 

 

         

 

.22.}: .

 

... L . s.\
.

T: .
\2 f

.0...

if...
3 .

‘ 31.....337 - . .

:23. an;

 

. 1'9. .
.. .2.» I..l.

5.5.5 33...»..éikﬂam .....b?a.u..r.».m.,..u. .

L

5:192

 

4|...

.i. {4.2

1. 9-...

1r .¢|.r

      

. .. . A...

4%“? <5. 3.2.5.
gun? ﬂung."
«an. «a...» .m...“ .5
32:5. . “x 1,:
$4.“. .. 1 ,1.

swunkma: 13...: w... .
a . 3

      

    

 

                 

              

      

     

  

                  

 

493.. .v. . z,
.59. . $42.2? .
«#5. . ﬂ. .4 Jun/n.1,...» .
, .w. 3 2 _ .1» A .1! .Ubr ..$...- 1‘}.
5.... . £3.22 .. .5 .34.“...th a
.mm 5-5 §#1.ﬂ£ 1.38m“ .71
'4. é '3 ..a $42.. . r . .
gm... 3.. snw%m m... ., . ,. .
.9 km. .1: 5...... m" ; .
a. t. bran? - ... . 1.5.3.2.“...an
. , . 4.4.3.33... i..........,..uu..,ii1 ;.
, 226W, . (22.3431... . 44.11.“.
.. . ; ....u.2..ds..n¢2.
fvhlhl...

  

           

    

2!. .39.
.32, 9....
r? .

e. .r.

L. .l- :5
o 9..

  

 

7L.

 

 

 

    

  

  

.. Hunk;
¢ 2 3...?
d...»

           

. .3 2.1.1...»
awaititcv
. ‘hﬁrll..- 0".v
.5...

 

lﬂvll‘tnll 1:3. . In).
. ~0v1‘l‘l1ﬂuutmnn‘bll Hit. .
I Irt.‘ .
n7

   

      

   

..uw.vuw.......nuan9«n.?f-
Ii. 1. I 7...: .i.
-.. 2 $8.“. .25: itvh t.
56L. um. . 3... F!!r:7nu.$ié. (D .2».
. {2,314,221.25 r. z
. 23:221.! battle..- .
I 1;»-..{2243'2212’}
.l. 5... an“ hoes... Hal's...

                          

v 4:42.):burﬂfﬁl. \I \

.205?

       

.......\.‘o2|»n..w2.nuv.3.. c!
-J,rbdvd._..vrﬁﬂ.1
2.3.1)... I...

. 21;!

 

1a.»;
3.01.. Orriv
.. rt; 3.1.3.]. In”:

   

Thims

HHIHIHIIHIII)HIHHUIHIHNHIIHllWllHIIHHIHHI

1293 01688 0555

This is to certify that the

dissertation entitled '

REA KNOWLEDGE ACQUISITION AND RELATED CONCEPTUAL
DATABASE DESIGN PERFORMANCE

presented by

Gregory James Gerard

 

has been accepted towards fulﬁllment
of the requirements for

Ph.D. Business Administration

degree in

 

 

(LLZLteuL¢-{:/¢R:<3tﬂgzz;

Major professor \J

 

Date 8/28/98

 

MSU is an Affirmative Action ’Equal Opportunity Institution 0-12771

 

 

LIBRARY

Michigan State
University

 

 

 

PLACE IN RETURN BOX
to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/” WM“

REA KNOWLEDGE ACQUISITION AND RELATED CONCEPTUAL DATABASE
DESIGN PERFORMANCE

By

Gregory James Gerard

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Accounting

1 998

ABSTRACT

REA KNOWLEDGE ACQUISITION AND RELATED CONCEPTUAL DATABASE
DESIGN PERFORMANCE

By-

Gregory James Gerard

This dissertation uses the experimental method to examine (l) the eﬁ‘ect of REA
conceptual modeling (Resources, Events and Agents; McCarthy 1982) experience on
REA template knowledge structure and attribute aggregation knowledge structure, and
(2) the effect of these knowledge structures on conceptual database design performance.
. 54 undergraduate and 47 graduate accounting information systems students (representing
inexperienced and experienced system designers) participated in the three tasks of the
study.

Task 1 involved the free recall of both an REA diagram and a random diagram to
infer REA template knowledge structure. Experienced participants recalled signiﬁcantly
(p < 0.0001) more diagram information than inexperienced participants in the REA
diagram condition. No signiﬁcant diﬁ‘erences between groups Ewas noted in the random
diagram condition. This suggests that experienced REA designers organize their
knowledge of business processes according to the REA template.

Task 2 involved the free recall of a randomized list of attributes to infer attribute
aggregation knowledge structure. High clustering scores indicated that the attributes
were being aggregated to entities. Experienced participants' clustering scores were

signiﬁcantly greater (p = 0.001) than inexperienced participants. This suggests that

experienced REA designers have a better developed schema for how attributes aggregate
to entities.

The third task involved the conceptual database design of a business enterprise.
Performance on this task was regressed on the knowledge structure measures from tasks 1
and 2 and additional measures of knowledge content and ability. Results of the
regression revealed that (l) REA knowledge structure is strongly associated (p < 0.001)
with conceptual design performance, and (2) attribute aggregation knowledge structure is
weakly associated (p = 0.07) with conceptual design performance while controlling for
knowledge content and ability. These results suggest that one of the determinants of
successful conceptual database design is the organization of knowledge in long-term
memory. Furthermore, the REA template appears to be a useful cognitive tool for

structuring knowledge.

ACKNOWLEDGMENTS

I would like to thank the members of my dissertation committee - Dr. William
McCarthy (Chair), Dr. Frank Boster, Dr. Joan Luﬁ, and Dr. Severin Grabski -- for their
time, guidance, and insights throughout the dissertation and doctoral program. I greatly
appreciate the opportunities I had to interact with these scholars.

I would like to thank members of Michigan State University's "Systems Group"
(in addition to Drs. McCarthy and Grabski): Guido Geerts, Julie Smith David, Robin
Poston, and especially Cheryl Dunn who has helped me in many ways.

I would like to thank Malcolm McLelland for the numerous interesting
discussions we have had, for our experiences in the doctoral program, but most of all for
the friendship.

I would also like to thank my family, my brother Chris, and my parents, James
and Mary Ann Gerard. My parents have provided constant support and encouragement.
Additionally, they gave me an appreciation for the value of education.

Most importantly, I would like to thank my wife, Jayne, whose love, patience, and

understanding were fundamental to my success.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... vii
LIST OF FIGURES ....................................................................................................... viii
1. INTRODUCTION ................................................................................................. l
1.1 Background ................................................................................................ 4
1.1.1 Conceptual Database Design ............................................................ 4
1.1.2 REA Accounting Model ................................................................. 10
1.2 Overview of Hypotheses .......................................................................... 12
1.3 Overview of Research Design and Experimental Tests ........................... 15
1.4 Organization of the Dissertation .............................................................. 18
2. THEORY AND HYPOTHESES DEVELOPMENT ........................................... 19
2.1 Review of Accounting and Information Systems Literature ................... 19
2.2 Review of Psychology Literature ............................................................. 26
2.3 Model and Hypotheses Development ...................................................... 31
2.3.1 Experience and REA Template Knowledge Structure ..................... 33
2.3.2 Experience and Attribute Aggregation Knowledge Structure ........ 36
2.3.3 Knowledge Structure and Performance .......................................... 40
2.4 Summary of Chapter Two ........................................................................ 41
3. RESEARCH METHOD ....................................................................................... 44
3.1 Research Method ..................................................................................... 44
3.2 Participants ............................................................................................... 45
3.3 Dependent Variables ................................................................................ 49
3 .4 Tasks and Procedures ............................................................................... 49
3.4.1 Task 1 .............................................................................................. 50
3.4.2 Task 2 .............................................................................................. 54
3.4.3 Task 3 .............................................................................................. 57
3.4.4 Supplemental Tasks ........................................................................ 65
3.5 Summary of Chapter Three ...................................................................... 71
4. RESULTS ............................................................................................................ 73
4.1 Tests for Order Effects ............................................................................. 74
4.2 Hypothesis Testing for REA Template Knowledge Structure ................. 75
4.3 Hypothesis Testing for Attribute Aggregation Knowledge Structure ..... 77
4.4 Hypothesis Testing for Relationship Between Knowledge Structure
and Conceptual Database Design Performance ....................................... 82
4.4.1 Speciﬁcation Test of Hypotheses 3 and 4 ....................................... 87
4.4.2 Ex post Analysis of Hypotheses 3 and 4 ......................................... 88
4.4.3 Ex post Analysis of Hypotheses 3 and 4 Based on Audit Class ..... 91

4.5 Summary of Chapter Four ....................................................................... 94
5. CONCLUSION .................................................................................................... 95
5.1 Summary of Research Questions, Theory, and Hypotheses .................... 95
5.2 Summary of Research Method and Results ............................................. 98
5.3 Contributions ......................................................................................... 100
5.4 Limitations ............................................................................................. 100
5.5 Future Research Directions .................................................................... 102
LIST OF REFERENCES ............................................................................................... 104

vi

LIST OF TABLES

Table 1 - ANOVA Tables for Practice Effects Tests ....................................................... 76
Table 2 - Means and Standard Deviations of the Number of Items Recalled

for Hypothesis 1 ............................................................................................... 78
Table 3 - AN OVA Table for Hypothesis 1 ....................................................................... 78
Table 4 - Summary of Group Simple Effects for Hypothesis 1 ....................................... 79
Table 5 - Means, Standard Deviations, Standard Errors, and Skewness of the

ARC Clustering Index for Hypothesis 2 .......................................................... 80
Table 6 - Means, Standard Deviations, Standard Errors of the Number of Attributes

Recalled for Hypothesis 2 ................................................................................. 82
Table 7 - Descriptive Statistics and Correlation Matrix (Pairwise Deletion) .................. 84
Table 8 - Descriptive Statistics and Correlation Matrix (Listwise Deletion) .................. 84
Table 9 - Multiple Regression for Hypotheses 3 and 4 ............................... i ..................... 85
Table 10 - Sensitivity Analysis of Differently Weighted Dependent Variable ............... 85
Table 11 — Multiple Regression (with interaction term) for Hypotheses 3 and 4 ............ 85
Table 12 - Multiple Regression (Speciﬁcation Test) for Hypotheses 3 and 4 ................. 87
Table 13 - Descriptive Statistics and Correlation Matrix (Listwise Deletion) ................ 89
Table 14 - Ex post Multiple Regression for Hypotheses 3 and 4 ..................................... 90
Table 15 - Descriptive Statistics and Correlation Matrix for Hypotheses 3 and 4 .......... 92
Table 16 - Ex post Multiple Regression for Hypotheses 3 and 4 (with Audit) ................ 93

vii

LIST OF FIGURES

Figure 1 - Information Systems Life Cycle ....................................................................... 5
Figure 2 - Data-driven Approach to Information Systems Design .................................... 6
Figure 3 - Components of Entity-Relationship Model ...................................................... 9
Figure 4 - REA Template ................................................................................................. 11
Figure 5 - REA Template Instantiation of Revenue Cycle (without attributes) .............. 13
Figure 6 - Antecedents and Consequences of Knowledge ............................................... 32
Figure 7 - Predicted Interaction for Hypothesis 1 ............................................................ 36
Figure 8 - Derivation of Normalized Tables ..................................................................... 38
Figure 9 - Experimental Tasks ......................................................................................... 46
Figure 10 - REA Diagram for Task 1 .............................................................................. 51
Figure 11 - Random Diagram for Task 1 ......................................................................... 52
Figure 12 - List of Attributes for Free Recall Task .......................................................... 55
Figure 13 - Design Task ................................................................................................... 58
Figure 14 - Solution to Design Task ................................................................................ 60
Figure 15 - Scoring Sheet ................................................................................................ 63
Figure 16 - Knowledge Content Test ............................................................................... 67
Figure 17 - Exit Survey .................................................................................................... 69
Figure 18 - Experience x Diagram Interaction ................................................................. 79

Figure 19 - Mean Levels of ARC Scores for Participants in Inexperieneed Versus
Experienced Groups, Independent Samples Design ...................................... 80

viii

Chapter One
INTRODUCTION

The study of knowledge structures has been of interest across many domains
including accounting, database design, compumr programming, physics, and chess (e.g.,
Nelson, Libby, and Bonner 1995; Weber 1996; Adelson 1981; Chi, Feltovich, and Glaser
1981; Chase and Simon 1973; de Groot 1965; de Groot 1966). This dissertation extends
results and research methods from the accounting and psychology literature regarding
knowledge structures to the ﬁeld of accounting information systems (AIS). While much
empirical evidence has been gathered about knowledge structures in the auditing context,
very little is known about the knowledge structures of information system designers,
speciﬁcally designers of REA (Resource Event Agent) accounting systems (McCarthy
1982)

AIS designers, much like auditors, have domain speciﬁc knowledge that allows
them to perform their jobs successfully. Knowledge of transaction cycles stored in their
long-term memory, for example, may partly determine the effectiveness and efﬁciency of
their AIS design. Therefore, in an effort to understand the underlying cognitive processes
related to AIS design, it is important to develop an understanding of how designers’
knowledge structures are organized. Since memory and knowledge are fundamental to
conceptual database design, this dissertation is a good starting point for investigating the
relationship between REA knowledge acquisition and task performance.

REA accounting systems maintain disaggregated enterprise information that
meets the needs of accountants and non-accountants in a shared data environment

(McCarthy 1982). The advantage of REA systems is that they overcome ﬁve

1

weaknessesI associated with traditional accounting systems: (1) tracking a subset of
business events considered to be accounting related; (2) capturing and processing data in
an untimely manner; (3) capturing limited dimensions of transactions such as date,
account involved, and dollar amount; (4) storing data at a high level of aggregation; and
(5) incurring costly control (auditing) costs (Walker and Denna 1997, 23). These
advantages, coupled with new powerful technology,2 are beginning to cause a shiﬁ in
corporate accounting systems. Walker and Denna (1997, 30) assert, “Corporations such
as Sears Roebuck, Alcoa, IBM, and others are using event-based systems concepts.”3
Since the REA accounting model is taught at major US. universities (as
evidenced by sales of REA related textbooks such as Hollander, Denna, and Cherrington
1996; Romney, Steinbart, and Cushing 1997), many accountants and accounting system
designers are familiar with it. However, there has been little cognitive research related to
the designers and users of REA systems. Although Dunn (1994) and Dunn and Grabski
(1997) examine cognitive issues related to the use of REA systems by accounting
students, the cognitive study of REA designers has not yet been addressed. It also is
possible that insights gained ﬁ'om research on REA designers can be used in the design of
systems, such as CREASY (Geerts and McCarthy 1996) and REACH (McCarthy and

Rockwell 1996) which are intended to aid less knowledgeable designers and users. This

research also can provide insights into education and training of accounting system

 

' Other criticisms of traditional accounting systems have been made by McCarthy (1982), Johnson and
Kaplan (1987), and Denna, Cherrington, Andros, and Hollander (1993).

2 One such technology is software developed by Price Waterhouse called Geneva VI'I'T". This technology
has the ability to process very large data queries which previously were computationally time-consuming
or not possible (Walker and Denna 1997, 27).

designers and potentially increase the generalizability of prior accounting and
psychological research.

