concluding that the client-preferred position is not appropriate. To test this hypothesis we compare the average absolute values of SHIFTPROB and SHIFTREC in the POS_BIAS condition to the average absolute values of these variables scores in the NEGA_BIAS condition. (15) As shown in Table 2, the mean absolute value of SHIFTPROB in the POS_BIAS condition (.0779) is greater than the mean absolute value of SHIFTPROB in the NEG_BIAS condition (.0239). This difference is in the predicted direction and is significant (t = 1.90, p < .03 one-tailed). Also, as presented in Table 2, the mean absolute value of SHIFTREC in the POS_BIAS condition (.79) is greater than the mean absolute value of the SHIFTREC in the NEG_BIAS condition (.44). While this difference is also in the predicted direction, it is not significant (t = 1.01, p < .16 one-tailed). Thus, H2 is supported when persuasiveness is measured by SHIFTPROB, but not when this construct is measured by SHIFTREC.
Interestingly, participants in the POS_BIAS condition increased their probability assessments of a client-preferred judicial outcome by almost eight percentage points (i.e., SHIFTPROB = .0779) after reviewing a staff-level memorandum concluding in favor of the investor treatment and citing only investor-outcome cases. (16) This increase is significantly greater than zero (t = 3.45, p < .01 one-tailed). Similarly, the average SHIFTREC score in the POS_BIAS condition shifted by an average of 0.58 units (from 3.31 to 3.89 on a 1-to-7 scale) in favor of the client-preferred investor treatment after reading the staff-level memo presented in the POS_BIAS condition. This increase is also significantly greater than zero (t = 1.93, p < .04 one-tailed). These results indicate that staff-level research memoranda biased in favor of a client-preferred but incorrect tax position influence supervisors' judgments away from the correct decision. These results are particularly compelling in light of the fact that, based on the manipulation check described earlier, participants recognized the fact that the memorandum they reviewed was biased in favor of the client-preferred position. Yet despite their recognition of this positive bias, their beliefs and recommendations still shifted in favor of the client-preferred position. In contrast, the mean values of SHIFTPROB and SHIFTREC are not significantly different from zero in the NEG_BIAS condition, indicating that the negatively biased memo had no significant effect on participants' judgments and recommendations.
Persuasiveness Descriptive Results
In addition to the comparisons required to test H1 and H2, we also compare the persuasiveness of the POS_BIAS memo to the persuasiveness of the UNBIASED memo for descriptive insights. As described above, the POS_BIAS memo is expected to be persuasive because it is consistent with the supervisor's advocacy objective. The UNBIASED memo is expected to be persuasive because it is consistent with the supervisor's accuracy objective. Because we have no basis for predicting which memo will be more persuasive, we present the results descriptively. The mean absolute value of SHIFTPROB is not significantly different (t = 0.17, p < .87) between the POS_BIAS (.0779) and UNBIASED (.0839) conditions. Similarly, the mean absolute value of SHIFTREC is not significantly different (t = 0.11, p < .92) between the POS_BIAS (.79) and UNBIASED (.83) conditions. However, when comparing the means of the actual (rather than absolute) values of SHIFTPROB and SHIFTREC across these conditions, it is clear that the POS_BIAS memo persuaded participants toward the incorrect but client-preferred position, whereas the UNBIASED memo persuaded participants toward the correct conclusion that the client-preferred position is not appropriate.
Hypotheses 3 and 4 concern the amount of additional work or "rework" supervisors require staff members to perform in order to complete the research assignment. As part of the data analysis we also consider the nature of the review comments made to the staff member. Hypothesis 3 predicts that supervisors reviewing negatively biased memos reaching conclusions inconsistent with client preferences will require a greater amount of rework than will supervisors reviewing unbiased memos reaching the same conclusion. Consistent with H3, the mean of REWORK is greater in the NEG_BIAS condition (3.94) than in the UNBIASED condition (3.28), and this difference is marginally significant (t = 1.35, p < .09 one-tailed). (17) In terms of the nature of the review comments, a greater proportion (Chi-squared = 5.46, p < .02) of supervisors in the NEG_BIAS condition (12 of 18 participants) wrote review comments requesting the staff person to search for more positive-outcome cases than in the UNBIASED condition (5 of 18 participants). (18) The amount and nature of rework requested by the supervisor provide evidence that when the supervisor reviews memos that reach correct conclusions inconsistent with client preferences and thus fail to achieve the advocacy goal, the supervisor places an emphasis on accomplishing the accuracy objective. The rework requested by the supervisor should help to mitigate negative confirmation bias by requiring staff to search for and analyze both positive and negative cases before reaching a final conclusion.
