Group Title: investigation of confirmatory bias in an audit setting
Title: An investigation of confirmatory bias in an audit setting
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00102768/00001
 Material Information
Title: An investigation of confirmatory bias in an audit setting
Physical Description: Book
Language: English
Creator: Church, Bryan K., 1959-
Copyright Date: 1986
 Record Information
Bibliographic ID: UF00102768
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: ltuf - AEN9854
oclc - 16137453

Full Text
















AN INVESTIGATION OF CONFIRMATORY BIAS IN AN
AUDIT SETTING: A CONCEPTUALIZATION
AND LABORATORY EXPERIMENT








BY






BRYAN K. CHURCH


A DISSERTATION PRESENTED TO THE
GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY



UNIVERSITY OF FLORIDA


1986































Copyright 1986

by

Bryan K. Church


















ACKNOWLEDGEMENTS

I appreciate the generous contribution of my dissertation committee

members: Michael Bamber, Joel Cohen, and Doug Snowball. Each member was

always more than willing to lend his time and effort. Special thanks

are extended to Doug Snowball who demonstrated superior insight and

patience in his role as chairman of the committee.

I acknowledge the accounting firms that provided participants for

the study: Arthur Andersen, Arthur Young, Coopers and Lybrand, Deloitte,

Haskins, and Sells, Ernst and Whinney, James Moore, Peat, Marwick, and

Mitchell, Price Waterhouse, Purvis Gray, and Touche Ross. Thanks are

extended to the practicing auditors who gave their time and effort as

participants in the study. Arnie Schneider and Doug Snowball made

invaluable contributions by arranging for firms to provide these

participants.

The pursuit of a Ph.D. is a very emotional and draining experience.

I thank my wife, Lucy, for her never-ending support and always "keeping

the faith." I also thank Bill Messier for convincing me to follow my

true research interests. Finally, credit must be given to Gary Fane for

encouraging me to pursue a career in academics.

















TABLE OF CONTENTS

Page

ACKNOWLEDGEMENTS . . . . . . . . . . . .. iii

LIST OF TABLES . . . . . . . . .. . . . vi

LIST OF FIGURES . . . . . . . .. . . . viii

ABSTRACT . . . . . . . .... . . . . ix

I. INTRODUCTION . . . . . . . . . . . 1

Statement of the Problem . . . . . . . . 2
Scope of the Dissertation . . . . . ...... 3
Organization of the Dissertation . . . . . . 4

II. RELATED LITERATURE, THEORETICAL MODEL, AND RESEARCH
HYPOTHESES . . . . . . . . . . . . 6

An Overview of Confirmatory Bias. . . ..... 7
Conditions Leading to Confirmatory Bias . . . . . 9
Motivational Factors o . . . . . . . .. 9
Cognitive Factors . . . .. . . . . 12
Situational Factors . . . .. . . . . .. 13
Factors Affecting the Auditor . . . . . . 14
Constrained Readiness for Search and Interpretation . 16
Information Search . . . . . . . . . . 17
Empirical Studies of Information Search . . . . .. 20
Logical-Problem Tasks . . . . .. . . . 20
Testing Hypotheses About Others . . . .. . 22
Information Interpretation . . . . . . . . 24
Selective Processing . . . .. . . . 26
Situational Attribution .. . . . . . . 27
Biased Assimilation . . . . . . . . . 28
Empirical Studies of Information Interpretation . . .. 28
Ignoring Disconfirming Cues . . . . . . .. 29
Explaining Away Disconfirming Cues . . . . .. 31
Formulation of the Hypotheses . . . . . . .. 34
Summary of the Model . . . . . . . 34
Elements Relevant to the Study . . . . . . 36
Research Hypotheses . . . . . . . . . 38










III. RESEARCH METHODOLOGY . . . . .

Experimental Design . . . . .
Subjects . . . . . . . .
Procedure . . .. . . . .
Task . . . . . . . . .
Testing the Hypotheses . . . .

IV. DATA ANALYSIS AND RESULTS . . . .

The Occurrence of Confirmatory Bias .
Manipulation Check for the Prime
Findings for HI . . . . .
Findings for H2 . . . . .
The Underlying Cognitive Mechanisms .
Findings for H3: Recall . . .
Findings for H3: Importance . .
Confidence . . . . . .

V. RESEARCH IMPLICATIONS AND LIMITATIONS .

Summary of the Experimental Study . .
Confirmatory Bias . . . . .
Cognitive Mechanisms . . . .
Implications . . . . . . .
Limitations . . . . . . .

Appendix


A. MATERIALS FOR SUBJECTS WHO WERE PRIMED FOR THE SRC ... .104

B. MATERIALS FOR SUBJECTS WHO WERE PRIMED FOR THE PPC ... .108

C. MATERIALS FOR SUBJECTS WHO WERE NOT PRIMED. . .. . .112

D. MATERIALS FOR SUBJECTS WHO WERE COMMITTED TO THEIR
HYPOTHESES . . . . . . . . .. . .. 116

E. MATERIALS FOR SUBJECTS WHO WERE NOT COMMITTED TO
THEIR HYPOTHESES . . . . . . . .. .. 122

F. MATERIALS FOR THE DEBRIEFING QUESTIONNAIRE . . . . 127

G. MATERIALS FOR RECALL QUESTIONS: PREDETERMINED ORDER AND
REVERSED ORDER . . . . . . . . ... ... .133


H. MATERIALS FOR IMPORTANCE QUESTIONS: PREDETERMINED ORDER AND
REVERSED ORDER . . . . . . . . . . .

BIBLIOGRAPHY . . . . . . . . . ... . .

BIOGRAPHICAL SKETCH . . . . . . . . . . . .


136

138

143

















LIST OF TABLES


Table

1. Compliance Tests and Sampling Results .

2. Contingency Table (Prime SRC vs PPC) .

3. Contingency Table (Control vs Treatment)

4. Chi-Square Goodness-of-Fit Test for Cycle


Selected


ANOVA for BUDHRS (n=63) . . .

ANOVA for BUDHRS (n=79) . . .

ANOVA for ADDHRS (n=32) . . .

ANOVA for ADDHRS (n=48) . . .

Mean Scores for ADDHRS . . .

ANOVA for ADDHRS (n=63) . . .

ANOVA for ADDHRS (n=79) . . .

ANOVA for ADDHRS by Cycle (n=63)

ANOVA for ADDHRS by Cycle (n=79)

T Tests for %ADDHRS-%BUDHRS (PPC)

T Tests for %ADDHRS-%BUDHRS (SRC)


Mean Scores for BUDHRS . . . . . . . . . .

Rank-Sum Means for COM Subjects . . . . . . . .

Rank-Sum Means for NCOM Subjects . . . . . . .

Rank-Sum Means for NCOM+ Subjects . . . . . . .

Subjects Who Followed a Systematic Pattern of Search . .


page

51

57

57

58

59

59

60

60

62

64

65

65

66


. . . . . . .










21. Summary Statistics for Recall . . . . . . . .. 81

22. Summary Statistics for Recall: Cycle Selected=SRC . . .. 82

23. Summary Statistics for Recall: Cycle Selected=PPC . . .. 83

24. Chi-Square Goodness-of-Fit Test for Corresponding Pairs . 84

25. Chi-Square Goodness-of-Fit Test for Recognition of Symmetry 85

26. Summary Statistics for Importance . . . . . ... 85

27. Summary Statistics for Importance: Cycle Selected=SRC . . 87

28. Summary Statistics for Importance: Cycle Selected=PPC . . 87

29. T Tests for SRC-PPC by Cycle . . . . . . .. 88

30. Mean Scores for Confidence by Commitment and Correctness
of Recall . . . . . . . . ... . . . 90

31. Mean Scores for Confidence by Cue Classification and
Correctness of Recall . . . . . . . .. 90



















LIST OF FIGURES


Figure

1. Conditions that Lead to Confirmatory Bias . . . .

2. Information Search . . . . . . . . .

3. Information Interpretation . . . . . . .

4. Frequency of Responses for ADDHRS . . . . . .

5. Frequency of Responses for BUDHRS: Cycle Selected=SRC .

6. Frequency of Responses for BUDHRS: Cycle Selected=PPC .

7. Frequency Distribution for Rank-Sum Combinations . .


page

. . 10

. . 18

. . 25


viii


* .

















Abstract of Dissertation Proposal Presented to the Graduate
School of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of Doctor of Philosophy


AN INVESTIGATION OF CONFIRMATORY BIAS IN AN
AUDIT SETTING: A CONCEPTUALIZATION
AND LABORATORY EXPERIMENT

By

Bryan K. Church


December 1986


Chairman: Douglas Snowball
Major Department: School of Accounting

A laboratory experiment was conducted to (1) investigate the

occurrence of confirmatory bias and (2) explore the cognitive

mechanisms underlying auditors' judgments. A theoretical model was

developed in which the conditions likely to lead to confirmatory bias

were identified. One particular condition, commitment, was examined in

the present study. Experimentally, subjects were required to allocate

audit effort between two transaction cycles that were characterized as

potential problem areas: the sales and receivables cycle and the

purchases and payables cycle. Next, subjects were asked to select one

of these cycles as being the most likely source for an unexpected

fluctuation in gross margin. Subjects who were committed to their

hypotheses were required to prepare a written memorandum justifying

their selections of a particular cycle. Subjects who were not committed










were not required to prepare this written justification. Subsequently,

subjects were presented with several pieces of audit evidence and then

asked to make another allocation of audit effort between the two cycles.

Lastly, a debriefing questionnaire was administered in which subjects

were asked to recall evidence and rate the importance of this evidence.

Subjects who were committed to their hypotheses were expected to show

signs of confirmatory bias. They were expected to allocate more

subsequent audit effort to the cycle selected than subjects who were not

committed. The results for the occurrence of confirmatory bias were

mixed. Subjects who selected the purchases and payables cycle (and were

committed to that cycle) exhibited confirmatory bias. Subjects who

selected the sales and receivables cycle did not show signs of the bias;

however, this finding may have been due to a ceiling effect. Recall and

importance measures were used to assess the underlying mechanisms.

These findings were also mixed. Subjects recalled the same amount of

confirming and disconfirming evidence, regardless of the cycle selected

or level of commitment; however, problems with this measure were

evident. In general, subjects assigned more importance to evidence

uncovered in the sales and receivables cycle than the purchases and

payables cycle. Only subjects who were committed to the purchases and

payables cycle did not follow this pattern.
















CHAPTER I


INTRODUCTION


External auditors are hired to attest to the fairness of a client's

financial-statement presentation. In particular, they are required to

conduct various audit tasks such that an opinion can be formed as to the

fairness of this presentation. The objective is for this opinion to

reflect the true state of the client. If the opinion does not reflect

this true state, auditors may be held liable at common law or under the

federal securities laws. Accordingly, the audit firm has an incentive

to issue an accurate and unbiased opinion.

An accurate and unbiased opinion requires that audit tasks are

conducted relatively free of judgmental biases. Felix and Kinney (1982)

have identified several audit tasks where these biases may arise (e.g.,

compliance tests of pertinent controls and substantive tests of

transactions and balances). However, these tasks concern the physical

nature of the audit and, as such, their identification does not shed

insight into "how" biases may occur. An alternative view of opinion

formulation focuses upon information-processing tasks. Five processing

tasks that may be associated with opinion formulation are hypothesis

generation, information search, information interpretation, audit

judgment, and audit decision. An examination of these tasks may provide









insight into the cognitive processes that lead to a judgment and, hence,

into "how" biases occur. The presumption is that understanding "how"

biases occur is essential if researchers are to develop decision aids

that minimize suboptimal responses.

This study focuses primarily on two of the five information-

processing tasks associated with opinion formulation: hypothesis

generation and information interpretation. The broad proposition is

that in the evaluation of information, individuals may exhibit a

tendency to interpret cues consistently with previously generated

hypotheses. This tendency is referred to as confirmatory bias.


Statement of the Problem

Confirmatory bias has been found in a variety of experimental

settings. The findings are fairly consistent in that subjects (1) fail

to consider alternative hypotheses and/or (a) ignore or explain away

disconfirming information. Despite the apparent pervasiveness of the

phenomenon of confirmatory bias, it does not necessarily follow that

auditors must be susceptible to this bias. In particular, professional

standards mandate an attitude of skepticism (AICPA, 1983, section

327.06). This attitude suggests that auditors should not accept

findings at face value but rather should question test results and seek

alternative explanations. The assumption is that auditors should remain

open minded in information search and interpretation. However, other

factors may override the auditor's mandated skepticism ("open

mindedness"). Specifically, motivational and cognitive factors (to be

discussed at some length in Chapter II) may affect auditor judgmental









processes and, in turn, lead to a closed-minded approach. To the extent

that "closed mindedness" results, there is an increased likelihood that

confirmatory bias will occur. The possibility that confirmatory bias

affects judgmental processes of external auditors warrants research

attention.

The external auditor constantly generates and tests specific

hypotheses in the process of assessing one general hypothesis (that a

client's true state is fairly represented by the respective financial

statements). Confirmatory bias suggests that audit evidence may be

interpreted consistently with initially generated hypotheses, even

though alternative interpretations may be more plausible. In this

sense, biased interpretation may lead to a suboptimal response.

Confirmatory bias may introduce ineffectiveness and/or inefficiency

in the opinion-formulation process. It may result in unjustified audit

opinions (ineffectiveness), in that the audit opinion does not reflect

an existing divergence between the client's true state and the

respective financial statements. Confirmatory bias also may result in

inefficient audits, whereby auditors maintain incorrect hypotheses

longer than is justified given the diagnosticity of the evidence

uncovered. Consequently, the likelihood that confirmatory bias affects

external auditors' judgments should be investigated.


Scope of the Dissertation

A laboratory experiment was conducted to assess the nature of

confirmatory bias in audit settings. In conjunction, the cognitive

mechanisms underlying auditors' judgments were explored. This










investigation of cognitive mechanisms provides an extension of the

previous literature. Early auditing research on judgmental biases

(Gibbins, 1977; Uecker and Kinney, 1977; Joyce and Biddle, 1981a; 1981b;

and Kinney and Uecker, 1982) adopted the black-box approach. That is,

researchers conducted tests of the existence or nonexistence of

particular biases without investigating the causal, cognitive mechanisms

that produce them. The present study differs from the earlier research

by seeking causal explanations (in terms of cognitive mechanisms) for

specific judgments.

