Group Title: effect of experience on the auditor's organization and amount of knowledge
Title: The effect of experience on the auditor's organization and amount of knowledge
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00102759/00001
 Material Information
Title: The effect of experience on the auditor's organization and amount of knowledge
Physical Description: Book
Language: English
Creator: Tubbs, Richard M., 1949-
Copyright Date: 1988
 Record Information
Bibliographic ID: UF00102759
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 - AFL2282
oclc - 19966131

Full Text











THE EFFECT OF EXPERIENCE ON THE AUDITOR'S
ORGANIZATION AND AMOUNT OF KNOWLEDGE















By

RICHARD M. TUBBS


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

1988

SOJE f LIBRARIES



































Copyright 1988

by

Richard M. Tubbs
















ACKNOWLEDGMENTS

Initially, I want to thank my committee members at

the University of Florida: Mike Bamber, Joe Alba, and Wes

Hutchinson. Their suggestions and comments were extremely

worthwhile. In addition, I would especially like to

express gratitude to my chairman, Bill Messier, for his

guidance throughout" my doctoral program.

I am extremely grateful to the following institutions

for the resources that they provided. First, the

Deloitte, Haskins, and Sells Foundation sponsored me by

means of a fellowship during the third and fourth years of

my doctoral program. Second, the Fisher School of

Accounting at the University of Florida furnished a

research grant and the access to students who served as

subjects. Third, ten offices of five "Big Eight" auditing

firms located in Florida supplied the auditors who served

as subjects and the facilities for conducting the

experiment.

The following individuals were of considerable help

during this project. The participants at workshops held

at the University of Florida, the University of

Connecticut, the University of Southern California,

Pennsylvania State University, the University of Illinois,

the University of North Carolina, the University of Iowa,


iii









and the University of Arizona supplied many useful

comments. JoAnne Glass provided expert assistance in the

coding of the recall protocols and in other aspects of the

project. Mike Zenor and Eric Olson furnished computing

advice that was extremely beneficial. John Lynch's

comments regarding the design of the study were also very

helpful.

Finally, I would like to thank my parents, Robert and

Jean Tubbs. They have provided a great deal of love,

encouragement, and guidance throughout my doctoral program

and throughout my life. Their help has been invaluable.

















TABLE OF CONTENTS

page

ACKNOWLEDGMENTS..................................... iii

ABSTRACT ..... ...................................... vii

CHAPTERS

1 INTRODUCTION.................................. 1

Background.................................... 1
Research Objective............................. 3
Motivation............ ......................... 4
Organization of the Dissertation.............. 5

2 THEORETICAL MODEL FOR AUDITOR'S KNOWLEDGE
REPRESENTATION.............................. 7

The Associative Network...... ................. 7
Auditor's Knowledge of Errors and
Irregularities ............................. 12
Hypotheses.................................... 19

3 METHODOLOGY................................... 24

Subjects........................ .............. 24
Task 1 Unconstrained Free Recall............ 25
Task 2 Conditional Prediction............... 37

4 RESULTS....................................... 45

Task 1...... .................................... 45
Task 2......................................... 58

5 SUMMARY AND DISCUSSION OF RESEARCH............ 77

Purpose of Research............................ 77
Discussion of Results.......................... 78
General Discussion............................. 81
Limitations................................... 84
Future Research ............................... 85

APPENDICES

A TASK 1 INSTRUMENT ............................. 88










B TASK 2 INSTRUMENT............................. 89

BIBLIOGRAPHY........................................ 98

BIOGRAPHICAL SKETCH ................................. 104
















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

THE EFFECT OF EXPERIENCE ON THE AUDITOR'S
ORGANIZATION AND AMOUNT OF KNOWLEDGE

By

Richard M. Tubbs

December 1988

Chairman: William F. Messier, Jr.
Major Department: Accounting

This study focuses on the manner in which auditing

experience affects the auditor's knowledge about a key

aspect in the audit process, errors and irregularities.

Since there are few audit judgment situations where a

"correct answer" can be known, this research provides two

possible benefits. First, identification of differences

between inexperienced and experienced auditors provides a

first step toward the training of inexperienced auditors.

Second, the knowledge of experienced auditors may provide

input for the construction of decision aids or expert

systems that can be employed in making day-to-day audit

decisions.

The ninety-five subjects who participated in this

experiment had varying levels of audit experience and

ranged from undergraduate auditing students to audit


vii









juniors, seniors, and managers. The subjects performed

two tasks. In the first task, subjects were allowed

fifteen minutes to recall as many possible types of errors

and irregularities that might occur in the sales,

receivables, and cash receipts cycle of a typical

wholesaling or manufacturing company. The purpose of this

task was to determine the quantity, quality, and

typicality of the auditors' knowledge about errors and

irregularities. In the second task, subjects were first

asked to estimate the probability of occurrence of eight

errors and irregularities. The subjects were then told

that a particular error (target) was discovered during the

audit, and were asked to reestimate the probability of

occurrence of the other seven errors. All eight items

served as targets for each subject. The purpose of this

task was to determine whether auditors are aware of

certain characteristics of errors and irregularities that

are related to causal reasoning.

The results demonstrate that differences in the

quantity, quality, and typicality of knowledge did exist

between auditors with different levels of experience.

Generally, undergraduate students and juniors did not

provide significantly different responses. Concurrently,

seniors and managers were not significantly different.

Auditors were very aware of two characteristics of errors

and irregularities that are related to causal

explanations, the internal control objective violated, and


viii









the department where the error occurred. The importance

of one characteristic, internal control objective

violated, was discovered to increase with experience.
















CHAPTER 1

INTRODUCTION

Background

A substantial amount of research in auditing (see

Bedard [1987]) and other areas (see Alba and Hutchinson

[1987]) has been devoted to explicating the relationship

between an individual's experience in a substantive area

and that individual's ability to perform tasks

successfully in that area (expertise). Many researchers

(e.g., Boritz [1985]) have employed experience as a

surrogate for expertise, implying that those who are

experienced are, in fact, expert. Other researchers

(e.g., Jacoby, et al. [1986]) have suggested that

experience and expertise are orthogonal constructs. Both

of these, however, are extreme views. While experience

may not necessarily lead to expertise, it is difficult to

imagine that one could exhibit expertise without first

having some experience in the area. More likely,

experience is a necessary but not a sufficient

prerequisite for expertise.

Experience within a substantive area normally leads

to the development of certain traits or cognitive

capacities (see Alba and Hutchinson [1987]). The

development of these traits or capacities generally allows









an individual to improve his/her task performance in that

substantive area. One of these traits is more complete

and more accurate knowledge structures.

A number of auditing researchers have attributed

differences in behavior to differences in knowledge.

Waller and Felix [1984] state that "the professional

auditor acquires a complex network of knowledge over his

or her years of experience . [which] strongly

influences his or her daily perceptions, evaluations, and

actions" (p. 383). Frederick and Libby [1986] suggest

that the expertise effects reported in their study are a

result of differences in the knowledge bases of auditors

and students regarding internal controls and financial

statement errors. Biggs, Mock, and Watkins [1988] found

that after identifying a potential problem in the audit of

the revenue cycle, managers were more selective than

seniors in increasing certain audit procedures. The

authors attributed this difference to the managers' deeper

and more complex memory schemata. Additionally,

researchers in other fields have demonstrated that

differences in the amount of knowledge within a domain

affect how individuals solve problems (e.g., Chi, Glaser,

and Rees [1982]) and how individuals encode and retrieve

information (e.g., Graesser and Nakamura [1982]).

Bedard [1987] states that "through experience, expert

auditors may have developed more complete knowledge,

better cross-referencing, and better memory organization"










(p. 21). Two auditing studies have examined this point.

Weber [1980] employed a recall task to examine the amount

and organization of recall of computer controls by EDP

auditors and students. He found that auditors recalled

more controls than students did and that the controls

recalled by the auditors were more organized into control

categories. Frederick [1986] also employed a recall task

to examine the amount and organization of recall of

internal controls in the purchasing and disbursements

cycle by auditors and students. He found that auditors

recalled more controls than did students, that auditors'

recall was more organized, and that auditors' recall was

more organized when presentation was organized by

transaction flow rather than by internal control

objective. Both of these studies imply that experience

leads to more and better-organized knowledge.


Research Objective

The objective of the present study is to examine

directly the differences between subjects with different

levels of audit experience in terms of their knowledge of

errors. This study differs significantly from the

existing auditing literature in the following respects.

First, although the studies by Biggs, Mock, and Watkins

[1988] and Frederick and Libby [1986] both imply that the

knowledge of errors is different for different levels of

experience, neither study examines this knowledge










directly. Second, the Weber [1980] and Frederick [1986]

studies examine differences in knowledge of controls

rather than errors. The importance of error and

irregularity identification to auditors is evident in the

recently issued Statement on Auditing Standards (SAS) No.

53 [American Institute of Certified Public Accountants

(AICPA), 1988] that deals specifically with the auditor's

responsibility for their detection. Third, this study

employs two paradigms for examining structure that have

not been used in the auditing literature: unconstrained

free recall and conditional prediction ratings. These

methods are arguably more direct than the methods

previously employed. Fourth, this study investigates more

levels of experience than previous auditing studies on

memory organization. Finally, although the Weber [1980]

and Frederick [1986] studies employed a statistical

methodology that allowed only a unidimensional measure of

organization (amount of clustering), there is much to

suggest that errors possess a variety of dimensions or

characteristics [Mautz and Sharaf, 1961]. Therefore, this

study employs methodologies that allow examination of the

multidimensional nature of the knowledge base.


Motivation

The theoretical motivations for this study are to

support the proposed link between audit experience and

differences in the auditors' knowledge base and to










demonstrate that experience leads to reasoning of a more

causal nature. There are also practical motivations for

this study. First, it has implications for auditor

training. The methods employed allow identification of

differences between experience levels. Training programs

can then be designed to make the knowledge base of the

novice more similar to that of the expert [Brown and

Stanners, 1983]. Retesting of novices can then be

employed to determine if the training program was

effective. Research in educational psychology (e.g.,

Diekhoff [1983]) has indicated that making relationship

judgments is a suitable and efficient way of testing the

understanding of concepts and the relationships between

concepts. Second, there is a growing interest in expert

systems in auditing (see Messier and Hansen [1987]). This

research should have implications for the development of

decision aids and expert systems. It should demonstrate

what types of conceptual relationships are learned with

experience. Also, it should allow for construction of a

system wherein the information is organized in the same

way that information is organized in the user's mind.

This makes the system easier to utilize [Durding, et al.,

1977].


Organization of the Dissertation

The remainder of this dissertation contains four

chapters. Chapter 2 present the theoretical model for







6



the manner in which an auditor's knowledge is organized

and the hypotheses to be investigated. Chapter 3 explains

the methodologies employed to test the hypotheses.

Chapter 4 presents the results of the experiments.

Finally, Chapter 5 contains a discussion of the results,

some implications of the study, and some areas for future

research.
















CHAPTER 2

THEORETICAL MODEL FOR AUDITOR'S KNOWLEDGE REPRESENTATION

This chapter presents the theoretical framework for

this study and the hypotheses to be investigated. The

first section presents a psychological model for the

manner in which knowledge of a domain may be organized in

memory and the manner in which experience within that

domain may affect this organization. Although this model

is only a metaphor for how knowledge is organized in

memory, it serves the purpose of suggesting testable

hypotheses regarding the effect of experience on knowledge

organization. The second section discusses the domain of

interest in this study, errors and irregularities. This

discussion is primarily based upon the professional

literature. The final section presents the hypotheses to

be investigated in this study. These hypotheses are based

on the psychological model presented in section 1 and the

domain knowledge presented in section 2.