Additionally, cognitive research involving system designers may provide insights
into design errors. The consequences of information system design errors are important
since any mistakes made in design will be difﬁcult and costly to correct. Boehm (1976)
suggests that post-implementation changes cost, on average, 75 times more than changes
made during the analysis and conceptual design stage. Banker, Datar, Kemerer, and
Zweig (1993) provide evidence that software maintenance costs are positively and
signiﬁcantly correlated with existing levels of software complexity. Furthermore, Banker
et al. suggest that many times complexity is a result of poor design decisions. This
implies that software maintenance costs could be reduced by more effective design.4

This dissertation employs a series of research tasks to examine the information
encoding and retrieval behavior of experienced and inexperienced REA designers. The
speciﬁc research questions addressed are: (1) does experience with conceptual modeling
of REA systems result in the organization of knowledge on an REA basis in long-term
memory; (2) does this experience help designers aggregate attributes to entities; and (3)
how do these two types of knowledge (REA and attribute aggregation knowledge) affect
REA design performance while controlling for knowledge content and ability? The ﬁrst

two research questions will examine the theoretical relationship between experience and

 

3 Walker and Denna’s (1997) use of the term “event-based systems” is synonymous with Dunn and
McCarthy’s (I997) “REA accounting” rather than Sorter’s (1969) “events accounting” (see Dunn and
McCarthy 1997 for an analysis of the differences).

‘ Effective design, however, does not necessarily eliminate complexity.

knowledge, whereas the third research question will examine the theoretical relationship
between knowledge and performance.
1.1 Background

The task of designing an information system for a company is a complex activity.
Batini, Ceri, and Navathe (1992) describe a typical information system life cycle (see
Figure 1) with the following seven phases: feasibility study, requirements collection and
analysis, design, prototyping, implementation, validation and testing, and operation.’ By
employing the experimental method, the feasibility study and requirements analysis can
be held constant, allowing inferences to be drawn regarding the design phase. The design
phase, speciﬁcally the design of REA systems, is of particular interest since this is the
initial documentation of the data requirements.
1.1.1 Conceptual Database Design

The study of knowledge structures and knowledge acquisition is directly
applicable to the domain of AIS, in particular an AIS design method based on the REA
accounting model. The REA accounting model was ﬁrst introduced by McCarthy (1982)
as an approach to designing event-based enterprise-wide information systems that meet
the needs of accountants and non-accountants in a shared data environment.

Part of the theoretical basis of the REA accounting model is derived from
database design theory. There are three fundamental stages of database design:

conceptual, logical, and physical (Batini et a1. 1992; see Figure 2). Conceptual database

 

’ The number of phases in the life cycle varies by author (e.g., Kendall and Kendall 1995; Davis and Olson
1985); however, the notion of a systems development life cycle provides a valuable, organized planning
ﬁamework.

 

 

Feasibility study

 

 

l

 

 

Requirements
collection &
analysis

 

 

l

 

 

 

 

Prototyping

 

 

Design

Figure 1

 

 

 

Implementation

 

1

Validation 8:
testing

 

1

Operation

 

Information Systems Life Cycle
(Adapted from Batini et al. 1992, 5)

 

 

— -—- Mam-5v amo- ﬁg“ .' '

 

Data requirements

 

 

Conceptual design

 

 

 

Conceptual diagram

Logical Design
Logical diagram

1

Physical Design.

 

 

 

 

 

 

 

 

Physical diagram .

Figure 2
Data-driven approach to information systems design
(Adapted from Batini et al. 1992, 7)

design is the process of taking real-world phenomena and documenting (modeling) the
corresponding data of interest. A key aspect of conceptual database design is that it is
independent of both the database management system (DBMS) class (i.e., a conceptual
model can be implemented in a variety of DBMS’s such as relational, hierarchical, and
network systems), and the speciﬁc DBMS (e.g., the Microsoft Access DBMS). The
process of logical design involves taking the conceptual schema (diagram) and converting
it into a logical diagram based on the intended DBMS class requirements.6 The process
of physical design involves taking the logical diagram and converting it into a physical
diagram that indicates how the database will actually be implemented based on the
chosen DBMS. The physical diagram, therefore, provides the details regarding how the
data will be physically stored and accessed (Batini et al. 1992). Since conceptual design
is the ﬁrst phase of database design and since each phase builds upon it in succession, it
can be argued that it is the most important phase. Any undetected errors made in this
phase will be implemented in the ﬁnal DBMS and costly to undo.

Conceptual database design is performed using a chosen data modeling formalism
(i.e., a diagramming technique). For the purposes of this dissertation, all uses of the
entity-relationship (E-R) model will be based on Batini et al. (1992), which includes
enhancements to Chen’s (1976) model.7 The four basic components of an ER model are

entities, relationships, attributes, and cardinalities (see Figure 3 for their corresponding

 

° The formal use of “schema” can be found in both the database design (“...a schema is a static, time-
invariant collection of linguistic or graphic representations that describe the structure of the data of interest,
such as that within one organization.” Batini et a1. 1992, 26) and the psychology literature (“Schemata are
data structures for representing the generic concepts stored in memory. They exist for generalized
concepts underlying objects, situations, events, sequences of events, actions, and sequences of actions...”
Rumelhart and Ortony 1977, 101). In this dissertation, “diagram” will be used in reference to the database
term, and “schema” will be used in reference to the psychology term.

graphical representations). “Entities” are representations of real-world objects or events
(e.g., “Sale” and “Cash Receipt” in Figure 3). “Relationships” (“S/CR”) represent
aggregations of (connections between) entities which in this case would provide
information on which cash receipts are related to which sales and vice versa.8
“Attributes” represent speciﬁc characteristics of entities or relationships. In this case the
attributes of the entity “Sale” are “Invoice Number,” “Invoice Date,” and “Invoice
Amount.” A “primary key” is used to uniquely identify each entity (e. g., “Invoice
Number” and “Remittance Advice Number”). Primary keys are graphically displayed as
attributes which are blackened.

The “cardinalities” represent intra—relationship constraints that specify how
entities participate in relationships. There are two types of cardinalities: minimum
(MINC) and maximum (MAXC). MIN C deﬁnes whether an entity’s participation in a
pairing relationship is optional (MINC = 0) or mandatory (MINC = 1). MAXC deﬁnes
the maximum number of times an entity can participate in a relationship. MAXC can
equal 1 (representing “at most one”) or N (representing “more than one” or inﬁnite
participation). In Figure 3, the (0, 1) cardinality (located next to “Sale”) means that there
can be a sale without a simultaneous cash receipt (allowing for credit sales; MINC = 0),
and there can be only one cash receipt for each sale (prohibiting installment sales;
MAXC = 1). The (0, N) cardinality (located next to “Cash Receipt”) means that there can

be a cash receipt ﬁom other events besides a sale (allowing for cash receipts

 

7 The E-R model is the leading data modeling formalism (Batini et a1. 1992).

' This would represent an “open-item” accounting method in which customer payments are matched
against a speciﬁc invoice or invoices. The elimination of this “S/CR” relationship would represent a
“balance-forward” accounting method in which customer payments are not matched against speciﬁc
invoices. Instead, payments are subtracted from the total outstanding balance.

 

 

 

 

(0,1)
(0, N)
(1,1)
(1, N)

. INVOICE NUMBER
INVOICE DATE
? INVOICE AMOUNT

 

 

 

Entity

Relationship

Attribute

Cardinality Patterns (Intra-relationship constraints)

. REMITTANCE ADVICE NUMBER
REMITTANCE ADVICE DATE
? REMITTANCE ADVICE AMOUNT

 

 

* (o, 1) (o, N)
CASH
SALE Q RECEIPT

Figure 3

Components of Entity-Relationship Model (Batini et al. 1992)

from ﬁnancing; MINC = 0),9 and a cash receipt can relate to many sales (one cash receipt
can pay for many different sales; MAXC = N). Cardinalities are important because they
can serve as internal control speciﬁcations.
1.1.2 REA Accounting Model

The REA accounting model is a design method for capturing and maintaining
accounting phenomena in shared data environments (e. g., databases). The general REA
accounting model is based on sets representing economic resources, economic events,
and economic agents (see Figure 4 for an example of the generalized REA accounting
model).lo During the stage of conceptual database design, the REA accounting model is
used as a general template for representing a company’s enterprise-wide information
needs (e.g., the revenue, acquisition, payroll, ﬁnancing, and conversion cycles).'l The
ﬁnal product of the conceptual database design stage is a conceptual diagram that
provides a global view of the database. This global view is created by integrating several
local diagrams (e.g., each REA accounting cycle forms the basis of a local diagram; the

integration of all cycles creates the global diagram).’2

 

9 Alternatively, this could represent a cash receipt without a simultaneous sale (allowing for prepayment;
MINC = 0).

'° When comparing the REA model and the Entity-Relationship model, it is important to understand that
the REA model is a domain speciﬁc approach to designing accounting systems. It is used to model
accounting events using a template (pattern) consisting of resources, events, and agents. The REA model
has been most commonly represented by Chen’s (1976) constructs of rectangles, diamonds, etc., although
other formalisms such as Nijssen’s Information Analysis Methodology (N 1AM) (N ijssen and Halpin 1989)
have been used. The B-R model is a generalized modeling notation which means it can be used to model
all kinds of phenomena.

“ In this dissertation it is assumed that there is a one-to-one correspondence with a single REA template
and a single accounting cycle (e.g., two REA templates equals two accounting cycles). Additionally, the
three-way control relationship has been modeled as two binary associations without loss of generality.
Also, relatively minor resource decrements (sometimes known as transaction costs) have been omitted in
this dissertation‘s examples (Geerts and McCarthy 1998).

'2 McCarthy (1982) describes this process as view modeling (creating the local diagram) and view
integration (creating the global diagram). More recently, Geerts and McCarthy (1997) advocate a top-

10

 

Economic

 

 

 

 

 

 

 

 

   
      

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Resource
Inside Agent
'articipan
Economic
Event (give)
articipan Outside
Agent
ualit
Outside
articipan Agent
nflow , Economic
Event (take)
articipan .
lnsrde Agent

 

 

 

 

 

Economic
Resource

 

 

 

Figure 4
REA Template
(Adapted from Dunn and McCarthy 1997, 3S)

 

down process based on value-added process hierarchies where the REA design process starts at a high level
of abstraction and moves toward lower levels of abstraction.

ll

The economic exchange is composed of a give event and a take event wherein
speciﬁc resources are consumed and acquired. To elaborate, an instantiation of the
general REA accounting model is shown in Figure 5 as a simple revenue cycle. Note the
relationship that exists: the give (consumption) event is the Sale; the take (acquisition)
event is the Cash Receipt. The resource given is the Inventory; the resource taken is
Cash. The internal and external agents participating in the Sale event are Salesperson and
Customer respectively; and the internal and external agents participating in the Cash
Receipt event are Cashier and Customer respectively. This is a simple example of the
model developed by McCarthy (1982); more complex models are in Geerts and
McCarthy (1994, 1997).

1.2 Overview of Hypotheses

The ﬁrst two hypotheses of this dissertation relate to how experience with
conceptual modeling of REA systems results in the organization of REA knowledge in
long-term memory. The third and fourth hypotheses (which are conditional on the results
of the ﬁrst two hypotheses) relate to how organization of REA knowledge affects
performance on the conceptual design of an REA system.

Hypothesis 1 states, “There will be an ordinal interaction between experience and
information format organization on knowledge structure.” The underlying variables of
interest are experience (the independent variable) and knowledge structure (the dependent
variable). Various accounting studies document the positive relationship between these
variables (e.g., Butt 1988; Libby and Frederick 1990; and Frederick 1991). However,
these accounting studies focused on auditor judgment performance. Similar research is

needed to assess AIS designer judgment performance.

12

 

 

Inventory

 

 

 

 

 

Cash

IO.N)

 

 

 

@—
(0.N)

(LN)

(1.1)

 

Salesperson

 

 

Sale
(give)

 

 

 

 

 

(O.N)

uality

IO.N)

 

 

Customer

 

 

 

 

 

Cash Receipt
(take)

 

 

 

Figure 5

 

Cashier

 

 

 

REA Template Instantiation of Revenue Cycle (without attributes)

13

Hypothesis 1 attempts to determine whether experience with conceptual modeling
of REA systems allows individuals to organize their long-term knowledge of business on
the basis of the REA framework. In examining the structuring of knowledge, this
hypothesis focuses on how entities and relationships are connected. It is proposed that
the REA template (involving entities and relationships) provides a basis for the
structuring of REA knowledge (this will be called REA template knowledge structure or
REATKS) since each REA template can map to a transaction cycle. Furthermore, this
structured knowledge can only be developed with experience. Several accounting studies
(Libby and Luft (1993) provide a good overview) document the use of transaction cycles
as one knowledge structuring mechanism that auditors use. This dissertation is the ﬁrst
paper to address the knowledge structuring of accounting system designers.

Hypothesis 2 states, “REA-related entity clustering increases with REA
experience.” This hypothesis is related to Hypothesis 1 in that it tests the similar
independent and dependent variables. However, it addresses the issue of how accounting
system designers aggregate attributes to entities. Attribute placement is important in
database design to ensure representation viability and minimize redundancy.
Experienced accounting system designers should have more well developed schemas for
how attributes aggregate to speciﬁc entities. In the accounting literature, Choo and
Trotrnan (1991) and Ricchiute (1992) provide empirical evidence on the link between
experience and schema acquisition.

Hypothesis 3 states, “Performance on the conceptual modeling task is positively
correlated with REA template knowledge structure.” Hypothesis 4 states, “Performance

on the conceptual modeling task is positively correlated with attribute aggregation

I4

knowledge structure.” Both of these hypotheses test the assertion that one of the
determinants of successful accounting system design is how knowledge is structured in A
memory. This link between knowledge structure and performance is of primary
importance to accounting researchers seeking to understand judgment performance and to
enhance the decision-maker’s efﬁciency or effectiveness. Since this dissertation makes
explicit speciﬁcations regarding the cognitive processes underlying REA conceptual
modeling, inferences can be drawn with respect to the effect of knowledge structure on
REA conceptual modeling performance.

1.3 Overview of Research Design and Experimental Tests

Data were gathered from an experiment conducted at Michigan State University.
Forty-seven graduate and ﬁﬁy-four undergraduate students participated in the study. This
sample was selected to ensure that the statistical tests had enough power to detect an
effect. Due to the nature of the experience factor (i.e., participants could not be assigned
to either the experienced group or inexperienced group), random assignment was not
possible (an inherent limitation of experience studies). Due to logistical constraints,
participants could not be randomly assigned to experimental sessions. However, all
experimental sessions were held at similar times, and there was almost an equal number
of participants in each of the task orders (which were counterbalanced).

The experiment can be conceptualized as three related tasks. Task 1 involves the
free recall of a random and an REA diagram; task 2 involves the free recall of a list of
randomized attributes; and task 3 involves performing a conceptual database design. The
data gathered in task 1 serves as a dependent variable (REA template knowledge

structure) in the test of Hypothesis 1, and an independent variable in the tests of

15

Hypotheses 3 and 4. The data gathered in task 2 serves as a dependent variable (attribute
aggregation knowledge structure) in the test of Hypothesis 2, and an independent variable
in the tests of Hypotheses 3 and 4. The data gathered in task 3 serves as a dependent
variable (design performance) in the test of Hypotheses 3 and 4.