Hypothesis 4 predicts that supervisors reviewing a positively biased memo incorrectly concluding in the direction of client preferences will request more rework of the staff member than will supervisors reviewing an unbiased memo correctly concluding against the client preferences. Consistent with H4, the mean REWORK scores were significantly higher (t = 3.17, p < .01) for supervisors in the POS_BIAS condition (4.53) than supervisors in the UNBIASED condition (3.28), suggesting that supervisors who reviewed the positively biased memorandum believed that the staff person should conduct more research. Also, a larger proportion (Chi-squared = 12.59, p < .01) of supervisors in the POS_BIAS condition (13 of 19 participants) than in the UNBIASED condition (2 of 18 participants) wrote review comments directing the staff person to locate and evaluate cases with negative outcomes. Thus, even when memoranda are biased in favor of the client-preferred position, supervisors' review comments may help to mitigate confirmation bias. However, this interpretation should not be widely generalized because, for reasons described earlier, the client-preferred position was clearly incorrect given the extreme client fact pattern used in this study. Further research is necessary to determine the extent to which supervisors' review comments might mitigate pro-client confirmation bias across a wider range of factual ambiguity.
DISCUSSION AND CONCLUSIONS
Tax professionals frequently provide their clients with recommendations and information regarding risks associated with alternative tax positions. When making these risk assessments and recommendations, supervising tax professionals must often rely on research conducted by staff members to resolve issues. Prior research indicates that staff-level search bias may lead to incorrect risk assessments and client recommendations (Cloyd and Spilker 1999; Hatfield 2000).
However, prior research does not consider the relative persuasiveness of positively biased, negatively biased, and unbiased staff-level research memoranda. Further, prior research does not consider how supervisors act to correct confirmation bias apparent in staff-level research memoranda. This paper contributes to the literature by providing empirical insights on these issues
Due to professional standards and environmental incentives, tax professionals are motivated to achieve both advocacy and accuracy goals when resolving client issues through the tax research process. While supervisors would prefer to accomplish both objectives during the research process, this is not possible when the client's preferred position does not have a "realistic possibility" of success if challenged by the IRS. We use such a scenario in this study, to create tension between the advocacy and accuracy objectives.
We predict and find that supervisors are more persuaded by an unbiased memo than by a negatively biased memo when both memos conclude that the client's preferred position is not appropriate. We also predict and find some evidence that supervisors are more persuaded by a positively biased memo that reaches an incorrect conclusion consistent with client preferences than by a negatively biased memo reaching a correct conclusion inconsistent with client preferences. The first result suggests that when the memos do not achieve the advocacy objective, supervisors find memos that support the accuracy objective as more persuasive than biased memos that do not. The second result suggests that when memoranda do not achieve the accuracy objective due to apparent bias, supervisors are more persuaded by those that offer encouragement of fulfilling the advocacy objective than by those that do not.
With regard to supervisors' review comments, we predict and find that supervisors reviewing biased memos require more rework of the staff member than supervisors reviewing unbiased memos. This result holds for both positively biased and negatively biased memos, suggesting that supervisors take steps to overcome apparent confirmation bias in memos whether they find the memos persuasive or not. We also report an exploratory analysis of supervisors' review comments, which indicates that they are directed toward mitigating confirmation bias, regardless of whether it favors the client-preferred position. However, future research is needed to address this issue more completely.
Our results both complement and extend the results of Hatfield (2001), who finds that supervisors give less weight to research reports believed to contain bias. Consistent with this result, we find that negatively biased memos are less persuasive than balanced, unbiased memos, when both memos correctly conclude against the client-preferred position. However, contrary to his prediction, Hatfield (2001) finds that client-favorable confirming reports are discounted just as much as client-unfavorable confirming reports. In contrast, we find that positively biased memos are more persuasive then negatively biased memos even though the client-preferred position is incorrect, which may lead the supervisor to make an incorrect recommendation. Finally, this study extends the Hatfield (2001) study by considering the extent to which supervisors request additional rework from staff members providing biased and unbiased research reports.