The Associative Network

The most commonly accepted model (metaphor) of how

knowledge within a domain is represented in long-term

memory is an associative network model (e.g., Anderson

[1983], Collins and Loftus [1975], Wickelgren [1981]). In









this model, knowledge is characterized as a network of

nodes interconnected by arcs. For every concept that is

represented in a person's memory, there exists a set of

features, attributes, or dimensions that represent or

encode that idea [Wickelgren, 1981, p. 24]. This set of

features is often referred to as a node. Exposure to a

new concept creates a node. Subsequent exposure to that

concept or thinking about that concept results in

activation of that node. Each node has an associated

strength which is an increasing function of the recency

and frequency of activation of that node [Hayes-Roth,

1977, p. 261]. An arc represents the linkage or the

association between nodes. When two nodes are first

associated, an arc between them is created. This arc may

vary in strength, depending on the degree of association.

Further associations of the two nodes strengthen the arc.

Obviously, some nodes will be more strongly linked than

others. The nature of these associations and their

respective strengths determine the structure of the

conceptual network.

The relatedness of two concepts in the network is,

therefore, based upon the strength of the connection

between them [Collins and Loftus, 1975]. Tversky [1977]

has posited (although not in a network context) that the

similarity of two concepts is an increasing function of

the features common to both concepts and a decreasing

function of the features unique to each concept. Some










typical features on which relatedness or similarity of

concepts might be based are appearance, function, causal

relationship, temporal contiguity, spatial contiguity,

covariation, and affect. Some features will be more

salient than others in determining the similarity of two

concepts. The features and the relative salience of these

features are dependent on the particular domain being

studied. Tversky [197/] states that the salience of a

feature is the result of its intensity (e.g., brightness,

loudness, size) and its diagnosticity (the power of that

feature to distinguish between a class of concepts).

The network metaphor can be extended to model how

long-term memory changes with experience. The network can

change in a number of ways. When a person learns a new

concept, a new node is created in memory. A person with

more experience in a substantive area should, therefore,

have more concepts in long-term memory [Hayes-Roth and

Hayes-Roth, 1975; Hutchinson, 1983; Murphy and Wright,

1984]. Considerable evidence exists that the most typical

concepts, or members of a category, are learned first and

are more frequently recalled later [Mervis and Pani, 1980;

Mervis and Rosch, 1981; Nelson and Nelson, 1978]. As

experience increases, more atypical concepts are learned.

Naturally, the creation of new nodes in the network is

accompanied by the creation of arcs between the new nodes,

and between the new nodes and old nodes. As previously









mentioned, recency and frequency of activation of a

particular node will increase the strength of that node.

Changes in the strength of the arcs connecting the

nodes can also occur with increases in experience

[Wickelgren, 1981, p. 28], which in turn produce a change

in the conceptual structure [Conover, 1982; Hayes-Roth and

Hayes-Roth, 1975; Homa, Rhoads, and Chambliss, 1979].

This change in strength can be caused by different

factors. First, the number of distinct levels that are

recognizable within a given feature increase as experience

is gained [Hutchinson and Alba, 1985, p. 16]. This can

cause two nodes to be either more or less related.

Second, features or relations are noticed that formerly

went unnoticed and/or the salience of certain features

changes with more experience in a domain. It has been

suggested that causal relationships become more noticed or

more salient in the conceptual network as experience

increases [Murphy and Medin, 1985, p. 304]. One result of

these changes in the strength of the arcs is the ability

to more easily distinguish concepts that belong to

different categories. Increased experience has been shown

to result in a more accurate and more complicated category

structure [Weber and Crocker, 1983].

Support for the idea that memory organization becomes

more causally related as experience is gained in a domain

comes from two areas: developmental psychology and

cognitive psychology. Studies in developmental psychology











have demonstrated that young children's knowledge is

organized along a functional or perceptual dimension,

while older children's knowledge is organized along a

semantic or conceptual dimension (e.g., Howard and Howard

[1977]; Melkman, Tversky, and Baratz [1981]). Chi and

Ceci [1987] argue that these differences in knowledge base

organization are a function of the subject's experience

within the domain rather than the subject's age, but these

two factors are typically confounded in these

developmental studies (for an exception, see Means and

Voss [1985]).

In cognitive psychology, a number of studies in a

wide variety of domains have demonstrated differences in

organization of memory between novices and experts. These

studies have examined organization of chess pieces [Chase

and Simon, 1973], math problems [Schoenfeld and Herrmann,

1982], computer programming concepts [Adelson, 1981],

physics problems [Chi, Feltovich, and Glaser, 1981], and

cold remedies [Hutchinson, 1983], among others. Most of

these studies suggest that a person shifts from a

"surface" structure to a "deep" or "abstract" structure as

s/he gains expertise. "Deep" structure may involve

underlying principles (e.g., Chi, Feltovich, and Glaser

[. 81]) or underlying ingredients [Hutchinson, 1983].

Voss, Greene, Post, and Penner [1983] suggest that as

learning within a domain increases, individuals construct










causal relationships among concepts and develop strategies

or procedures for solving different problems. Chi,

Glaser, and Rees [1982] support this suggestion by

demonstrating that experts possess more procedural

knowledge and more knowledge about the conditions under

which certain procedures should be applied.

This section presented the most commonly accepted

psychological model for how knowledge within a domain is

organized in long-term memory. The model was extended to

demonstrate that domain experience increases the amount of

knowledge and alters the organization of knowledge.

Empirical support was presented that suggests that

knowledge organization becomes more causally related as

experience increases.


Auditor's Knowledge of Errors and Irregularities

This section discusses the substantive domain being

investigated in this research, errors and irregularities.

First, definitions of errors and irregularities are

presented. Second, the importance of errors and

irregularities in the audit process is emphasized. Third,

characteristics or features of errors and irregularities

are identified.

Definitions

Different sources in the auditing literature define

the concept of errors in different ways. Mautz and Sharaf

[1961] use the general term, irregularity, to refer to










"any departure from truth in the financial statements or

accounting records or any deviation from established and

duly authorized and established company policies" (p.

118). SAS No. 53, The Auditor's Responsibility to Detect

and Report Errors and Irregularities [1988], however,

distinguishes between errors and irregularities. Errors

are:

unintentional misstatements or omissions of
amounts or disclosures in financial statements.
Errors may involve mistakes in gathering or
processing accounting data from which the
financial statements are prepared, incorrect
accounting estimates arising from oversight or
misinterpretation of facts, and mistakes in the
application of accounting principles relating to
amount, classification, manner of presentation, or
disclosure [AICPA, 1988, pp. 1-2].

Irregularities, on the other hand, refer to:

intentional misstatements or omissions of amounts
or disclosures in financial statements.
Irregularities include fraudulent financial
reporting undertaken to render financial
statements misleading, sometimes called management
fraud, and misappropriation of assets, sometimes
called defalcations. Irregularities may involve
acts such as the following: manipulation,
falsification, or alteration of accounting records
or supporting documents from which financial
statements are prepared; misrepresentation or
intentional omission of events, transactions, or
other significant information; and intentional
misapplication of accounting principles relating
to amounts, classification, manner of
presentation, or disclosure [AICPA, 1988, p. 2].

Importance in the Audit Process

The importance of errors and irregularities in the

audit process is emphasized in SAS No. 53. This document


states that:









the auditor should assess the risk that errors and
irregularities may cause the financial statements
to contain a material misstatement. Based on that
assessment, the auditor should design the audit to
provide reasonable assurance of detecting errors
and irregularities that are material to the
financial statements [AICPA, 1988, pp. 2-3].

Assessing the risk that errors and irregularities have

caused material misstatement in the financial statements

requires:

the auditor to understand the characteristics of
errors and irregularities . and the complex
interaction of those characteristics. Based on
that understanding, the auditor designs and
performs appropriate audit procedures and
evaluates the results [AICPA, 1988, p. 3].

This document makes it clear that errors and

irregularities play an essential role in the audit

process.

There are many situations during an audit in which

the auditor must rely on his/her knowledge of errors and

irregularities. Two situations, in particular, are: (1)

when the auditor evaluates accounting control during the

planning stage of the audit and (2) during the performance

of compliance and substantive testing when an error or

irregularity is discovered. Auditing standards suggest

the following four-step procedure for evaluating

accounting control:

a. Consider the types of errors and
irregularities that could occr. b. Determine
the accounting control procedures that should
prevent or detect such errors and irregularities.
c. Determine whether the necessary procedures are
prescribed and are being followed satisfactorily.
d. Evaluate any weaknesses i.e., types of
potential errors and irregularities not covered by









existing control procedures to determine their
effect on (1) the nature, timing, or extent of
audit procedures to be applied and (2) suggestions
to be made to the client [AICPA, 1986, AU Section
320.65].

It seems as if the auditor's knowledge of errors and

irregularities and the relationship of the errors and

irregularities to accounting control procedures are vital

to the evaluation of accounting control. In many cases

the auditor may rely on an audit manual or decision aids

rather than his/her memory, but total dependence on an

audit manual or decision aids seems unlikely. SAS No. 53

states that detection of an irregularity, even an

immaterial one, "requires consideration of the

implications for the integrity of management or employees

and the possible effect on other aspects of the audit"

[AICPA, 1988, p. 15]. If the effect of the irregularity

is suspected to be material, the auditor should "consider

the implications for other aspects of the audit" (p. 11)

and "attempt to obtain sufficient competent evidential

matter to determine whether, in fact, material

irregularities exist and, if so, their effect" (p. 11).

This statement assumes that upon discovery of an error or

irregularity, the auditor possesses sufficient knowledge

about the potential sources and effects of errors or

irregularities to proceed appropriately. Nevertheless,

examination of a number of audit manuals provided only

very general guidelines to follow in such a situation.

One aspect of the present study is to investigate whether










this knowledge about the source and effects of errors and

irregularities varies with experience.

Identifying Features

A first step in exploring the organization of

different error or irregularity types in long-term memory

is the identification of the features that errors and

irregularities possess. Mautz and Sharaf [1961] identify

six characteristics or attributes. Materiality, the first

characteristic, may be different in regard to

irregularities from the normal idea of materiality in that

not only does the amount of an irregularity contribute to

its materiality but also its type. For example, an

instance of management fraud may be material regardless of

its amount (p. 119). The intent of an irregularity is a

second characteristic. Unintentional irregularities may

be mechanical mistakes, errors in principle in making

accounting judgments, or errors of bias in presenting

financial information. Intentional irregularities may be

made to hide incompetence or carelessness, to cause or

conceal a shortage, or to mislead financial statement

readers (p. 120). A third characteristic is the

relationship of the irregularity to internal control.

Some irregularities are perpetrated within the scope of

the internal control system; some are perpetrated by

circumventing the system; and some are beyond the scope of

the system (p. 122). A fourth characteristic is the









influence of the irregularity on the financial statements.

Irregularities can affect the balance sheet by overstating

assets and understating liabilities, by understating

assets and overstating liabilities, or by improper

description. Irregularities can affect the income

statement by overstating or understating income, or by

improper description. In addition, irregularities can

affect both the balance sheet and income statement (p.