Hypothesis 1 is tested using a mixed within-subjects factorial design, described in

Chapter Three, with one between-subjects factor (experience) and one within-subjects

factors (diagram type). This hypothesis predicts (1) that experienced participants will

 

recall signiﬁcantly more items of information than inexperienced participants when
presented with an REA diagram and (2) that there will be no signiﬁcant difference
between these groups when presented with a Random diagram (i.e., an ordinal
interaction). This hypothesis is an analog to the studies (discussed in Chapter Two) of de
Groot (1965, 1966), Chase and Simon (1973), Egan and Schwartz (1979), and
McKeithen, Reitrnan, Rueter, and Hirtle (1981) which present random and non-random
stimuli to test similar hypotheses.

Hypothesis 2 is tested using free recall of a random list of attributes in a 2 x l
paired-groups design. This hypothesis predicts that experienced participants will achieve
a signiﬁcantly higher ARC (adjusted ratio of clustering) index (Roenker, Thompson, and
Brown 1971) than inexperienced participants. In other words, when presented with a
random list of attributes, experienced participants are more likely to cluster their recall of
attributes in terms of the entities that they represent. The experienced participants will
invoke a schema allowing them to overcome short-term memory processing limitations.

Hypotheses 3 and 4 are tested using multiple regression analysis where conceptual

database design performance is regressed on both the REA template knowledge structure

16

measure obtained in task 1 and the attribute aggregation knowledge structure measure
obtained in task 2. Successful conceptual database design requires the correct structuring
of entities, relationships, and cardinalities, as well as the correct placement of attributes.
Therefore, it is expected that knowledge structure (both REA and attribute aggregation)
will be positively correlated with conceptual database design performance while
controlling for knowledge content and ability. The relationship between knowledge
structure and system design performance has not been investigated in the accounting,
information systems, or psychology literature. Accounting research has examined the
relationship between knowledge structure and audit judgment performance. However,
very few accounting studies (e.g., Nelson et al. 1995) have used knowledge structure as
an independent variable. This is one of the main contributions of the dissertation.

In summary, the primary statistical analysis techniques employed will be AN OVA
to test for order effects and Hypothesis 1, an independent-samples t test to test Hypothesis
2, and bivariate correlation and multivariate regression to test Hypotheses 3 and 4. The
results obtained suggest that experienced REA designers have signiﬁcantly different
(better developed) knowledge structures than inexperienced REA designers with respect
to both REA structuring knowledge (consistent with Hypothesis 1) and attribute
aggregation structuring knowledge (consistent with Hypothesis 2). However, it appears
that better developed REA structuring of knowledge is signiﬁcantly associated with REA
conceptual modeling performance (consistent with Hypothesis 3), whereas attribute
aggregation structuring knowledge is weakly associated (marginally consistent with

Hypothesis 4).

l7

1.4 Organization of the Dissertation

The rest of this dissertation consists of four chapters. Chapter Two covers related
literature spanning the disciplines of accounting, information systems, and cognitive
psychology. Additionally, the hypotheses are developed within the context of Libby and
Luﬁ’s (1993) model of the Antecedents and Consequences of Knowledge. Chapter Three
describes the research method used including participants, independent variables, and
experimental tasks. Chapter Four presents the results. Chapter Five concludes the
dissertation with an overall summary, and with a discussion of contributions, limitations,

and future research directions.

18

Chapter Two
THEORY AND HYPOTHESES DEVELOPMENT

This chapter ﬁrst reviews pertinent literature and then develops the formal
hypotheses to be tested. The literature review consists of an analysis of relevant studies
in accounting, information systems, and psychology. The review of the accounting
literature in Section 2.1 will primarily emphasize major ﬁndings from audit studies. The
psychology review in Section 2.2 will primarily emphasize the role of domain speciﬁc
(other than accounting) knowledge and chunking. In Section 2.3 Libby and Luﬁ’s (1993)
model of the Antecedents and Consequences of Knowledge, coupled with an analysis of
the speciﬁc knowledge needed for conceptual database design, is used to motivate the
hypotheses. Section 2.4 closes Chapter Two with a brief summary.
2.1 Review of Accounting and Information Systems Literature

The effects of experience and knowledge on audit task performance have been
well documented in the accounting literature (Libby 1995; Libby and Luﬁ 1993; Bimberg
and Shields 1984). As discussed in Libby and Luﬁ (1993), the knowledge-related
accounting literature has grown from early correlational studies between experience and
performance where knowledge was inferred (called the “expertise paradigm”), to more
recent studies where experience, ability, knowledge, and performance are all related in
one model (called the Antecedents and Consequences of Knowledge which is discussed
later in the chapter). This section will detail some of the notable ﬁndings of links
between (1) experience and knowledge and (2) knowledge and performance in an audit
setting. Related studies from the information systems literature, though few, will also be

discussed.

l9

 

One of the earlier studies in accounting was Butt’s (1988) study of auditors’
acquisition of frequency knowledge. This study is noteworthy because it actually
involves the manipulation of experience in an experimental setting (something rarely
done). Student and auditor participants were presented with direct experience
(“individual presentations of ﬁnancial statement errors”) and indirect experience
(“summary data”) stimuli (Butt 1988, 316). Results - which were consistent across
groups —- indicated that more accurate frequency judgments were associated with direct
experience and that less accurate frequency judgments were associated with indirect
experience.

Choo and Trotrnan (1991) used a schema-based framework to examine recall,
inference, and predictive judgment of experienced and inexperienced auditors. Subjects
in this study were required to make going-concem judgments, with the expectation that
the knowledge structures of experienced auditors would be signiﬁcantly more developed
than those of the inexperienced auditors. Both experience levels had similar levels of
recall performance with typical going-concem items. However, recall performance of
atypical going-concern items became worse for inexperienced auditors and better for
experienced auditors (and the difference between groups was signiﬁcant). These results
are consistent with the theory that experienced auditors’ knowledge structures are better
deﬁned.

In another going-concem setting, Ricchiute (1992) found evidence that going-
concem problems were more likely to be identiﬁed when the working-papers were
presented in causal order rather than the typical order which allows for convenient

compiling of audited ﬁnancial statements. Audit work-papers are usually assembled in a

20

manner corresponding to major segments of a ﬁnancial audit. However, this usually
means that causal evidence of a going-concem problem is not grouped together, thereby
increasing the cognitive load of the auditor trying to construct the causal order. In
Ricchiute’s (l 992) experiment, going-concem problems were more likely to be identiﬁed
when the working-papers were presented in causal order rather than the convenient order.
This result is in agreement with the notion that the auditors possess some type of going-
concern causal schema (i.e., knowledge organization) which is apparently interacts with
the working-paper presentation.

Frederick (1991) obtained a particularly interesting result regarding experience
and knowledge structure. Using a 2 x 2 between-subjects free recall experiment in which
internal controls were presented either organized by audit objective or by transaction
cycle, Frederick found that experienced auditors had signiﬁcantly better recall in the
transaction cycle condition versus the audit objective condition, and inexperienced
auditors had insigniﬁcant differences between the two stimuli.

The series of knowledge structure studies by Nelson et al. (1995) and Bonner,
Libby, and Nelson ( 1996) represent a signiﬁcant contribution to this literature. Nelson et
al. (1995) suggest that when a dominant knowledge structure does not match a particular
task, dysfunctional behavior may occur. Based on evidence from Frederick, Heiman-
Hoffrnan, and Libby (1994), indicating that experienced auditors are likely to have
objective-dominant knowledge structures (as opposed to transaction cycle-dominant),
Nelson et al. (1995) conducted an experiment involving the task of audit planning (a task
which is organized by transaction cycle). Results indicate that “auditors’ estimates of

conditional frequencies of the form P(cycle error | objective) were inﬂuenced by the

21

frequencies presented to them in the experiment, while their estimates of conditional
frequencies of the form P(objective error I cycle) were not.” (Nelson et al. 1995, 28).
Additionally, auditors’ allocation of audit hours was “inﬂuenced by experimental
frequencies when auditors distributed hours across cycles for each objective, but not
when they distributed hours across objectives for each cycle” (Nelson et al. 1995, 28).
The next step for these authors was to determine whether a decision aid could
address the weakness of the knowledge structure/task structure mismatch described
above. In a similar test of conditional probability judgments, Bonner et al. (1996)
developed and tested the effectiveness of two decision aids: a “checklist aid” and a
"decomposition-and-mechanical-aggregation aid.” The objective of the checklist aid is to
facilitate knowledge retrieval whereas the objective of the decomposition aid is to
facilitate both knowledge retrieval and aggregation. Although results indicate that both
aids were beneﬁcial to auditors, the decomposition aid was superior to the list aid.
Results also suggest that once the decomposition aid was used, the use of the list aid
made no improvement in judgment; however the incremental use of the decomposition
aid with the list aid did improve judgments. Since these two aids deal with different
cognitive processes it is assumed that both retrieval and aggregation processes cause the
difﬁculties underlying the knowledge structure/task structure mismatch. The most
interesting contribution of this paper is stated best by the authors: “Thus, our
development of decision aids which target speciﬁc hypothesized cause(s) of a judgment
difficulty not only enabled us to test whether the difﬁculty can be alleviated, but also to
test (indirectly) our underlying theory for why the difficulty occurs” (Bonner et a1. 1996,

237)

22

The free recall method was used by Weber (1980) to investigate experienced EDP
auditors’ consensus of computer controls. Ten controls were selected from each of the
following categories: management and organizational, data preparation, input, processing,
and output. The ﬁfty controls were then randomized and read aloud after which the free
recall commenced. The experienced EDP auditors outperformed the inexperienced
auditors in terms of both total controls recalled and clustering of controls recalled.

Of the four constructs in Libby and Luﬁ’s (1993) model, ability has proved the
most difﬁcult for researchers to measure. Bonner and Lewis (1990) were the ﬁrst to
consider individual differences in abilities as determinants of performance. However,
their proxy for ability was measured only by having individuals respond to a small
number of GRE questions. The difficulty in ﬁtting ability into a model is that researchers
need to specify exactly what abilities are required for the task, just as they need to specify
exactly what cognitive processes are involved. Unfortunately, specifying and measuring
the cognitive processes has proven more manageable than specifying and measuring the
abilities.

The study of the antecedents and consequences of knowledge has started
expanding outside of the audit context. According to Libby and Luft (1993, 446),
researchers in other areas of accounting should follow “... a similar approach to that taken
in the audit literature, beginning with an analysis of key attributes of the settings and task
requirements.” Recent evidence shows that other areas of accounting are beginning this
research. Libby, Trotrnan, and Zimmer (1987), Moser (1989), and Maines (1990) have
all examined ﬁnancial analysis; Bonner, Davis, and Jackson (1992), Cloyd (1994), and

Spilker (1995) have studied tax judgments; and Dearrnan and Shields (1997) have

23

expanded this research into a managerial accounting setting. This present dissertation
begins research into knowledge related to accounting information system design.

While there have been many accounting studies of knowledge, there have been
few information systems or computer science studies. One of the earliest studies of the
design of computer software consisted of an observational study and two experiments
(Malhotra, Thomas, Carroll, and Miller 1980). The observational study was performed
by videotaping an actual designer interacting with a client (or user). By analyzing
videotapes of the study participants, it was determined that the dialogs involved “cycles”
of the following six “states”: (1) goal statement; (2) goal elaboration; (3) (sub) solution
outline; (4) (sub) solution elaboration; (5) (sub) solution explication; and (6) agreement
on (sub) solution. In an interesting conclusion, Malhotra et al. (1980, 126) note, “The
solutions proposed in a cycle do not, however, always match the requirements discussed
in it.”

In the Malhotra et al. software design experimentl3 four participants were given
the task of writing requirements for a query system. One participant gave a simple
function set with simple syntax. A second participant gave a complex function set with
complex syntax. A third participant suggested the use of a query language (e.g., Query-
By-Example), and a fourth participant gave a menu of queries to choose from. These
vastly diﬂ’erent designs led the authors to conclude “the sub-goals suggested by people

seem idiosyncratic and to depend strongly on past experience” (Malhotra et al. 1980,

 

‘3 The other experiment is not discussed here, since its purpose was to examine the trade-offs between
originality and practicality in the design process. Some of the participants who claimed to be interested in
originality did worse on the practicality of their design (without a corresponding increase in originality)
and vice versa (Malhotra et al. 1980).

24

128). The apparent implication is that there is a link between experience and knowledge
organization. The two main weaknesses of Malhotra et al.’s study were (1) the sample
size was too small to make any generalizations about designers, and (2) cognitive
psychology methods were not employed.

The only study that has looked at the effect of experience on conceptual modeling
in database design is Batra and Davis (1989)." While their protocol analysis revealed
some interesting similarities and differences between the novice and expert participants,
the task employed did not test any domain-speciﬁc knowledge such as REA. A major
beneﬁt of this present dissertation is that the task environment is rich in accounting
domain knowledge. Additionally, Batra and Davis had their participants ﬁrst elicit
requirements from a simulated user and then proceed with the task of designing the data
model. Therefore, the tasks of information requirement analysis and data model design
were merged into one task. While it is possible, and even likely, that the skills of
information requirement analysis and data model design are related, an experimental
setting should allow for one of these tasks to be held constant, resulting in unambiguous
inferences about the other.

Conceptual database design errors made by novices were investigated by Batra
and Antony (1994b). By analyzing the results of novice conceptual database diagrams,
Batra and Antony were able to develop a model of the causes of design errors. Two types

of errors were identiﬁed: errors due to biases and errOrs due to incomplete knowledge.

 

" Related studies have looked at the effects of different data models on individual modeling performance
(Batra and Antony 1994a, Batra et al. 1990, Jarvenpaa and Machesky 1989, Shoval and Even-Chaime
1987, Juhn and Naumann 1985, and Brosey and Shneiderrnan 1978) how designers deal with problem
representation (Srinivasan and Te’eni 1995), biases related to structural constraints (Siau, Wand, and
Benbasat 1997), and optional properties versus subtyping (Bodart and Weber 1997).

25

The biases were assumed to be due to the use of an anchoring heuristic, whereas the
incomplete knowledge errors were assumed to be due to the novices’ insufﬁcient
knowledge. To expand on this model, Batra and Antony performed a second experiment
using a verbal protocol analysis. This analysis revealed that bias error could be
decomposed into literal translation and anchoring errors. Furthermore, the anchoring
heuristic was being driven by data saturation, an outcome irrelevant learning system,
availability, or order effects.

Weber (1996) examined chunking in a database setting trying to determine if it
helped people differentiate between entities and attributes. The results showed that both
the novice and the expert database designers “appear to be using a distinction between
entities and attributes to facilitate their recall of items” (Weber 1996, 158). While this
provides evidence to answer Weber’s research question, it leaves open the question of
how does the ability to chunk information affect database design performance?