The purpose of this study, therefore, is to provide insight into

(1) the existence or nonexistence of confirmatory bias in audit settings

and (2) the cognitive mechanisms that underlie various audit judgments.

As a result, this study is descriptive in nature. Such descriptive

research facilitates the establishment and implementation of normative

(optimal) models of judgment. In particular, the identification of

suboptimal processes by descriptive research ultimately may lead to the

development of decision aids to correct for them.


Organization of the Dissertation

The remainder of the dissertation is organized into five chapters.

Chapter II presents the relevant prior research, a general theoretical

model, and specific research hypotheses. The prior research serves as a

foundation for the development of the theoretical model. The model

characterizes the occurrence of confirmatory bias and, as such,

identifies (1) certain conditions that may lead to the bias and (2)

cognitive mechanisms that may underlie judgments when these conditions










are present. Elements of this model are used to draw specific research

hypotheses. The laboratory experiment undertaken to test the hypotheses

is described in considerable detail (encompassing descriptions of the

design, subjects, procedure, and the task) in Chapter III. Chapter IV

presents the analysis and the results of the study. The chapter also

relates the results to the research hypotheses. Chapter V presents a

summary of the results, discusses the implications of these results,

highlights the limitations inherent in the methodology, and provides

suggestions for future research.

















CHAPTER II

RELATED LITERATURE, THEORETICAL MODEL, AND RESEARCH HYPOTHESES


Confirmatory bias has been examined in a variety of experimental

settings, and for the most part, the findings of previous research have

indicated that the bias does occur. The research has provided limited

structure (for an exception see Fischhoff and Beyth-Marom, 1983),

however, as to "when" and "how" the bias occurs, and an objective of

this chapter is to provide such a structure. The chapter presents a

theoretical model which was developed using prior research as a

foundation. The model provides the basis for discussion of (1) the

conditions that are likely to lead to confirmatory bias and (2) the

processes that are likely to underlie confirmatory bias. Specific

research hypotheses are formulated from this model.

The chapter is organized into five sections. First, an overview of

the bias is presented. The second section discusses the performance of

information-processing tasks and identifies certain conditions that may

bias the performance of these tasks. This section provides an

understanding of "when" confirmatory bias occurs. The third and fourth

sections present a more detailed examination of those phases of

processing tasks (information search and information interpretation) in

which the bias is likely to arise. The emphasis of these two sections

is on providing an understanding of "how" confirmatory bias occurs. The










final section summarizes the general model and identifies specific

elements that are used to draw the research hypotheses. The hypotheses

are also formally stated in this section.


An Overview of Confirmatory Bias

Confirmatory bias results from a predisposition to confirm rather

than disconfirm. Individual inquiry is biased toward the fulfillment of

expectations (Deighton, 1983). Schustack and Sternberg (1981) have

suggested that the bias affects (1) the particular information to which

individuals attend and (2) the interpretation applied to this particular

information. In other words, confirmatory bias may arise in both

information search and information interpretation. Only the latter

context is examined empirically in this study. Bias originating in

information search is considered beyond the scope of the study and is

not examined directly. Nevertheless, both processing tasks are

discussed in the development of the theoretical model, in order to

provide a more complete framework for the current study and for future

research in the area.

In the first case, confirmatory bias may arise in the task of

information search. Individuals may formulate hypotheses and then

search for information that confirms these hypotheses, while ignoring

information that disconfirms them. Einhorn and Hogarth (1978) concluded

that individuals have more difficulty handling disconfirming information

than confirming information. Apparently, less cognitive effort is

associated with the latter. Bruner et al. (1956) similarly argued that

individuals are more ready or more prepared to search for information










that confirms a hypothesis. This type of search strategy allows for

direct tests of existing hypotheses, whereas alternative disconfirmingg)

search strategies only allow for indirect tests. Hoch (1984) suggests

that judgmental habits are likely to be based on the former.

In the second case, confirmatory bias may arise in the task of

information interpretation. Individuals may tend to interpret

information consistently with their hypotheses rather than

inconsistently. Bruner (1957) suggested that individual hypotheses

establish a basis for information interpretation. Individual's attempt

to produce a "fit" between existing hypotheses and attended information,

and this may lead to a biased interpretation of disconfirming

information. The information may be discounted or assimilated toward

existing hypotheses. The argument proposed by Bruner et al. (1956) for

information search can also be extended to information interpretation.

Individuals are simply more ready or more prepared to interpret

information consistently with their hypotheses. In this sense,

inconsistent disconfirmingg) interpretations of information are not as

likely to be considered.

In both cases discussed above, motivational and/or cognitive

factors are likely to underlie confirmatory bias. Individuals may want

to confirm their hypotheses and/or they may be more ready to confirm

them. The next three sections of this chapter discuss these factors and

how they lead to confirmatory bias. The related literature (theoretical

and empirical) is also discussed.










Conditions Leading to Confirmatory Bias

Confirmatory bias arises when individuals exhibit "closed

mindedness" in information search and/or interpretation. The presence

of "closed mindedness" may be due to motivational and/or cognitive

factors. In addition, situational factors may heighten the effects of

these factors and, hence, indirectly influence the presence of "closed

mindedness." When "closed mindedness" is indeed present, it reduces or

constrains an individual's readiness for search and/or interpretation.

The entire process is depicted in Figure 1, which provides a framework

for the discussion below.


Motivational Factors

The relationship between individual motivations and "closed

mindedness" has been addressed frequently within the psychological

literature. In particular, two competing motives or tendencies are

identified within this literature: consonance and curiosity. Consonance

stems from cognitive-consistency theory and suggests "closed

mindedness." Festinger (1957) has asserted that individuals strive to

maintain a consonance or consistency among the various elements of their

cognitive systems. Inconsistent cognitions arouse an unpleasant

psychological state, and as a result, behavior is designed to achieve

consistency--which is psychologically pleasing. Consistent cognitions

also serve to enhance individual "self image" (Aronson, 1976). McGuire

(1966, p. 37) has indicated that "an individual motivated by consonance

will have a penchant for stability, redundance, familiarity,

confirmation of expectations, and avoidance of the new and the



























Motivational
Factors
(Commitment)


Constrained
Readiness
for
Search


Situational
Factors
(Time Pressure)


S"Closed
q Mindedness"


Cognitive
Factors
(Salience)


SConstrained
Readiness
for
Interpretation


Figure 1: Conditions that Lead to Confirmatory Bias









unpredictable." In short, such individuals desire a structured and

ordered picture of the world.

Curiosity, on the other hand, suggests "open mindedness" and is the

focus of cognitive-complexity theory. Berlyne (1960) has asserted that

individuals display exploratory behavior to attain an understanding of

the unknown and the unpredictable. This type of behavior is likely to

arise from a low level of arousal, i.e., boredom. McGuire (1966, p.37)

has asserted that "an individual motivated by curiosity takes pleasure

in the unexpected, wants to experience everything, shows alternation

behavior, and finds novelty rewarding."

Specific factors may affect an individual's motive to act. "Closed

mindedness" is likely to occur when an individual's motive for

consonance is heightened. Commitment (defined by Kiesler, 1971, as a

psychological state that creates a resistance to cognitive change) is an

example of a factor that leads to a heightening of this motive (see

Bazerman et al., 1984). Individuals who are committed are likely to

feel a need to justify their positions both to themselves and to others

(Staw, 1981). Internal justification is necessary for cognitive

consistency (Festinger, 1957; Aronson, 1976). External justification is

necessary for an appearance of competence (Fox and Staw, 1979).

Individuals who publicly support (or commit to) particular positions are

less likely to subsequently withdraw their support for fear of public

embarrassment, an unpleasant psychological state. These individuals are

more likely to advocate their chosen positions and view these positions

favorably in terms of the surrounding environment.









Cognitive Factors

Figure 1 indicates that cognitive factors also may lead to "closed

mindedness." These factors involve the accessibility of specific items

from long-term memory (LTM). Items that are accessed are initially

activated. Activation is a state of excitation (achieved either

consciously or unconsciously) that dimishes over time. It is necessary

but not sufficient for the accessibility of particular items. The state

of excitation is triggered by environmental cues such that items

associated with these cues are likely to be remembered. The likelihood

that an item will be activated and subsequently accessed may be directly

affected by the recency and frequency with which it has been accessed

previously. Items that have been accessed recently may still be

activated, and as such they may be relatively easy to access: the

association between these items and environmental cues is at a

heightened level because of recent access. Repeated activation also

serves to strengthen the association between certain items and

environmental cues. Items that have been accessed frequently then may

be relatively easy to activate. As a result, individuals may display

static tendencies. They may not adequately adapt to new situations and,

hence, may exhibit "closed mindedness"--in the form of a bias towards

items accessed recently and frequently.

The salience of an item also may affect LTM accessibility.

Salience is defined as the prominence or conspicuousness associated with

a particular item. Salience strengthens the association between a

particular item and environmental cues. Tversky and Kahneman (1973;

1974) indicated that individuals are more likely to access salient









events than non-salient ones. Taylor et al. (1979) concluded that

salience effects are highly generalizable. Consequently, "closed

mindedness" may follow--in the form of a bias towards items that are

salient. Similarity and cognitive effort are two other factors that may

affect LTM accessibility and increase the likelihood that items will be

accessed which are consistent with expectations. A cue that appears

similar to an expectancy-consistent cue is likely to be interpreted in

that manner, even though in fact it may not be similar (Bruner, 1957).

This interpretation is more readily accessible than an alternative one,

and as such, it requires less cognitive effort (Einhorn and Hogarth,

1978). Items that are consistent with expectations are likely to be

activated when a specific expectation is formed. A strong association

is likely to exist between these items and expectations. In contrast,

inconsistent items are not as likely to be activated since the

association between these items and expectations is likely to be weak.

These items require more cognitive effort to access. As a result,

individuals may exhibit "closed mindedness" in the form of a bias

towards items that are expectancy-consistent.


Situational Factors

Situational factors may have an impact on motivational and/or

cognitive factors such that "closed mindedness" follows. In particular,

these factors may heighten an individual's motive for consonance and/or

restrict LTM accessibility. Situations where task demands exceed

individual capabilities are likely to produce these effects. Time

pressure is an example of a factor that may lead to excess task demands.










An individual may be motivated to simplify a situation in order to reach

a quick solution, and this simplification may involve an increased

awareness of expectancy-consistent cues and a decreased awareness of

expectancy-inconsistent cues. In addition, time pressures may serve to

heighten the accessibility of specific items from LTM, i.e.,

expectancy-consistent items. These items typically allow for the

fastest and most direct test of existing hypotheses. As such, "closed

mindedness" is likely to follow.


Factors Affecting the Auditor

Biased judgments are likely to follow to the extent that "closed

mindedness" results. As shown in Figure 1, commitment, salience, and

time pressure are specific factors that may lead to "closed mindedness"

and, hence, may have an impact on judgments, including those made by

auditors. From a psychological perspective, the effects associated with

these factors are likely to be pervasive. From an auditing perspective,

the effects associated with these factors may surface or they may be

mitigated by professional training. Although the end result is an

empirical question, plausible scenarios may be construed where these

factors have adverse (undesirable) effects on the audit process. The

possibility of an undesirable outcome (inefficiency and/or third-party

litigation) underlies the need to examine the relationship between

factors leading to "closed mindedness" and the audit process.

Commitment to a hypothesis may result in "closed mindedness"

through an effect on auditor motivations. In particular, if auditors

are required to disclose their hypotheses to others (especially









supervisors) before specific audit work is conducted, they may have

increased incentives to maintain these hypotheses. Fox and Staw (1979)

have found that individuals feel a need to justify actions disclosed to

others. This need only serves to heighten incentives to maintain the

correctness of previous actions (or a reluctance to change one's

position). For auditors, these incentives may be heightened further by

the performance-evaluation process, since how quickly auditors identify

problem areas may directly affect evaluations of performance. As a

result, auditors may want to believe in the correctness of their initial

hypotheses, and they may be reluctant to reject these hypotheses.

The salience of a hypothesis may lead to "closed mindedness"

through an effect on the specific audit areas that are perceived as

potential problems. In an audit investigation, some areas are more

likely to "stand out" and fall under closer inspection than others. For

example, areas that have been focal points of recent court cases should

be very salient. Areas where the auditor has observed major

discrepancies in other audits should also be very salient. Auditors may

unwittingly focus their scope of investigation in these salient areas,

even though the evidence strictly may not justify that focus.

Situational factors, particularly time pressures, may lead to

"closed mindedness" through an effect on auditor motivations and/or the

recognition of potential problem areas. They may generally heighten the

effects of the motivational and cognitive aspects associated with

information evaluation. Time pressures may be such that auditors

operate in a setting where task demands exceed individual capabilities.

Accordingly, quick, simple, and viable solutions may be sought.









Auditors may be motivated to seek the most efficient, but not

necessarily the most effective, solutions. As such, these motivations

may lead to a reduction in the auditor's scope of investigation.

Time pressures also are likely to affect the specific audit areas

that are viewed as potential problems. Libby (1984) found that the

likelihood auditors will investigate particular areas is affected by the

recency and frequency associated with discrepancies that have been

uncovered in these areas. Time pressures serve to heighten this

likelihood. That is, auditors may limit their areas of investigation to

those areas that have been examined only very recently and/or very

frequently. Therefore, the auditor's scope of investigation may be

focused in these areas.


Constrained Readiness for Search and Interpretation

"Closed mindedness" increases the likelihood that confirmatory bias

will occur. Specifically, it leads to a constrained readiness for

search and a constrained readiness for interpretation, as shown in

Figure 1. These states of readiness reflect the motivational and

cognitive factors discussed above. In particular, these states will

have an impact on what hypotheses are generated, what information cues

are uncovered, what information cues are attended, and what information

interpretations are provided. When "closed mindedness" is present, the

likelihood is increased that fewer and more focused hypotheses will be

generated; information cues will be searched and attended to that are

consistent with (confirm) these hypotheses; and that information will

tend to be interpreted in a fashion consistent with these hypotheses.










Information Search

Information search is defined as the task of actively seeking

evidence to formulate hypotheses or test hypotheses. This task is

affected by hypothesis generation and readiness for search. The

information-search process is presented in Figure 2. Although

information search is not examined directly in the present study, a

discussion is included to provide a more complete framework for

understanding confirmatory bias.