Although our results are likely to generalize to other tax and accounting settings that are influenced by client advocacy, discretion should be used because the results presented in this paper are limited to the tax setting and task examined in the reported experiment. Several limitations of the experiment point to suggestions for future research. First, while we gave supervisors a base rate for the investor/dealer cases and the factors used by the courts to distinguish between investor/dealer situations, we did not allow them to perform external information searches during the experiment. Future research could examine how staff-level confirmation bias affects supervisors' information search behavior, which would provide additional evidence regarding whether staff-level confirmation bias is mitigated or perpetuated by the review process. Second, this study had supervisors review memoranda ostensibly prepared by staff-level accountants. Future research could observe actual pairings of supervisors with subordinates to see if other forms of communication also affect supervisors' judgments and decisions with regard to confirmation bias in a more interactive setting. Third, this study examined only the initial stages of the review process. Future research should address issues relating to later iterations of the review process.
Description of Experimental Procedures in the Sequence Performed
A. Background Phase
1) General instructions to participants
2) Review client facts indicating investor preference
3) Review authoritative background summarizing investor/dealer factors
4) Preliminary likelihood assessment of the courts concluding the client is an investor
5) Preliminary recommendation regarding the client's tax position
6) Provide demographic information
B. Review Phase
1) Introduction to review task
2) Review one of three research memorandums (Experimental Manipulation)
3) Confidence rating
4) Rating of additional work required
5) Explanation of review points
6) Final likelihood assessment of the courts concluding the client is an investor
7) Final recommendation regarding the client's tax position
C. Post-Experimental Phase
1) Manipulation checks (Client Preference: investor versus dealer; Rating of staff bias)
2) Assessment of task realism
3) Advocacy instrument
TABLE 1 Demographic Information by Experimental Condition Mean (Standard Deviation) Negatively Unbiased/ Positively Overall Biased Balanced Biased (n = 55) (n = 18) (n = 18) (n = 19) Months (b) 51.16 54.00 45.44 53.89 (29.41) (32.37) (18.28) (35.23) Research 3.58 3.39 3.39 3.95 Supervision (c) (1.52) (1.46) (1.42) (1.68) Real Estate 8.93 12.00 8.78 6.16 Experience (d) (11.13) (11.20) (10.80) (11.21) Issue-Specific 2.24 2.56 2.33 1.84 Experience (e) (1.41) (1.72) (1.41) (1.01) Advocacy (f) 46.67 46.83 48.28 45.00 (7.43) (7.56) (7.57) (7.75) Mean (Standard Deviation) F-statistic (a) (p-value) Months (b) 0.50 (0.61) Research 0.83 Supervision (c) (0.44) Real Estate 1.29 Experience (d) (0.28) Issue-Specific 1.25 Experience (e) (0.29) Advocacy (f) 0.91 (0.41) (a) The F-statistic, based on a one-way ANOVA in which experimental condition is treated as a single factor with three levels, tests the hypothesis that at least one of the cell means is significantly different from the one of the other cell means. (b) Months of work experience as a tax professional. (c) Self-rating of experience supervising research projects (1 = No experience; 7 = Extensive experience). (d) Percentage of chargeable hours spent working for clients in the real estate industry or on real estate transactions. (e) Self-rating of experience dealing with investor versus dealer issues (1 = No experience; 7 = Extensive experience). (f) Aggregate score on nine-item measure of taxpayer advocacy developed by Davis and Mason (2003) and further tested by Mason and Levy (2001). This scale has a theoretical range from 9 to 63. TABLE 2 Descriptive Statistics on Dependent