123). A fifth characteristic of irregularities is their

extent of concealment. An open irregularity is one in

which no attempt at concealment has been made.

Concealment may be accomplished by manipulation of records

or by manipulation of documents. Manipulation of records

includes omission of entries, false entries, false footing

or other calculations, and false postings or other

bookkeeping procedures. Manipulation of documents

includes destruction of documents, preparation of false

documents, and alteration of legitimate documents (p.

125). The sixth characteristic identified by Mautz and

Sharaf is the auditor's responsibility for discovery.

Auditors have little responsibility to detect immaterial

items; errors for which there is no record; rare, non-

reoccurring items; and well-concealed errors caused by

collusion. There are also irregularities, on the other

hand, which any alert auditor should discover. In between

these two extremes are irregularities for which the

auditor's responsibility is not clear-cut (pp. 130-131).










SAS No. 53 [AICPA, 1988], which deals with the

auditor's responsibility to detect errors and

irregularities, identifies five characteristics related to

detection which have a great deal of overlap with those of

Mautz and Sharaf. The first characteristic is

materiality. Although an audit provides no assurance of

detecting immaterial errors and irregularities, discovery

of an immaterial irregularity might have a significant

effect on the audit while discovery of an immaterial error

would not be significant. A second characteristic is the

level of management or employees involved. Irregularities

can be either caused by employees, low-level management,

or upper-level management. A third characteristic is the

extent and skillfulness of concealment by the perpetrator

of the irregularity in order to reduce the likelihood of

detection. The fourth characteristic is the relationship

of the error or irregularity to the presence or absence of

specific control procedures. A final characteristic is

the financial statement effect of the error or

irregularity. Errors involving overstatement and errors

that affect both financial statements are generally easier

to detect (pp. 15-17).

There are other features which might influence memory

organization of errors. Libby [1985] demonstrated that

auditors employ a transaction cycle basis for

categorization of errors. He attributed this to the fact









that errors are categorized by transaction cycle in most

firms' training and procedural literature. Libby also

found some support for classification by general error

type (e.g., cutoff, unrecorded, etc.). The particular

audit procedure that uncovers an error is another possible

feature around wh-ch errors may be organized. Within a

transaction cycle there are various functions within which

certain errors might occur. Therefore, function might

serve as an organizing feature. Lastly, certain pairs of

errors are more likely to co-occur than other pairs. This

co-occurrence of certain error types also seems like a

possible organizing feature. Many of these features,

however, are not independent of each other.

This section has defined errors and irregularities,

stressed their importance in the audit process, and

identified their features. This research project will be

confined to the study of errors and irregularities within

one transaction cycle for two reasons. First, the study

of a more limited domain adds to the manageability of the

project. Second, previous research in auditor's memory

organization [Libby, 1985; Libby and Frederick, 1988] has

already examined errors across all transaction cycles.


Hypotheses

An auditor's knowledge about errors within a

particular transaction cycle develops with experience.

The knowledge of an auditing student or novice auditor is









probably confined to knowledge gained from textbooks and,

perhaps, training sessions. It seems reasonable to assume

that novices would have exposure to a standard control

system and be aware of how documents flow within the

cycle. This would suggest that novices would be aware of

the typical errors described in textbooks, where these

errors occur in the document flow of the cycle, and,

perhaps, the relationship of these errors to standard

internal controls. A study by Frederick and Libby [1986]

reported some probabilistic judgment results that are

consistent with the idea that students have no knowledge

(or knowledge that is not as organized as experienced

auditors) of the relationship between errors and internal

control weaknesses.

The auditor's amount and organization of knowledge of

errors is a function of specific audit experiences,

discussions of other audits with one's associates, the

supervision and review of one's work by superiors, case

materials used in training programs, the following of

audit plans, and the use of audit guides. Advanced level

activities such as the supervision of subordinates and the

actual designing of audit plans will reinforce and,

perhaps, enhance one's knowledge and organization of

knowledge about errors.

The exact manner in which experience affects the

auditor's amount and organization of knowledge about

errors is an empirical question. Some general











propositions can be made, however, based on the

psychological model that was presented earlier. First, as

experience is gained in auditing, both the quantity and

quality of knowledge about errors and irregularities will

increase. The auditor will become aware of more errors.

Atypical errors will be learned in addition to the typical

errors known by novices. Also, misunderstandings about

errors and irregularities are less likely to occur for two

reasons: the auditor's concept of errors and

irregularities becomes more defined, and the ability to

determine whether certain errors and irregularities occur

in a particular transaction cycle improves. This will

naturally occur through exposure to more errors and more

error types. The following three hypotheses are,

therefore, proposed:

H1: The number of different types of errors and
irregularities known by the auditor will increase
with experience.

H?: The auditor is less likely to have
misunderstandings about errors and irregularities
as experience increases.

H3: A less experienced auditor will be aware of
typical errors and irregularities; a more
experienced auditor will be aware of both typical
and atypical errors and irregularities.

Second, all of the previously mentioned characteristics of

errors and irregularities will probably gain in salience

with experience. For example, it is logical to assume

that more experienced auditors would better realize which

error types are more likely to be material, which error










types are related to certain internal control procedures,

and which error types are discovered by certain audit

procedures. However, certain characteristics will gain in

salience relative to the other characteristics. The

psychological literature referred to earlier would suggest

that those features that are related to causal

explanations would gain in relative salience. Those

features that are important in memory-based decisions are

likely to gain in relative salience, and the relative

salience of a characteristic is likely to be dependent

upon the judgmental situation. The quote from auditing

standards dealing with the planning stage of the audit

cited earlier (p. 14) suggests that once the auditor

considers the types of errors and irregularities that

could occur, s/he should determine the internal controls

necessary to prevent or detect these mistakes and then

perform audit tests to determine if these controls are

present. Similarly, the earlier discussion dealing with

the discovery of an error or irregularity during

compliance or substantive testing implies that the auditor

should be aware of the control procedure that was absent

in order for the error or irregularity to occur and the

likely perpetrator. In both situations, the reasoning

required of the auditor is causal in nature. Therefore,

the following four additional hypotheses are proposed.

They focus on the causal nature of the auditor's knowledge

base.










H4: The control procedure necessary to prevent or
detect an error or irregularity is a feature that
is salient to auditors.

H5: The control procedure necessary, to prevent or
detect an error or irregularity is a feature that
becomes relatively more salient with experience.

H6: The likely perpetrator of an error or
irregularity -s a feature that is salient to
auditors.

H7: The likely perpetrator of an error or
irregularity is a feature that becomes relatively
more salient with experience.

The preceding hypotheses will be empirically examined as

described in the subsequent section. In addition,

analysis of a more exploratory nature will be conducted in

order to identify any other features of errors and

irregularities, such as those identified in the

professional literature, that auditors attend to.















CHAPTER 3

METHODOLOGY

A laboratory experiment consisting of two separate

tasks was conducted to test the research hypotheses that

were developed in Chapter 2. This chapter discusses the

methodology of the experiment in three sections. The

first section describes the subjects and some of the

details of the administration of the experiment. The

second section discusses the following aspects of task 1:

previous uses of the methodology, the hypotheses to be

tested, and the exploratory analysis to be performed. The

third section discusses the following aspects of task 2:

previous uses of the methodology, the hypotheses to be

tested, and the exploratory analysis to be performed.


Subjects

Subjects included seventy-two auditors from ten

offices of five different Big Eight firms located in

north-central Florida. Participating offices were

requested to provide equal numbers of juniors, seniors,

and managers who had experience in the audits of

manufacturing and/or wholesaling companies. This resulted

in a sample of twenty-nine juniors, twenty-one seniors,

and twenty-two managers. Twenty-three student subjects









also participated. The students were enrolled in an

introductory auditing class and had just completed their

study of the sales, receivables, and cash receipts cycle

when they participated in the experiment. Mean (median)

years of auditing experience for the subjects were:

students 0.0 (0.0), juniors 1.04 (1.00), seniors -

3.40 (3.00), and managers 6.71 (6.30). One junior-level

subject was deleted from the analysis in task 2 because

s/he accidentally skipped a page of the instrument.

The experiment was performed in small groups for the

auditors at their offices and at the University of Florida

for the students. The researcher was present during all

administrations of the experiment. Subjects were assured

beforehand that their identity and the identity of their

firms would remain anonymous, and the importance of

independent work was stressed. After completing the

experimental tasks, a post-experimental questionnaire was

administered to gather selected demographic information.

The total experiment took approximately forty-five minutes

to administer.


Task 1 Unconstrained Free Recall

Previous Research

Task 1 employs an unconstrained free recall task.

This task has frequently been used in the study of

semantic memory [e.g., Bousfield and Sedgewick, 1944;

Bousfield and Barclay, 1950; Johnson, Johnson, and Mark,









1951; Indow and Togano, 1970; Rubin and Olson, 1980;

Gruenewald and Lockhead, 1980; Herrmann and Pearle, 1981;

Hutchinson, 1983; Millsap and Meredith, 1987]. In an

unconstrained free recall task, the subject is given the

name of a category, such as "animals", and is asked to

respond with as many examples of that category as possible

within a fixed period of time. Responses may be either a

word or a phrase depending upon the category. The timing

and the sequence of responses are then examined. The

timing of responses is typically recorded as either the

time between successive responses or the number of

responses occurring within specific time intervals. The

temporal data are used to construct mathematical models of

recall. The number and +he sequence of responses are used

to examine the amount and organization of memory for the

category.

In the experiments that have employed this paradigm,

two general phenomena have been observed. First, when the

cumulative number of responses, N(t), is plotted against

the amount of time since recall began, t, the curve is

concave in shape and approaches an asymptote, N(o), as

time increases. This asymptote represents total available

items in long-term memory. Some studies (e.g., Bousfield

and Sedgewick [1944], Indow and Togano [1970]) have

proposed that this curve is exponential in nature. Other

studies (e.g., Gruenewald and Lockhead [1980]) have found

that data were better fit by a hyperbolic function rather










than by an exponential function. Second, responses occur

in semantically related clusters [Bousfield and Sedgewick,

1944]. Gruenewald and Lockhead [1980] suggest that the

subject remembers a semantic field, or subcategory, within

a category and then responds with instances of that

semantic field. For example, if the category is

"animals", the subject might first think of the semantic

field of mammals. Then the subject would recall all of

the mammals that s/he could remember. Then the subject

might think of reptiles, and recall instances of reptiles.

This process would continue until a time limit was reached

or until the subject could think of no more semantic

fields.

Empirical research (e.g., Gruenewald and Lockhead

[1980], Herrmann and Pearle [1981]) has identified other

characteristics of the recall protocol. First, the time

between clusters, Tb, increases as the task progresses.

Second, the time between items within a cluster, Tw, is

relatively constant. Third, the number of items within a

cluster, Wc, is relatively constant across time.

Method

Two preliminary items were dealt with before subjects

performed the first task. First, each subject was shown a

non-auditing example of a subject's unconstrained free

recall protocol. The purpose of doing this was to

familiarize the subject with the type of task that s/he









would be performing and the specific directions that

needed to be followed. Second, the subjects were asked to

write down a phrase that was read aloud by the

experimenter and to record the time at completion of

writing. This was repeated using three different phrases.

The experimenter recorded the time at which he finished

reading each phrase. This allowed the experimenter to

measure each subject's individual rate of writing. This

rate was used in subsequent analysis of the task.