2.2 Review of Psychology Literature

This dissertation investigates the role of “chunking” (also known as “recoding” or
“grouping”) in facilitating the organization of knowledge about REA systems. Miller’s
(1956) classic paper provides evidence that the human short-term memory has a limited
capacity of seven plus or minus two units of information. A unit can be a letter, a word, a
phrase, or a number, and its deﬁnition varies by individual and situation (Simon 1974).
Chunking is the ability to group these-units in a meaningful fashion, allowing humans to
overcome a processing limitation. For example, a telephone number broken into three

chunks (517-355-7486) is easier to remember then a string of ten digits (5173557486).

26

Chunking allows multiple units to be reduced to a single unit (in the telephone number
example, ten units become three chunks of single units).

Even though Miller’s (1956) paper is considered a classic, Bousﬁeld (1953)
actually did earlier work on memory organization (which Bousﬁeld called “clustering”).
Bousﬁeld conducted a free recall experiment involving words from four categories
(animals, names, professions, and vegetables) presented to subjects in random order.
Even though the words were presented randomly, subjects had a tendency to recall the
words according to the categories at levels above chance. Bousﬁeld (1953, 237) stated
that “This implies the operation of an organizing tendency.” The main difference
between chunking and clustering is that chunking reduces multiple units to a single unit,
but clustering creates links between units without necessarily reducing them to a single
unit.

It appears that the tendency to organize knowledge is a fundamental instinctive
cognitive process. Tulving (1962) conducted a multi-trial free recall experiment using
sixteen words in sixteen different randomized lists. The results suggested “... that the
subjects do impose a sequential structure on their recall, that this subjective organization
increases with repeated exposures and recall of the material, and that there is a positive
correlation between organization and performance” (Tulving 1962, 352).

An interesting extension of the chunking ﬁndings examines the effect of domain
speciﬁc knowledge on chunking. In studies of chess experts," de Groot (1965, 1966) and

Chase and Simon (1973) showed that they appear to have a superior memory for domain

 

'5 The review of expert-novice studies provides insights into experience effects. Experience is a necessary
(but not sufficient) condition for expertise (Davis and Solomon 1989).

27

speciﬁc patterns, although there are hardly any other differences (e.g., depth of search or
moves considered) in comparison to weaker players. De Groot hypothesized that the
superior memory for domain speciﬁc patterns was attributable to differences in
knowledge of board positions between novice and expert players. Since expert chess
players study previous games and develop certain strategies for the opening and closing
of the game, they are more likely to store previous board positions in some schematic
fashion.

De Groot gave individuals presentations of either “random” or “normal” board
positions for a brief period of time. Then he removed the board positions and asked the
participants to reconstruct the positions. The main result was that the chess masters’
recall accuracy for normal board positions was approximately 90%, but the novices’
accuracy was only around 41%. However, when a random board position was used,
novice and master performance was similar. This ruled out the possibility of the experts
having superior recall ability. Rather the chess masters were able to remember normal
board positions of several pieces as a single chunk.

Chase and Simon (1973) replicated and extended De Groot’s ﬁndings to test the
chunking hypothesis. Their results “suggest that the superior performance of stronger
players (which does not appear in random positions) derives from the ability of those
players to encode the position into larger perceptual chunks, each consisting of a familiar
subconﬁguration of pieces. Pieces within a single chunk are bound by relations of mutual
defense, proximity, attack over small distances, and common color and type” (Chase and

Simon 1973, 80).

28

Evidence of chunking was also found by Egan and Schwartz (1979) in
experiments using symbolic circuit drawings. Skilled technicians signiﬁcantly
outperformed unskilled technicians on a recall task. However, when given randomized
circuit drawings, there was no signiﬁcant difference between the skilled and unskilled
technicians. Egan and Schwartz (1979, I49) propose that a possible explanation of their
results could be that “skilled subjects identify the conceptual category for an entire
drawing, and retrieve elements using a generate-and-test process.”

In a computer science setting, McKeithen et al. (1981) employed a free recall
method to examine knowledge organization of computer programmers using an
ALGOL W program. Participants received either a scrambled or coherent version of a
computer program. Experts recalled signiﬁcantly more mean lines of code than novices
in the coherent condition; no differences between groups were found in the scrambled
condition. Results indicated that novices relied primarily on mnemonic techniques, that
intermediates relied on common-language associations, and that experts grouped words
by ALGOL W ﬁmctionality (McKeithen et al. 1981).

Adelson (1981) provided additional evidence that domain-speciﬁc computer
science experience can change knowledge structures over time. Adelson (1981)
performed a multi-trial free recall experiment with a group of novices (ﬁve under-
graduates who had just completed an introductory course in computer programming) and
with a group of experts (ﬁve teaching fellows for the same course). Sixteen lines of PPL
(Polymorphic Programming Language) code were presented randomly (one line at a
time) as the stimulus which could be organized procedurally (into three distinct

programs) or syntactically (e.g., by IF statements, FOR statements, etc.). The experts

29

recalled signiﬁcantly more lines of code than did the novices across all nine trials.
Results indicate that novices used a syntax-based organization, whereasexperts used a
more abstract hierarchical organization based on principles of program function.

The results of McKeithen et al. (1981) and Adelson (1981) conﬁrm the classic
ﬁndings of Chi et a1. (1981) who conducted a study of novice (undergraduate students)
and expert (Ph.D. students) physicists. The ﬁndings indicated that, even though both
groups form a schema containing the surface features of the problem, the experts’
schemas are based on underlying physics principles whereas the novices’ schemas are
based solely on surface features. In other words, the experts’ view of the problem is
based on principles such as Newton’s laws, whereas the novices’ view is based on things
mentioned in the problem such as an inclined plane or the literal use of terms such as
“friction.”

The above examples from the domains of database design, chess, and electronics
are particularly relevant to this dissertation because they involve chunking of
picture-like visual stimuli e.g., garneboards or diagrams. REA design involves the
development of a diagram which is a map of an enterprise’s information architecture.
Since the REA accounting model (template) is highly generalizable, patterns become
apparent after repeated template instantiations.

The major limitation of the psychology studies reviewed above (except the chess
studies) is that they failed to investigate whether differences in knowledge structure were
relevant to task performance. For example, Adelson (1981) found that novices and expert
computer programmers organize their knowledge differently, but did not take the

additional step to ascertain if these differences made the experts better computer

30

programmers. Since behavioral accounting research is concerned speciﬁcally with the
accounting task environment, this additional step is necessary.
2.3 Model and Hypotheses Development

It is worth reviewing the research questions of this dissertation before moving on
to the development of the hypotheses. The speciﬁc research questions addressed are: (1)
does experience with conceptual modeling of REA systems result in the organization of
knowledge on an REA basis in long-term memory; (2) does this experience help
designers aggregate attributes to entities; and (3) how do these two types of knowledge
(REA and attribute aggregation knowledge) affect REA design performance while
controlling for knowledge content and ability?4

In order to provide a framework in which to analyze the research questions, Libby
and Luﬁ’s (1993) model of Antecedents and Consequences of Knowledge (Figure 6) is
used. According to this model, experience and ability affect knowledge, and knowledge
and ability affect performance. In terms of investigating how REA design knowledge is
acquired, two levels of experience are to be examined: (1) undergraduate accounting
students enrolled in their ﬁrst AIS class, and (2) graduate accounting and MBA students
who are enrolled in a Master’s level AIS class. Both classes place heavy emphasis on
REA conceptual modeling. Additionally, all graduate students have completed either the
undergraduate AIS class or its graduate equivalent. These groups are being chosen in
order to maximize control over participants’ experience with REA conceptual modeling

and to investigate early stages of knowledge acquisition.

31

   
 
 

 

 

 

Performance

 

    

I Experience I

 

 

 

 

Ability
V

  

 

Figure 6
Antecedents and Consequences of Knowledge
(adapted from Libby and Luft 1993, 433)

It is important to distinguish knowledge content from knowledge structure.
“Knowledge content” means what information is stored in memory (which is actually
how cognitive psychologists deﬁne “knowledge”). “Knowledge structure” means that
some type of organization (e.g., hierarchical, temporal, causal) is placed on the
knowledge content. For example, two auditors may have similar knowledge content for
internal controls, yet organize and link the controls differently in their memories by
transaction cycle versus by audit objective (Frederick 1991). In this dissertation,
knowledge content will be measured to determine if any differences exist between
inexperienced and experienced participants. If no differences exist, then the effects of
knowledge structure -- the main variable of interest -- can be determined. However, even

if differences are noted, they can at least be controlled for in the regression equation.

 

'6 In this dissertation, motivation is assumed to be randomly distributed. It is also assumed that the
incentive provided to the participants induced similar motivation across inexperienced and experienced

groups.

32

Two measures of knowledge structure are proposed: one for REA template knowledge
and the other for attribute aggregation knowledge. Both knowledge structure constructs
will be measured with a free recall task. This dissertation seeks to determine if these
knowledge structures are correlated with REA conceptual modeling performance while
controlling for knowledge content and ability.
2.3.1 Experience and REA Template Knowledge Structure

It is proposed that the REA accounting model provides a schema template which
will aid knowledge organization in long-term memory. Since the REA accounting model
is highly generalizable, it may serve as a good basis for the organization of knowledge.
At a high level of abstraction, the REA accounting model can be thought of in terms of
three groupings: (1) the resource - event grouping (consisting of a stock-ﬂow); (2) the
event - event grouping (consisting of an increment and a decrement or give and take); and
(3) the event - agent grouping (consisting of the control relationship with the event, the
internal agent, and the external agent). It is the union of these groupings (i.e., the REA
template) that makes this model semantically expressive. At lower levels of abstraction
these same groupings can be seen in terms of instantiated accounting cycles (i.e., the
revenue,'7 acquisition, conversion, ﬁnancing, and payroll cycles). Frederick (1991)
provides evidence that experienced auditors organize their knowledge of internal controls
according to transaction cycles, and the use of the generalizable REA template should

facilitate knowledge organization.

 

'7 Refer to Chapter One for an example of how the general REA template (Figure 4) is instantiated as a
simple revenue cycle (Figure 5).

33

Extended experience with REA design will reinforce these knowledge structures.
From a knowledge acquisition point of view, different parts of an accountant’s experience
lead to different knowledge structures (and knowledge content, which is held constant in
this dissertation). Students learning REA are not going to organize their accounting
knowledge according to REA; rather, they are likely to organize according to accounts in
terms of journal entries (Frederick and Libby 1986). Although REA systems have the
ability to generate ﬁnancial statements conforming with generally accepted accounting
principles, the REA design process is not focused on accounts. This paradigm shift away
from accounts (Walker and Denna 1997) is likely to interfere with the ability to organize
REA knowledge in memory. Additionally, inexperienced students may haVe difﬁculty
organizing their knowledge according to this schema template because they typically also
have to acquire new knowledge regarding conceptual modeling and database design.
Attempts at simultaneous encoding of these two types of knowledge may cause
confusion.

Since Miller (1956) provides evidence that working memory is constrained to
seven plus or minus two units of information, the ability to recall large amounts of
information depends on chunking ability and long-term memory. The REA template
illustrated previously in Figure 5 has eight entities, seven relationships, fourteen
minimum cardinalities and fourteen maximum cardinalities for a total of 43 pieces of
information. A person with journal entry knowledge should have no trouble recalling the
entities and relationships, and perhaps a few cardinalities. A person with REA template

knowledge should have little difﬁculty recalling the entities, relationships, and

34

cardinalities since they will be able to access their long-term memory (i.e., knowledge
structure) to facilitate chunking and memory will not be overburdened.

When two or more REA templates are joined together, complexity increases and
people with REA template knowledge should outperform people with only journal entry
knowledge. For example, two templates (including all entities and relationships) would
be two chunks for a person with an REA template knowledge structure, and they would
still have memory left over to chunk the cardinalities. The person with the journal entry
knowledge structure, however, would likely encode just two entities and a relationship as
one chunk.

It is expected that the results of de Groot (1966) and Chase and Simon (1973) will
be replicated.” With the REA diagram, experienced REA designers will recall more
entities, relationships, and cardinalities than will inexperienced REA designers. There
will be no difference between experienced and inexperienced REA designers’
performance with a random diagram. This hypothesis is formalized below.

H1: There will be an ordinal interaction between experience and information

format organization on knowledge structure (see Figure 7).

It is important to note that although Hypothesis 1 involves a replication of prior
ﬁndings (which appear to be robust), it is necessary to perform this empirical test to
gather data which will be used in the tests of Hypotheses 3 and 4 (which are the true

hypotheses of interest for accounting researchers).

 

" Since there is a large difference in the experience levels of chess experts and novices, it is unclear
whether this result should replicate with the smaller differences in experience between undergraduate and
graduate students. However, VanLehn (1989) suggests that the ﬁndings are robust. Regardless, this
replication is necessary in order to evaluate the relationship between knowledge structure and conceptual
database design performance.

35

Recall

 

 

 

4‘

 

 

 

 

 

 

lnexperienced ’
Group
Random REA
Diagram Diagram
Figure 7

Predicted Interaction for Hypothesis 1

2.3.2 Experience and Attribute Aggregation Knowledge Structure

In addition to the structuring of the REA template, another crucial part of the
design task is properly aggregating entities to attributes. Attribute aggregation involves
taking individual data elements (attributes) and aggregating them to the appropriate
entity. Entities can be thought of as “things” and attributes can be thought of as “features
of things.” For example, the attributes “invoice number,” “date,” “amount,” and
“shipping date” aggregate to the entity “sale.” Actually, it is these attributes which will
embody all extensional information (speciﬁc instances of the attributes) in the database.

Errors in attribute aggregation can lead to unnormalized (inefﬁcient and
ineffective) database design and use. According to Batini et al. (1992, 143), “The normal
forms... are intended to keep the logical structure of the data clean... by alleviating the

problems of insertion, deletion, and update anomalies, which cause unnecessary work

36

because the same changes must be applied to a number of data instances, as well as the
problem of accidental loss of data or the difﬁculty of representing given facts.” Consider

,9 66

the following simple example: if the attributes “bank account number, account
balance,” “bank name,” “bank address,” and “bank zip code” were aggregated to the
entity “cash” there would be redundant information store in the database (the same bank
address and zip code shows up for every instance of bank name; see the unnormalized
table in Figure 8). This creates great inefficiencies if, for example, First National’s
address changed. Normalized tables solve this problem (Codd I972). The best solution
is to have two entities, one for “cash” and one for “bank.” Then the address change only
needs to be made in one place, and there is far less chance for crippling redundancy.
There are (at least) two ways that attributes can be organized in memory. The
ﬁrst involves clustering together groups of attributes that aggregate to a particular entity

’9 ‘6

(e. g., the attributes “employee number, employee name,” and “employee address”
aggregate to the entity “employee”). It may seem obvious that these attributes aggregate
to the entity “employee” because they all contain the word “employee.” However, it is
necessary to describe the attributes this way. If the word “employee” was not used, the
attributes “number,” “name,” and “address” might apply to other entities, such as

3’ ‘6

“invoice number, customer name,” or “warehouse address.” This suggests another way

to organize knowledge about attributes: functional similarity. For example, the attributes

,9 66

“inventory item number, customer order number,” and “customer num ” could be

clustered together as “number” even though they would aggregate to three different

37

Unnorrnalized table

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Derivation of Normalized Tables

38

Cash

Bank account Account Bank name Bank address Bank zip

number balance code

01234 $ 3,000 First National 201 Main Street 48823

01235 $ 2,500 First National 201 Main Street 48823

01355 $ 9,000 First National 201 Main Street 48823

01356 $ 1,500 First National 201 Main Street 48823

01357 $ 3,500 First National 201 Main Street 48823

22235 $20,000 Spartan Bank 1 Spartan Road 48823

22236 $10,000 Spartan Bank 1 Spartan Road 48823

22237 $15,500 Spartan Bank 1 Spartan Road 48823

22301 $18,000 Spartan Bank 1 Spartan Road 48823

22302 $17,500 Spartan Bank 1 Spartan Road 48823
Normalized tables

Cash Bank

Bank Account Bank name Bank Bank Bank zip

account balance name address code

number

01234 $ 3,000 First National First 201 Main 48823

01235 $ 2,500 First National National Street

01355 $ 9,000 First National Spartan 1 Spartan 48823

01356 $ 1,500 First National Bank Road

01357 $ 3,500 First National

22235 $20,000 Spartan Bank

22236 $10,000 Spartan Bank

22237 $15,500 Spartan Bank

22301 $18,000 Spartan Bank

22302 $17,500 Spartan Bank

Figure 8

 

entities (inventory, customer order, and customer). Therefore, it is possible that for less
experienced designers, functional similarity clustering increases."