Initially, search is undertaken or hypotheses are generated. No

predictions are made as to which courses of action will be taken.

Little empirical evidence is available and a priori reasoning does not

lead to any strong expectations. Whatever the course taken, the end

result is that hypotheses will be generated, as shown in Figure 2.

Information search that is undertaken initially (preliminary search) is

affected by readiness for search. Readiness for search determines the

specific types of cues that are attended. This readiness may be

constrained as the result of motivational and/or cognitive factors.

Hoch (1984) has suggested that individuals prefer information

framed in terms of positive instances rather than negative ones.

Positive instances allow for direct tests of hypotheses, whereas

negative instances allow for indirect tests. Individuals are likely to

have had much more practical experience with direct tests rather than

indirect tests. Hoch (1984) presents the example of a husband who

asserts that he knows the kinds of clothes his wife likes. The husband

is not likely to test this belief indirectly by buying his wife clothes

that he thinks she will dislike. To maintain marital harmony, he will
















Preliminary
Search


Readiness
for
Search


2 Hypothesis
Generation '


Yes


v
Information
Cues


New
Hypotheses?




No


Readiness
for
Interpretation


V
Information
Interpretation


Figure 2: Information Search


Judgment?






Yes




Search
is
Complete









buy her clothes that he thinks she will like, which is a direct test of

the belief. For practical reasons, instances that lead to direct tests

may be much more accessible than instances that lead to indirect tests

and as such readiness for search may be constrained. If readiness for

search is constrained, fewer types of cues are attended.

If hypothesis generation is undertaken initially, it is also

affected by readiness for search, as shown in Figure 2. In particular,

this readiness determines the specific hypotheses generated: the number

of hypotheses and the focus of these hypotheses. If this readiness is

constrained, hypothesis generation will be restricted.

After hypothesis generation is completed, information search is

undertaken. Figure 2 indicates that this task involves seeking out

various information cues. Previous research (Wason, 1960; 1968; Mynatt

et al., 1977; 1978; Snyder and Swann, 1978; and Snyder and Campbell,

1980) has found that in certain situations, search is likely to be

biased towards uncovering cues that are consistent with initial

hypotheses. In this sense, confirmatory bias may originate in

information search.

Subsequent to encountering information cues, they are interpreted.

At this point, search activity either continues or it is completed.

Figure 2 indicates that if search continues, additional information cues

may be sought or hypothesis generation may be undertaken again. In the

former, search is still directed towards the initial hypothesis. In the

latter, this direction is changed or modified (not shown in Figure 2).

If this direction is changed, new hypotheses will be accessed,

initial hypotheses rejected, and subsequent search will not be a









function of these initial hypotheses. In contrast, if the direction of

search is modified, old hypotheses will be reformulated, not necessarily

rejected, and subsequent search will be some function of these old

hypotheses. In this case, search will be focused in a specified area

such that confirmatory bias may still be likely to arise in this

processing task. Subsequently, the processes discussed above may be

repeated until search is completed or abandoned.


Empirical Studies of Information Search

Empirical studies of information search generally involve one of

two tasks. In one case, subjects are required to complete a logical

problem. They are asked to verify a rule or determine a rule that

underlies a set of instances. In a second case, subjects are required

to test hypotheses about other people or firms. Both types of studies

are reviewed below.


Logical-Problem Tasks

In an early study, Wason (1960) presented subjects with a task

described as structurally simple but deceptively difficult. Subjects

were required to determine the rule that accounted for the numerical

sequence 2-4-6. The correct rule was simply three ascending numbers.

While some subjects guessed this rule immediately, others became fixated

on their initial choices and took much longer. These results may be

driven by the association between the rule initially accessed and the

correct rule. The association between two similar rules is likely to be









much stronger than that between two dissimilar rules. Subjects who

guessed the correct rule quickly may have activated this rule or a

similar rule at a very early stage. As such, these subjects may have

had little difficulty activating and subsequently accessing the correct

rule. Subjects who took much longer, on the other hand, may have

initially activated an incorrect rule that was dissimilar (to the

correct rule). In this case, these subjects may have had problems

activating the correct rule, which would account for why they took much

longer to guess this rule.

Wason (1968) presented a similar task in which subjects were

required to select the necessary instances to verify the truth of a

given rule. The results were almost identical to those reported in the

earlier study (Wason, 1960). Mynatt, Doherty and Tweney (1977; 1978)

extended Wason (1960) with a much more complex task. The results,

however, did not change. The results of these studies also may be

driven by the association between the rule initially accessed and the

correct rule.

Finally, Hoch and Tschirgi (1983) found that task performance could

be improved (search could be made more efficient) if concrete (thematic)

rather than abstract (symbolic) materials were presented. Earlier

studies used primarily abstract materials. The use of concrete

materials should enable subjects to associate the experimental task with

mundane experiences. Such experiences are likely to be stored in LTM,

whereas abstract rules usually must be inferred. Instances that are










stored are generally easier (less cognitive effort is involved) to

activate than instances that are inferred. Consequently, the use of

concert materials is likely to improve task performance.

Although the findings in this area suggest that subjects have a

tendency to follow a confirmatory strategy, most of the studies

discussed employed abstract tasks. The results of the study by Hoch and

Tschirgi (1983) indicate that this tendency may be diminished when the

task is framed in mundane terms. In other words, subjects may perform

more efficiently and/or effectively in the real world than in an

artificial, experimental setting. The reason is that subjects have more

experience with the former than the latter. With an abstract task,

subjects are likely to follow a strategy that involves less cognitive

effort, which suggests a confirmatory strategy.


Testing Hypotheses About Others

Snyder and Swann (1978) required subjects to formulate

question-asking strategies to assess the extraversion or introversion

(one was specified) of a target person. Subjects selected 12 of 26

questions from a list provided by the experimenters. This list included

questions framed in terms of confirming, disconfirming, and neutral

instances. The main finding was that subjects showed a definite

preference for confirming questions over both disconfirming and neutral

ones. Snyder and Campbell (1980) repeated this study with a slight

variation and found similar results. In both studies, subjects employed

direct tests of hypotheses: they searched for positive, confirming

instances. Although the results suggest that subjects followed a










confirming strategy, these findings also suggest that subjects followed

a strategy consistent with judgmental habit. A strategy consistent with

judgmental habit is one that allows individuals to adapt satisfactorily

(without adverse consequences) to the environment.

Trope and Bassok (1982) provided another extension of the study by

Snyder and Swann (1978). They presented subjects with two hypotheses,

instead of one, and asked them to assess both hypotheses. Subjects did

not show a preference for attending to confirming information. These

results may have been driven by the fact that the two hypotheses were

made salient. Subjects may have employed a direct test for each

hypothesis. While subjects attended to information that disconfirmed a

particular hypothesis, this information also confirmed the alternative

hypothesis. Subjects may have searched for information that confirmed

each hypothesis.

Lastly, Kida (1984) extended Snyder and Swann (1978) in an audit

setting. Audit partners and managers investigated the going-concern

status of an hypothetical firm. One half of the subjects were asked to

determine if this firm would remain viable for two more years. The

other half were asked to determine if this firm would fail within two

years. The results were mixed. Subjects who assessed failure attended

primarily to confirming information, whereas subjects who assessed

viability attended to both confirming and disconfirming information.

However, the consequences are associated with a negative outcome (loan

default) are more severe than those associated with a positive outcome.

A question of going-concern status is inherently framed as a failure

question to protect against a negative outcome. This hypothesis is










salient even when it is not presented. Consequently, a confirming

strategy was followed in the failure condition and not in the viability

condition.

The findings in this area suggest that subjects seek information

that is consistent with salient hypotheses. In each of the studies

discussed, one or more hypothesis was presented and, consequently, made

salient by the experimenter. However, other hypotheses also may have

been salient if subjects had experience with the task. Personal

experience may have caused the activation and subsequent access of

hypotheses that were not presented. As such, personal experience may

account for the mixed results that have been uncovered as to the

occurrence of confirmatory bias. That is, the bias has been found to

occur only when one hypothesis is salient and not otherwise.


Information Interpretation

Information interpretation is defined as the task of assigning

meaning to specific information cues. This task is affected by

readiness for interpretation and the particular information cues

encountered in search. That is, individuals are more ready or more

prepared to interpret some information cues than others. The

information-interpretation process is presented in Figure 3, which

provides a general overview for the discussion below.

Figure 3 indicates that readiness for interpretation will affect

"how" specific information cues are interpreted. These interpretations

will tend to be biased when this readiness is constrained and unbiased

otherwise. If interpretation is biased, the processing of disconfirming




















Readiness for
Interpretation


I No
Information Constrained Unbiased
Cues -Readiness? Interpretation


Explain Away
Disconfirming
Cues


Processing
Requirements
High?


Yes Ignore
Disconfirming
Cues








n t-


Biased
I nterpretatioz


Figure 3: Information Interpretation










cues will be affected. In particular, these cues will be ignored or

explained away depending upon the processing requirements of the

situation. As shown in Figure 3, disconfirming cues will be ignored

when these requirements are high and explained away otherwise.

Whichever processing phenomenon occurs, confirmatory bias is likely to

result.

The cognitive mechanisms that may underlie the bias include

selective processing, situational attribution, and biased assimilation.

The former involves ignoring disconfirming cues while the latter two

involve explaining away these cues.


Selective Processing

Selective processing occurs when disconfirming cues are ignored or

overlooked in information interpretation. These cues are weighted with

a factor of zero. The mechanism should arise when processing

requirements are stringent and readiness for interpretation is

constrained. Individuals are likely to restructure situations where

processing requirements are high (Bruner et al., 1956; Elstein et al.,

1978) in order to reduce task demands. That is, the mechanism provides

for a simplification in processing disconfirmingg cues are not

processed) and enables individuals to satisfactorily cope with the

situation. Wright (1974) found that subjects attended to less

information when processing requirements were high. In this sense,

selective processing is a mechanism that provides for restructuring.










Situational Attribution

Situational attribution is a cognitive mechanism by which

disconfirming cues are explained away. Under situational attribution,

the apparent inconsistency associated with these cues is recognized and,

subsequently, attributed to various other factors. These cues are

discounted, and their negative impact is significantly reduced.

Motivational factors are likely to lead to this processing mechanism.

That is, individuals desire or want to explain away disconfirming cues.

The mechanism is more likely to be employed when processing requirements

are not high (task demands do not exceed individual capabilities) and

readiness for interpretation is constrained.

If disconfirming cues are explained away as discussed above, they

are likely to be processed at a deeper level than confirming cues.

Level of processing refers to the degree of elaboration in

interpretation, where greater "depth" implies a greater degree of

semantic or cognitive analysis (Craik and Lockhart, 1972, p. 675).

Disconfirming cues likely require further elaboration (by comparison to

confirming cues) in order to explain the apparent inconsistency between

them and existing hypotheses. This further elaboration suggests that

disconfirming cues may take longer to process than confirming cues

(Hastie and Kumar, 1979; and Hastie, 1980). As such, situational

attribution is not expected to be employed when processing requirements

are restrictive.









Biased Assimilation

Biased assimilation is another cognitive mechanism by which

disconfirming cues are explained away. In this case, however, the

apparent inconsistency associated with the disconfirming cues is not

fully recognized or appreciated. Instead, these cues are (1) viewed as

unimportant (neutral) or (2) assimilated to fit existing hypotheses. In

either case, specific cues are not interpreted as being inconsistent

with expectations. Cognitive factors are likely to lead to this

processing mechanism. That is, individuals are less likely to activate

and subsequently access inconsistent interpretations. The mechanism is

most likely to be used when processing requirements are not high and

readiness for interpretation is constrained.

If disconfirming cues are explained away without a recognition of

their apparent inconsistency, they are likely to be processed at a

similar level to that of confirming cues. In fact, these cues may be

processed as if they are confirming cues. Although these assimilated

cues will serve to support expectations, they will not provide a basis

for these expectations. Since these cues will provide weak rather than

strong support for expectations, they are less likely to be examined

when processing requirements are high.


Empirical Studies of Information Interpretation

Empirical studies of information interpretation are reviewed below.

These studies are divided into two categories: studies where

disconfirming cues are ignored and studies where these cues are

explained away.









Ignoring Disconfirming Cues

Information interpretation may involve a selective processing of

cues consistent with initial hypotheses. In this case, selectivity is

likely to produce a biased (superior) recall of confirming cues. Zadny

and Gerard (1974) conducted a study to test this hypothesis. Subjects

were given an expectation and asked to watch a skit. Subsequently, they

were asked to recall facts presented in the skit. The authors found

that subjects recalled significantly more expectancy-consistent

(confirming) facts than expectancy-inconsistent disconfirmingg) facts.

Subjects were likely to have activated characteristics associated with

their expectations when these expectations were formulated (presented).

When expectancy-consistent cues were encountered, subjects were likely

to have matched these cues with their expectations. In the subsequent

recall task, subjects' expectations were likely to have cued

expectancy-consistent items, and as such, these items were more likely

to be recalled than expectancy-inconsistent items.

Rothbart et al. (1979) performed a similar study. However, one

half of the subjects were given an expectation before information was

presented, whereas the other half were given an expectation after

information was presented. Recall was superior for expectancy-

consistent cues only in the before condition. Subjects in the before

condition were likely to have matched expectancy-consistent cues with

their expectations when these cues were encountered. Subjects in the

after condition, on the other hand, were likely to have processed all









cues in a similar manner since they were not initially given an

expectation. Consequently, the findings suggest that selectivity occurs

in encoding rather than recall.

Other studies (i.e., Srull, 1980; Hastie, 1980) have found superior

recall for unexpected or novel information. Unexpected information,

however, is not necessarily disconfirming. These cues may simply appear

novel and unrelated, and as such, they may stand out (the cues may be

salient). In addition, the presence of some disconfirming cues may

actually be expected. Individuals may anticipate that a certain

position will not be supported completely by the evidence (exceptions

may be expected).

Hastie and Kumar (1979) found that subjects recalled unexpected

cues at least as well as expected cues. However, subjects were

initially told that a recall task would be administered. This may have

increased their awareness and their level of processing of unexpected

(and negative) cues. Zadny and Gerard (1974) and Rothbart et al. (1979)

did not inform subjects that recall would be tested. Additionally,

Hastie and Kumar found an inverse relationship between the accuracy of

recall and the number of disconfirming cues that were presented. But

decreasing the number of disconfirming cues should increase their

novelty, which in turn should heighten the conspicuousness associated

with these cues. As a result, recall of disconfirming cues should

improve as these cues become more novel.