In the actual task, subjects were allowed fifteen

minutes to list, in order of recall, as many possible

types of errors or irregularities that might occur in the

sales, receivables, and cash receipts cycle of a typical

wholesaling or manufacturing company. The subjects were

provided with lined sheets of paper to record the errors

and irregularities with a column on the right side of the

paper to record the time (see Appendix A for a copy of the

instrument). The time clock was a videotape which

displayed the number of seconds that had passed since the

beginning of the task. Therefore, it ran from 000 to 900.

This task was not totally unrealistic for an auditor

because, other than not being able to use an audit manual,

it is similar in nature to the consideration of errors and

irregularities that occurs at the planning stage of the

audit. The method of recording the time was not found to

be unnecessarily intrusive by the subjects, and it has

been employed in previous research [Hutchinson, 1983].










A fairly comprehensive list of forty-six errors and

irregularities that could occur in the sales, receivables,

and cash receipts cycle of a wholesaling or manufacturing

firm was identified by the experimenter by reviewing

various auditing textbooks and the audit manuals of five

Big-Eight firms (see table 1) before the experiment was

administered. Each item in each subject's recall protocol

was then coded as either one of the forty-six errors or

irregularities on the list or as an intrusion. The

intrusions were of two types: (1) items that referred to a

different cycle or (2) items that did not meet the

definition of an error or irregularity.

Coding was performed by the experimenter and a

master's of accounting student with experience in the

audits of manufacturing firms. Each coder independently

coded each subject's recall protocol. Coming was somewhat

difficult due to differences in individual response style.

Nevertheless, inter-rater reliability or the total number

of agreements divided by the total number of items across

all subjects was .843. Items on which the coders

disagreed were subsequently reconciled by means of

discussion.

In order to calculate the inter-item response times

between two items, it was necessary to know the time at

which the subject completed writing the former item and

the time at which the subject started writing the latter










TABLE 1

Errors and Irregularities and Their Frequency of Occurrence
---------------------------------------------


Item Number Item

1. Orders accepted and goods shipped to poor
credit risks or unauthorized customers.

2. Orders accepted at terms other than those
authorized by management.

3. Order specifications may not be met as to
type and quantity.

4. Excessive and unwarranted granting of
credit under kickback arrangement.

5. Sales invoice incorrectly priced or
extended.

6. Management, employees, or third parties
receive products or services without being
billed or at unauthorized reduced rates.

7. Customers billed at incorrect amounts.

8. Billings recorded but goods not shipped.

9. A/R aged incorrectly; potentially
uncollectible amounts not recognized.

Error in recording sales in Sales Journal.
10. Not recorded(shipped not billed).
11. Recorded twice.
12. Wrong amount.
13. Wrong period.

14. Error in posting Sales/Accounts
Receivable to General Ledger.

15. Fictitious sales.

16. Failure to record recoveries on accounts
written off.


Frequency

49


24


19


1


29


10



15

22

65



49
11
16
48

45


54

7










TABLE 1--continued


Error in posting accounts receivable to
subsidiary ledger.
17. Wrong amount. 5
18. Wrong period. 8
19. Wrong account. 21
20. Not recorded. 12
21. Recorded twice. 2

22. Checks not properly endorsed upon 15
receipt.

23. Accompanying checks may not agree with 20
prelist from mailroom.

24. Cash or checks lost. 3

25. Cash receipts not deposited on a timely 17
basis.

Error in recording of cash receipts in Cash
Receipts Journal.
26. Not recorded. 21
27. Recorded twice. 2
28. Wrong amount. 9
29. Wrong period. 21

30. Cash receipts recorded as sales rather 3
than applied to A/R.

31. Collections on account misappropriated. 47

Error in posting cash receipts to Accounts
Receivable subsidiary ledger.
32. Not posted. 20
33. Posted twice. 2
34. Wrong amount. 7
35. Wrong account. 42
36. Wrong period. 3

37. Lapping. 41

38. Accounts improperly written off to cover 21
misappropriation of receipts.

39. A/R recorded at less than full amount; 0
customers remit in full and difference
misappropriated.










TABLE 1--continued


40. Returns or other allowances misclassified 44
or not recorded.

41. Returns not recorded in proper period. 7

42. Credits issued for returns or allowances 37
not earned or otherwise not according to
company policy.

43. Customers fail to pay within discount 1
period and remit in full; payments
recorded as if discount earned, amount of
discount misappropriated.

44. Records may be destroyed, stolen, or 14
lost.

45. Overstatement of revenues where "right of 11
return" or warranties exist.

46. Mathematical errors in computing a 4
customer's balance.









item. Since only completion times were provided by the

subject, starting times were estimated in the following

manner. Based on the aforementioned exercise where the

experimenter read aloud phrases and the subjects wrote

them down, a writing rate was calculated for each subject.

This writing rate was multiplied by the length of the

subject's response and subtracted from the completion time

for that item to arrive at an estimated starting time.

Hypotheses to Be Tested

Hypothesis 1 states that the number of different

types of errors and irregularities known by the auditor

will increase with experience. The recall protocols

(i.e., list of items recalled) indicate the number of

errors and irregularities retrieved by the auditors.

Intrusions and repetitions of earlier mentioned items were

not counted. This number serves as a surrogate for the

number of items known and serves as the dependent

variable. This is a reasonable surrogate because the

number of items recalled is considered to be a function of

the number of items known by the researchers who employ

this paradigm (e.g., Bousfield and Sedgewick [1944]). The

hypothesis was tested by performing a test of ordered

alternatives on the number of items recalled for the

members of each of the four levels. Hypothesis 1 suggests

that this trend would be positively related to experience.

A distribution-free test [Jonckheere, 1954] was employed

for the following two reasons. First, a parametric trend










analysis requires that the independent variable be

quantitative in nature and that the sizes of the intervals

between treatment levels can be specified [Kirk, 1982, p.

150]. Certainly this requirement is not met for level of

experience, the independent variable in this experiment.

Second, the distribution free test does not require

normally distributed treatment populations or homogeneity

of error variance among treatments as are required by the

parametric test.

Hypothesis 2 states that the auditor is less likely

to have misunderstandings about errors and irregularities

as experience increases. For this hypothesis, the number

of items that are coded as intrusions for each subject

serves as the dependent variable. This hypothesis was

tested by performing a test of ordered alternatives on the

number of intrusions. The expectation was that this trend

would be negatively related to experience.

Hypothesis 3 states that less experienced auditors

will be aware of typical errors and irregularities whereas

more experienced auditors will be aware of both typical

and atypical errors and irregularities. In order to test

this hypothesis, the frequency with which each specific

error or irregularity occurred in the recall protocols of

all the auditors was recorded (see table 1). These

numbers were then divided by the number of subjects in

order to arrive at the percentage of subjects who









generated that particular error or irregularity. The

percentage of subjects who generated an item is referred

to as item dominance and has often been employed as a

surrogate for typicality (e.g., Loftus [1973]; Mervis,

Catlin, and Rosch [1976]; and Barsalou [1985]). Item

dominance has been found in these studies to be highly

correlated with other measures of typicality like goodness

of example ratings and frequency of instantiation ratings.

Two depen_-.nt measures were considered for testing this

hypothesis. First, an average item dominance score was

determined for each subject by taking the sum of the

dominance scores for the items that the subject recalled

and dividing by the number of items. Second, the minimum

item dominance score was determined for each subject. The

hypothesis was tested by performing a test of ordered

alternatives on these variables. The expectation was that

this trend would be negatively related to experience since

typical items are given high numeric ratings and atypical

items are given low numeric ratings.

Exploratory Analysis

In addition to testing hypotheses 1 through 3,

analysis of a more exploratory nature was performed upon

the data from experiment 1. This was done to determine

whether any dimensionality or related clusters that

correspond to the features of errors and irregularities

identified in Chapter 2 existed in the data. Similarity

matrices were constructed for each subject in the










following manner. First, items of low frequency

(mentioned by less than four subjects) were eliminated

from consideration. Because these items had been

mentioned by so few subjects, no temporal distances

existed between these items and the majority of the other

items. Nine of the forty-six items were in tLns category.

Second, for each subject the temporal distance between

every pair of items in the subject's recall protocol was

calculated. Repetitions were included in this analysis.

Therefore, when multiple distances between the same two

items existed, the shortest distance was retained. Third,

the reciprocal of each of these distances was taken in

order to convert the distances into similarities. Fourth,

zeros were assigned as similarities to all pairs of errors

or irregularities that did not co-occur in a subject's

recall protocol. In this manner, a 37 x 37 symmetric

similarity matrix, with zeros on the diagonal, was

constructed for each subject.

Because subjects recalled different items and

different numbers of items, individual differences

multidimensional scaling and individual differences

additive clustering were not employed. Rather, the

maximum similarity for each pair of items was calculated

across all subjects and across each group of subjects.

Two-way ordinal multidimensional scaling was performed on

these matrices using the ALSCAL procedure [SAS Institute










Inc., 1986]. Two-way additive clustering was also

employed in analyzing this data using the OVERCLUS

procedure [SAS Institute Inc., 1986]. In this manner any

dimensionality or clustering in the data could be

identified for the four different experience levels and

for the subjects in general.


Task 2 Conditional Prediction

Previous Research

Task 2 employs a conditional prediction task. While

the first task examined differences in the amount, the

typicality, and the organization of memory, the second

task examines experience differences that are more related

to causal reasoning and procedural knowledge. A similar

task was employed by Johnson and Tversky [1984] to study

the mental representations of a specific domain of

knowledge, the prevalent causes of death. In the

experiment by Johnson and Tversky, the subjects were first

asked to estimate the number of people in the United

States who died each year from eighteen different risks.

The subjects were then given the following instructions:

Suppose you were to learn that many more people
die each year in the U.S. from than you
had estimated. Please indicate whether or not you
would increase your estimate for each of the
following causes of death, given the new
information (p. 56).

Every subject followed these instructions for various

target risks and made seventeen judgments relative to the

other risks each time. The proximity between risks was










defined as "[P(XIY) + P(YJX)]/2, where P(XIY) is the

percentage of subjects who wished to increase their

estimate of risk X when told that they had underestimated

risk Y" (pp. 57-58).

A proximity or distance matrix was created from this

conditional prediction data. It was found that the

proximity matrix created by conditional prediction data

was highly correlated with a proximity matrix created by

similarity ratings which were made by a different group of

subjects. Johnson and Tversky indicate that this high

correlation suggests that similarity "may play an

important role in predicting people's responses to new

risks or to new evidence about risk" (p. 68). The data

were also evaluated using four different models:

principal-components factor analysis, ordinal

multidimensional scaling, and hierarchical and

nonhierarchical tree models. It was discovered that the

conditional prediction data was better described (in terms

of R2) by the tree models than by the multidimensional

scaling or factor analysis models.

Method

The forty-six errors and irregularities referred to

in task 1 were classified according to a number of

dimensions including internal control objective violated,

department where the error occurred, accounts affected,

and the audit procedure appropriate to discover the error.









The dimensions of primary interest were the internal

control objective violated and the department where the

error occurred. Because of the nature of the task and the

large number of responses required, only a subset

consisting of eight items from the original list of errors

and irregularities was included. The eight items were

chosen to satisfy the following criteria. First, two

errors each originated in the following four departments:

sales order, billing, general accounting, and cash

receipts. Second, two errors each could have violated the

following four internal control objectives: validity,

authorization, completeness, and valuation. Third, there

was no ambiguity for these errors as to classification

upon either dimension: department or internal control

objective. Two experienced auditing professors confirmed

the classification of these eight items on these two

dimensions and noticed no ambiguity in these

classifications. Fourth, no two errors could have both

the same department and the same internal control

objective in common.