Experienced REA designers should have more well-developed schemas for
attribute aggregation. Since inexperienced auditors have a journal entry knowledge
structure (Frederick and Libby 1986), and inexperienced REA designers have been
exposed to the same classroom training, the inexperienced REA designers are likely to
have a journal entry knowledge structure. This knowledge is composed of only a limited
set of accounting features (e.g., date, account, and amount). Since REA systems capture
multi-dimensional features and store them in a disaggregated manner, the experienced
REA designer will develop a schema for which attributes aggregate to which entities.
This will be tested by examining clustering (Bousﬁeld 1953) in recall of attributes.20
Bousﬁeld (1953) found that individuals presented with a randomized list of animals,
names, professions, and vegetables recalled according to these groups (i.e., clustered) at a
greater than chance expectation.2| Weber (1996) presented individuals with a conceptual
diagram that did not differentiate between entities and attributes, and he found that there
was a tendency to recall entities before attributes.22 This suggests that database designers
differentiate between entities and attributes. In the dissertation participants will only
recall attributes. Therefore, the following hypothesis is tested:

H2: REA-related entity clustering increases with REA experience.

 

'9 This contention is not formalized as a hypothesis due to lack of theory.

2° See Weber (1980) and Shields, Solomon, and Whittington (1996) for related accounting papers on
experience and clustering.

2' Even when participants are allowed to choose as many categories as they feel is necessary, participants
usually preferred 5 plus or minus 2 (Mandler 1967) which is remarkably similar to Miller’s (1956) ﬁnding.

39

2.3.3 Knowledge Structure and Performance

Performance in this study will be measured by having participants design an REA
conceptual model. Speciﬁcally, participants will be given a written narrative describing a
company’s operations and a list of data elements. Subjects will be asked to perform their
conceptual modeling using the Batini et al. (1992) entity-relationship modeling
formalism. Experts (tenured AIS faculty members) will aid in setting performance
standards on this design task and the free recall tasks. A correctness score will be
calculated based on all correct entities, relationships, attributes, and cardinalities.
Different types of ability will be controlled for by using ACT scores, and scores obtained
on the Group Embedded Figures Test (GEFT) (Oltrnan, Raskin, and Witkin 1971).23

Successful performance on the task of REA conceptual modeling involves:
(.1) deriving the structure of the diagram (i.e., an REA template or templates) and
cardinalities; and (2) aggregating the attributes to the correct entities. If mistakes are
made in deriving the structure of the diagram, inefﬁcient relations could be implemented.
For example, connecting resources and agents in a relationship is something not normally
done, and Figure 5 shows that there is no relationship between inventory and salesperson.
If information was needed on which salesperson sold which inventory, a query could be

derived (since there are relations between salesperson and sale, and sale and inventory).

 

22 Weber (1996) used conceptual schema diagrams based on NIAM (N ijssen and Halpin 1989) in which
entities and attributes are denoted by the same construct. E-R (Chen 1976) conceptual schema diagrams do
differentiate between entities and attributes.

2’ The GEFT is a test of ﬁeld dependence/independence in which the participant has to locate and trace a
target stimulus shape within a larger and more complex ﬁgure. Dunn and Grabski (1998) have shown this
test to be positively correlated with REA conceptual modeling performance. Speciﬁcally, ﬁeld
independent undergraduates performed signiﬁcantly better at conceptual modeling tasks than ﬁeld
dependent undergraduates, after controlling for grade point average. There was no signiﬁcant difference in
performance between these groups on non.conceptual modeling tasks.

40

An example of potential effects of incorrect cardinalities is illustrative. Note that

Figure 5 has a maximum cardinality from salesperson to sale of “N” (interpreted as a
salesperson can relate to many sales). If the cardinality was replaced with a “1,” this
would be interpreted as “a salesperson can participate in at most one sale.” Implementing
this design error in the database would result in the salesperson not being allowed to enter
more than one sale (i.e., the database would not physically allow it). REA template
knowledge structure can help prevent these problems because a default schema will serve
to guide the modeling process. An example of the problems caused by incorrect or
unnormalized attribute aggregation was given above. Knowledge of how attributes
aggregate can prevent the problem of unnormalized data.

It is expected that if supportive evidence is obtained for H1 and H2, these
knowledge structures should have a positive affect on the task of designing an REA
system. Therefore, the following hypotheses are tested:

H3: Performance on the conceptual modeling task is positively correlated with

REA template knowledge structure.
H4: Performance on the conceptual modeling task is positively correlated with
attribute aggregation knowledge structure.
2.4 Summary of Chapter Two

The study of individual decision making has grown to be a signiﬁcant branch of
the accounting literature. The use of .cognitive psychology theory has been centrally
important to advancing the understanding of decision making performance in the
accounting task environment. Information systems researchers have likewise been

interested in the cognitive links to system design performance. This dissertation takes an

41

interdisciplinary approach by drawing on accounting, information systems, and
psychology literature to develop and test theory related to accounting information system
design, speciﬁcally REA conceptual design.

The Libby and Luft (1993) framework is used to outline the linkages between the
constructs of interest (i.e., experience, knowledge structure, knowledge content,
performance, and ability). Hypotheses land 2 examine the link between experience and
knowledge structure. Individuals in the early stages of REA knowledge acquisition are
likely to have journal entry knowledge structure (Frederick and Libby 1986). However,
with REA conceptual modeling experience, the REA template should serve as an
organizing mechanism in long-term memory. Therefore, experienced designers should
have an REA template knowledge structure. Hypothesis 1 states, “There will be an
ordinal interaction between experience and information format organization on
knowledge structure.” Likewisz. REA conceptual modeling experience should allow
designers to develop a schema for how attributes aggregate to entities. Hypothesis 2
states, “REA-related entity clustering increases with REA experience.”

While Hypotheses 1 and 2 examine the effect of experience on knowledge
structure, these hypotheses are the ﬁrst step to the central issue: does knowledge structure
affect REA conceptual modeling performance? Hypotheses 3 and 4 examine the link
between knowledge structure and REA conceptual modeling performance. Since two
steps - (l) deriving the structure of the entities and relationships, and (2) properly
aggregating attributes to entities -- are critical to effective system design, the structuring

of this knowledge should be a necessary cognitive process. Hypotheses 3 (and 4) state,

“Performance on the conceptual modeling task is positively correlated with REA

42

template (attribute aggregation) knowledge structure.” Chapter Three describes the

research design employed to evaluate these hypotheses.

43

Chapter Three
RESEARCH METHOD

Chapter Three describes the research method used to gather data to test the
hypotheses developed in Chapter Two. An overview of the research method is outlined
in Section 3.1. Details regarding the study participants are given in Section 3.2. The
dependent variables are deﬁned in Section 3.3. An overview of the tasks (including
independent variables) and procedures is in Section 3.4. Section 3.5 concludes the
chapter.
3.1 Research Method

In the psychology literature, knowledge is deﬁned as information stored in
memory. Since knowledge cannot be directly observed, psychologists have developed
indirect methods (e.g., free recall experiments, reaction time, protocol analysis) which
allow inferences about knowledge. This dissertation utilizes the free recall method to
determine knowledge structure. Ashcraft (1994, 671) deﬁnes free recall as “... the
memory task in which the list items may be recalled in any order, regardless of their order
of presentation.”24 In their use of the free recall method, psychologists usually” measure
the dependent variable as the number of items recalled correctly. The independent
variable is usually measured as the type of list items (Ashcraft 1994, 218).

As mentioned in Chapter One, this dissertation’s research method can be

decomposed into three tasks26 (tasks 1 and 2 measure knowledge structure; related to

 

2‘ The assumption is that recall order corresponds to the location of information in knowledge structure.
2’ Other dependent variables that have been used include: order, speed or organization of recall (Ashcraﬁ
1994, 218).

2" These tasks will be discussed in greater detail in Section 3.4.

44

Hypotheses l and 2, task 3 measures conceptual database design performance and relates
to Hypotheses 3 and 4). To obtain a measure of REA template knowledge structure, task
1 employs ﬁee recall of diagrams in a 2 x 2 design (Figure 9, Panel A) with one between-
subjects factor (experience) and one within-subjects factor (diagram type). Task 1 allows
for inferences to be drawn regarding the effects of experience on REA template
knowledge structure. To obtain a measure of attribute aggregation knowledge structure,
task 2 employs free recall of a list of attributes in a 2 x 1 paired-groups design (Figure 9,
’ Panel B). Task 2 allows for inferences to be drawn regarding the effects of experience on
attribute aggregation knowledge structure. Task 3 (Panel 9, Figure C) involves the
conceptual database design of an REA system. The performance on task 3 will be
regressed on the knowledge structure variables measured in tasks 1 and 2. This allows
for inferences to be drawn regarding the effects of knowledge structure on performance.
3.2 Participants

Two groups of participants were used in this dissertation: undergraduate (N = 54)
and graduate (E = 47) information systems students attending Michigan State University.
This sample was selected to ensure that the statistical tests had enough power to detect an
effect (i.e., without the use of students it would be difficult to assemble a large group of
conceptual database modelers). Additionally, since this is the ﬁrst study of knowledge
structures and accounting system designers, it was decided that it would be important to
hold work experience constant and examine knowledge acquisition derived solely from
classroom (i.e., training) experience. The advantage of using students in this study is a

high level of control over the participants’ experiences with REA.

45

Panel A: REA Template Knowledge Structure Measure (Task 1)

 

 

 

 

 

Diggram
Random REA
Inexpenenced
REA
Experienced
REA

 

 

 

 

Note: Diagram type is a within-subjects factor, whereas experience level is a
between-subjects factor.

Panel B: Attribute aggregation knowledge structure measure (Task 2)

 

Attribute List

 

Inexpenenced
REA

 

 

Experienced
REA

 

 

 

Panel C: Design Performance Measure (Task 3)

 

Design Task

 

lnexpeﬂenced
REA

 

 

Expenenced
REA

 

 

 

Note: Four different orders were used in the study.
1. REA diagram, attribute list, random diagram, design
2. Random diagram, attribute list, REA diagram, design
3. Design, REA diagram, attribute list, random diagram
4. Design, random diagram, attribute list, REA diagram

Figure 9
Experimental Tasks

46

Participants were asked to volunteer for the study. Students were not required to
participate in the study. Those that chose to participate received class credit worth six
percent of their ﬁnal grade; those that chose not to participate received the six percent
foregone credit in the form of points reallocated to their ﬁnal exam. The instructors of
these classes agreed to these terms. Even though participation was not mandatory, the
majority chose to participate (47 out of 52 graduate students and 56 out of 60
undergraduate students).27 It is unclear why students chose not to participate, in the study.
Therefore it is assumed that the choice to participate or not was not driven by any
systematic factor. It is also assumed that motivation is randomly distributed across
groups (this will be discussed further in Chapter F our).

Due to the manner in which experience was proxied (i.e., undergraduate versus
graduate student) random assignment was not possible (an inherent limitation of
experience studies).28 Due to logistical constraints, participants could not be randomly
assigned to treatments. However, all experimental sessions were held at similar times,
and there was almost an equal number of participants in each of the task orders (which
were counterbalanced).

The undergraduate participant group had 30 males and 24 females with an average
age of 21.2 years; the graduate participant group had 31 males and 16 females with an

average age of 25.1 years. It appears that none of the participants had work experience

 

2’ The data was collected over two separate sessions (one session for the free recall tasks and another for
the conceptual data modeling task). Two undergraduate students participated only in the data modeling
session. This data was not used. One undergraduate student participated only in the free recall tasks. This
data was used for testing the ﬁrst Hypotheses 1 and 2, but not Hypotheses 3 and 4. Therefore, there were
really 54 undergraduate students used in the data analysis. The students who attended only one session
received credit worth three percent of their ﬁnal grade with the remaining three percent reallocated to their
ﬁnal exam.

47

involving conceptual database design (although the majority of the participants did not
answer the exit survey question regarding auditing and/or information system design
experience). Four undergraduates and two graduates reported having audit internships.
One graduate was working as an auditor. Four graduates reported being IS interns.

For the undergraduate students (who were enrolled in their ﬁrst accounting
information systems class), the experiment was administered during the latter part of the
ﬁrst half of the semester. The rationale for this choice was to try to observe the
participants during the early stages of knowledge acquisition. The experiment was
administered at a point in the semester where the participants should have obtained the
knowledge necessary to complete the study. Professors who taught the undergraduate
AIS class veriﬁed that the students would have the knowledge content at that point in the
semester. A test of knowledge content, discussed later in the chapter, was also used to
see if the undergraduate participants had the knowledge necessary to complete the
experiment.

For the graduate students (who were enrolled in a graduate accounting
information systems class dealing with conceptual modeling), the experiment was
administered near the end of the semester. The rationale for this choice was to maximize
the amount of experience obtained by the participants. All graduate student participants
had taken the same undergraduate AIS class (or its graduate equivalent) as the
undergraduate student participants. The REA framework was the main focus of both the

undergraduate and graduate AIS classes, and the E-R model was the chosen

 

2' However, see Butt (1988) for an alternative way to measure experience.

48

representation formalism. In summary, the level of conceptual modeling experience was
likely to be homogeneous within groups.
3.3 Dependent Variables

The dependent variable in tasks 1 and 2 is knowledge, speciﬁcally knowledge
structure. Knowledge structure was measured by a free recall method using diagrams and
a list of data attributes for the respective tasks. In task 1, REA template knowledge
structure is derived by summing the total information units recalled.29 In task 2, attribute
aggregation knowledge was derived by calculating an ARC index (discussed below). The
dependent variable in task 3 is performance on a conceptual database design task. The
performance measure was agreement-based in accordance with a recognized expert in the
ﬁeld of accounting database design.