The findings in this area suggest that expectations may lead to an

encoding bias, which may affect the processing and subsequent recall of

information cues. In particular, expectancy-consistent cues may be the










focus of attention, whereas expectancy-inconsistent cues may be

overlooked. Expectancy-consistent cues are likely to be matched with

existing expectations, and this matching is likely to facilitate the

activation and subsequent access of these cues. That is, the

expectation triggers activation which enhance the likelihood of

accessibility. In conjunction, although novel and/or unexpected cues

are not likely to be matched with expectations, these cues may stand out

due to their nature. As such, they may be processed at a deeper level

than otherwise bland and unimportant cues. A greater depth of

processing suggests that these cues are further elaborated, which

increases the number of cues that are associated with these cues.

Consequently, the likelihood that these cues will be activated and

subsequently accessed is increased.


Explaining Away Disconfirming Cues

Individuals may explain away disconfirming cues such that the

inconsistency associated with these cues is reduced or eliminated.

Hayden and Mischel (1976) presented subjects with information that was

both consistent and inconsistent with initial expectations.

Subsequently, subjects' interpretations of this information were

elicited. The authors found that subjects either (1) treated

inconsistent information as actually being consistent or (2) dismissed

this information as unimportant. Disconfirming information was

explained away. Subjects were likely to have activated characteristics

associated with their expectations when expectations were formulated.

They may have matched information subsequently encountered with their









expectations. Subjects' expectations were likely to have guided

information interpretation. Simply put, inconsistent interpretations

were not as likely to have been activated as consistent interpretations.

Moreover, subjects may have been motivated to interpret information

consistently with their expectations. Staw (1981) has asserted that

individuals perceive consistent behavior much more favorably than

inconsistent behavior. Subjects may have been motivated to present a

consistent picture of themselves to the experimenter.

Elstein et al. (1978) investigated a range of medical tasks in

which physicians were required to diagnose the condition of hypothetical

patients. The results showed that physicians generated hypotheses early

and tended to explain away evidence that contradicted these hypotheses.

In this case, motivational factors may have driven the results. Subjects

may have felt committed to their initial hypotheses. After all,

subjects performed tasks at which they were experts. They may have felt

a need to maintain their hypotheses in order to exhibit professional

competence to the experimenter. Consequently disconfirming evidence was

explained away.

Lord et al. (1979) conducted a study to assess individual attitudes

on capital punishment. Subjects were presented with information that

both supported and opposed capital punishment. The authors found that

subjects tended to accept information that was consistent with their

attitudes and reject information that was inconsistent. Again,

motivational factors may have driven the results. Capital punishment

may be perceived as an emotional issue, and in general, such issues are

likely to evoke strong feelings. Individuals who had strong beliefs










(either pro or con) about capital punishment were recruited to

participate in the study. These individuals were likely to have been

committed to their beliefs. Subjects may have felt a need to interpret

information consistently with existing attitudes in order to justify

(maintain cognitive consistency) their beliefs to themselves.

Lastly, Darley and Gross (1983) showed subjects a videotape of a

child taking an academic test. The authors labeled this child as being

from either a "high" or "low" socioeconomic background. After viewing

the tape, subjects were given an evaluation form to complete. The child

labeled as coming from a high socioeconomic background was rated above

her grade level, whereas the child labeled as coming from a low

socioeconomic background was rated below her grade level. Subjects also

cited evidence to support their conclusions. In particular, subjects

exposed to the "high" background label indicated that the child answered

more questions correctly, exhibited more instances of positive behavior,

and took a more difficult test than subjects exposed to the "low"

background level. Subjects were likely to have activated

characteristics associated with the label that was presented. That

label was likely to have guided information interpretation. Subjects

may have been more likely to access consistent interpretations than

inconsistent ones. As such, cognitive factors may have driven the

results.

The findings in this area suggest that subjects may have a tendency

to dismiss disconfirming information. These findings may be driven by

motivational and/or cognitive factors. Subjects may have felt compelled

to attribute inconsistent information to factors that were not relevant









to existing expectations. Alternatively, subjects may have been more

likely to activate and subsequently access consistent interpretations

than inconsistent interpretations.


Formulation of the Hypotheses

Specific research hypotheses are drawn from elements of the model

presented above. A test of the entire model is beyond the scope of the

dissertation. The present study is restricted to examining the effect

that commitment has on information interpretation. The broad

proposition is that this factor induces individuals to interpret

information consistently with existing hypotheses.

Before the research hypotheses are presented, the theoretical model

is briefly summarized. The model must be clearly understood at this

point in order to follow the development of the hypotheses. The

decision to investigate certain elements of the model is also discussed.

This discussion provides a justification for inclusion of these elements

in the study. Then, the research hypotheses are developed and formally

stated.


Summary of the Model

A theoretical model was developed to explain the occurrence of

confirmatory bias. Motivational and/or cognitive factors are likely to

underlie the bias. In particular, these factors may lead to "closed

mindedness," which establishes a necessary condition for the occurrence

of the bias. "Closed mindedness," in turn, leads to a constrained

readiness for search and interpretation. This constrained readiness

serves to focus what information cues are attended and how these cues









are interpreted. As a result, confirmatory bias may arise in

information search or information interpretation.

Confirmatory bias arises in information search as a tendency for

individuals to seek information cues that are consistent with existing

hypotheses rather than inconsistent. This type of strategy provides a

direct test of hypotheses. From a practical standpoint, individuals

usually have more experience with direct tests than indirect tests. The

empirical evidence in this area indicates that individuals search in a

focused (confirming) direction when (1) alternative hypotheses are not

apparent (cognitive factor) and/or (2) alternative hypotheses are not

preferred or desired (motivational factor).

Confirmatory bias arises in information interpretation as a

tendency for individuals to interpret information cues consistently. In

this case, the bias affects how disconfirming cues are processed. These

cues will be ignored or overlooked when processing requirements are

high. Otherwise, these cues will be explained away. Disconfirming cues

may be discounted (the inconsistency associated with these cues is

reduced) or assimilated to fit existing hypotheses (the inconsistency

associated with these cues is eliminated). The empirical evidence in

this area indicates that individuals ignore or explain away

disconfirming cues when (1) alternative interpretations are not apparent

(cognitive factor) and/or (2) alternative interpretations are not

preferred or desired (motivational factor).










Elements Relevant to the Study

The present study investigates the proposition that commitment to a

hypothesis leads to confirmatory bias arising in information

interpretation. Commitment was defined earlier as a psychological state

that establishes a reluctance to change one's position. Wicklund and

Brehm (1976) suggest that commitment leads to a motive for consonance as

long as responsibility can be assigned for particular actions or

judgments. In otherwords, responsibility provides for a crucial link

between commitment to a hypothesis and "closed mindedness." The authors

further assert that responsibility encompasses two components: choice

and forseeability. In particular, (1) individuals are free to choose

particular actions or judgments and (2) the consequences associated with

these actions or judgments are (or hindsight indicates that they should

be) reasonably foreseeable.

Responsibility can be readily assigned in the audit environment

simply by examining work papers. Included in a client's work papers are

(1) the audit procedures applied, (2) the tests performed, (3) the

information obtained, and (4) the pertinent conclusions reached in the

engagement (AICPA, 1983, section 339.03). These papers also allow for

the evaluation of an auditor's performance. In fact, professional

standards require a critical review of work performed and conclusions

drawn at every level of supervision (AICPA, 1983, section 210.05). As a

result, audit judgments are publicly disclosed and subject to

examination. These conditions are likely to induce commitment to one's

judgment or position.










Choice and forseeability are also implicit in the audit environment

as underlying components of responsibility. Although audit procedures

may be somewhat constrained, judgments are usually left to the auditor's

discretion. For the most part, the auditor is free to interpret

evidence as desired. In conjunction, the consequences associated with

judgments may be reasonably foreseeable. The structure of an audit is

such that forseeability is likely to follow. The audit process is so

interrelated that in many cases particular judgments dictate certain

consequences. Hence, choice and forseeability may facilitate a feeling

of responsibility, which in turn may lead to commitment.

Commitment may affect information search and/or information

interpretation such that confirmatory bias results. Although both tasks

assume important roles in the audit process, the present study focuses

on information interpretation. This processing task may be particularly

pertinent in light of existing audit technology. The inclusion of

decision aids in the audit program may sufficiently limit (on a routine

basis) the extent of search decisions that must be made. These aids may

structure and direct (restrict choice in) information search more than

information interpretation. Confirmatory search has been examined to

some degree in Libby (1984), while confirmatory interpretation has not

been addressed empirically in an auditing context. As such, this study

seeks to provide a foundation for explaining confirmatory bias that

arises in information interpretation (in an auditing context).










Research Hypotheses

Three specific research hypotheses are developed in the discussion

that follows. The first hypothesis is concerned with the general

occurrence of confirmatory bias. The second hypothesis posits a

relationship between the occurrence of the bias and specific behavior.

The third hypothesis concerns the cognitive mechanisms underlying

auditors' judgments.

Individuals who are committed to their hypotheses are more likely

to interpret information consistently with these hypotheses than

individuals who are not committed. Commitment to a hypothesis leads to

"closed mindedness" which is likely to affect information

interpretation. For individuals who are committed, inconsistent

information threatens existing hypotheses and creates unease--the

resulting psychological state is unpleasant. These individuals are

likely to have an intrinsic want or desire to confirm their hypotheses.

They may even go to great lengths (blatantly biased interpretation) to

dismiss inconsistent information. The findings in Lord et al. (1979)

and Elstein et al. (1978) may be attributed to commitment. In both

studies, subjects interpreted information as being largely consistent

with their hypotheses regardless of the nature of the information

(consistent or inconsistent). Subjects in Lord et al. (1979) were

likely to have been committed to their preexisting attitudes, and

subjects in Elstein et al. (1978) were likely to have been committed to

their expert diagnoses.

Auditors may also be susceptible to the effects of commitment. As

discussed earlier, this factor is likely to be present in an audit










setting. Audit judgments are fully documented and subject to review.

As a result, auditors are likely to feel responsible for their

judgments. This feeling of responsibility, in turn, is likely to lead

to commitment. Auditors may be compelled to maintain their positions.

They may bias their interpretations of inconsistent evidence. Formally,

the research hypothesis is

HI: Auditors who are committed to their hypotheses are more
likely to interpret evidence consistently with these
hypotheses than auditors who are not committed.

When the information to be attended to is prescribed, individuals

who are committed to their hypotheses are likely to show a preference

for examining potentially confirming information before other

information, whereas individuals who are not committed are not likely to

show any preference. Individuals who are committed have a greater need

to confirm their hypotheses than individuals who are not committed.

They are more likely to seek supportive (confirming) information as

opposed to any other information. Although individuals do not actually

know if an information cue is confirming until it has been examined,

they are likely to seek cues that are expected to be (potentially)

confirming. Elstein et al. (1978) have found such search behavior using

experienced internists as subjects. In that study, subjects were likely

to have been committed to their diagnoses and, as such, searched

primarily for information that would confirm these diagnoses.

Commitment may also have an effect on search when search is

prescribed. In the case of prescribed search, the particular

information cues to be attended to are predetermined (or designated).

Individuals have no control over the information that is encountered;










however, they may control the sequence in which it is encountered.

Individuals who are committed may initially avoid potentially

disconfirming information for fear that they have chosen a totally

inappropriate position. Therefore, these individuals are more likely to

examine potentially confirming information before examining potentially

disconfirming or neutral information. Confirming information is likely

to bolster existing hypotheses and strengthen individual positions.

Individuals who are committed are likely to feel less restrained about

examining potentially disconfirming information after potentially

confirming information has been examined. In contrast, individuals who

are not committed are not likely to have an intrinsic need to confirm

their hypotheses, and as such, they are not likely to show a preference

for the sequence in which information is examined.

Prescribed search may be particularly relevant in the present audit

environment. The inclusion of decision aids in the audit program may

largely constrain information search: the direction of search (the

specific cues to be examined) may be prescribed. As discussed above,

the presence of commitment is likely to influence the sequence in which

evidence is examined and interpreted (assuming that information is

interpreted at the time that it is attended). The research hypothesis

is

H2: When search is prescribed, auditors who are committed to
their hypotheses are likely to show a preference for
examining potentially confirming evidence before other
evidence, whereas auditors who are not committed are not
likely to show a preference.

Individuals who are committed to their hypotheses are more likely

to discount disconfirming information than individuals who are not









committed. Situational attribution is the underlying cognitive

mechanism when disconfirming information is discounted. Individuals who

are committed to their hypotheses have a need to explain away

disconfirming information, although they are not blind to this

information. In other words, the inconsistency associated with

disconfirming information (and existing hypotheses) is recognized and

considered. The presence of commitment is likely to compel individuals

to dismiss (explain away) this inconsistency: the inconsistency is

likely to be attributed to a factor that is not relevant to the decision

at hand. The empirical literature has provided little direct evidence

as to the cognitive mechanisms underlying confirmatory bias, although

the findings in Lord et al. (1979) and Elstein et al. (1978) may be

explained by a process of discounting disconfirming information. The

results in both studies indicate that disconfirming information was

likely to have been explained away, but the particular mechanism

underlying this process was not assessed. Other research (i.e., Hayden

and Mischel, 1976; Darley and Gross, 1983) simply has investigated the

occurrence of confirmatory bias (this was the stated purpose) without

making a distinction between the factors (cognitive and motivational)

that drive the bias. Nonetheless, commitment is likely to trigger

situational attribution as an underlying cognitive mechanism.

Auditors who are committed to their hypotheses are also likely to

discount disconfirming evidence. The ability to discount this evidence

is necessary for auditors to justify their positions. Gibbins (1984)

has reported that auditors feel a need to be able to justify their





42



positions. The presence of commitment only serves to heighten this

need. Formally, the research hypothesis is

H3: Auditors who are committed to their hypotheses are more
likely to discount disconfirming evidence than auditors
who are not committed.

















CHAPTER III


RESEARCH METHODOLOGY


A laboratory experiment was conducted to test the specific research

hypotheses developed in Chapter II. Simply put, the study investigates

the proposition that in auditing contexts, a motivational factor

(commitment) may lead to confirmatory bias arising in information

interpretation. Specifically, auditors who are committed to their

hypotheses may be more likely to interpret evidence consistently with

these hypotheses than auditors who are not committed. The cognitive

mechanisms that underlie auditors' judgments are also explored.