Each subject was presented with the list of eight

errors and irregularities. The subjects were asked to

estimate the probability of occurrence of each error and

irregularity, P(Ei), where i=l to 8, for a typical

company. Each subject was then told that during an audit

a particular (target) error or irregularity, Ej, where j=l

to 8, was discovered. The subject was then asked to re-










estimate the probability of occurrence of the other seven

(judged) errors and irregularities, P(EilEj), where ifj.

All eight items served as targets for each subject (see

Appendix B for a copy of the instrument). The order of

the targets was randomized, and the judged errors and

irregularities were presented alphabetically. Therefore,

each subject made fifty-six (8 x 7) conditional

predictions. This data was used to form an 8 x 8 matrix

with no entries along the diagonal for each subject. The

entries in the matrix, referred to as the subject's

prediction matrix, is the conditional probability of the

error minus the "base rate" probability of occurrence of

that error, [P(EilEj)-P(Ei)], where ifj. This measure was

chosen because it was thought to capture the auditor's

change in belief upon the receipt of new information.

Another measure considered was the conditional probability

of the error divided by the "base rate" probability of

occurrence of the error, [P(EijEj)/P(Ei)]. This

alternative measure was not employed because a great deal

of the "base rate" probabilities were equal to zero. The

task is fairly realistic because it is analogous to the

situation in which either an error or irregularity is

discovered during testing, and the auditor has to decide

how to alter or extend the auditing procedures to find

other related errors or irregularities.








Hypotheses to Be Tested

Hypothesis 4 states that the control procedure

necessary to prevent or detect an error or irregularity is

a feature that is salient to auditors. Because there is

an almost one-to-one correspondence between errors and

internal control procedures, a less specific category,

internal control objective, was chosen as a surrogate for

internal control procedure. Control objective or audit

objective have been employed by other researchers as

categories around which internal controls [Frederick,

1986] and financial statement errors [Libby and Frederick,

1988] are organized. Because the set of eight errors

contains two errors each from four internal control

objective categories, each subject's prediction matrix

will contain eight entries in which both the target error

and the judged error share the same internal control

objective. Each of these entries is referred to as Pi

where i=l to 8. In the other forty-eight entries, the

target error and the judged error do not share the same

internal control objective. Each of these entries is

referred to as Nj where j=l to 48. The dependent measure

for testing this hypothesis is the difference between the

mean of the entries for which the target and the judged

error share the same objective and the mean of the entries

for which the target and judged errors do not share the

same objective, [1/8 Z8Pi 1/48 E48Nj]. This difference









in means is referred to as Do. This measure is, in

essence, a planned comparison because the expectation is

that the related cells will have higher values than the

unrelated cells. Hypothesis 4 was tested by performing a

test of location on the Do's of the members of each of the

four levels. The expectation was that the treatment

effect would be greater than zero for each level.

Hypothesis 5 states that the control procedure

necessary to prevent or detect an error or irregularity is

a feature that becomes relatively more salient with

experience. D,, the difference between the mean of the

entries for which the target and the judged error share

the same objective and the mean of the entries for which

the target and judged errors do not share the same

objective serves as the dependent variable. Hypothesis 5

was tested by performing a test of ordered alternatives on

the Do's of the members of each of the four levels. The

expectation was that this trend would be positively

related to experience.

Hypothesis 6 states that the likely perpetrator of an

error or irregularity is a feature that is salient to

auditors. Because the duties assigned to specific

personnel vary across companies, department was chosen as

a surrogate for the likely perpetrator. The discovery of

an error that occurred in a specific department should

lead to a search for other possible errors in the same

department. This hypothesis was tested in much the same










manner as hypothesis 4. The dependent variable, in this

case, is the difference between the mean of the entries

where the target and judged errors share the same

department and the mean of the entries where the target

and the judged errors do not share the same department.

This difference in means is referred to as Dd. Hypothesis

6 is tested by performing a test of location on the Dd's

of the members of each of the four levels. The

expectation is that the treatment effect would be greater

than zero for each level.

Hypothesis 7 states that the likely perpetrator of an

error or irregularity is a feature that becomes relatively

more salient with experience. Dd, the difference between

the mean of the entries for which the target and judged

errors share the same department and the mean of the

entries for which the target and the judged errors do not

share the same objective serves as the dependent variable.

Hypothesis 7 is tested by performing a test of ordered

alternatives on the Dd's of the members of each of the

four levels. The expectation is that this trend would be

positively related to experience.

Exploratory Analysis

As was the case with task 1, analysis of a more

exploratory nature was performed. Similarity matrices

were constructed for each subject in the following manner.

Instead of using each subject's prediction matrix, a









symmetric measure of similarity between items was

employed, as was done by Johnson and Tversky [1984]. This

measure was the average of the two corresponding items

from above and below the diagonal of the subject's

prediction matrix, 1/2([P(EilEj)-P(Ei)] + [P(EjlEi)-

P(Ej)]), where ifj. Individual differences

multidimensional scaling and individual differences

additive clustering were employed to analyze the data from

each level and from all the levels combined using the

ALSCAL and the OVERCLUS procedures. In this manner any

dimensionality or clustering in the data would be

identified.
















CHAPTER 4

RESULTS

This chapter presents the results of the experiment.

Operationally, the following null hypothesis is being

tested throughout the experiment (except in hypotheses 4

and 6)

Ho: Tstudents = Tjuniors = Tseniors = Tmanagers

where T is the treatment effect associated with each

level. The first section presents the results related to

task 1. This includes the tests of hypotheses 1 through 3

and the exploratory analysis. The second section presents

the results related to task 2. This includes the tests of

hypotheses 4 through 7 and the exploratory analysis.


Task 1

Hypothesis 1

The purpose of testing hypothesis 1 is to see if the

number of errors and irregularities known by the auditor

increases with experience. The dependent variable is the

number of errors or irregularities recalled by the

subjects. The alternative hypothesis is

Ha: Tstudents I Tjuniors < Tseniors < Tmanagers

where at least one inequality is strict. The mean

(median) numbers of errors and irregularities retrieved by









the different levels of subjects are presented in table 2.

The large sample approximation of the Jonckheere Test (see

Hollander and Wolfe [1973]) on the four levels yielded a Z

statistic equal to 2.652. This allows rejection of the

null hypothesis of equal treatment effects in favor of the

alternative hypothesis at a probability level of .004.

This is indicative of a strongly significant experience

effect. In order to examine which treatment effects were

significantly different from each other, the large sample

approximation of the Wilcoxon Rank Sum Test (see Hollander

and Wolfe [1973]) was performed on all six pairwise

comparisons. The following treatment effects were

significantly different (or close to significance):

students versus seniors (Z=1.277; p=.101), students versus

managers (Z=2.489; p=.006), juniors versus seniors

(Z=1.471; p=.071), and juniors versus managers (Z=2.609;

p=.005). Therefore, students and juniors seem to form one

group, and seniors and managers form another.

Hypothesis 2

The purpose of testing hypothesis 2 is to determine

if misunderstandings about errors and irregularities

decrease with experience. The dependent variable is the

number of items recalled by the subjects that were coded

as intrusions. The alternative hypothesis is

Ha: Tstudents > Tjuniors > Tseniors > Tmanagers

where at least one inequality is strict. The mean

(median) numbers of intrusions of the different levels of










TABLE 2

Experimental Results


Mean
(Median)


Dep. Var.



Items recalled



Intrusions



Average item
dominance


Minimum item
dominance


4-5



6-7


Students Juniors Seniors Managers



9.00 8.69 10.33 11.27
(8.00) (9.00) (10.00) (11.50)


8.30 7.31 7.76 6.27
(9.00) (7.00) (6.00) (5.00)


0.388 0.375 0.370 0.371
(0.371) (0.357) (0.353) (0.371)


0.147 0.123 0.121 0.103
(0.147) (0.084) (0.105) (0.084)


2.57 2.33 4.40 5.84
(0.98) (1.53) (3.85) (6.84)


4.86 4.12 3.64 3.92
(1.46) (2.48) (3.54) (3.02)









subjects are presented in table 2. The large sample

approximation of the Jonckheere test on the four levels

yielded a Z statistic equal to 1.899. This allows

rejection of the null hypothesis of equal treatment

effects across levels at a probability level of .029.

This is indicative of a significant experience effect.

The large sample pproximation of the Wilcoxon Rank Sum

Test was performed on the six pairwise comparisons, and

the following treatment effects were significantly

different (or close to significance): students versus

juniors (Z=1.543; p=.061), students versus seniors

(Z=1.239; p=.108), and students versus managers (Z=1.835;

p=.033). Therefore, students seem to form one group, and

juniors, seniors and managers form another. Perhaps most

of the reduction in misunderstandings about errors and

irregularities takes place early in the auditor's career.

Hypothesis 3

The purpose of testing hypothesis 3 is to determine

if auditors learn more atypical errors and irregularities

as they gain experience. The dependent measures are the

subject's average item dominance and minimum item

dominance. The alternative hypothesis is

Ha: Tstudents > Tjuniors ? Tseniors L Tmanagers

where at least one inequality is strict. The mean

(median) average item dominance scores and the minimum

item dominance scores of the different levels of subjects

are presented in table 2. The large sample approximation









of the Jonckheere Test on the four levels of average item

dominance yielded a Z statistic equal to 1.049. The

probability level associated with this Z statistic is

.147. Therefore, the null hypothesis cannot be rejected.

Nevertheless, when the large sample approximation of the

Wilcoxon Rank Sum Test was performed on the six pairwise

comparisons, the following treatment effects were

significantly different: students versus juniors (Z=1.437;

p=.075) and students versus seniors (Z=1.551; p=.060).

The large sample approximation of the Jonckheere Test on

the four levels of minimum item dominance yielded a Z

statistic equal to 1.769. This allows rejection of the

null hypothesis of equal treatment effects across levels

at a probability level of .038. This is indicative of a

significant experience effect. The large sample

approximation of the Wilcoxon Rank Sum Test was performed

on the six pairwise comparisons, and the following

treatment effects were significantly different: students

versus juniors (Z=1.513; p=.065), students versus seniors

(Z=1.402; p=.081), and students versus managers (Z=2.129;

p=.017). Regardless of which dependent measure is

employed, students seem to form one group, and juniors,

seniors, and managers form another. Both dependent

measures indicate that a trend which is negatively related

to experience does exist, but it seems as if many of the

atypical errors are learned fairly early in the auditor's

career.










Exploratory Analysis

In order to determine any dimensionality or related

clusters that exist in the data, multidimensional scaling

and additive clustering were performed on the maximum

similarity matrices between the thirty-seven items. Table

3 and figure 1 present the two-dimensional ordinal scaling

solution for the maximum similarity matrix of all subjects

combined. Stress, or badness of fit, for this solution

was .333 and R2 was .419. Solutions with additional

dimensions improved R2 (stress): 3 dimensions .514

(.247), 4 dimensions .542 (.202). However, additional

dimensions did not aid in interpreting these solutions.

Therefore, only the two-dimensional solution is presented.