It is important to note that, although the REA template and the attribute
aggregation knowledge structures are measured as dependent variables in tasks 1 and 2,
they become independent variables in task 3. In other words, it is necessary to ﬁrst
determine a measure of knowledge structure before the relationship between knowledge
structure and task performance can be examined. The measurement of these variables is
further discussed in the following section.

3.4 Tasks and Procedures
This dissertation consists of three main tasks, two supplemental tasks, and an exit

questionnaire. Tasks 1 and 2 are used to measure knowledge structure, whereas task 3 is

 

’9 In this dissertation an information unit is either an entity, a relationship, a minimum or maximum
cardinality, or an attribute.

49

used to measure design performance. As stated above, the link between knowledge
structure and conceptual database design performance is the main research question.
3.4.1 Task 1

In task 1 each group of participants received an REA diagram and a random
diagram (see Figures 10 and 11). This is conceptually analogous to the chess studies of
de Groot (1966) and Chase and Simon (1973) although those studies used actual
chessboards rather than diagrams. Both diagrams were represented using the E-R model
(Batini et al. 1992). The REA diagram was based on the REA accounting model and
consisted of two connected REA templates (representing the revenue and acquisition
cycles). The random diagram was the same structure as the REA diagram except all of
the labels were randomized.30 In other words, both treatments looked similar (i.e., had
the exact same interconnected rectangles and diamonds) but the REA diagram conformed
to the REA accounting model. Cardinalities were also included in each diagram. Since
the design performance in task 3 was to be regressed on the task 1 free recall measure
(REA template knowledge structure), it was necessary to make diagram type a within-

subjects factor.3|

 

30“

Randomization” was attempted by (1) putting the labels on pieces of paper (one label per piece of
paper); (2) placing the pieces in a box; (3) drawing the labels one at a time; and (4) placing them on the
diagram. After this process was complete, the diagram was examined to make sure nothing related to the
REA diagram (in some cases it was necessary to rearrange the labels). After the experiment was completed
it was noted that the relationship between “Accounts Payable Clerk” and “Cash Disbursement” was present
in both diagrams (although the cardinalities between them were different). This appears to be an
insigniﬁcant problem and does not affect the results obtained. If anything it should bias against supporting
the hypothesis. Furthermore, a truly random diagram would have links that could appear on the REA
diagram.

3' If a between-subjects design had been used with task 1, only half of the participants would be used in the
task 3 analysis. Task 3 involved regressing design performance on the knowledge measures from tasks 1
and 2. A between-subjects design on task 1 would only provide knowledge measures for the group that
received the REA diagram not the random diagram.

50

 

 

 

I” (1.1)

. Purchase
(1,1) ‘0'") . Order

Cash «on

Disbursement? ' . . M)

 

(1,1

   
      

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  
  

 

 

 

   

 

 

 

 

 

(our (1.1) (1.1) .
Accounts J Purchase (1,1) .
Payable Clerk 7 (0,") 7 (ON)
° (LN)
h
Clerk I0,N) (ON)
(0,") ,
O ,_ (0,") . (0M Inventory
Warehouse ‘0 N (0,") .
(1,1) (1 N) l Salesperson (0,"
’ (em
| Cash Receipt W Sale (1.1)
RM) 7
( (1,1) Customer
I0.N) Order

 

 

   
     

IO,N) (1,1)

 

 

 

1,

(CM

Cashier I

 

 

Shipping Clerk]
(0 ,

 

 

 

Customer

I (or)

 

 

Figure 10
REA Diagram for Task 1

51

 

 

 

 

(1'1) (1'1) 7
P bIe Clerk

(our , ‘1'" Disbursement
I0,llll

. p . . (1.11

 

 

o

 

(1,1)

 

 

 

(1.1)

. Warehouse
(O,N) (0.")

(UN) Receiv’ng
Clerk

(1.1) (0.") .

 

 

 

 

(0,10

(1,1)

l (1,1) ‘
Inventory W Salesperson
‘0' (our
(0.") (0,")
(ON)

I Customer I
I

Shipping curl

 

 

 

 
 

 

 

 

 

 

 

Cash ReceipTl

(1,1)

      

 

 

 

n .

  

   

 

(O.N)

 

 

Figure 11
Random Diagram for Task 1

52

 

The tasks were counterbalanced to control for order effects (see Figure 9). Half of
the participants performed tasks 1 and 2 before 3, and half performed the opposite.
Additionally, within tasks 1 and 2, half of the participants did the REA diagram recall,
followed by the attribute recall, followed by the random diagram recall; the other half did
the Opposite. Tasks 1 & 2 were conducted on the same day and task 3 on a different day.
This was due to (1) the length of administration and (2) the need to separate the
knowledge measure tasks from the performance tasks to try to minimize any carryover
effects. Scores of total number of items recalled correctly and incorrectly were calculated
for this task. Each entity, relationship, minimum cardinality, and maximum cardinality
was treated as one item recalled. For all tasks, participants were given complete
instructions and opportunities to ask questions.

The diagram ﬁee recall was administered in the following manner.. Participants
were ﬁrst given a diagram “face down.” The participants were then instructed to leave
the papers “face down” until told to turn them over. Upon turning the papers over, the
participants were given three minutes to study the diagram. Instructions were given
indicating that they had three minutes to study the diagram, after which they would be
told to turn their papers over and begin recalling as much of the diagram as possible.
Participants were given up to ten minutes to perform the recall (although no one used the
full ten minutes). These procedures are similar to those used by Weber (1996).

However, the participants in Weber’s study were singly run through the experiment. For
this dissertation it was necessary to administer the experiment in groups of around 15
participants. This was essential to control knowledge acquisition. The amount of time it

would take to administer the experiment to over 100 participants could lead to the

53

problem of the last participants having more experience (and thus more knowledge
acquisition) than the ﬁrst participants.

Recall that task 1 employs a 2 x 2 design with one between-subjects factor
(experience) and one within-subjects factor (diagram type). The structural model for this
ANOVA can be speciﬁed by the following linear model:

Yijk = u + a} + Bk + (aB)jk + (nB)ik/j + Sijk

where Yijk = the ith score at jth level of experience and kth level of diagram type

it = the overall mean of the population
of = the treatment (main) effect of experience at level Aj
Bk = the treatment (main) effect of diagram type at level Bk

(aB)jk = the interaction effect between Aj and Bk

(Timur/j = the nested effect between the B treatments and subjects within Aj

Sijk = uncontrolled sources of variability
Hypothesis testing was conducted to determine if (orB)jk > 0. These results and related
analyses are presented in Chapter Four.
3.4.2 Task 2

The purpose of task 2 was to measure attribute aggregation knowledge

structure using a free recall method. Inexperienced and experienced groups received the
same treatment (see Figure 12). The list of attributes was derived by the following
manner. First, entities from the revenue and acquisition cycles were identiﬁed. Second,
ﬁve attributes were assigned to each entity. Attributes were carefully assigned to

minimize obvious clustering (i.e., the entity “vendor” being composed of the attributes

54

Proposed List:
total items on purchase order

reorder point

terms of sale

quantity on hand

date of bank account opening
invoice number

dollar amount of receipt
bank account number
phone number of vendor
address of customer

dollar amount of purchase order
vendor number

bank account balance
condition of items received
name of bank

customer number

dollar amount of invoice
quantity of items received
accounts payable balance
total items on sale

accounts receivable balance
name of vendor

date of purchase order
description of inventory item
name of customer

date of invoice

address of vendor

inventory item number

date of receipt

unit cost of inventory
receiving report number
phone number of customer
terms of order

bank account type
purchase order number

Progosed clustering results:

Cash

Customer

Inventory

Purchase

Purchase

Order

Sale

Vendor

Figure 12

bank account number

name of bank

bank account balance

bank account type

date of bank account opening

customer number

name of customer

address of customer

phone number of customer
accounts receivable balance

inventory item number
description of inventory item unit
cost of inventory

quantity on hand

reorder point

receiving report number
date of receipt

dollar amount of receipt
quantity of items received
condition of items received

purchase order number

date of purchase order

dollar amount of purchase order
total items on purchase order
terms of order

invoice number

date of invoice

dollar amount of invoice
terms of sale

total items on sale

vendor number

name of vendor

accounts payable balance
address of vendor

phone number of vendor

List of Attributes for Free Recall Task

55

9, 66 99 6‘

“vendor number, vendor name,” “vendor address, vendor phone number,” and
“vendor contact”). Third, each attribute was assigned a number from one to thirty-ﬁve
and then a computer generated random process was used to derive the list in Figure 12.32
The administration procedures were similar to those used by Weber (1980) and
originally conceived by Bousﬁeld (1953), with one modiﬁcation: instead of being read a
list of attributes, participants visually studied a list of attributes.” The list of attributes
were drawn from accounting related entities (as categories) used in REA models. It was
expected that if the experienced participants had more well-developed schemas, they
would cluster according to entities signiﬁcantly more than the inexperienced participants.
The attribute list free recall was administered in the following manner (which was
almost identical to the diagram free recall procedures). Participants were each given a list
of attributes “face down.” The participants were instructed to leave the papers “face
down” until they were told to turn them over. Upon turning the papers over, the
participants were given four minutes to study the list. Instructions were given indicating
that they had four minutes to study the list, after which they would be told to turn their
papers over and begin recalling as much of the diagram as possible. The instructions also
speciﬁed that the words had to be written in a list straight down the page. Participants

were given up to ten minutes to perform the recall (although no one used the full ten

minutes).

 

’2 As a result of the random process, there were a few instances of attributes ﬁom the same entity being
adjacent in the list. These instances were removed to generate the list in Figure 14.

’3 This modiﬁcation is a result of pilot testing. A pilot study was conducted in which attributes were read
and then recalled. Participants, on average, only recalled around eleven attributes. This seemed to limit
opportunities to observe potential clustering.

56

To evaluate whether the participants’ recall protocol exhibited clustering a
measure of clustering had to be derived. This measure of clustering was determined
using an adjusted ratio of clustering (ARC) index developed by Roenker et al. (1971).
This index is deﬁned as: 1

ARC = [R - E(R)] / [max R - E(R)]
where R = total number of observed category repetitions
E(R) = expected (chance) number of category repetitions
= [(25 n? )/N] -1;
n,- = number of items recalled from category i
N = total number if items recalled
max R = maximum number of category repetitions (total ntunber of items recalled
minus the number of categories present in the recall)
The ARC measures degree of clustering with a range from - 1 (no clustering) to 1
(clustering) with zero being clustering due to chance.
3.4.3 Task 3

The critical research question in this dissertation hinges on task 3, which was used
to examine the relationship between knowledge and performance. Participants were
given a narrative describing a particular enterprise and a list of relevant attributes (see
Figure 13 for problem; Figure 14 for solution) and asked to design an REA data model.
As mentioned above, successful performance of this task involves: (1) reading the
description of the enterprise and deriving the structure of the diagram (i.e., an REA
template or templates); (2) mapping the cardinalities; and (3) aggregating the attributes to

the correct entities. The theory described in Chapter Two suggests that knowledge

structure is a fundamental cognitive process related to the task of conceptual modeling.

57

Given the following narrative, prepare an entity-relationship diagram including all
cardinalities and attributes. Use only the attributes given; do not add or subtract

any attributes. For convenience, the relationships in your diagram can be numbered
rather than assigning names.

University Video

University Video is a privately owned store specializing in videotape rentals. In addition to videotape
rentals University Video sells various kinds of candy. The store has ten employees including the
owner; normally only three or four are working at a time. New employees are to be entered into the
database as soon as they are hired (before they actually start working). When customers enter the
store they pass an employee who is available to answer questions about the different videotapes.
Only customers who have rented videotapes or bought candy are to be entered into the database.
Customers usually rent one or two videotapes and buy one or two boxes of candy; however,
sometimes a customer wants only candy or only videotapes. The average customer rents videotapes
once a week but there are a few customers who rent videotapes every day. When customers rent
videotapes and/or buy candy, the product is scanned at the cash register, and a sale timestamp is
assigned. During day hours employees average only a few transactions per hour. However, during
evening hours employees process an average of thirty sales and/or cash receipts per hour. Since any
employee can check out videotapes and/or sell candy, the owner wants to know which employee was
responsible for a given transaction.

To acquire new videotapes the owner searches one of the many videotape vendors en the Internet
and electronically places an order. The owner typically orders several videotape titles at a time from a
vendor. Vendors are to be entered into the database before any purchases occur. The videotapes
usually arrive the next day and one of the employees prepares a receiving report (with a unique report
number) listing the date and value of the purchase. Only one vendor is listed on each receiving report.
University Video often orders videotapes from the same vendors. Upon receipt, all purchased
videotape titles get cataloged with a unique videotape catalog number. It is necessary to catalog the
videotapes by number rather than title since there can be different movies with the same title (for
example, “Cape Fear” was ﬁlmed in 1961 and another version was ﬁlmed in 1991). If the videotape
was purchased before, then the employee simply veriﬁes the catalog information. Associated with
each videotape catalog number is the title, rating, running length, replacement cost, and rental fee.
For example, the title “Batman” is rated “PG-13,” has a running length of 126 minutes, has a
replacement cost of $45.00, and a rental fee of $3.00. Multiple copies of a videotape title can be
purchased; in fact, it is necessary to purchase multiple copies of the latest blockbuster movies to be
competitive. Employees must keep track of the quantity purchased. Simultaneous with the catalog
number assignment, each individual videotape is also assigned a unique videotape identiﬁer number
(however, the owner is not interested in tracking the labor involved in this process). The videotape
identiﬁer number is used to track when and who rents that videotape. For each videotape, the owner
wants to know how many times it was rented. Each videotape has only one movie release on it.

Figure 13
Design Task

58

The store owner has a policy of allowing tape rentals and candy sales to be paid for at the end of the
month (i.e., a tape can be scanned at the cash register without a cash receipt at that time) or the tape
rental and candy sales can be paid for immediately. The videotape rental period is for three days. .
When videotapes are returned to the store they are immediately scanned and assigned a return time.
If a customer rents two or more videotapes at the same time, they do not have to be returned at the
same time. Many customers have returned one videotape on time and another one late, even though
both videotapes were rented at the same time. The database will compare the return time to the time
of the rental and assess a $3.00 late fee per videotape if necessary. The owner only wants to keep
track of total late fees per rental. This information will automatically be updated when the returned
tapes are scanned.

Various candy vendors stOp by the store intermittently to check the stock of candy. If the stock of any
candy product is low, the vendor stocks it. When the candy is stocked one of the employees ﬁlls out a
new delivery form (with a unique number) including the date. Each delivery form lists only one vendor.
The owner is not interested in tracking the sale of individual boxes of candy (for example, the second
box of M&M’s on the shelf), just the kind of candy being sold (for example, M&M's). For each kind of
candy the owner needs to know the quantity purchased. Each kind of candy has a unique product
number. The owner also wants to have a description, a price, and a count of quantity on hand for
each kind of candy. This information is to be entered into the database after the purchase occurs.
Candy vendors can be entered in the database even if University Video never purchased from them.