External auditors participated in the study. They were asked to

complete a three-part task involving audit planning and internal

control. The primary part of the task required subjects to formulate

and assess a hypothesis about a potential problem area. Subjects were

allowed to select one of two hypotheses. Subjects in the commitment

group were required to justify their selections, while subjects in the

noncommitment group were not required to provide justifications.

Subsequently, all subjects were asked to examine several pieces of audit

evidence in light of their hypotheses. The evidence presented was not

conclusive about the problem area. Subjects then were asked to make a

final assessment concerning their hypotheses. The extent to which final










assessments favored subjects' hypotheses was used as an indicator of the

occurrence of confirmatory bias.

The remainder of this chapter is organized into five sections. The

experimental design subjects, procedure, and task employed are discussed

in the first four sections. The specific (operational) variables used

to assess the research hypotheses are identified in the final section.


Experimental Design

A 2x2 factorial design was used. The two independent variables

were the priming of a hypothesis and level of commitment. The priming

of a hypothesis is a subtle method used to increase the salience of that

hypothesis. All treatment subjects were primed for one of two

hypotheses, where each hypothesis was identified as a potential problem

area. Subjects were required to select one of these hypotheses as

being the most likely source for an unexpected fluctuation in gross

margin. The manipulation was intended to induce subjects to select the

hypothesis that had been primed. The generalizability of the study is

enhanced if a reasonable number of subjects select each hypothesis. A

control group was also included, where subjects were not primed for a

possible hypothesis. This group would allow for a comparison between

primed and unprimed subjects.

The commitment variable was the primary variable of interest. A

high level of commitment should lead to a reluctance to reject one's

position. One half of the treatment subjects were included in a

commitment (COM) group and the other half were included in a

noncommitment (NCOM) group. Subjects in the COM group were required to










prepare a written justification indicating why they selected a

particular hypothesis. Subjects in the NCOM group were not required to

prepare this justification.


Subjects

The managing partners from sixteen public accounting offices agreed

to assist in obtaining subjects for this study. The offices represented

six Big Eight and two local firms throughout Florida. Subjects with a

minimum of one year's audit experience were requested. Although

subjects were required to assume the role of an audit senior, the task

judgments were rather straightforward. The feeling was that experienced

staff personnel would be familiar with these judgments, and as such,

they were included in the sample.

Seventy-nine practicing external auditors took part in this study.

Their audit experience ranged from three months to 7.50 years, with an

average of 2.48 years. Seventy-five subjects had at least one year's

audit experience, and fifty-eight subjects had received professional

certification.


Procedure

Arrangements were made with the managing partners of the

participating offices for coordination of the task. In all cases, the

task was administered at specific firm offices. It was administered to

subjects in groups of two to six and took approximately 40 minutes to

complete.

In terms of procedure, subjects at each office were gathered and

taken to an enclosed area. Subjects initially were asked to write their










names on a sheet of paper that was passed around. They were told that

their names were being collected for the researcher's personnel records;

however, knowledge of individual names was necessary for the commitment

manipulation (discussed further below). Next, the task materials were

distributed. Subjects were asked to complete two booklets. They were

informed that the second booklet would not be passed out until the first

one had been completed.

Subsequently, a brief, verbal introduction to the task was given.

Subjects were told that the objective of the study was to obtain

information as to how various judgments are made that affect audit

planning and internal control. In addition, they were informed that

different versions of the task had been distributed. They were told of

these differences in very general terms ("tasks are similar but not

identical") to eliminate any possible confusion. Confusion may have

arisen if subjects had not been informed of these general differences,

but noticed that their tasks were not the same as other subjects' tasks.

However, specific differences (what each manipulation entailed) were not

likely to be apparent and, as such, demand characteristics of the

experiment should have been unaffected. Finally, subjects were thanked

and told to begin.


Task

The experimental task consisted of three parts. In part one,

subjects were primed for a particular hypothesis. In part two, a

hypothetical case setting was presented. In the final part, subjects










were asked to complete a debriefing questionnaire. All three parts are

discussed below.

Initially, subjects were presented with internal-control

information for a particular transaction cycle. This information was

very general and did not relate to any specific firm. Pretests

indicated that specific information may lead to serious demand effects.

The same information was presented in three different forms (see

Appendix A): a narrative form, a checklist form, and a grouped (by

internal-control strengths and weaknesses) form. Subjects were asked to

indicate which form they preferred. Subjects were also asked to

evaluate the overall strength of the internal-control system. This

evaluation was included to ensure that subjects read the content of the

information presented as opposed to simply examining the form. The

content was intended to prime a particular hypothesis, and as such, it

needed to be read. The priming procedure was very subtle; however,

protests indicated that a subtle prime was necessary, otherwise subjects

were likely to see right through the procedure. The content of the

prime was also manipulated (see Appendices A, B, and C). One half of

the treatment subjects received information about the sales and

receivables cycle (SRC), while the other half received information about

the purchases and payables cycle (PPC). In both cases, the prime was

intended to affect the subsequent selection of a hypothesis. Subjects

in the control group received information about the payroll and

personnel cycle. This cycle did not represent a possible hypothesis

and, as a result, should not have affected the subsequent selection. At

this point, part one of the task was completed.










In part two, subjects were asked to play the role of an audit

senior (see Appendices D and E). They were provided with background

information on a hypothetical client. In particular, they were told

that the client had had noncompliance problems with several internal

controls. These problems had occurred primarily in the SRC and the PPC.

In addition, subjects were informed that a total of 100 man hours had

been budgeted for the investigation of these two cycles. They were

asked to allocate the total number of hours between the two cycles;

however, subjects were told that the alternatives were limited to the

following allocation schemes:

65% PPC 60% PPC 55% PPC 50% PPC 55% SRC 60% SRC 65% SRC
35% SRC 40% SRC 45% SRC 50% SRC 45% PPC 40% PPC 35% PPC

These alternative allocation schemes captured the responses of nearly

90% of pretestt) subjects who performed the task without any constraints

(that is, could allocate hours in anyway desired). The primary reason

for constraining the choice of allocation schemes in the main experiment

was that subjects had relatively little firm-specific information to use

as a basis for their allocations. A secondary reason for this

constraint was to prevent subjects from choosing a very extreme

allocation scheme (more than 65% for one cycle and less than 35% for the

other cycle). The choice of such an allocation scheme would virtually

prohibit a finding of confirmatory bias, given the experimental task.

However, the selection of a very extreme allocation scheme should not be

expected realistically (given all the facts) when assigning hours

between the SRC and the PPC--both cycles are important. The allocation

schemes selected were taken as a measure of subjects' prior beliefs.









Subjects were informed that an unexpected fluctuation in gross

margin had been uncovered during preliminary analytical review. They

were told that their objective should be to determine the source of this

fluctuation. Subjects were asked to indicate which of the two specified

areas, SRC or PPC, was more likely to underlie the fluctuation.

Subjects' choices were used as a basis for identifying their hypotheses.

After completing this procedure, subjects then returned the materials to

the experimenter in exchange for materials presented in a second

booklet.

At this time, commitment was manipulated. Subjects in the

commitment group were asked to prepare a written memo explaining why

they selected a particular cycle as being the most likely source for the

unexpected fluctuation in gross margin. They were told that their

explanations would be discussed, at a later date, with representatives

from their firms or offices (as part of a second research project). To

induce commitment, each subject's name and firm affiliation appeared at

the top of the page on which subjects wrote their explanation. Names

and affiliations had been handwritten by the experimenter while the

subjects were working with the first booklet. Commitment is likely to

arise since subjects are overtly taking a position which is to be

reviewed by their peers. Subjects in the noncommitment groups were not

required to prepare a written memo, and their names and firm

affiliations were not included on the second booklet.

All subjects had been provided with a large envelope labeled "Other

Materials" and a plastic container at the beginning of the task.

Subjects were informed that the large envelope contained ten small









envelopes relating to compliance-test results for ten internal-control

improvements that had been adopted by management. In particular, four

improvements had been made in the SRC, four in the PPC, and two in the

payroll and personnel cycle (PYC). Nonstatistical sampling had been

conducted to assess the reliability of these controls. Each sampling

result was included in a small envelope labeled by the control being

tested. Subjects were instructed to open the large envelope and examine

the sampling results one at a time. They were told that the sampling

results could be examined in any order, but that they could not refer

back to earlier sampling results. Subjects were instructed to place

each small envelope in the plastic container after the result had been

examined. This container was used to maintain the order that sampling

results were examined. The order that each result was attended could

not be mixed up after it was placed in this container.

The sampling results that were used in the task are presented in

Table 1. Each result is referred to as an information cue. The SRC and

PPC cues should be relevant for uncovering the source of the unexpected

fluctuation in gross margin. The PYC cues should be irrelevant or

neutral. Exceptions in the SRC or the PPC may have a direct affect on

gross margin, whereas exceptions in the PYC will not have such an

effect. The SRC and PPC cues were also symmetrical. That is, the tests

conducted in each cycle and the corresponding sample sizes and exception

rates were the same. A careful examination of Table 1 reveals this

symmetry. Since the cues were symmetrical, they should not have given a

clear indication as to the source of the unexpected fluctuation. The

cues were intended to be inconclusive.










TABLE 1: Compliance Tests and Sampling Results


1. Test: A sample of 60 sales invoices was examined for proper
authorization.
Result: Four exceptions were found. These invoices were not
approved in accordance with company policy.
2. Test: A numerical sequence of 40 sales invoices was examined to
account for the integrity of this sequence.
Result: Two exceptions were found. In this case, duplicate sales
invoices were uncovered.
3. Test: A sample of 50 sales invoices was examined for internal
verification.
Result: One exception was found. This invoice was not initialled
to ensure a proper recording in the sales journal.
4. Test: A sample of 70 shipping reports was examined for internal
verification.
Result: Five exceptions were found. These reports were not
initialled to ensure that the terms listed were the same
as those listed on sales invoices.
5. Test: A sample of 60 purchase orders was examined for proper
authorization.
Result: Four exceptions were found. These orders were not
approved in accordance with company policy.
6. Test: A numerical sequence of 40 purchase orders was examined
to account for the integrity of this sequence.
Result: Two exceptions were found. In this case, purchase orders
were missing.
7. Test: A sample of 50 vendors' invoices was examined for
internal verification.
Result: One exception was found. This invoice was not initialled
to ensure a proper recording in the purchases journal.
8. Test: A sample of 70 purchase orders was examined for internal
verification.
Result: Five exceptions were found. These orders were not
initialled to ensure that the terms listed were the same
as those listed on vendors' invoices.


9. Test:

Result:


10.


A sample of 40 time cards was examined for proper
authorization.
No exceptions were found.


Test: A sample of 60 payroll transactions was examined for
internal verification.
Result: Seven exceptions were found. These transactions were not
initialled to ensure that employees' earnings records
were correct.










In terms of the analysis, the SRC and PPC cues were classified as

confirming or disconfirming depending on the cycle selected as the most

likely source for the unexpected fluctuation in gross margin. Cues that

involved the same cycle as the one selected were classified as

confirming and cues that involved the other cycle were classified as

disconfirming. For example, if a subject selected the SRC then cues 1-4

from Table 1 were classified as confirming and cues 5-8 were classified

as disconfirming. Cues 9 and 10 were always classified as neutral,

regardless of the cycle selected.

After examining the sampling results, subjects were informed that

an additional 20 man hours had been budgeted for further investigation

of the SRC and the PPC. Pretests indicated that subjects perceived 20

additional hours as being appropriate for the investigation: the mean

and median responses of pretest subjects were 21.6471 and 20.0000 hours,

respectively. Subjects in the main study were told that the additional

hours were budgeted as the result of a growing concern over the

unexpected fluctuation in gross margin. They were asked to allocate

these additional hours between the two cycles. Again, subjects were

told that their supervisors would only approve certain allocations. The

allocation schemes (and the underlying reasons for their use) were the

same as presented earlier. The number of additional hours allocated to

the cycle selected as the most likely source for the unexpected

fluctuation (20 hours x % hours allocated to the cycle selected) was

used as the primary dependent variable in the data analysis. At this

point, the second part of the task was completed.










In part three, subjects were asked to complete a debriefing

questionnaire (see Appendix F). The first few questions were intended

to collect general information. After these questions had been

answered, subjects were asked to indicate whether (1) more exceptions

were uncovered in the SRC, (2) more exceptions were uncovered in the

PPC, or (3) the same number of exceptions were uncovered in both cycles.

The general impressions that subjects formulate as to the most likely

source for the unexpected fluctuation in gross margin should be captured

in their responses to this question. Next, ten cued-recall questions

were presented. Subjects were given a list of the controls

(improvements) that had been tested and asked to indicate the number of

exceptions that were uncovered in each of the sampling results. The

presentation of this list was reversed to allow for a test of a possible

order effect (see Appendix G). Subjects also were asked to indicate

their levels of confidence in these answers. These responses were given

on a seven-point scale anchored by very low confidence and very high

confidence.

Subsequently, subjects were asked to indicate how important each

sampling result could have been in producing the unexpected fluctuation

in gross margin. Again, they were given a list of the ten controls that

had been tested and asked to respond. As with recall, the presentation

of this list was reversed (see Appendix H). Subjects' responses were

given on a seven-point scale anchored by extremely unimportant and

extremely important.

Finally, subjects were asked to provide an allocation of the total

120 hours (the 100 originally budgeted plus the 20 additionally










allotted) given all the information available to them. These responses

may provide insight into subjects' responses for additional hours.

Again, subjects' responses were constrained to the seven allocation

schemes.


Testing the Hypotheses

The research hypotheses are assessed in terms of the laboratory

experiment described above. Commitment to a hypothesis was the primary

independent variable. Subjects in the COM group were required to

prepare a written memo justifying their selections of hypotheses.

Subjects in the NCOM and control groups were not required to prepare a

written justification. The particular dependent measures used to test

the hypotheses are discussed below.

Research hypothesis one states that auditors who are committed to

their hypotheses are more likely to interpret evidence consistently with

these hypotheses than auditors who are not committed. The extent to

which information cues were interpreted consistently with existing

hypotheses was inferred from subjects' allocations of additional audit

hours. Subjects who were committed to their hypotheses should have

allocated more additional hours to the cycle selected than subjects who

were not committed.