It is extremely difficult to label the dimensions, but

many of the items that are close in proximity seem to be

related in terms of department, internal control

objective, or account affected. This is supported by the

additive clustering results which are reported

subsequently. Scaling solutions were also derived for the

four experience levels separately, but no additional

insight was achieved.

The general (all experience levels combined) five

group additive clustering solution had an R2=.439.

Additional groups only increased R2 marginally (e.g. 6

groups R2=.467, 7 groups R2=.493). The first four

groups were easily interpretable (see table 4). Group 1










TABLE 3

MDS Plot Symbols and Coordinates Task 1
--- --------------------------------------------_--------


Coordinates


Error Number


Plot Symbol


Dimension 1

-0.5041
0.2438
1.1801
-0.9594
1.8987
1.1750
1.0602
-0.1759
-0.1511
-0.6018
0.3737
-0.2531
-0.8188
0.0572
2.0217
-0.6808
-2.0022
0.7094
-0.5956
1.3038
-1.4266
0.2126
0.0017
-0.9535
-0.8453
0.1651
-0.4526
1.8578
0.0216
0.0393
0.5866
0.1087
1.7109
-0.4071
-1.4727
-1.2470
-1.1804


Dimension 2

0.7505
1.4168
0.0836
0.6687
0.0983
1.3019
-0.2552
0.5240
-0.3622
-1.8132
-1.1866
-0.6476
0.6162
-0.4362
0.5123
-2.1495
0.0824
-1.0679
-1.8157
0.4172
0.1017
1.5633
-1.1327
-1.4866
0.7217
-0.4276
-0.8475
-0.5129
-0.4710
1.0891
0.7830
-0.2639
0.3902
0.6230
-0.3049
1.3 58
2.0704










DIM2---------------------------------------------------------
DIM2 2
2
2.0+




M
1.5+
1 2
6

U
1.0+


1
4P
D Y
0.5+ 8




H L
0.0+----------------------------


K X


3 5
-------------- -------------


9 EQ
T
C

R


N


AJ

-2.0+
I G
----+----+----+----+----+----


B










----+----+----+----+----+--


-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
DIM1


FIGURE 1


MDS Plot Task 1


-0.5+





-1.0+





-1.5+











TABLE 4

Overclus Solutions Task 1
----------------------;~-------------------------

Exp. Level Group Members R2

General 1 10 40 0.15

2 40 42 0.24

3 1 2 42 0.29

4 26 37 0.33

5 5 8 9 10 13 15 26 28 31 0.44
32 35 37 40

Students 1 10 40 0.36

2 1 2 5 0.45

3 12 17 0.50

4 26 28 35 40 0.55

Juniors 1 26 37 0.14

2 5 28 0.18

3 1 7 9 10 12 14 15 19 20 0.35
23 26 31 35 40

4 9 18 44 0.39


Seniors









Managers


9 10 11 13 15 31 35 40

2 3 8 10

12 19 35

9 14 22 23 2- 32 35 44


40 42

8 10

9 13 15 29 31 35 37

1 42


0.13

0.20

0.25

0.30


0.21

0.29

0.39

0.45










includes item 10 (Not recording sales in the Sales

Journal) and item 40 (Returns or other allowances

misclassified or not recorded). This group can be

interpreted as a "failure to record" group. Group 2

includes item 40 and item 42 (Credits issued for returns

and allowances not earned or otherwise not according to

company policy). This group is related to returns and

allowances, and the two items might be considered

opposites. Group 3 includes item 1 (Orders accepted and

goods shipped to poor credit risks or unauthorized

customers), item 2 (Orders accepted at terms other than

those authorized by management), and item 42. These items

are all related to the internal control objective of

authorization. Group 4 contains item 26 (Not recording

cash receipts in Cash Receipts Journal) and item 37

(Lapping). These two items are almost identical, except

that the former is an error and the latter is an

irregularity. Group 5 includes item 5 (Sales invoice

incorrectly priced or extended), item 8 (Billings recorded

but goods not shipped), item 9 (A/R aged incorrectly;

potentially uncollectible amounts not recognized), item 10

(Sales not recorded in sales journal), item 13 (Sales

recorded in wrong period), item 15 (Fictitious

sales/receivables), item 26 (Cash receipts not recorded in

Cash Receipts Journal), item 28 (Cash receipts recorded at

wrong amount), item 31 (Collections on account










misappropriated), item 32 (Cash receipts not posted to A/R

subsidiary ledger), item 35 (Cash receipts posted at wrong

amount), item 37 (Lapping), and item 40 (Returns or other

allowances misclassified or not recorded). This group is

not quite as interpretable as the other groups, but most

of the items deal with a failure to record or post.

Separate four group solutions were also analyzed

using OVERCLUS (see table 4). Additional groups only

increased explanatory power marginally (e.g., average R2

across four levels: five group .485, six group .543,

seven group .56). As can be seen in table 4, the

OVERCLUS procedure produces clusters or groups of variable

size that can overlap. Comparison of all of the

experience levels is extremely difficult, particularly

because of the size of some of the clusters for juniors

and seniors. However, the comparison of the least

experienced level (students) and the most experienced

level (managers) is more manageable and illustrates some

interesting differences. The groups for students are

related to internal control objectives and departments.

The first group contains items 10 (Not recording sales in

the Sales Journal) and 40 (Returns or other allowances

misclassified or not recorded). These items are both

related to the internal control objective of completeness.

The second group contains items 1 (Orders accepted and

goods shipped to poor credit risks or unauthorized









customers), 2 (Orders accepted at terms other than those

authorized by management), and 5 (Sales Invoice

incorrectly priced or extended). All three of these items

involve the sales order department. The third group

includes item 12 (Sales recorded at wrong amount in Sales

Journal) and item 17 (Accounts receivable posted at wrong

amount to subsidiary ledger). These items are related to

valuation. The fourth group contains items 26 (Cash

receipts not recorded in Cash Receipts Journal), 28 (Wrong

amount of cash receipts recorded in Cash Receipts

Journal), 35 (Cash receipts posted to wrong account of

Accounts Receivable subsidiary ledger), and 40. These

items all involve mistakes in recording or posting.

The groups for managers, on the other hand, have

different interpretations. Group 1 includes items 40 and

42 (Credits issued for returns and allowances not earned

or otherwise not according to company policy). These

items both deal with returns and allowances and are

opposites in the sense that in item 40 the return is

earned but not recorded, while in item 42 the return is

recorded but not earned. Group 2 consists of item 8

(Billings recorded but goods not shipped) and item 10

(Sales not recorded in Sales Journal). These items both

deal with shipping and billing, and one violates the

completeness objective while the other violates the

validity objective. As with group 1, these items can be

thought of as opposites. Group 3 contains items 9 (A/R







57



aged incorrectly; potentially uncollectible amounts not

recognized), 13 (Sales recorded in wrong period), 15

(Fictitious sales), 29 (Cash receipts recorded in wrong

period), -1 (Collections on account misappropriated), 35

(Cash receipts posted to wrong subsidiary ledger account),

and 37 (Lapping). This group seems to contain items which

are of a fairly serious nature which suggest either

defalcations or management override. Group 4 includes

items 1 and 42, which are both authorization errors.

The groups for students were related to internal

control objectives (completeness, valuation) and

departments (sales order). These clusters could also be

interpreted as being related to "key words" such as

"failure to record" (group 1), "wrong amount" (group 3),

and "failure to record or post" (group 4). One of the

groups for managers was related to an internal control

objective (authorization), and one group was related to a

function (returns and allowances). However, the manager's

solutions also focus on opposites and the seriousness of

the error. If the students are grouping simply on "key

words", then the manager's groupings seem to demonstrate a

more substantive understanding of how certain errors and

irregularities are related. These interpretations lend

some support to the argument that novices organize around

"surface" features while experts organize around "deep"

features (e.g., Adelson [1981]).










Task 2

Hypothesis 4

The purpose of testing hypothesis 4 is to determine

if the control procedure necessary to prevent or detect an

error or irregularity is a feature that is salient to

auditors. The dependent measure is Do, the difference

between the mean of the entries where the target and the

judged error share the same internal control objective and

the mean of the entries where the target and judged errors

do not share the same objective. The null hypothesis

being tested is

H0: Ti = 0

where T is the treatment effect for the ith group, and the

groups are the four experience levels. The alternative

hypothesis is

Ha: Ti > 0.

The large sample approximation of the Wilcoxon Signed Rank

Test (see Hollander and Wolfe [1973]) was performed on

each level. The null hypothesis of no treatment effect

was rejected for all four levels: students (Z=2.677;

p=.004), juniors (Z=2.391; p=.008), seniors (Z=2.937;

p=.002), and managers (Z=3.945; p=.000). Therefore,

control procedure seems to be a salient feature.

Hypothesis 5

The purpose of testing hypothesis 5 is to determine

if the control procedure necessary to prevent or detect an

error is a feature that becomes more salient with











experience. The dependent measure is Do, the difference

between the mean of the entries where the target and the

judged error share the same internal control objective and

the mean of the entries where the target and judged errors

do not share the same objective. "he alternative

hypothesis is

Ha: Tstudents < Tjuniors I Tseniors < Tmanagers

where at least one inequality is strict. The mean

(median) values of Do for the different levels of subjects

are presented in table 2. The large sample approximation

of the Jonckheere Test yielded a Z statistic equal to

2.750. This allows rejection of the null hypothesis at a

probability level of .003. This is indicative of a highly

significant experience effect. The Wilcoxon Rank Sum Test

was performed on all pairwise comparisons, and the

following treatment effects were significantly different:

students versus seniors (Z=1.375; p=.085), students versus

managers (Z=2.634; p=.004), juniors versus seniors

(Z=1.586; p=.056), and juniors versus managers (Z=2.541;

p=.006). Therefore, students and juniors seemed to form

one group, and seniors and managers seemed to form

another.

Hypothesis 6

The purpose of testing hypothesis 6 is to determine

if the likely perpetrator of an error or irregularity is a

feature that is salient to auditors. The dependent









variable is Dd, the difference between the mean of the

entries where the target error and the judged error share

the same department and the mean of the entries where the

target and judged error do not share the same department.

The null hypothesis being tested is

H0: Ti = 0,

where T is the treatment effect for the ith group, and the

groups are the four experience positions. The alternative

hypothesis is

Ha: Ti > 0.

The large sample approximation of the Wilcoxon Signed Rank

Test was performed on each level. The null hypothesis of

no treatment effect was rejected for all four levels:

students (Z=3.011; p=.001), juniors (Z=3.689; p=.000),

seniors (Z=2.589; p=.005), and managers (Z=3.036; p=.001).

Therefore, the likely perpetrator seems to be a salient

feature.

Hypothesis 7

The purpose of testing hypothesis 7 is to determine

if the likely perpetrator of an error or irregularity is a

feature that becomes more salient with experience. The

dependent variable is Dd, the difference between the mean

of the entries where the target error and the judged error

share the same department and the mean of the entries

where the target and judged error do not share the same

department. The alternative hypothesis is

Ha: Tstudents Tjuniors I Tseniors < Tmanagers










where at least one inequality is strict. The mean

(median) values of Dd for the different levels of subjects

are presented in table 2. The large sample approximation

of the Jonckheere test on the four levels yielded a Z

statistic equal to 0.169. The probability level

associated with this statistic is .433. Therefore, the

null hypothesis of no treatment effect can not be

rejected. This result and the results of the test of

hypothesis 7 imply that likely perpetrator is an obvious

feature.