The owner also wants to track customers’ accounts receivable balances and vendors' accounts
payable balances. The owner is the only employee allowed to write checks. Checks can be applied to
one or more purchases of videotapes or candy; however, each purchase of videotape or candy can
relate only to one check. Checks cannot be written without affecting a bank account, and each check
can affect at most one bank account. University Video maintains several bank accounts. The account
information is entered into the database immediately upon its opening, even though no activity with the
account may have occurred. The same bank account can be used for many cash receipts or cash
disbursements. Occasionally there are cash receipts from financing. The owner deposits cash
receipts from sales/rentals at the end of the day. All vendors are paid at the end of the month.

Data Elements (do not add or delete any)

- candy quantity purchased - vendor address - employee name
- sale dollar amount - vendor name - bank account number
- accounts receivable balance - videotape rating - candy quantity on hand
- cash receipt timestamp - vendor number - candy description
- employee number - videotape replacement cost - bank account balance
- total late fee - customer address - videotape catalog number
- number of times rented - receiving report number - videotape title
- cash receipt amount - sale timestamp - receiving report value
- delivery date - hire date - videotape identiﬁer number
- vendor phone number - receiving report date - delivery form number
- check amount - candy product code - bank name
- check number - videotape running length - accounts payable balance
- candy price - customer number - customer name
- check date - return time - rental fee
- outstanding balance per rental
- videotape quantity purchased
Figure 13 (cont’d)

59

.3: 5.8: s .323.

 

 

 

 

 

           

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

 

  

 

          

 
 
  

.26.

2 8.5..
2.5 SE A» 3:23 «Seen; :80 Q/ 100 >230
oEazm. oEez xcem / :30 no.5 >050 OX >950.
oo>eEEw 0.]
5.32800
026.95 .BEsz >230
.3832 E3933 :30 . e
s a .z o. 80
8 2 .5 839a >28
EaeE< 3.80m
. v :30
2 Z \ anueoEF
.. _ 8.30m . .zd.
coco a o ozone. 3:330 . .26. an 8.32".
2.3. 4 < O/ n o r. 58
3203‘ 8.50.330 A . -
oEuz .eESwao 0/ 35330 .26. a E: 3.
.enEaz 88330 .26. on 2
.26.
. EucoEoﬁm
: : 2.9
.35: can. ecceim 9.55350 0 _ 2...... 530m

eon. Bu.— .Se... C

EneE< 5.00 gum L E
QEBmeEF 03w

 

 

ensues—.5

 

 

totem aw

L

36:. .e .BEaz

:3. 38265

 

A5330 3 2.5...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

   

  
   

 

 

 

 

     

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

33:05.“. AZ. 5 ~30 EoEoouAQom A on > .C
3:520 33035 A. can. .35: 0] @2300
A26. 595.. 9:25: 33805
, 8 2.0. 95a
2.50:3 tone: 9:233. A . c
Sun 2.: r) 2:...
8.5 tone: 0:353: OIII eEmz ,/
omucouzm eo>oEEm O/ oo>oAaEw .enEaz
ABEsz tone: 9:333. .\ mango—05 .3532 00350
eAF. — A oe>o_aEm
. A26. A26. .
coco-em oAnn>aa 3:386. AP 0. AZ 3
.onEaz ace—E Buco> AZ .0. AF. 2 Eco
\Mu>.e>__oo
3223 Beco> .. AZ. 0. .
A V/ Loeco> T AZAO. A .5 3205.". .3532
oEuz ceece> AX A306 . \AVxA F .9 >960 E8".
AA. OT >.e>__oo
cﬁEaz botco> EoEomShmE \AZ.oA AZ.oA .
we
Bun. x0050 ﬂ 5 0 83:05.“. , e
anE< x030 $55.50 > 50 / AF. 5
. :oo .28 f .2. :
.mnEaz x850 AF 5 8...; >930
35.5 53.2830 0/
E382 , A26. >950 0/ >950
:80 A/ 250 38035
, 8:02.... >980
page... 2
Esooo< :80 pence: moEE. Ac conEaz a F

.3852 5:353 330005

61

The choice of which type of company to model -- in this case, a videotape rental
enterprise -- was made to ensure that all of the participants would understand the business
to be modeled. Revenue and acquisition cycles were used in this problem since both of
the knowledge structure measures relate to these cycles. The use of a narrative and list of
data attributes was consistent with how data modeling problems are presented in the
undergraduate and graduate AIS classes. A designer in the “real world” would have to
interview users to determine the data needs of the business. This process can be held
constant in an experimental setting. However, to simulate some of the complex “real
world” features (e. g., users providing irrelevant information) of obtaining information
from users, additional information (e. g., distracters") were written in the narrative. If the
experienced designers have well developed schemas (such as REA template knowledge
structure), it should reduce the complexity associated with the narrative.

In performing the scoring, each participant’s solution was compared to a copy of
the correct solution (seen in Figure 14). Everything that was correct was marked with a
green pen on the copy of the solution; everything that was incorrect was marked with a
red pen on the copy of the solution. Based on this task performance, a correctness score
was derived. To keep the scoring consistent across participants, a scoring sheet was used
(see Figure 15). One scoring sheet was used for each participant. On the right-hand

column of the scoring sheet the number of instances per error type were recorded. The

 

3‘ An example of one of the distracters is the following sentence near the beginning of the narrative: “The
average customer rents videotapes once a week but there are a few customers who rent videotapes every
day.” The surface semantics of this sentence may lead an inexperienced designer to think that this is
information about cardinalities.

62

University Video Score Sheet

 

 

- 10 each

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Entity omitted

Attribute of entity omitted - 4 each
Attribute of relationship omitted - 4 each
Relationship omitted - 6 each
Minc omitted - 1 each
Maxc omitted - 1 each
Entity incorrect (inc. no attributes) - 10 each
Attribute incorrect on entity - 4 each
Attribute incorrect on relationship - 4 each
Relationship incorrect - 6 each
Minc incorrect - 1 each
Maxc incorrect - 1 each

 

 

Sale and Rental two entities

not a problem if
modeled correctly

 

 

 

 

 

 

 

Candy and Video Vendor two not a problem if
entities modeled correctly
Purchase one entity - 10 each

Video one entity - 10 each

 

Notes:

Figure 15
Scoring Sheet

63

 

instances were then weighted by the scoring per error type given in Figure 15.35 These

weights were recommended by a professor with several years of expertise in grading
conceptual data models. The weighting scale takes into account the fact that some errors
have greater consequences than other errors. For example, an incorrect or omitted entity
is a severe design error because entities are implemented as tables in a database.
Relationships are associations between entities. An error with a relationship is not as
important as an error with an entity because it usually results in an inefficient
implementation rather than a total loss or misrepresentation of data. Omitted or incorrect
attributes result in data loss or misrepresentation, but the problem is much smaller at the
attribute level than the entity level. Incorrect cardinalities affect the constraints on
relationships which is the least serious of the design problems mentioned. Sensitivity
analysis, using both (1) an ordinal ranking of error types, and (2) a ranking where all
errors types are equally weighted, is discussed in Chapter Four of this dissertation.

The sum of all of the weighted errors was then calculated (giving negative scores
where the more errors made, the more negative the score). To convert this to a
correctness score, the most negative score was transformed to zero (e.g., if the worst
score was a - 200, the formula was correctness score = (200 - raw score). This
transformed score was then used in the regression on the knowledge structure (free recall)

measures from tasks 1 and 2 to test Hypotheses 3 and 4.

 

3’ The “notes” section was used to note any errors that did not ﬁt in the above categories. Two such errors
were found. The ﬁrst was not identifying a primary key. The second was identifying an incorrect primary
key. Both of these error types resulted in a deduction of two points.

64

Recall that the results obtained in task 3 will be. regressed on the knowledge
structure measures obtained in tasks 1 and 2. The linear model for this regression can be
speciﬁed by the following linear model:

SDP = [5,, + leEATKS + BzAAKS + B3KC + [3,GEFT + c
where SDP = System Design Performance score ﬁom task 3
REATKS = REA template knowledge structure score from task 1
AAKS = Attribute aggregation knowledge structure score from task 2
KC = Knowledge content score from knowledge content test

GEFT = Group Embedded Figures Test score

Hypothesis testing was conducted to determine if [3, > 0 and/or [32 > 0. These results and
related analyses are presented in Chapter Four.
3.4.4 Supplemental Tasks

Before administration of task 3, participants took the Group Embedded Figures
Test or GEFT (Oltrnan, Raskin, and Witkin 1971). The GEFT'3 is a booklet of embedded
ﬁgures consisting of one practice section (two minutes in duration) and two actual
sections (ﬁve minutes in duration per session). The selection of target ﬁgures is given on
the back of the booklet. Each embedded ﬁgure question has a target ﬁgure embedded
within a more complex ﬁgure. Participants are allowed to ﬂip to the back of the booklet
as often as necessary, and they are asked to trace the outline of the target ﬁgure in each
problem. After completion of each section, participants were told to stop working and

wait until signaled to begin the next section. This task was used to proxy for spatial

65

abilities which may be related to conceptual data modeling performance (Dunn and
Grabski 1998).

At the end of the second experimental session participants completed a knowledge
content test (see Figure 16) and an exit questionnaire (see Figure 17). As mentioned in
Chapter Two, it is important to distinguish between knowledge content and knowledge
structure. "Knowledge content" means what information is stored in memory.
"Knowledge structure" means that some type of organization (e. g., hierarchical, temporal.
causal) is placed on the knowledge content. It was assumed that inexperienced and
experienced participants had the knowledge content necessary to perform the conceptual
design task and a test was conceived to test this assumption. Even if differences exist in
knowledge content, this measure could be used as a control variable in the regression
equation. It should be noted that even though the experienced participants had classroom
exposure to advanced conceptual modeling concepts (e.g., classiﬁcation, aggregation,
generalization), this type of knowledge was excluded from this study by design. The goal
was to have participants with two levels of experience and the knowledge content
necessary to perform the tasks. Then any differences noted can be attributed to
knowledge structure and not to knowledge content.

The knowledge content test was used to determine whether or not participants had
an understanding of the basic constructs used in REA conceptual modeling. Question I

asked what four basic constructs represented.” This question tested for knowledge of the

 

3" A sample GEFT booklet is not included in this dissertation due to copyright restrictions. The booklet can
be obtained from Consulting Psychologists Press, Inc., 3803 E. Bayshore Road, Palo Alto, CA 94303.

’7 Symbol la. represents an entity, 1b. represents a relationship, 1c. represents an attribute, and Id.
represents a primary key attribute.

66

Survey Questions ldentifer Code:
1. Describe what the following symbols represent:

a.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 16
Knowledge Content Test

67

2. Give an example of a resource, an event, an internal agent, and an external agent.

Resource

 

Event

 

Internal Agent

 

External Agent

 

3. Out of the nine cardinality sets (minimum, maximum) below, which of the following
are incorrect? Circle all of the incorrect pairs.

(0,0) (0.1) (0N) (1.0) (1.1) (IN) (N0) (N1) (MN)

4. Fill in the default cardinalities for the following entity-relationship diagram:

 

 

Sale 1 \src/ I Customer

 

 

5. Describewhateachofthecardinalitiesmeanintheabovediagram. Youranswer
should include descriptions of both of the minimum and both of the maximum
cardinalities that you ﬁlled in above. You can write on the back of this page if necessary.

 

 

 

 

 

 

 

Figure 16 (cont’d)

68

PARTICIPANT INFORMATION
(This data is confidential and will only be used in conjunction with this project).

Age: Sex:

 

Instructor’s name for ACC 321 (or ACC 821):

 

Undergraduate GPA: __

Graduate GPA: __

Undergraduate accounting GPA:

Graduate accounting GPA:

Undergraduate accounting course grades received:
Principles of Financial (ACC 201)

Principles of Managerial (ACC 202)
Intermediate I (ACC 300)

Total number of accounting classes completed:
Are you enrolled in, or have you completed, an auditing class?
Number of undergraduate credits completed

Number of graduate credits completed '

SAT scores

Overall: Verbal: Quantitative:
ACT scores

Overall: English: Math: Reading: Science:
GMAT scores

Overall: Verbal: Quantitative:

Please describe any work experience you have related to auditing and/or information
system design. Include a description of responsibilities and number of years worked.
You can use the back of this sheet if necessary.

Figure 17
Exit Survey

69

modeling grammar. Question 2 asked for examples of a resource (e.g., inventory), event
(e.g., sale), internal agent (e.g., salesperson), and external agent (e.g., customer). This
question tested for knowledge of components of the REA model. Question 3 asked for
the identiﬁcation of all incorrect cardinality pairs.38 This question tested for knowledge
of what cardinality patterns could and could not be used. Question 4 asked for the default
cardinality patterns. Usually (I) an event has to have one and only one agent, and (2) an
agent does not have to participate with every event, but can participate with many.
Question 5 tested for knowledge of the use of cardinalities.

The knowledge content test was scored on a dichotomous scale with 1 point given
for correct answers, 0 points given for incorrect answers. Question 1 was worth four
points (one point per construct). Question 2 was worth four points (one point per
resource, event, internal and external agent). Question 3 was worth ﬁve points (one point
for each incorrect cardinality pair identiﬁed). Question 4 was worth four points (one
point for each minimum and maximum cardinality). Question 5 was worth four points
(one for each cardinality described correctly).

The exit questionnaire (Figure 17) was used to obtain basic demographic data and
ability information. This information was self reported and hence, possibly unreliable.
However, the participants were assured of anonymity and conﬁdentiality. This beneﬁt to
the participants outweighed the potential error of measurement in the exit questionnaire.

The experimental instruments and procedures described in this chapter were pilot

tested. This provided valuable insights regarding how much time should be allocated in

 

3‘ The incorrect pairs are as follows: (0,0); (1,0); (N,0); (N,l); and (N,N).

70

the free recall experiments. The pilot tests revealed that only minor modiﬁcations were
needed to the experimental instruments and procedures.

Finally, the participant’s informed consent was obtained at the outset of the study.
Participants were told that they were participating in a study of the effects of experience
on the acquisition of REA modeling knowledge, and given estimates regarding the
amount of time the experiment would take. They were informed that their participation
was voluntary and that they could withdraw ﬁ'om the experiment at any time without
penalty. As mentioned above, participants were told that their performance would be
kept anonymous and conﬁdential. They were also informed of the compensation (i.e., the
six percent class credit explained above) and the name of the researcher and dissertation
chair. At the conclusion of the experiment participants were given a letter from the
researcher describing the objectives of the research and thanking them for their
participation.

3.5 Summary of Chapter Three

This chapter described the research method used to gather data to test the
hypotheses developed in Chapter Two. The research method involved three tasks. The
ﬁrst task used free recall of diagrams to infer REA template knowledge structure. The
second task used free recall of a list of attributes to infer attribute aggregation knowledge
structure. The third task required participants to perform a conceptual database design.
A distinguishing feature of the experiment was that the conceptual modeling task was a
realistic abstraction of what professional conceptual database designers would perform.

Supplemental tasks were used to obtain demographic data. The tasks were

71

counterbalanced to control for order effects. Chapter Four summarizes the data collected

in the experiment and describes the statistical tests of the hypotheses.

72

Chapter Four
RESULTS

This chapter describes the statistical techniques used to test the hypotheses
developed in Chapter Two. The data used in these statistical tests were gathered from
undergraduate and graduate students at Michigan State University as outlined in Chapter
Three. The chosen or level in this dissertation was .05. All numbers (except the p-values)
were rounded to the nearest hundredth.