Measures by Condition Negatively Unbiased/ Positively Biased Balanced Biased (NEG_BIAS) (UNBIASED) (POS BIAS) (n = 18) (n = 18) (n = 19) Panel A: Dependent Measures for H1 and H2 Probability of Investor Ruling by Courts (a) Preliminary (before .2744 .3244 .3326 viewing memo) Final (after viewing .2517 .2472 .4105 memo) SHIFTPROB (b) Mean -.0228 -.0772 .0779 Standard Deviation (.0720) (.1191) (.0984) Recommendation (c) Preliminary 2.72 3.50 3.31 Final 2.72 2.67 3.89 SHIFTREC (d) Mean 0.00 -0.83 0.58 Standard Deviation (0.97) (1.34) (1.30) Absolute Values (e) SHIFTPROB Mean .0239 .0839 .0779 Standard Deviation (.0716) (.1142) (.0984) SHIFTREC Mean 0.44 0.83 0.79 Standard Deviation (0.86) (1.34) (1.18) Panel B: Dependent Measures for H3 and H4 REWORK (f) Mean 3.94 3.28 4.53 Standard Deviation (1.66) (1.27) (1.12) (a) The response scale for each probability measure ranged from 0.00 = Certain to be considered a Dealer to 1.00 = Certain to be considered an Investor. (b) SHIFTPROB is the final probability estimate minus the preliminary probability estimate. A positive (negative) value indicates a shift in favor of investor (dealer) treatment. (c) The response scale for each recommendation measure ranged from 1 = Strongly Recommend Dealer (ordinary income) to 7 = Strongly Recommend Investor (long-term capital gain). (d) SHIFTREC is the final recommendation minus the preliminary recommendation. A positive (negative) value indicates a shift toward investor (dealer) treatment. (e) The absolute values of SHIFTPROB and SHIFTREC are used to test H2. (f) REWORK is the amount of additional work the reviewer would require the staff person to conduct before completing the assignment. The response scale ranged from 1 = No Rework to 7 = Extensive Rework. TABLE 3 Summary of Hypothesis Tests Hypothesis Test (a) Test Statistic (b) p-value (c) H1 [SHIFTPROB.sub.UNBIASED] > t = 1.66 .05 [SHIFTPROB.sub.NEG_BIAS] (-0.0772 > -0.0228) H1 [SHIFTREC.sub.UNBIASED] > t = 2.14 .02 [SHIFTREC.sub.NEG_BIAS] (-0.83 > 0.00) H2 [SHIFTPROB.sub.POS_BIAS] > t = 1.90 .03 [SHIFTPROB.sub.NEG_BIAS] (.0779 > .0239) (d) H2 [SHIFTREC.sub.POS_BIAS] > t = 1.01 .16 [SHIFTPROB.sub.NEG_BIAS] (0.79 > 0.44) (d) H3 [REWORK.sub.NEG_BIAS] > t = 1.35 .09 [REWORK.sub.UNBIASED] (3.94 > 3.28) H4 [REWORK.sub.POS_BIAS] > t = 3.17 .01 [REWORK.sub.UNBIASED] (4.53 > 3.28) (a) See Table 2 for variable definitions. The hypothesis tests compare selected cell means as reported in Table 2. The subscripts on each variable name refer to the experimental condition as follows: UNBIASED, NEG_BIAS, and POS_BIAS. (b) The t-statistics for both tests of H1 have 35 degrees of freedom. The t-statistics for both tests of H2 have 18 degrees of freedom. (c) All p-values are one-tailed. (d) The absolute values of the SHIFTPROB and SHIFTREC scores are compared for testing H2 because the memos are expected to shift participants' responses in opposite directions. The POS_BIAS memo is expected to shift participants' beliefs toward the client's preferred position, while the NEG_BIAS memo is expected to shift their beliefs away from the client's preferred position.
The authors thank Jean C. Bedard, Andy Cuccia, Diana Falsetta, Marlys Lipe, Arnie Wright, two anonymous reviewers, and workshop participants at the 2003 AAA Annual Meeting, University of Oklahoma, University of Nebraska, and NEBARS for their helpful comments. We also thank Bill Kulsrud, Gene Willis, and Freda Mulhall for arranging access to the participants at the AICPA's National Tax Education Program.
(1) Prior research provides evidence of confirmation bias in various accounting decision making contexts including taxation (e.g., Cloyd and Spilker 1999; Cuccia et. al 1995; Johnson 1993) and auditing (Bedard and Biggs 1991; Church 1991; Hackenbrack and Nelson 1996; McMillian and White 1993) and other business-related contexts including marketing (Russo et al. 1996) and managerial decision making (Boiney et al. 1997).