Exploratory Analysis

In order to determine any dimensionality or related

clusters that exist in the data, individual differences

multidimensional scaling and individual differences

additive clustering were performed. Table 5 and figure 2

present the two-dimensional individual differences ordinal

scaling solution for all the subjects combined. Stress

for this solution was .306 and R2 was .276. Solutions

with additional dimensions improved R2 (stress): three

dimensions .384 (.233), four dimensions .450 (.161).

However, additional dimensions did not aid in the

interpretation of these solutions. Therefore, only the

two dimensional solution is presented. This solution

yields dimensions that are quite interpretable. The

dimension that runs from the northwest corner to the

southeast corner is related to the department .n which the









error occurred. The order of items along this dimension

is 2, 5, 1, 8, 3, 6, 7, and 4 (the figures presented are

slightly distorted because the scale is different on the

two axes). The northwest end of the dimension includes

errors that are from the cash receipts (1 and 5) and

general accounting (2 and 8) departments. The southeast

end of the dimension includes errors that are from the

sales order (4 and 7) and billing (3 and 6) departments.

The dimension that runs from the southwest corner to the

northeast corner seems to be related to the severity of

the problem. The order of the items along this dimension

is 1, 6, 7, 5, 2, 4, 3, and 8. The southwest end includes

items that are more likely to be classified as

irregularities (1, 5, 6, and 7). The northeast end

includes items that are more likely to be considered

errors (2, 3, 4, and 8).

Scaling solutions were also derived for the four

positions separately. The same two dimensions present in

the general solution are present in the senior and manager

solution (see tables 8-9 and figures 5-6). For seniors

the dimension that runs from the southeast corner to the

northwest corner (ordering: 2, 5, 1, 8, 3, 4, 6, 7) is

related to departments, and the dimension that runs from

the northeast corner to the southwest corner (ordering: 1,

5, 7, 6, 2, 4, 3, 8) is related to severity of the

problem. For managers the dimension that runs from the

northwest corner to the southeast corner (ordering: 2, 5,










1, 8, 3, 6, 7, 4) represents departments, and the

dimension that runs from the southwest corner to the

northeast corner (ordering: 6, 1, 5, 7, 4, 2, 3, 8)

represents severity of the problem.

These two dimensions are not as clearly present in

the solutions of the student and junior levels (see tables

6-7 and figures 3-4). For students the dimension that

runs from north to south (ordering: 8, 7, 2, 4, 3, 1, 5,

6) seems to be related to severity since items 1, 5, and

6, which are all irregularities, are at the south end.

The west to east dimension (ordering: 2, 1, 7, 5, 8, 6, 3,

4), however, is difficult to interpret. For juniors the

dimension that runs from northeast to southwest (ordering:

5, 1, 2, 8, 3, 7, 6, 4) can be interpreted as representing

departments. However, the southeast to northwest

dimension (ordering: 7, 2, 4, 1, 5, 6, 3, 8) is difficult

to interpret. Therefore, the more experienced subjects

seem to be aware of two features of errors and

irregularities while the less experienced subjects seem to

be aware of only one.

The general four-group individual differences

additive clustering solution, with R2=.428, yielded

interpretable groups (see table 10). Addition of a fifth

group added only .04 to R2, and addition of subsequent

groups added even less to the explanatory power of the

solution. Therefore, the additional groups were ignored.









The items in group 1 i,, 5) are both irregularities, and

both occur in the cash receipts department. The items in

group 2 (3, 8) are non-cash errors that violate the

validity objective. Group 3 contains two items (2, 5)

that both result in overstatement of accounts receivable.

The items in group 4 (1, 6) are irregularities that result

in an understatement of accounts receivable. Therefore,

the subjects seem to group on features such as severity

(error versus irregularity), department, internal control

objective, and account affected. Clustering solutions

were also obtained for the four positions separately (see

table 10). These solutions contain many of the same

groups that were identified in the general solution. For

instance, all four experience level solutions contained a

group with items 3 and 8. This would suggest that

subjects from all experience levels search for items that

violate the validity objective upon discovery of an item

that violates the validity objective. Both the student

level and the manager level solutions contained a group

with items 1 and 5. This group could be interpreted as an

irregularities group or a cash receipts department group.

On the other hand, junior and senior solutions contain a

group with items 1, 5, and 6. This cluster can be

interpreted as an irregularities cluster. A plausible

interpretation for this observation is that upon discovery

of an irregularity during an audit, juniors and seniors

are more likely than students and managers to search for







65



additional irregularities. Further comparisons of the

solutions yield no obvious insights to differences between

experience levels.







66


TABLE 5

MDS Plot Symbols and Coordinates Task 2 General


Coordinates
Coordinates


Error Number


Plot Symbol


Dimension 1

-1.3267
-0.9283
0.9051
1.4516
-1.2871
-0.0174
0.4124
0.7903


Dimension 2

-0.3160
0.9848
1.0723
-0.4219
0.2501
-1.4297
-1.3900
1.2505







67


DIM2 ---------------------------------------------------------
I
2.0+





1.5+

8

3
1.0+ 2





0.5+

5


0.0+-------------------- ---------------------------


1
4
-0.5+





-1.0+




6 7
-1.5+





-2.0+

----+----+----+----+----+----+----+----+----+----+----+--
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
DIM1

FIGURE 2


MDS Plot Task 2 General







68


TABLE 6

MDS Plot Symbols and Coordinates Task 2 Students
-------------------------------------------------------

Coordinates

Error Number Plot Symbol Dimension 1 Dimension 2

1 1 -0.8363 -1.1686
2 2 -1.2822 0.6632
3 3 1.4022 0.2811
4 4 1.6577 0.2940
5 5 -0.5079 -1.2238
6 6 0.4469 -1.2645
7 7 -0.6605 1.1463
8 8 -0.2201 1.2722







69



DIM2-----------------------------------------------------

2.0+





1.5+

8

7
1.0+


2

0.5+

3 4


0.0+-------- ---------------- ---------------------------





-0.5+





-1.0+

1 5
6

-1.5+





-2.0+

----+--------------------------------------------
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
DIM1

FIGURE 3


MDS Plot Task 2 Students







70


TABLE 7

MDS Plot Symbols and Coordinates Task 2 Juniors
-- ----------------------------------------------------


Coordinates

Error Number Plot Symbol Dimension 1 Dimension 2

1 0.9913 0.9087
2 2 1.3055 -0.4127
3 3 -1.1878 0.5567
4 4 -0.3931 -1.6327
5 5 0.8867 1.0978
6 6 -1.2628 -0.3162
7 7 0.63 -1.2633
8 S -0.97,_ 1.0617







71


DIM2---------------------------------------------

2.0+





1.5+




8 5
1.0+
1


3
0.5+





0.0+------------------------------------------------------



6
2
-0.5+





-1.0+


7

-1.5+
4




-2.0+

--------+-----------------------------+----------------
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
DIM1

FIGURE 4


MDS Plot Task 2 Juniors







72


TABLE 8

MDS Plot Symbols and Coordinates Task 2 Seniors


Coordinates


Error Number


Plot Symbol


Dimension 1


1.2675
1.1542
-0.6977
-1.3969
1.3102
-0.4559
-0.5695
-0.6119


Dimension 2

0.4778
-0.6825
-1.2222
-0.0708
0.0290
1.3466
1.4727
-1.3506











DIM2-------------------------------------------

2.0+





1.5+ 7

6


1.0+





0.5+





0.0+---------------------------- ------------5--------------
4



-0.5+

2


-1.0+

3

8
-1.5+





-2.0+

---+----+----+----+----+----------+----+----------------
-2.5 -2.0 -1.5 -1.0 -0.5 -.0 0.5 1.0 1.5 2.0 2.5
DIM1

FIGURE 5


MDS Plot Task 2 Seniors






74


TABLE 9

MDS Plot Symbols and Coordinates Task 2 Managers


Coordinates

Error Number Plot Symbol Dimension 1 Dimension 2

1 1 -1.3220 -0.2400
2 2 -0.6388 0.9539
3 3 0.7800 1.1231
4 4 1.1744 -1.0162
5 5 -1.2376 0.2924
6 6 -0.6821 -1.2761
7 7 1.0132 -1.1455
8 8 0.9129 1.3086











DIM2--------------------------------------------------

2.0+





1.5+

8

3
1.0+ 2


0.5+

5


0.0+--------------------------

1


-0.5+





-1.0+



6

-1.5+





-2.0+

---------2.5 -2.0 -1.5 ----.0 -0.5---
-2.5 -2.0 -1.5 -1.0 -0.5


- ---------------------------












4
7












-----+----+----+----+------
0.0 0.5 1.0 1.5 2.0 2.5
DIM1


FIGURE 6


MDS Plot Task 2 Managers










TABLE 10

Overclus Solutions Task 2


Exp. Level

General









Students









Juniors









Seniors









Managers


Group

1

2

3

4


Members

1 5

3 8

2 5

1 6


R2

0.16

0.32

0.38

0.43


0.16

0.29

0.38

0.45


0.22

0.36

0.48

0.54


0.20

0.41

0.49

0.57


0.16

0.34

0.43

0.50
















CHAPTER 5

SUMMARY AND DISCUSSION OF RESEARCH

This chapter presents a summary and discussion of the

results of this study. The first section reiterates the

purpose of this research study. The second section

discusses the results associated with the quantity and

quality of knowledge, typicality, and salient features of

errors and irregularities. The third section serves as a

general discussion of the results and some implications of

the results. The final two sections present some

limitations of this research project and some ideas for

future research.


Purpose of Research

The stated objective of this research was to directly

examine differences in the knowledge base for errors

between subjects with different levels of experience. The

theoretical motivation for conducting this examination was

to support the hypothesized link between audit experience

and differences in the auditor's knowledge base and to

demonstrate that experience leads to reasoning of a more

causal nature. The practical motivation was that

identification of differences between experience levels

would be a first step toward the reduction of these









differences either by training, decision aids, or expert

systems.


Discussion of Results

Quantity and Quality of Knowledge

The number of errors and irregularities known by the

auditor was found to increase with experience in the test

of hypothesis 1. Students and juniors seemed to form one

group, while seniors and managers seemed to form another

in terms of significant differences. This can be

explained by the fact that the juniors only had mean

(r .dian) experience of 1.04 (1.00) years. The juniors are

similar to the students in the respect that they have not

had to plan an audit of this cycle, while it is probable

that many of the seniors and managers have.

It is surprising that this effect was not even

stronger. This could have been caused by the fact that

the students and juniors can still easily retrieve the

textbook list of errors and irregularities from their

college auditing course. Having students who just

completed their study of the cycle being examined serve as

subjects is likely to have biased this test against

finding a result.

As suggested by hypothesis 2, the number of

intrusions recalled by the subjects did decrease with

experience. In this case, students seemed to form a group

by themselves by recalling significantly more intrusions









than the other experience levels. Apparently, students

have yet to form clear-cut boundaries between different

transaction cycles and still have some misunderstandings

about what an error or an irregularity is.

Typicality

As suggested by hypothesis 3, more experienced

auditors did appear to be aware of more atypical errors

and irregularities. In terms of significant difference,

students seemed to form one group, and juniors, seniors,

and managers formed another. The fact that the results

are stronger when minimum item dominance is employed as

the dependent measure rather than avc .age item dominance

is not totally surprising. All levels of experience were

expected to know the typical items. Only the knowledge of

atypical items was expected to increase with experience.