Research hypothesis two states that when search is prescribed,

auditors who are committed to their hypotheses are likely to show a

preference for interpreting confirming evidence before other evidence,

whereas auditors who are not committed are not likely to show a

preference. Subjects were required to examine ten information cues.










Eight cues were classified as confirming and disconfirming according to

the cycle (hypothesis) selected by each subject. The other two cues

were always classified as neutral. The sequence in which information

cues were examined was determined by the order that sampling results

(contained in small envelopes) were placed in the plastic container.

Subjects who were committed to their hypotheses should have examined

confirming cues before other cues, whereas subjects who were not

committed should not have shown any preference for the sequence in which

cues were examined.

Research hypothesis three states that auditors who are committed to

their hypotheses are more likely to discount disconfirming evidence than

auditors who are not committed. The cognitive mechanisms underlying

subjects' judgments were inferred from their responses to recall and

importance questions. Disconfirming evidence is elaborated and

processed at a deeper level than confirming evidence when it is

discounted (a motivational factor drives this process). In turn, the

weight assigned to disconfirming evidence is reduced. Hastie (1980) and

Hastie and Kumar (1979) have suggested that recall will be superior for

information processed at a deeper level as compared to a shallower

level. Subjects who were committed to their hypotheses should have

recalled more disconfirming cues than subjects who were not committed.

These subjects also should have assigned less importance to

disconfirming evidence than subjects who were not committed.

















CHAPTER IV


DATA ANALYSIS AND RESULTS


This chapter describes the analysis and presents the results of the

study. The first section investigates the occurrence of confirmatory

bias. The second section explores the cognitive mechanisms underlying

auditors' judgments. Each of the sections explains the results in terms

of the specific research hypotheses.


The Occurrence of Confirmatory Bias

Treatment subjects were primed for one of two hypotheses. The

prime was intended to affect the cycle (hypothesis) selected by each

subject. The dependent measures, however, were not expected to differ

(between subjects) as a result of the cycle selected. Level of

commitment was also manipulated. This manipulation was expected to

affect (1) subjects' allocations of additional audit hours and (2) the

sequence in which subjects examined and interpreted confirming and

disconfirming cues. The remainder of the section is organized as

follows: first the results of a manipulation check for the prime are

presented; and then the findings for HI and H2 are discussed.


Manipulation Check for the Prime

An extensive manipulation check was performed to determine the

effects) of the prime. The success of this manipulation (inducing










subjects to select the cycle that had been primed) was assessed with a

contingency table (Mendenhall et al., 1981, p. 564). A X statistic was

computed and used to test for independence between the cycle primed and

the cycle selected. The test results show that subjects selected

hypotheses independently of the cycle that was primed (see Table 2).

Subjects were not affected differently by the nature of the prime (SRC
2
or PPC). A second contingency table was constructed and another X

statistic computed to test for differences in the selection patterns of

control unprimedd) subjects and treatment (primed) subjects. The

results show no differences (see Table 3). The prime did not affect the

cycle selected, and as a result, the manipulation was unsuccessful.

This finding suggests that the cycle selected should be examined in

subsequent analysis to test for differences between the SRC and the PPC.


TABLE 2

Contingency Table (Prime SRC vs PPC)

Prime SRC Prime PPC Significance Level
Select SRC 12(11.6825)* 11(11.3175) NS
Select PPC 20(20.3175) 20(19.6825)
*Numbers in parentheses represent expected cell means.

TABLE 3

Contingency Table (Control vs Treatment)

Control Treatment Significance Level
Select SRC 3(5.2658)* 23(20.7342) NS
Select PPC 13(10.7342) 40(42.2658)
*Numbers in parentheses represent expected cell means.

A chi-square goodness-of-fit test was also performed to determine

if one hypothesis was more likely to be selected than the other

hypothesis. The results show that subjects were more likely to select










the PPC than the SRC (see Table 4). This finding applies even though

the same number of treatment subjects were primed for the PPC as the

SRC.


TABLE 4

Chi-Square Goodness-of-Fit Test for Cycle Selected

Select SRC Select PPC Significance Level

Observed Cell
Frequencies 26 53 0.005
Expected Cell
Frequencies 39.50 39.50

Although the prime did not affect the cycle selected, it may have

affected the allocation scheme chosen for originally budgeted hours. In

fact, subjects were asked to choose this scheme before selecting a

hypothesis. A one-way analysis of variance (ANOVA) was conducted to

test for a main effect of the prime. The number of budgeted hours

allocated to the cycle selected (BUDHRS) was used as the dependent

variable. The following model was run on the treatment subjects:

y.. = u + a. + E..
1 1 1]
where

Y.. is the jth observed response (BUDHRS) for the ith level of the
13
prime,

u is the population mean,

a. is the main effect of the ith level of the prime, and

E.. is the random error associated with the jth response.

The results show that the prime did not have an effect on subjects'

responses for BUDHRS (see Table 5). Subjects who were primed for the

SRC and the PPC then were collapsed into one group, and the analysis was










repeated for all subjects. This analysis allowed for a comparison

between treatment (primed) subjects and control unprimedd) subjects.

Once again, the results show that the prime did not have an effect (see

Table 6).


TABLE 5

ANOVA for BUDHRS (n = 63)

Source DF SS F value PR > F
Prime 1 21.9038 0.28 0.5983
Error 61 4763.8105 -
R2 = 0.0046

TABLE 6

ANOVA for BUDHRS (n = 79)

Source DF SS F value PR > F
Prime 1 121.0697 1.57 0.2143
Error 77 5946.6518 -
R2 = 0.0200

As a precaution, the ANOVAs discussed above were run again with the

number of additional hours allocated to the cycle selected (ADDHRS) as

the dependent variable. The objective was to determine if the prime had

an effect on subjects' responses for ADDHRS. The COM treatment subjects

were excluded from the analysis. Their responses were expected to

differ as a result of the commitment manipulation. First, the ANOVA was

conducted for NCOM treatment subjects. No differences were found (see

Table 7). As a result, these subjects were collapsed into one group,

and the ANOVA was repeated for NCOM subjects and control subjects. As

before, no differences were found (see Table 8). The prime (SRC, PPC,

or unprimed) did not affect subjects' responses for ADDHRS.










TABLE 7

ANOVA for ADDHRS (n = 32)

Source DF SS F value PR > F
Prime 1 5.2813 1.65 0.2086
Error 30 95.9375 -
R2 = 0.0522

TABLE 8

ANOVA for ADDHRS (n = 48)

Source DF SS F value PR > F
Prime 1 3.7604 1.04 0.3130
Error 46 166.2188 -
R2 = 0.0221

Though the prime generally had no affect, one anomaly was

uncovered. Control subjects' responses for ADDHRS were significantly

correlated with BUDHRS, whereas treatment subjects' responses for ADDHRS

were much less correlated with BUDHRS. The Pearson correlation

coefficients between ADDHRS and BUDHRS for control, NCOM treatment, and

COM treatment subjects, were .7963 (p<.0002), .3762 (p<.0365), and

-.2876 (p<.1167) respectively. No obvious explanation for this finding

is evident.


Findings for HI

Hypothesis one states that auditors who are committed to their

hypotheses are more likely to interpret evidence consistently with these

hypotheses than auditors who are not committed. Operationally, HI

suggests that subjects who are committed will allocate more additional

hours to the cycle selected than subjects who are not committed. A

two-way ANOVA was conducted to assess this hypothesis. Level of

commitment was the primary variable of interest. A significant effect










was expected for this factor. The cycle selected was also included in

the analysis, but no significant effect was expected for this factor.

The variable was added to the model, as a blocking factor, after the

priming manipulation was determined to be unsuccessful. The number of

additional hours allocated to the cycle selected was used as the

dependent variable (ADDHRS). Although subjects' responses were provided

in terms of percentages, ADDHRS was calculated simply by multiplying the

number of additional hours available by the percentage allocation

chosen. As an example, if a subject selected the SRC and chose 60%

SRC/40% PPC as an allocation scheme for additional hours, the dependent

variable would be 12 hours (20 hours x 60%). Since subjects' responses

were constrained to one of seven allocation schemes, the dependent

variable ranged from 7 hours to 13 hours. Hypothesis one suggests that

ADDHRS will be greater for subjects who are committed than those who are

not committed. Mean scores for ADDHRS are presented in Table 9. The

frequency of responses is shown in Figure 4. On average, subjects who

were committed to their hypotheses allocated more additional hours to

the cycle selected than subjects who were not committed. Subjects also

allocated more additional hours to the SRC than the PPC. Further

analysis was conducted to determine if these differences were

significant.










TABLE 9

Mean Scores for ADDHRS


COM NCOM CONTROL Totals

Select SRC 11.3636 10.5833 12.3333 11.1154
n=ll n=12 n=3 n=26

Select PPC 10.8500 9.1060 9.7692 9.9245
n=20 n=20 n=13 n=53

Totals 11.0323 9.6563 10.2500
n=31 n=32 n=16

Before ANOVA was performed (and H1 assessed), the appropriateness

of the method given the data was considered. Subjects' responses for

ADDHRS do not appear to be normally distributed (see Figure 4), and

normality is an underlying assumption of ANOVA. However, the F test for

equality of factor level means is largely unaffected (either in terms of

the level of significance or power of the test) by lack of normality.

The robustness of the F test in the face of departures from normality is

well documented (see Neter and Wasserman, 1974, pp. 513-514). In

addition, a test was performed to check for homogeneity of variances

between treatment groups. Scheffe's test was used since it is not

sensitive to departures from normality (Winer, 1971, p. 219). The

calculated test statistic was 2.1023 and the critical value for

F 90(3,12 was 2.6100. Accordingly, the null hypothesis (that the

variance between treatment groups are the same) was not rejected.





63




FREQUENCY

19 *
I*
18 *
I *
17 *
I*
16 *
I *
15 *
*
14 *
I *
13 *
I *
12 *
i *
11 *
I *
10 *
7- *
9 *
I *
8- *
*
7 *
j *
6- *
j *
5 *
*
4 *
I *
3 *
I *
2- *
S *
1- *
ADDHR
7 8 9 10 11 12 13


Figure 4: Frequency of Responses for ADDHRS










The following model was run:

Y. = u + a. + 8. + (aB). + .
ijk i j 1j ijk
where

Y.k is the kth observed response (ADDHRS) for the ith level of the

cycle selected and the jth level of commitment,

u is the population mean,

a. is the main effect of the ith level of the cycle selected,

8. is the main effect of the jth level of commitment,

(aB).. is the interaction effect, and

..k is the random error associated with the kth response.

The model was run for (1) treatment subjects and (2) all subjects

(control subjects who were included in this analysis were all

partitioned into the noncommitment treatment group). The results were

similar in both cases. Commitment and the cycle selected each had a

significant effect (see Tables 10 and 11). The two analyses were

repeated without the four subjects who had less than a year's audit

experience. The results were unaffected by the exclusion of these

subjects.


TABLE 10

ANOVA for ADDHRS (n = 63)

Source DF SS F value PR > F
Commitment 1 23.3459 7.92 0.0066
Select 1 14.5415 4.94 0.0302
Commitment
x Select 1 3.4288 1.16 0.2851
Error 59 173.8121 -
R2 = 0.2171










TABLE 11

ANOVA for ADDHRS


(n = 79)


Source DF SS F value PR > F
Commitment 1 15.4432 4.96 0.0290
Select 1 18.2458 5.86 0.0179
Commitment
x Select 1 4.6884 1.50 0.2238
Error 75 233.6652 -
2
R = 0.1861

Although HI predicts a significant effect for commitment, a

significant effect for the cycle selected was not expected.

Accordingly, univariate analyses were undertaken to investigate the

specific effect of commitment for each cycle. The main finding was that

commitment had a significant effect only when the PPC was selected (see

Tables 12 and 13). These results did not change when the four subjects

with less than a year's audit experience were excluded from the sample.

The findings for each univariate analysis are discussed further below.


Source
Commitment
Error
2
R = 0.0660


Source
Commitment
Error
R2 = 0.1976
R = 0.1976


TABLE 12

ANOVA for ADDHRS by CYCLE (n = 63)

Cycle Selected = SRC (n = 23)

DF SS F value
1 3.4944 1.48
21 49.4621 -


Cycle Selected = PPC (n = 40)

DF SS F value
1 30.6250 9.36
38 124.3500 -


PR > F
0.2367
-


PR > F
0.0041










TABLE 13

ANOVA for ADDHRS by CYCLE (n = 79)

Cycle Selected = SRC (n = 26)

Source DF SS F value PR > F
Commitment 1 1.1751 0.47 0.4977
Error 24 59.4788 -
R2 = 0.0194
Cycle Selected = PPC (n = 53)

Source DF SS F value PR > F
Commitment 1 27.5117 8.06 0.0065
Error 51 174.1864 -
R2 = 0.1364

When the PPC was selected, commitment had a significant effect;

however, a comparison of cell means was necessary to determine if this

result is consistent with H1. The COM subjects should have assigned

more additional hours to the cycle selected than the NCOM subjects. The

cell mean for COM subjects was compared with the means for two groups of

NCOM subjects: treatment subjects (NCOM) and treatment plus control

subjects (NCOM+). The results indicated that the mean for COM (10.8500)

was significantly greater (p<.05) than the means for NCOM (9.100) and

NCOM+ (9.3636) subjects. This finding is consistent with HI.

Secondary analysis was performed to investigate the extent to which

subjects interpreted information cues as being either consistent or

inconsistent with their hypotheses. Comparisons were made between the

percentage of additional hours (%ADDHRS) allocated to the cycle selected

(made after the information cues were presented) and the percentage of

budgeted hours (%BUDHRS) allocated to that cycle (made initially after a

cycle was selected as the most likely source for the unexpected

fluctuation). The difference between these responses (%ADDHRS-%BUDHRS)










may provide insight into the consistency or inconsistency of information

interpretation. A positive difference (%ADDHRS-%BUDHRS>O) suggests that

interpretation is strongly consistent, whereas a negative difference

(%ADDHRS-%BUDHRS<0) suggests that interpretation is strongly

inconsistent. No difference between these responses (%ADDHRS-%BUDHRS=0)

is ambiguous as to interpretation of information cues. In this case,

interpretation may be slightly consistent, slightly inconsistent, or

indifferent. T tests were conducted to determine if %ADDHRS-%BUDHRS was

significantly different from zero for subjects in the COM, NCOM, and

NCOM+ groups, respectively. The results are shown in Table 14. The COM

subjects showed a marginally-significant, positive difference. A closer

inspection of the data revealed that %ADDHRS-%BUDHRS was negative for

five of the twenty subjects in this group, and in each case, this

difference was at least 10. These responses were likely to have

contributed to the lack of an interaction effect in the overall

analysis. The difference was positive for all other subjects except

one, where it was zero. A sign test (Conover, 1980) was performed to

determine if this difference was more likely to be positive than

negative. This proposition was accepted at a significance level of .05.