(2) Hatfield (2001) examines whether staff-level confirmation bias, which a supervisor might infer from the consistency between the staff person's initial opinion and research conclusion, affects the weight the supervisor gives the research report. Some of the research questions we examine are similar to those addressed by Hatfield (2001). However, there are several methodological differences between the studies. For example, a primary difference between these studies is the ambiguity of the correct conclusion to the tax issue imbedded in the experimental materials. Hatfield (2001) examines professionals' responses to an ambiguous tax issue while we examine professionals' responses to an unambiguous tax issue for which the client-preferred position is insupportable. As a consequence, results from Hatfield (2001) may not generalize to the setting we consider. Along these lines, because the correct outcome of the issue used in the Hatfield (2001) study is ambiguous, it is not clear that confirmation bias will necessarily lead to an incorrect recommendation, and could even be within the scope of client advocacy encouraged by AICPA standards.
(3) Because our experiment involves only the initial round of supervisors' review comments, the extent to which a staff person's subsequent research corrects initial confirmation bias is beyond the scope of this study.
(4) Auditing research reports that (1) the review process reduces but does not eliminate judgmental errors (Ramsay 1994; Trotman 1985), (2) familiarity with the preparer results in more effective reviews (Asare and McDaniel 1996), (3) real-time reviews lead to better communication and fewer errors than do sequential reviews (Messier and Tubbs 1994), (4) different levels of reviewer are more effective in eliminating different types of errors (Ramsay 1994), (5) evidence included in working papers affects superiors' decisions (Ricchiute 1999), (6) superiors are overly optimistic in assessing the knowledge of their staff and may detrimentally rely on the incorrect judgments of their staff (Kennedy and Peecher 1997), and (7) reviewers recall information inconsistent with the conclusion while preparers recall information consistent with the conclusion (Libby and Trotman 1993),
(5) The taxpayer penalty for positions that lack "substantial authority" is 20 percent of the amount by which taxable income is understated (IRC [section] 6662). This penalty may be avoided if the tax position and facts underlying the position are explicitly disclosed in the tax return. Although the regulations do not provide a numeric definition of "substantial authority," they state that this standard is "less stringent than the more likely than not standard" (i.e., 50 percent likelihood of the position being upheld), but "more stringent than the reasonable basis standard" (Treas. Reg. [section] 1.6694-4(d)(2)).
(6) The "realistic possibility" standard used by the AICPA is similar to that defined in Treas, Reg. [section] 1.6694-2(b)(1) in that it is more stringent than the "reasonable basis" standard (used prior to the Revenue Reconciliation Act of 1989) and less stringent than the "substantial authority" standard of IRC [section] 6662. However. unlike the Treasury Regulations, the AICPA's Interpretation of Statement No. 1 explicitly states that this standard "cannot be expressed in terms of percentage odds."
(7) Both participant groups, AICPA and Big 5, were randomly assigned to and equally dispersed across experimental conditions.
(8) The AICPA participants received approximately $15 each as compensation. In order to encourage participation in the study, these participants were given a choice between receiving $15 with certainty or $10 with certainty plus a lottery ticket. The lottery pool was $100 plus $5 for each subject who chose the lottery option. The lottery pool was split (1/6, 1/3, and 1/2) among holders of the first three tickets drawn. Participants from the Big 5 accounting firm received a fiat fee of $20. Because compensation was based on participation rather than performance in the study, we see no reason why differing compensation schemes should affect our results. We unable to verify whether reported results for our hypothesis tests are influenced by the compensation scheme because cell sizes are too small to derive meaningful results from a full factorial design where compensation scheme is treated as a second factor.
(9) As reported in Cloyd and Spilker (1999, 315), a panel of four experts considered the fact pattern and determined that the client-preferred position had less than a 13.75 percent chance, on average, of being sustained by the courts (range of 10 percent to 20 percent). This probability is substantially below the one-in-three reasonable possibility standard necessary to avoid preparer penalties ([section] 1.6694-2(b)(1)).
(10) The total number of cases, as well as the mix of dealer-outcome and investor-outcome cases, is based on an actual keyword search of the CCH Tax Research Network using the following search string "real estate development 'and' ordinary income 'or' capital gain."
(11) We did not include an experimental condition in which the memo concluded in favor of investor treatment but cited both investor-outcome and dealer-outcome court cases. Because the client fact pattern so strongly suggests that dealer treatment is appropriate, the inclusion of dealer-outcome cases in such a memorandum would have made it very difficult to conclude in favor of investor treatment.
(12) Although investor treatment is inappropriate given the client facts used in this study, it should be noted that, after conducting information searches that were heavily skewed in favor of investor-outcome cases, 46 percent of the participants in Cloyd and Spilker's (1999, 315) study 2 recommended investor treatment for the same set of client facts.