When average item dominance is employed as the dependent

measure, the fact that all experience levels recall the

typical items tends to reduce the possibility of finding a

significant result. For this reason a measure which

ignores the typical items, like minimum item dominance, is

likely to be more sensitive in distinguishing between

experience levels.

Features of Errors and Irregularities

Based on a review of the auditing literature, it was

expected that the control procedure necessary to prevent

or detect an error or irregularity and the likely

perpetrator of an error or irregularity would be features











that would be salient to auditors. These expectations

were confirmed for all levels of experience in the tests

of hypotheses 4 and 6. It was corroborated by the

exploratory analysis that these two features or their

surrogates, control objective and department, were

salient. The exploratory analysis also suggested that

three other features were important to some auditors. One

of these was the seriousness of the error. This is

related to two of the characteristics mentioned in the

professional literature, materiality and intent. This

feature was most clearly seen in task 2, and it is logical

that an auditor would search for other serious errors if

s/he had discovered a serious error during the audit. A

number of experienced subjects in post-experimental

discussions indicated that if an error or irregularity was

discovered during an audit that could possibly have been

the result of management override, the audit would be

expanded to search for other examples of management

override. Another feature that appeared to be important

was the account affected by the error or irregularity.

This is related to the influence of the item on the

financial statements, a characteristic mentioned by Mautz

and Sharaf [1961] and SAS No. 53 [1988]. A final feature

that some of the subjects seemed to group on was

opposites. This is a natural characteristic of concepts

within some domains, and the use of this dimension by a









subject demonstrates a somewhat sophisticated knowledge of

the subject area.

It was also expected that the two features, control

procedure and likely perpetrator, would become more

salient with experience because these features are related

to causal explanations. The test of hypothesis 5

demonstrated that control procedure or control objective

did, indeed, become more salient as experience increased.

Once again, the students and juniors formed one group

while the seniors and managers formed another in terms of

significant differences. The test of hypothesis 7 did not

demonstrate that the perpetrator or the department of the

error or irregularity became more salient as experience

increased. This dimension is apparently a fairly obvious

one and, even though it is extremely important to all

auditors (as demonstrated in the test of hypothesis 6), it

is likely that the level of salience of this feature to

the auditor reaches an asymptote early in ones career.

Therefore, perpetrator or department might be thought of

as a "surface" feature and control procedure or control

objective might be thought of as a "deep" feature of

errors and irregularities.


General Discussion

A number of general conclusions can be made in regard

to the results and how these results relate to previous

research. First, differences between experience levels









have definitely been identified. Although this is

consistent with the previous literature, this research

utilized methods with a limited history in the study of

experience. The unconstrained free recall paradigm has

been employed once previously to study experience

[Hutchinson, 1983], and the conditional prediction

paradigm has never been employed in the study of

experience to my knowledge. Second, for the most part,

students and juniors are not significantly different from

each other in their judgments. The same can be said for

seniors and managers. This is not totally consistent with

previous auditing research in which possessing experience

often was more important than the amount of experience

(see Ashton and Kramer [1980]). Third, the finding

related to quantity of knowledge is consistent with the

previous literature (e.g., Hutchinson [1983]). Fourth,

subjects with more experience did appear to be more aware

of atypical items. Typicality of response across

experience levels has not been previously examined.

Fifth, two features of errors and irregularities that are

related to causal explanation, internal control objective

violated and the department where the error occurred, were

found to be s -ient. A number of other features that were

salient were also identified: opposites, the severity of

the problem, and the account affected. Previous auditing

research has identified transaction cycle [Libby, 1985]

and audit objective violated [Libby and Frederick, 1988]










as characteristics around which errors are organized.

Finally, task 2 demonstrated that one feature, internal

control objective violated, appeared to increase in

salience with experience while another feature, department

where the error occurred, did not appear to increase in

salience with experience. The exploratory analysis

indicated that another feature, severity of the problem,

was more salient to more experienced subjects. Libby and

Frederick [1988] employing a part-list cuing paradigm

found evidence of categorizing errors on the basis of

transaction cycle for experienced auditors but not for

students. Therefore, this study offers a great deal of

additional insight to the existing literature regarding

experience (especially experience in auditing).

Whether these results offer any prescriptions for

practice is a difficult question to answer. However,

inasmuch as no absolute criterion exists for most audit

judgments, it is reasonable to assume that most auditing

firms would prefer that their inexperienced auditors make

decisions that are similar to the decisions made by their

experienced auditors. For inexperienced auditors to make

judgments that are more similar to the judgments of

experienced auditors, it seems that the inexperienced

auditors must be made aware of more errors and

irregularities and the need to put more weight on certain

features of errors and irregularities when an audit









situation similar to the situation in task 2 occurs. This

might be accomplished by training, by construction of a

decision aid or an expert system, or by a more explicit

description of the procedure to follow in the audit manual

when a situation such as the one modeled in task 2 occurs.


Limitations

A number of limitations of this research exist.

First, because the independent variable (experience level)

is observed rather than manipulated, sample selection bias

might provide an alternative explanation for the results.

It is possible that the auditors who are promoted are more

intelligent or more motivated in the experimental setting.

If this is the case, this is a serious limitation. This

limitation might be partially remedied by either observing

the same subjects over time as they changed level or by

performing some type of covariate analysis in which

measures of intelligence or motivation are gathered for

the subjects. Second, perhaps an alternative method of

measuring typicality should have been employed in the

testing of hypothesis 3. Item dominance scores were

derived from the responses of the subjects, and then were

used in evaluating the typicality of the subjects' recall

protocols. An independently determined typicality index

might be preferable. This limitation could be dealt with

either by calculating item dominance scores for an

independent group or by having an independent group rate










the forty-six errors and irregularities on typicality.

Third, although the eight items employed in task 2 were

carefully selected and verified as unambiguous in terms of

the two dimensions of interest, it is possible that

alternative dimensions present in these items are driving

the results in the tests of hypotheses 4 through 7. This

was one of the reasons for also performing MDS and

clustering analysis on the task 2 data. This exploratory

analysis confirms the importance of control procedure and

likely perpetrator as relevant features, but also suggests

that other features such as seriousness and account

affected are important. The tests performed, therefore,

were based on the assumption that these alternative

features were randomized across the set of items employed

and were not confounded with control procedure and likely

perpetrator. Finally, some of the interpretations of the

dimensions and the cluster groups by the researcher might

have been inappropriate. This is often a problem in this

type of exploratory research, but the interpretations

discussed were supported by two experienced auditing

professors.


Future Research

This project suggests some possibilities for future

research. First, the results of the exploratory analyses

indicate that auditors focus on other dimensions of errors

and irregularities besides the internal control objective









violated and the department where the error occurred.

These dimensions include opposites, the severity of the

problem, and the account affected. Explicit testing for

the existence of and the effect of experience upon these

dimensions is a possible extension of this study. Second,

a longitudinal study in which subjects performed the two

experimental tasks at different points in their careers

would eliminate the possibility of a sample selection

bias. Third, since the audit structure employed by

different firms is a current topic of interest (see

Cushing and Loebbecke [1986]), one could examine whether

any firm effects exist in the development of knowledge

structure. This might aid in the explanation of how the

knowledge structures develop (or at least how the work

environment affects this development). Fourth, the

investigation of how experience in auditing affects the

development of cognitive capacities other than the

knowledge base is possible. Other possible cognitive

capacities affected by experience include: cognitive

effort, the ability to analyze information, elaboration

upon given information, and the ability to remember [Alba

and Hutchinson, 1987]. Fifth, in order to complete the

theoretical link between experience and expertise, the

effect of the identified differences in the knowledge base

on behavior and performance of auditors in realistic audit

settings could be examined. Finally, one might test to

determine whether training or decision aids can eliminate







87



differences in knowledge structure. This has been

demonstrated in other fields [Brown and Stanners, 1983]

and would certainly be of interest to practitioners.
















APPENDIX A


TASK 1 INSTRUMENT

List as many possible errors or irregularities as you can
that could occur in the sales, receivables, and cash
receipts cycle of a typical wholesaling or manufacturing
company (Be specific). You have 15 minutes to complete
this task. After recording each item, look at the time
clock in front Df the room and record the number of
seconds that .h -e passed before continuing.

Errors or Irregularities Time


(Additional space on following pages)
















APPENDIX B


mASK 2 INSTRUMENT

Assume that you are part of the audit team examining the
sales, receivables, and cash receipts cycle of a typical
wholesaling or manufacturing firm. You may assume that
the company's internal control system has been judged to
be reliable in the past.

For such a client, what is your judgment of the
probability (on a scale from 0 to 100) that a material
error or irregularity of the following type could occur
during the year under examination? [You might think of
this as a judgment of how many such clients out of 100
would exhibit a material error or irregularity of the
following type.]


1. Customers failed to pay within the discount
period and remitted the payment in full.
Nevertheless, discounts were approved, and the
amount of the discounts were misappropriated. 1.

2. Accounts receivable were aged incorrectly;
potentially uncollectible amounts were not
recognized. 2.

3. Billings were recorded, but goods were not
shipped. 3.

4. Customer order specifications were not met
as to type and/or quantity. 4.

5. Lapping occurred. 5.

6. Management, employees, or 3rd parties
received goods without being billed. 6.

7. Orders were accepted in violation of the
company's credit policies. 7.

8. Revenues were recorded in the current period
when they should have been recorded in the next
period. 8.









Treat pages 2-9 as totally independent situations.

Now assume that during the audit of the sales,
receivables, and cash receipts cycle of the client
described on page 1, the following material error or
irregularity was discovered.


Orders were accepted in violation of the company's credit
policies.


In light of this discovery, what now is your judgment of
the probability (on a scale from 0 to 100) that a material
error or irregularity of the following type could occur
during the year under examination?

You may look back at page 1 if you wish.



1. Customers failed to pay within the discount
period and remitted the payment in full.
Nevertheless, discounts were approved and the
amount of the discounts were misappropriated. 1.

2. Accounts receivable were aged incorrectly;
potentially uncollectible amounts were not
recognized. 2.

3. Billings were recorded, but goods were not
shipped. 3.

4. Customer order specifications were not met
as to type and/or quantity. 4.

5. Lapping occurred. 5.

6. Management, employees, or 3rd parties
received goods without being billed. 6.

7. Revenues were recorded in the current period
when they should have been recorded in the next
period. 7.









Treat pages 2-9 as totally independent situations.

Now assume that during the audit of the sales,
receivables, and cash receipts cycle of the client
described on page 1, the following material error or
irregularity was discovered.


Accounts receivable were aged incorrectly; potentially
uncollectible amounts were not recognized.


In light of this discovery, what now is your judgment of
the probability (on a scale from 0 to 100) that a material
error or irregularity of the following type could occur
during the year under examination?

You may look back at page 1 if you wish.


1. Customers failed to pay within the discount
period and remitted the payment in full.
Nevertheless, discounts were approved, and the
amount of the discounts were misappropriated. 1.

2. Billings were recorded, but goods were not
shipped. 2.

3. Customer order specifications were not met
as to type and/or quantity. 3.

4. Lapping occurred. 4.

5. Management, employees, or 3rd parties
received goods without being billed. 5.

6. Orders were accepted in violation of the
company's credit policies. 6.

7. Revenues were recorded in the current period
when they should have been recorded in the next
period. 7.




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