The COM subjects were more likely to interpret evidence consistently

with their hypotheses than inconsistently.


TABLE 14
T Tests for %ADDHRS-%BUDHRS (PPC)

Group n Means Std Dev T PR > 1T1
COM 20 6.2500 14.5886 1.92 0.0705
NCOM 20 -4.7500 10.5724 -2.01 0.0589
NCOM+ 33 -2.4242 9.3643 -1.49 0.1468










The NCOM subjects showed a marginally-significant, negative

difference for %ADDHRS-%BUDHRS. An examination of the data indicated

that this difference was negative for twelve subjects, positive for five

subjects, and zero for three subjects. A sign test was performed to

determine if %ADDHRS-%BUDHRS was more likely to be negative than

positive. The proposition is accepted at a marginal level of

significance (p<.0835). This finding is consistent with the result of

the t test. Finally, %ADDHRS-%BUDHRS was not significantly different

from zero for NCOM+ subjects. The inclusion of control subjects drives

this result. The difference between %ADDHRS and %BUDHRS was zero for

seven control subjects, positive for three control subjects, and

negative for three control subjects.

For subjects who selected the SRC, commitment did not have a

significant effect. The mean responses for ADDHRS were 11.3636,

10.5833, and 10.9333 for COM, NCOM, and NCOM+ subjects, respectively.

The differences were not statistically significant. This finding is not

consistent with HI. Secondary analysis was conducted to determine if

%ADDHRS-%BUDHRS was significantly different from zero for COM, NCOM, and

NCOM+ subjects. The results of t tests are shown in Table 15. The

difference between %ADDHRS and %BUDHRS was exactly zero for COM

subjects. The difference was marginally significant and negative for

NCOM subjects and insignificant for NCOM+ subjects. Sign tests were

also performed. The difference between %ADDHRS and %BUDHRS was no more

likely (p<.10) to be positive than negative for COM subjects. The

difference was positive for four subjects, negative for five subjects,

and zero for two subjects. In contrast, the difference was more likely










to be negative than positive for NCOM subjects at a significance level

of .05. The difference was positive for one subject, negative for seven

subjects, and zero for four subjects. Finally, the difference was no

more likely (p<.10) to be positive than negative for NCOM+ subjects.

The inclusion of control subjects drives this result also. The

difference between %ADDHRS and %BUDHRS was positive for two control

subjects and zero for one control subject.


TABLE 15

T Tests for %ADDHRS-%BUDHRS (SRC)

Group n Means Std Dev T PR > 1T1
COM 11 0.0000 12.4499 0.00 1.0000
NCOM 12 -5.0000 9.5346 -1.82 0.0966
NCOM+ 15 -2.6677 10.1536 -1.02 0.3263

Subjects' responses for ADDHRS were likely to be affected by their

responses for BUDHRS. Nonexperimental factors may have caused subjects

who selected the SRC to respond differently for BUDHRS than subjects who

selected the PPC. Mean scores for BUDHRS are presented in Table 16.

The distribution of responses for subjects who selected the SRC and the

PPC are shown in Figures 5 and 6, respectively. Responses that were

subsequently revised upward (%ADDHRS-%BUDHRSO0), downward

(%ADDHRS-%BUDHRSO0), and not at all (%ADDHRS-%BUDHRS=0) are also

identified in Figures 5 and 6.










TABLE 16

Mean Scores for BUDHRS


COM NCOM CONTROL Totals

Select SRC 56.8182 57.9167 55.0000 57.1154
n=ll n=12 n=3 n=26

Select PPC 48.0000 50.2500 47.6923 48.7736
n=20 n=20 n=13 n=53

Totals 51.1290 53.1250 49.0625
n=31 n=32 n=16

For the most part, subjects allocated at least as many budgeted

hours to the SRC as the PPC. Moreover, eighteen of twenty-six subjects

who selected the SRC (including seven of eleven COM subjects) gave an

original allocation of at least 60% SRC/40% PPC (the second most extreme

allocation). Only one of these eighteen subjects subsequently revised

the chosen allocation scheme upward. Subjects who selected the SRC

apparently allocated such a large percentage of budgeted hours to that

cycle that no "room" was left for upward adjustment. On the other hand,

only eleven of fifty-three subjects who selected the PPC gave an

original allocation as extreme as 60% PPC/40% SRC. An inspection of

Figures 5 and 6 indicates that responses were much more evenly

distributed for subjects who selected the PPC than those who selected

the SRC. Consequently, subjects who selected the PPC had "room" for

either upward or downward adjustment.

The Pearson correlation coefficients between %BUDHRS and

%ADDHRS-%BUDHRS for subjects who selected the SRC and the PPC were

-.7081 (p>.0001) and -.5958 (p>.0001), respectively. Subjects who chose

a relatively high %BUDHRS for the cycle selected were less likely to











FREQUENCY
13 0
0

12 0
O

11 0
0

10 0
0
O

9- N
N



N

7 N
N
N

6 N
N
N

5 N 0
N O

4 N O
N O

3- N N N
N N N

2- P 0 P N N
P O P N N

1- P P P P P N
P P P P P N

35 40 45 50 55 60 65 BUDHRS


represents
represents
represents


responses
responses
responses


subsequently revised upward.
subsequently revised downward.
that were not subsequently revised.


Figure 5: Frequency of Responses for BUDHRS: Cycle Selected=SRC











FREQUENCY
11 -



10 -



9 -



8 -



7 -



6 -



5 -



4-



3 -



2 -


1 -


P represents responses
N represents responses
0 represents responses

Figure 6: Frequency of


subsequently revised upward.
subsequently revised downward.
that were not subsequently revised.

Responses for BUDHRS: Cycle Selected=PPC


O 0
0 0

0 0 O
O 0 0

O O 0
0 0 0

O N N
O N N

N N N
N N N

N N N
N N N

P P N
P P N

P P N
P P N

P P P
P P P

P P P
P P P

P P P
P P P

40 45 50


65 BUDHRS









increase that percentage than were subjects whose %BUDHRS were lower.

Consequently, the lack of a commitment effect for subjects who selected

the SRC may be due to a ceiling effect.

As a precaution, the reasons that subjects provided for selecting a

particular cycle were examined to assess the validity of the commitment

manipulation. The COM subjects who selected the SRC may not, in fact,

have been committed to their hypotheses. These subjects may have been

reluctant to select a particular cycle since very little background

information had been given. As such, they may have hedged themselves in

their written responses. They may have expressed doubt in their initial

selections and/or they may have given reasons supporting both cycles.

An independent rater read the responses for the thirty-one COM subjects

and coded them as (1) committed to the SRC, (2) committed to the PPC, or

(3) not committed to a particular cycle. Twenty-six subjects were coded

as committed to the cycle that they initially selected. The other five

subjects were coded as not committed to a particular cycle. Two of

these subjects selected the SRC and the remaining three selected the

PPC. The univariate analyses discussed above were repeated deleting

these five subjects from the sample. The results, however, were

unaffected. Commitment had a significant effect for subjects who

selected the PPC and no effect for subjects who selected the SRC.

While the results were consistent with HI when the PPC was

selected, an alternative explanation to HI should be acknowledged.

Subjects' allocations of additional hours may have been compensating for

original allocations that were too extreme in favor of the cycle that










was not selected. This explanation can be assessed by examining

subjects' allocations of total hours (TOTHRS), which were collected at

the end of the task. These allocations should indicate the cycle that

subjects favor (the one that they believe should be assigned more hours)

after viewing all of the evidence. A compensating response was defined

as a response where ADDHRS favored (more hours were assigned to) a

different cycle than both TOTHRS and BUDHRS. Only four out of

seventy-nine subjects did not favor the same cycle for ADDHRS and

TOTHRS, suggesting that subjects were not offsetting extreme responses

for BUDHRS. The four inconsistent responses were classified into two

groups. Two responses appeared to be ambiguous: subjects favored the

same cycle for BUDHRS and ADDHRS and the other cycle for TOTHRS. The

other two responses appeared to be compensating: subjects favored the

same cycle for BUDHRS and TOTHRS and the other cycle for ADDHRS. The

analysis discussed above was repeated deleting these observations from

the sample. Three versions of the analysis were conducted: without

ambiguous responses, without compensating responses, and without all

inconsistent responses. In all three cases, the basic results were the

same. Consequently, this alternative explanation is rejected.


Findings for H2

Hypothesis two states that when search is prescribed, auditors who

are committed to their hypotheses are likely to show a preference for

examining potentially confirming evidence before other evidence, whereas

auditors who are not committed are not likely to show a preference. For

analysis, the hypothesis is broken down into two parts, and each part is










examined separately. The first part of the hypothesis (H2A) suggests

that subjects who are committed will examine potentially confirming cues

before other cues (potentially disconfirming and neutral). The second

part of the hypothesis (H2B) suggests that subjects who are not

committed will not show a preference for the sequence in which

information cues are examined. While the information cues are referred

to as confirming and disconfirming in the discussion that follows, it

should be remembered that during search the cues are only potentially

confirming and disconfirming. That is, subjects do not examine a

complete cue until it is selected and interpreted.

A randomization procedure (Conover, 1980) was used to assess both

H2A and H2B. Subjects were presented with ten information cues: four

for the SRC, four for the PPC, and two for the PYC (payroll and

personnel cycle). Four cues could be classified as confirming and four

could be classified as disconfirming. For example, for subjects who

selected the SRC, the cues for the SRC were considered confirming and

those for the PPC were considered disconfirming. The PYC cues were

always considered neutral. Cues were ranked from 1 to 10 in the order

to which they were attended. The rank scores for confirming and

disconfirming cues then were summed. For example, if a subject selected

the SRC and examined the cues for that cycle in the 3rd, 4th, 7th, and

8th positions, a rank sum of 22 (3+4+7+8) was assigned for confirming

cues. A low rank sum indicates that the particular cues were given
10
early attention. A total of ( ) or 210 possible rank-sum combinations

were available, where rank sums ranged from 10 (1+2+3+4) to 34

(7+8+9+10). The randomization procedure assumes that each possible










rank-sum combination has an equal chance of being chosen. A

distribution of the 210 possible combinations was enumerated. The

frequency distribution is shown in Figure 7. The tails of the

distribution were examined to assess H2. Rank sums located in the lower

tail indicate that particular (confirming or disconfirming) cues were

attended before other cues, whereas rank sums located in the upper tail

indicate that particular (confirming or disconfirming) cues were

attended after other cues. The cut off for a significance level of .05

in the lower tail is a rank sum of 14. The cut off for a significance

level of .05 in the upper tail is a rank sum of 30. Each cut-off value

represents one of the twelve (210 x .05) most extreme possibilities

located in the tails of the distribution (see Figure 7). These cut-off

values were compared with rank-sum means calculated for (1) COM subjects

to assess H2A and (2) NCOM and NCOM+ subjects to assess H2B.

The first part of H2 is not rejected if the rank-sum mean for

confirming cues (of COM subjects) is found to be 14 or less. A mean of

14 or less would indicate that these cues were examined before other

cues. Although H2A is only concerned with the sequence in which

confirming cues are attended, the sequence in which disconfirming cues

are attended is also investigated for completeness. Means of 20.1613

and 20.7419 were calculated for confirming cues (of COM subjects) and

disconfirming cues (of COM subjects), respectively. The finding for

confirming cues is not consistent with H2A. While disconfirming cues

are not considered in H2A, the finding indicates that subjects did not

show a preference for the order that these cues were attended. The mean

is above the lower cut off and below the upper cut off. Subsequently,






77











FREQUENCY

18 *
I *
17- *
I *
16 * *
I a *
15 * *
I * * *
14- * *
13 * * *
13 * * * *
I * * * *
12- * * * *
I * * * *
11- * * * *
7 * * * *a
10- * * * * *

5- * * * * * *

8 -- *






4- * * * * * * *



3- * * * * * * * *






Rank
1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3B8m
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4

Figure 7: Frequency Distribution for Rank-Sum Combinations










each cycle was analyzed separately. Again, COM subjects did not show a

preference for attending to confirming cues or disconfirming cues (see

Table 17). Since confirming cues were not examined before other cues,

H2A is rejected.


TABLE 17

Rank-Sum Means for COM Subjects

SRC PPC

Confirming 19.5455 20.3000
Disconfirming 21.6366 20.2500

The second part of H2 is not rejected if the rank-sum means for

confirming and disconfirming cues (of NCOM and NCOM+ subjects) are found

to be between 14 and 30. Means between 14 and 30 would indicate that

subjects do not have a preference for the sequence in which cues are

attended. Means of 19.6875 and 19.4792 were calculated for confirming

cues of NCOM and NCOM+ subjects, respectively. Means of 21.7500 and

21.5417 were calculated for disconfirming cues of NCOM and NCOM+

subjects, respectively. These findings are consistent with H2B. All

means are between 14 and 30. In addition, each cycle was analyzed

separately; however, the results did not change (see Tables 18 and 19).

Since subjects who were not committed did not show a preference for the

sequence in which cues were examined, H2B is not rejected.


TABLE 18

Rank-Sum Means for NCOM Subjects


SRC PPC

Confirming 19.0006 18.5000
Disconfirming 20.6667 21.7000










TABLE 19

Rank-Sum Means for NCOM+ Subjects


SRC PPC

Confirming 18.8000 19.6970
Disconfirming 21.2000 21.3030

Subsequent analysis was undertaken to explain the findings

discussed above. In particular, an attempt was made to explain why

commitment did not have an effect. Cues may have been examined in order

of importance. The importance scores assigned to cues were correlated

with the order that they were attended. The Spearman correlation

coefficient for all subjects was -.0581 (p<.1450). Apparently,

importance did not affect the order that cues were attended. A closer

inspection of the data revealed a systematic pattern of search for some

subjects. Forty-four subjects attended to cues by cycle. All of the

cues in each cycle were attended in sequence. Moreover, six other

subjects attended to cues by cycle with the exception of one cue. This

cue was not attended in sequence. The remaining subjects did not show

any discernible patterns of search. Subjects who did show a systematic

pattern of search were classified according to (1) treatment group (COM,

NCOM, and control) and (2) the cue class (confirming, disconfirming, and

neutral) that was examined first (see Table 20). An examination of

Table 20 does not reveal an obvious explanation for why some subjects

followed a systematic pattern of search and others did not.