(13) Hatfield (2001) finds that supervisors consider the staff member's conclusion to be less objective when it is consistent with their pre-search opinion. However, professionals in that study made this assessment with knowledge of only the pre- and post-search opinions of the staff member, but did not have access to the reasoning underlying the pre- and post-search opinions.
(14) On average, participants from the Big 5 firm rated their research supervision experience higher than AICPA participants (4.17 versus 3.30, p < .05), We expected this difference given the different training programs these participants were attending. On average, participants from the Big 5 firm had less real estate experience (3.67 versus 11.49, p < .01) and less issue-specific experience (1.44 versus 2.62, p < .01) than AICPA participants. These differences simply reflect the nature of the tax practices in which participants were employed.
None of the bivariate correlations between these demographic variables and the dependent measures are statistically significant. In sensitivity tests, we included each demographic variable as a covariate in ANCOVA models used to test our hypotheses. In only one instance does a demographic variable have a significant effect on the dependent measure, as described in footnote 17.
(15) The absolute values of the SHIFTPROB and SHIFTREC scores are compared because the memos are expected to shift participants' responses in opposite directions. The POS_BIAS memo is expected to shift participants' beliefs toward the client's preferred position, while the NEG_BIAS memo is expected to shift their beliefs away from the client's preferred position.
(16) An absolute shift in probability assessment of eight percentage points represents a meaningful influence on supervisors' beliefs for two reasons. First, the average shift is 23.4 percent of the average initial probability assessment, which represents a substantial shift from participants' initial anchors. Second, even a small shift in a probability assessment can have an important impact on the tax professionals' final recommendations if that shift causes the probability assessment to cross one of several important thresholds established by the Internal Revenue Code, Treasury Regulations, and Circular 230. For example, the Internal Revenue Code imposes a preparer penalty of $250 on professionals who fail to disclose a position taken on a return that does not meet the "realistic possibility test," which the Treasury Regulations ([section] 1.6694-2(b)(1)) define as one-in-three chance (33 percent) of judicial success if the issue went to court. The Internal Revenue Service (Circular 230 Paragraph 10.34) also imposes a stiffer penalty on professionals failing to meet the "realistic possibility standard" (33 percent)--disbarment from practicing before the Internal Revenue Service. Finally, there are several additional standards (e.g., more likely than not [50 percent], substantial authority [between 33 and 50 percent], and reasonable basis [20 percent]) relating to underpayment penalties that are defined in the Treasury Regulations ([section] 1.6692-4(d)(2); [section] 1.6694-3(b)(3)), but relate to taxpayers rather than tax professionals. Consequently, a relatively small absolute shift in probability assessment could make an important difference in the recommended tax reporting strategy.
(17) As described in footnote 14, we performed sensitivity analyses using ANCOVA to control for potential effects of the demographic variables described in Table 1 in testing hypotheses. Months of experience is positively related to REWORK (t = 3.18, p < .01) in a one-way ANCOVA model using NEG_BIAS versus UNBIASED as the grouping factor. Including this covariate in the analysis increases the significance of the grouping factor in our test of H3 (t = 2.03, one-tailed p < .03).
(18) A graduate research assistant and one of the authors independently coded participants' written review comments.
American Institute of Certified Public Accountants (AICPA). 1988. Statement of Responsibilities in Tax Practice No. 1. New York, NY: AICPA.
Asare, S., and L. McDaniel. 1996. The effects of familiarity with the preparer and task complexity on the effectiveness of the audit review process. The Accounting Review 71 (April): 133-159.
Ashley, K. 1990. Modeling Legal Argument: Reasoning with Cases and Hypotheticals. Cambridge, MA: The MIT Press.
Bedard, J. C., and S. F. Biggs. 1991. Pattern recognition, hypotheses generation, and auditor performance in an analytical task. The Accounting Review 66 (July): 622-642.
Boiney, L., J. Kennedy, and P. Nye. 1997. Instrumental bias in motivated reasoning: More when more is needed. Organizational Behavior and Human Decision Processes 72 (l): 1-24.
Church, B. 1991. An examination of the effect that commitment to a hypothesis has on auditors' evaluations of confirming and disconfirming evidence. Contemporary Accounting Research 7 (2): 513-534.
Cloyd, C. B., and B. C. Spilker 1999. The influence of client preferences on tax professionals' search for judicial precedents, subsequent judgments, and recommendations. The Accounting Review 74 (July): 299-322.
Cuccia, A., K. Hackenbrack, and M. Nelson. 1995. The ability of professional standards to mitigate aggressive reporting. The Accounting Review 70 (April): 227-248.
Davis, J., and D. Mason. 2003. Similarity and precedent in tax authority judgment. The Journal of the American Taxation Association 25 (Spring): 53-71.
Hackenbrack, K., and M. W. Nelson. 1996. Auditors' incentives and their application of financial accounting standards. The Accounting Review 71 (January): 43-59.
Hatfield, R. 2000. The effect of accountability on the evaluation of evidence: A tax setting. Advances in Taxation 12: 105-125.
--. 2001. The effect of staff accountant objectivity in the review and decision process: A tax setting. The Journal of the American Taxation Association 23 (Spring): 61-74.
Johnson, L. 1993. An empirical investigation of the effects of advocacy on preparers' evaluations of judicial evidence. The Journal of the American Taxation Association 15 (Spring): 1-22.
Kennedy, J., and M. Peecher. 1997. Judging auditors technical knowledge. Journal of Accounting Research 35 (2): 279-293.
Kunda, Z. 1990. The case for motivated reasoning. Psychological Bulletin 108 (3): 480-498.
Libby, R., and K. Trotman. 1993. The review process as a control for differential recall of evidence in auditor judgments. Accounting, Organizations and Society 18 (6): 559-574.
Lord, C., L. Ross, and M. Lepper. 1979. Biased assimilation and attitude polarization: The effect of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology (November): 2098-2109.
Mason, D., and L. Levy. 2001. The use of latent constructs methodology in behavioral accounting research: The measurement of taxpayer advocacy. Advances in Taxation 13: 123-139.
McMillian, J., and R. White. 1993. Auditors' belief revisions and evidence search: The effect of hypotheses frame, confirmation bias, and professional skepticism. The Accounting Review 68 (3): 443-465.
Messier, W., and R. Tubbs. 1994. Recency effects in belief revision: The impact of audit experience in the review process. Auditing: A Journal of Theory & Practice 13 (1): 57-72.
Ramsay, R. 1994. Senior/manager differences in audit work paper performance. Journal of Accounting Research 23 (Spring): 127-135.
Ricchiute, D. 1999. The effect of audit seniors' decisions on working paper documentation and on partners' decisions. Accounting, Organizations and Society 24: 155-171.
Rich, J., I. Solomon, and K. Trotman. 1997. Multi-auditor judgment/decision making research: A decade later. Journal of Accounting Literature 16: 86-126.
Roberts, M. 1998. Tax accountants' judgment/decision-making research: A review and synthesis. The Journal of the American Taxation Association 20 (1): 78-121.
Russo, J., V. Medvec, and M. Meloy. 1996. The distortion of information during decisions. Organizational Behavior and Human Decision Processes 66 (1): 102-110.
Shafir, E., I. Simonson, and A. Tversky. 1993. Reason-based choice. Cognition 49:11-36.
Shields, M. D., I. Solomon, and K. D. Jackson. 1995. Experimental research on tax professionals' judgment and decision making. In Behavioral Tax Research: Prospects and Judgment Calls, edited by J. S. Davis. Sarasota, FL: American Taxation Association.
Snyder, M., and W. Swann. 1978. Hypothesis-testing processes in social interaction. Journal of Personality and Social Psychology (November): 1202-1212.
Trotman, K. 1985. The review process and the accuracy of auditor judgments. Journal of Accounting Research 23 (2): 740-752.
John A. Barrick is an Assistant Professor at Brigham Young University, C. Bryan Cloyd is an Associate Professor at the University of Illinois at Urbana--Champaign, and Brian C. Spilker is a Professor at Brigham Young University.
Submitted: July 1999
Accepted: October 2003
|Printer friendly Cite/link Email Feedback|
|Author:||Barrick, John A.; Cloyd, C. Bryan; Spilker, Brian C.|
|Publication:||Journal of the American Taxation Association|
|Date:||Mar 22, 2004|
|Previous Article:||Firm valuation effects of the expatriation of U.S. corporations to tax-haven countries.|
|Next Article:||Does representation matter in IRS office audits?|