Nevertheless, the finding that a majority of subjects followed a


systematic pattern is notable.





80



TABLE 20

Subjects Who Followed a Systematic Pattern of Search

COM NCOM CONTROL
Confirming 11 6 2

Disconfirming 8 5 2

Neutral 3 4 3


The Underlying Cognitive Mechanisms

Further analysis was performed to explore the cognitive mechanisms

underlying subjects' judgments. Responses to recall and importance

questions were investigated. For recall, subjects recorded the number

of exceptions that were uncovered in each compliance test. These

responses were subsequently coded as correct or incorrect. For

importance, subjects responded on a seven-point scale, anchored by

extremely important and extremely unimportant, indicating the weight

that they would assign to each compliance-test result. Responses to

recall and importance questions were classified as confirming or

disconfirming, depending on the cycle selected. As an example, if a

subject selected the SRC, responses to questions about this cycle were

classified as confirming and responses to questions about the PPC were

classified as disconfirming. Subjects' responses were summed for each

classification and grouped by level of commitment. This procedure

resulted in four groups for each set of responses. The analysis

involved a comparison of cell means. The specific means that were

compared will be discussed below.

Subjects' responses for confidence in recall were also

investigated. As above, the analysis involved a comparison of cell










means. It will be remembered that the model developed in Chapter II

does not provide a basis for any predictions as to effects on

confidence, and hence this analysis is purely exploratory.


Findings for H3: Recall

Hypothesis three states that auditors who are committed to their

hypotheses are more likely to discount disconfirming evidence than

auditors who are not committed. This evidence is likely to be processed

at a deeper level than confirming evidence when auditors are committed.

Operationally, H3 suggests that subjects who are committed will be more

likely to recall disconfirming cues than confirming cues. The

hypothesis also suggests that these subjects will be more likely to

recall disconfirming cues than subjects who are not committed. The

number of confirming (CONF) and disconfirming (DONF) cues recalled

correctly was summed and grouped by level of commitment. Summary

statistics are shown in Table 21. As with earlier analysis, two

noncommitment groups are presented: NCOM and NCOM+. NCOM included

treatment subjects and NCOM+ included treatment plus control subjects.


TABLE 21

Summary Statistics for Recall

Mean Std Dev Std Error
COM/DONF 1.5484 1.0276 0.1846
COM/CONF 1.4839 1.0605 0.1905
NCOM/DONF 1.5323 0.8793 0.1554
NCOM/CONF 1.5625 0.9136 0.1615
NCOM+/DONF 1.5417 0.9216 0.1330
NCOM+/CONF 1.4375 0.8482 0.1224









Hypothesis three suggests the following relationships for recall:

COM/DONF > COM/CONF

COM/DONF > NCOM/DONF,

and COM/DONF > NCOM+/DONF.

T tests were used to test for differences between these groups. No

statistically significant differences (p<.10) were found. Subjects

recalled the same number of confirming and disconfirming cues,

regardless of level of commitment. This finding is not consistent with

H3. The analysis was repeated for each cycle separately, but the

results did not change. Once again, commitment did not have an effect.

Summary statistics for recall by the cycle selected are shown in Tables

22 and 23. A subsequent analysis was performed to assess effects of the

order in which recall questions were presented. One half of the

subjects had been given a predetermined order, and the other half had

been given this order reversed. A t test revealed no significant

differences (p<.10) between the two groups. It appears that the order

in which recall questions were presented did not have an effect on

recall.


TABLE 22

Summary Statistics for Recall: Cycle Selected=SRC

Mean Std Dev Std Error
COM/DONF 0.9090 1.0445 0.3149
COM/CONF 0.9090 0.5312 0.2506
NCOM/DONF 1.3333 0.8876 0.2562
NCOM/CONF 1.5833 0.7930 0.2289
NCOM+/DONF 1.4000 0.8281 0.2138
NCOM+/CONF 1.4000 0.8281 0.2138










TABLE 23

Summary Statistics for Recall: Cycle Selected=PPC

Mean Std Dev Std Error
COM/DONF 1.9000 0.8522 0.1906
COM/CONF 1.8000 1.0563 0.2362
NCOM/DONF 1.6500 0.8751 0.1957
NCOM/CONF 1.5500 0.9987 0.2233
NCOM+/DONF 1.6060 0.9663 0.1682
NCOM+/CONF 1.4545 0.8693 0.1513

Although the findings were not expected, ex post investigation

suggests a plausible explanation. Recall was likely to be affected by

the symmetry associated with the information cues. Cues were recalled

in symmetrical pairs. Similar tests were conducted for each cycle, and

the exception rates were the same for each type of test. If subjects

noticed this symmetry, they may have recalled confirming and

disconfirming cues in corresponding (symmetrical) pairs. Subjects only

had to remember one exception rate in order to correctly recall two

cues. In addition, the general nature of these cues may have heightened

the associated symmetry. Very little firm-specific information was

provided.

The results indicate that subjects recalled the same number of

confirming and disconfirming cues -- regardless of the cycle selected or

level of commitment. On average, subjects recalled 3 of 8 cues

(excluding the neutral cues). A total of 237 (79 subjects x 3 cues

recalled per subject) cues were recalled correctly. Of this total, 150

cues were recalled in corresponding pairs. A chi-square goodness-of-fit

test was conducted to determine if cues were more likely to be recalled

in pairs. A test statistic of 16.75 was calculated. The cut off for a

significance level of .005 was X =7.88. Subjects tended to recall cues










in corresponding pairs (see Table 24), and hence, symmetry appears to

have had an effect on recall.


TABLE 24

Chi-Square Goodness-of-Fit Test for Corresponding Pairs

Recall By Recall Not Significance
Pair By Pair Level
Observed Cell
Frequencies 150 87 0.005
Expected Cell
Frequencies 118.50 118.50

Subjects apparently recognized cue symmetry to some degree;

however, they may not have noticed that all cues (except neutral cues)

were symmetric. Subjects had been asked to indicate if the number of

exceptions uncovered in the SRC and the PPC, respectively, was the same

or different. If symmetry was obvious, subjects should have responded

that the same number of exceptions was found. A chi-square

goodness-of-fit test was performed to assess this proposition. The

results indicate that symmetry was not obvious. Subjects did not

recognize the symmetry associated with all cues (see Table 25).

Nevertheless, the fact that subjects recalled cues in symmetric pairs

reduces the inferences that can be drawn from this measure. Cues

recalled correctly were not necessarily processed at a deeper level than

cues recalled incorrectly.











TABLE 25

Chi-Square Goodness-of-Fit Test for Recognition of Symmetry

Same Number Different Number Significance
of Exceptions of Exceptions Level

Observed Cell
Frequencies 22 57 .005
Expected Cell
Frequencies 39.50 39.50


Findings for H3: Importance

Importance measures were also used to assess H3. Auditors who are

committed to their hypotheses are more likely to discount (assign less

importance to) disconfirming evidence than auditors who are not

committed. Operationally, H3 suggests that committed subjects will be

more likely to assign less importance to disconfirming cues than

confirming cues. The hypothesis also suggests that these subjects will

be more likely to assign less importance to disconfirming cues than

subjects who are not committed. The importance scores assigned to

confirming (CONF) and disconfirming (DONF) cues were summed and grouped

by level of commitment. Summary statistics are presented in Table 26.

As above, two noncommitment groups are included: NCOM and NCOM+.


TABLE 26

Summary Statistics for Importance

Mean Std Dev Std Error
COM/DONF 18.7742 3.8791 0.6967
COM/CONF 19.7742 3.9976 0.7180
NCOM/DONF 18.6250 3.9574 0.6996
NCOM/CONF 17.8750 4.2331 0.7483
NCOM+/DONF 18.1250 3.7566 0.5422
NCOM+/CONF 17.6170 4.3316 0.6318










Hypothesis three suggests the following relationships for

importance:

COM/DONF < COM/CONF

COM/DONF < NCOM/DONF,

and COM/DONF < NCOM+/DONF.

T tests were used to test for differences between these groups. No

predicted differences (p<.10) were found. Subjects apparently assigned

equal importance to confirming and disconfirming cues, regardless of

level of commitment. This finding is not consistent with H3.

Differences between groups were also investigated for each cycle

separately. Summary statistics are presented in Tables 27 and 28. For

the most part, the results were not consistent with H3; however, one

predicted difference was noted. The COM subjects who selected the SRC

assigned less importance (p<.0115) to disconfirming cues than confirming

cues. No other predicted differences (p<.10) were found. The

possibility that the order in which importance questions were presented

had an effect also was considered. As with recall, one half of the

subjects were given these questions in a predetermined order, and the

other half were given this order reversed. A t test was performed to

assess effects of order of presentation. No significant differences

(p<.10) were found. The order in which importance questions were

presented apparently did not have an effect.










TABLE 27

Summary Statistics for Importance: Cycle Selected=SRC

Mean Std Dev Std Error

COM/DONF 17.2727 3.7441 1.1289
COM/CONF 20.9091 3.3303 1.0041
NCOM/DONF 16.1667 4.0862 1.1796
NCOM/CONF 18.6667 3.6265 1.0469
NCOM+/DONF 15.9333 3.8816 1.0022
NCOM+/CONF 18.4000 4.2561 1.0989

TABLE 28

Summary Statistics for Importance: Cycle Selected=PPC

Mean Std Dev Std Error

COM/DONF 19.6000 3.7892 0.8473
COM/CONF 19.1500 4.2708 0.8550
NCOM/DONF 20.1000 3.1271 0.6992
NCOM/CONF 17.4000 4.5814 1.0244
NCOM+/DONF 19.1212 3.2954 0.5737
NCOM+/CONF 17.2500 4.3847 0.7751

Further analysis was conducted with respect to the importance that

subjects assigned to various cues. Table 27 indicates that subjects who

selected the SRC tended to rate confirming (SRC) cues as more important,

whereas Table 28 shows that subjects who selected the PPC tended to rate

disconfirming (SRC) cues as more important. Regardless of the cycle

selected or level of commitment, subjects appear to have assigned more

importance to the SRC cues than the PPC cues. T tests were performed to

determine if the importance assigned to SRC cues minus the importance

assigned to PPC cues was significantly different from zero for COM,

NCOM, and NCOM+ subjects, respectively. Except for COM subjects who

selected the PPC, the groups tended to assign more importance to the SRC

cues (see Table 29). These results are consistent with the earlier










finding that subjects originally allocated more hours to the SRC than

the PPC.


TABLE 29

T Tests for SRC-PPC by Cycle

Cycle Selected = SRC

Mean Std Dev T PR > 1T1
COM 3.6364 3.9057 3.09 0.0115
NCOM 2.5000 5.7446 1.51 0.1598
NCOM+ 2.4667 5.2897 1.81 0.0925

Cycle Selected = PPC

Mean Std Dev T PR > 1T1
COM 0.4506 4.1861 0.48 0.6362
NCOM 2.7000 4.4022 2.74 0.0129
NCOM+ 1.9688 4.4550 2.50 0.0179

Subjects may have generalized their responses for importance

questions. They may not have considered the task in responding to these

questions since very little firm-specific information was provided. In

addition, the form that these questions were asked may have encouraged a

generalized response. Subjects were asked to indicate how important

each test result could have been in explaining the fluctuation in gross

margin, but the exception rates given in the task were not provided with

the questions. This characteristic confounds importance with recall.

Subjects may only have been able to respond in very general terms.

If subjects did in fact respond in general terms, the number of

inferences that can be made from subjects' importance scores will be

reduced. Alternatively, the predictions made for importance may be

incorrect. The COM subjects who selected the PPC appeared to exhibit

confirmatory bias, even though their responses for importance questions










were not consistent with H3. That hypothesis suggested that subjects

who are committed will discount disconfirming cues individually as these

cues are being attended. Instead, subjects may have discounted

disconfirming cues on the whole rather than individually. Accordingly

confirmatory bias may have occurred in information aggregation as

opposed to information processing (as predicted). The present study did

not investigate the aggregation phase. Further research is needed to

assess this phase of interpretation.


Confidence

Confidence was also investigated in the present study. However,

this investigation was purely exploratory since the model developed in

Chapter II does not provide the basis for any predictions. Responses

for confidence in recall were summed and grouped by correctness of

recall: correct or incorrect. Mean scores of 4.4301 and 3.6828 were

calculated for cues recalled correctly and incorrectly, respectively. A

t test was performed to compare these mean scores, and a marginally

significant difference (p<.0750) was found. Subjects tended to show

more confidence in cues recalled correctly than incorrectly.

Next, confidence scores were grouped by level of commitment. Mean

scores of 4.0471, 3.8907, and 4.0235 were calculated for COM, NCOM and

NCOM+ subjects, respectively. Again, t tests were conducted to compare

mean scores. No differences were found at a significance level of .10.

Cues recalled correctly and incorrectly were examined separately, and

this analysis again revealed no significant commitment effect. Mean










scores are shown in Table 30. In sum, level of commitment did not

affect subjects' responses for confidence.


TABLE 30

Mean Scores for Confidence by Commitment and Correctness of Recall

Correct Incorrect
Recall Recall
COM 4.3936 3.5817
NCOM 4.4646 3.7821
NCOM+ 4.5175 3.7292

Lastly, confidence scores were grouped by cue classification:

confirming or disconfirming. Mean scores of 3.9917 and 3.8250 were

calculated for confirming and disconfirming cues, respectively. A t

test was performed, and no significant difference (p<.10) was found. As

above, this analysis was repeated for cues recalled correctly and

incorrectly, respectively. Mean scores are presented in Table 31. The

results indicate that cue classification did not affect subjects'

responses for confidence.


TABLE 31

Mean Scores for Confidence by Cue Classification and
Correctness of Recall

Correct Incorrect
Recall Recall

Confirming Cues 4.5556 3.6818
Disconfirming Cues 4.1381 3.7261




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - Version 2.9.7 - mvs