Hypothesis I predicted that there would be an ordinal interaction between
experience and information format organization (i.e., diagram type) on knowledge
structure. Analysis using a mixed two-factor within-subjects AN OVA revealed a
signiﬁcant interaction between experience and diagram type, If (1,99) = 45.08, p < .0001.
Additionally, tests for simple effects revealed that experienced participants recalled
signiﬁcantly more of the REA diagram than inexperienced participants, E (1, 99) = 56.99,
p < 0.0001. Tests for simple effects revealed no signiﬁcant difference in recall between
groups with the random diagram.

Hypothesis 2 predicted that REA-related entity clustering would increase with
REA experience. Analysis using an independent-samples t test on derived ARC scores of
attributes recalled revealed a signiﬁcant difference between experienced and
inexperienced participants, 1 (99) = -3. l 7; p_ = 0.001 (one-tailed test).

Hypothesis 3 predicted that performance on the conceptual modeling task would
be positively correlated with REA template knowledge structure. Hypothesis 4 predicted

that performance on the conceptual modeling task would be positively correlated with

73

attribute aggregation knowledge structure. Multiple regression analysis regressing
design performance on REA template knowledge structure, attribute aggregation
knowledge structure, knowledge content, and ability revealed signiﬁcant regression
coefﬁcients for REA template knowledge structure (2 < .01), attribute aggregation
knowledge structure (2 < .01), and knowledge content (2 < .05).

These results suggest that experience with REA conceptual modeling results in
knowledge organization. The knowledge organization exhibited is consistent with the
notion that the REA template is a useful aid in knowledge organization. The results also
suggest that having REA template and attribute aggregation knowledge structures
increases conceptual database design performance.

The remainder of this chapter presents a detailed analysis supporting these
conclusions. Section 4.1 illustrates tests for order effects; if no order effects are noted,
then the data can be aggregated and analyzed. Sections 4.2, 4.3, and 4.4 describe the
hypothesis testing for REA template knowledge structure, attribute aggregation
knowledge structure, and the relationship between knowledge structure and conceptual
database design performance. Section 4.5 completes this chapter with a summary.

4.1 Tests for Order Effects

In an attempt to disentangle treatments effects from practice effects, the
experimental treatments were counterbalanced. Two sets of instruments were ordered:
(1) participants recalled either the REA or the random diagram ﬁrst; and (2) participants
either performed the system design task or the diagram recalls ﬁrst (this was also
described in Chapter Three and Figure 9). Due to logistical constraints (i.e., the length of

time administering the experiment; the experiment being held outside of normal class

74

hours, etc.) participants were not randomly assigned. Participants were given several
different times at which they could attend the experimental sessions. There is no reason
to believe that any participants were systematically biased in their selection of
experimental sessions. In quasi-experiments, it is necessary to “explicate the speciﬁc
threats to valid causal inference that random assignment rules out and then in some way
deal with these threats” (Cook and Campbell 1979, 6).

A 2 x 2 AN OVA (Order 1 x Order 2)” was calculated to test for practice effects.
Table 1 shows the AN OVA results for the three dependent variables: REA template
knowledge structure (Table 1, Panel A), attribute aggregation knowledge structure (Table
1, Panel B), and system design performance (Table 1, Panel C). Additionally, the control
variable random diagram recall (Table 1, Panel D) from task 1 is also included. This
analysis revealed that no signiﬁcant order effects were noted, at a = .05. Therefore, the
data were combined for hypothesis testing purposes.
4.2 Hypothesis Testing for REA Template Knowledge Structure

Hypothesis 1 states, “There will be an ordinal interaction between experience and
information format organization on knowledge structure.” This hypothesis was tested
using a mixed design with one between-subjects factor (experience), and one within-
subjects factor (diagram type). The dependent variable for this task is the total recall
score (as described in Chapter Three, one point was awarded for each entity, relationship,

minimum cardinality, or maximum cardinality recalled).

 

’9 Order 1 is whether the REA diagram or Random diagram is recalled ﬁrst. Order 2 is whether the
diagrams are recalled ﬁrst or the system design is performed.

75

Table 1
ANOVA Tables for Practice Effects Tests

Panel A: Dependent Variable REATKS (REA template knowledge structure)

 

Source _cﬁ _S§ _M_S_ F Pr > F R2

 

 

 

 

 

 

Order 1 1 1.01 1.01 0.00 0.9719 .00
Order 2 1 1,869.22 1,869.22 2.32 0.1313 .02
Order 1 x Order 2 1 1,601.15 1,601.15 1.98 0.1622 .02
Within groups 96 77,477.73 807.06

Total 99 80,949.11

m: N = 100.

Panel B: Degndent Variable AAKS (attribute aggrpgation knowledge structure)
Source gl_j Si M=S ; Pr > F R2
Order 1 1 0.39 0.39 2.97 0.0882 .03
Order 2 1 0.00 0.00 0.01 0.9368 .00
Order 1 x Order 2 1 0.38 0.38 2.89 0.0922 .03
Within groups 96 12.65

Total 99 13.42

M3131 N = 100

Panel C: Dependent Variable SDP (System Design Performance)

 

 

 

 

 

 

Source ____1 g; M_$_ E Pr > F R2

Order 1 1 4,513.42 4,513.42 0.75 0.3881 .01
Order 2 1 957.51 957.51 0.16 0.6905 .00
Order 1 x Order 2 1 4,452.74 4,452.74 0.74 0.3913 .01
Within groups 96 576,305.45

Total 99 586,229.12

m: _N = 100

Panel D: Control Variable RDR (random diagram recall)

 

 

 

Source ____f §__s M__§ E Pr > F R2
Order 1 1 2.15 2.15 0.02 0.8905 .00
Order 2 1 9.81 9.81 0.09 0.7689 .00
Order 1 x Order 2 1 15.61 15.61 0.14 0.7109 .00
Within groups 96 10,843.78 -

Total 99 10,871.35

 

Note: ﬂ = 100

76

Table 2 shows the means and standard deviations of the total number of items
recalled by experience level and diagram type. The AN OVA analysis, summarized in
Table 3, revealed a signiﬁcant interaction between experience and diagram type,

F (1,99) = 45.08, p < .0001. The interaction is such that there is no difference between
inexperienced and experienced group on the random diagram recall, but the experienced
group recalled signiﬁcantly more than the inexperienced group on the REA diagram
recall. This interaction is graphically depicted in Figure 18. Additionally, tests for
simple effects (illustrated in Table 4) revealed that experienced participants recalled
signiﬁcantly more of the REA diagram than inexperienced participants, E (1, 99) = 56.99,
p < 0.0001. No signiﬁcant differences were revealed in recall between groups with the
random diagram, E (1, 99) = 1.46, p > .23. These obtained results are consistent with
Hypothesis 1.“
4.3 Hypothesis Testing for Attribute Aggregation Knowledge Structure

Hypothesis 2 states that, “REA-related entity clustering increases with REA
experience.” This hypothesis was tested using an independent-samples t test calculated
on the mean ARC index scores. This test revealed a signiﬁcant difference between
experienced and inexperienced participants, 1 (99) = -3.17; p = 0.001 (one-tailed test).
The sample means are shown in Table 5 and Figure 19 which illustrate that experienced

participants scored signiﬁcantly higher on the ARC index than did inexperienced

 

‘° It is reasonable to assume that the individual treatment populations were drawn from normal
distributions. Using the Wilk-Shapiro test on the univariate variables of interest, the null hypothesis (5,:
the sample was drawn from a normally distributed population) could not be rejected. A review of the
histograms conﬁrmed that the normality assumption is reasonable.

77

Table 2
Means and Standard Deviations of the Number of Items Recalled for Hypothesis 1

 

 

 

 

 

 

 

 

 

 

 

 

 

Inexperienced Experienced Group
Group
REA Diagram M 31.05 65.34
(S__12) (14.06) (29.80)
Random Diagram M 21.00 23.51
(§_=_Q) (6.75) (13.48)
N_otg: N = 101
Table 3
ANOVA Table for Hypothesis 1
(Signiﬁcant Interaction)
Source g1 §__S= L _F_ Pr > F
Between Subjects 100 51,231.40
Experience (E) 1 17,011.00 17,011.00 49.21 0.0001
Residual Between 99 34,220.40 345.66
Within Subjects 101 74,364.10
Diagram (D) 1 33,824.36 33,824.36 120.22 0.0001
E x D 1 12,685.00 12,685.00 45.08 0.0001
Residual Within 99 27,854.74 281.36
Total 201 125,595.50

 

Note: N =101.

78

 

 

 

Number of Items Recalled

 

 

 

 

”-0- Inexperienced Group ;

 

 

 

o , _ +Experienced Group 3
Random REA
Diagram Diagram
Diagram Type
Figure 18

Experience x Diagram Interaction

Table 4

Summary of Group Simple Effects for Hypothesis 1

 

Inexperienced vs. Experienced

 

 

 

 

E (91) Pr > F
REA Diagram F (1, 99) = 56.99 0.0001
Random Diagram F (1, 99) = 1.46 0.2303

 

 

 

NptezN=101

79

 

 

 

 

 

 

 

Table 5
Means, Standard Deviations, Standard Errors, and Skewness of the ARC Clustering
Index for Hypothesis 2
Inexperienced Group Experienced Group
(N = 54) (N = 47)
ARC Score M 0.30 0.53
S_D (0.37) (0.32)
_S_E [0.05] [0.05]
Skewness {0.1 0} {-0.60}

 

 

 

ARC Index

 

Inexperienced Experienced Group
Group

Figure 19
Mean Levels of ARC Scores for Participants in Inexperienced versus Experienced
Groups, Independent Samples Design

80

participants (for experienced participants, M = 0.53, _SQ = 0.32; for inexperienced
participants, M = 0.30, S_D = 0.37). The standard deviations of the scores is reduced from
the inexperienced to experienced group (S_D = .37 versus ﬂ = .32). The skewness
moves from slightly positive in the inexperienced group to negative in the experienced
group (skewness = .10 versus skewness = - .60). These statistics are also consistent with
the theory that members of the experienced group have better developed schemas than
members of the inexperienced group. These obtained results are consistent with
Hypothesis 2."

As part of ex post analyses, an independent-samples t test was calculated on the
mean number of attributes recalled. This test revealed a signiﬁcant difference between
experienced and inexperienced participants, 1 (99) = -2.41; p < 0.05 (two-tailed test). The
sample means are shown in Table 6, which illustrates that experienced participants scored
signiﬁcantly higher on the number of attributes recalled than did inexperienced
participants (for experienced participants, M = 19.30, §_l_3_ = 3.97; for inexperienced
participants, M = 17.33, _S_Q = 4.20). Therefore, out of a total possible list of attributes of
35, experienced participants recalled 55% on average, while inexperienced participants
recalled 50%. Even though the difference is statistically signiﬁcant, this difference may

not be practically important. The E statistic revealed that the homogeneity of variance

 

" It is reasonable to assume that the individual treatment populations were drawn from normal
distributions. Using the Wilk-Shapiro test on the univariate variables of interest, the null hypothesis (L1,:
the sample was drawn from a normally distributed population) could not be rejected. A review of the
histograms conﬁrmed that the normality assumption is reasonable.

81

Table 6
Means, Standard Deviations, and Standard Errors of the Number of Attributes

 

 

 

Recalled for Hypothesis 2
Inexperienced Group Experienced Group
(N. = 54) (N = 47)
Number of Items M 17.33 19.30
Recalled SD (4.20) (3.97)
i [0.57] [0.58]
Skewness {.07} {.00}

 

 

 

 

assumption is tentatively accepted, If (53, 46) = 1.12, p > 0.05.

Further analysis of the ARC index scores indicated there were 43 positive and 4
negative ARC index scores from the 47 experienced participants (i.e., 91% and 9%
respectively) and 41 positive and 13 negative ARC index scores from the 54
inexperienced participants (i.e., 76% and 24% respectively). This analysis provides
further evidence which is consistent with the expectation that experience has a positive
effect on the structuring of attribute aggregation knowledge.

4.4 Hypothesis Testing for Relationship Between Knowledge Structure and
Conceptual Database Design Performance

Hypothesis 3 states, “Performance on the conceptual modeling task is positively
correlated with REA template knowledge structure.” Hypothesis 4 states, “Performance
on the conceptual modeling task is positively correlated with attribute aggregation
knowledge structure.” These hypotheses were tested using both correlation and multiple

regression analysis.42

 

‘2 P values will be based on one-tailed test unless noted.

82

 

Table 7 presents means, standard deviations, and Pearson correlations using
pairwise deletion. The correlations indicated three variables associated with system
design performance: REA template knowledge structure (g = .55), attribute aggregation
knowledge structure (I = .26), and knowledge content (x = .33). These correlations were
in the predicted direction and statistically signiﬁcant at p < .001, p < .01, and p < .001 ,
respectively. These results are consistent with Hypotheses 3 and 4. The correlation
between the Group Embedded Figures Test and system design performance is not
statistically signiﬁcant (p = .18, p = .08). Table 8 presents a similar analysis of Pearson
correlations using listwise deletion. The correlations are similar; the only real difference
is the correlation between system design performance and knowledge content (; = .32) is
signiﬁcant at (p < .01 rather than .001).

Parameters of the regression model were estimated using ordinary least squares
where system design performance was regressed on REA template knowledge structure,
attribute aggregation knowledge structure, knowledge content, and ability (GEFT).
These four variables accounted for 34% of the variance in system design performance,
E (4, 94) = 11.84, p < .001, adjusted R2 = .31.

Table 9 shows the parameter coefﬁcient estimates, standard errors, and
standardized coefficients. REA template knowledge structure and knowledge content
both have signiﬁcant standardized coefﬁcients. The sign of REA template knowledge
structure is in the predicted direction. No prediction was made regarding the sign of
knowledge content -- it is a maintained hypothesis that this sign is positive. REA

template knowledge structure has a greater standardized coefﬁcient of .44 (p < .001),

83

Table 7
Descriptive Statistics and Correlation Matrix (Pairwise Deletion)

 

 

 

Correlations
Variable N M Q SDP REATKS AAKS KC GEFI'
(N) (N) (bl) (bl) (N)
SDP 100 220.38 77.00 1.00
(100)
REATKS 101 47.01 28.43 .55“ 1.00
(100).- (101)
AAKS 101 0.41 0.37 :26” .27“ 1.00
(100) (101) (101)
KC 101 20.02 1.23 .33'“ .35” .03 1.00
(100) (101) , (101) (101)
GEFT 100 13.75 3.92 .18 .26“ .03 06 1.00

(99) (100) (100) (100) (100)

 

Note: N < 101 for SDP, and GEFT due to missing data.
*p<.05,**p<.01,***p<.001

Table 8
Descriptive Statistics and Correlation Matrix (Listwise Deletion)

 

Correlations

 

Variable N M _S__Q SDP REATKS AAKS KC GEFT

 

SDP 99 219.60 77.02 1.00
REATKS 99 46.47 28.30 .55“ 1.00

AAKS 99 0.40 0.37 .27“ .27“ ' I 1.00
KC 99 20.04 1 .20 .32” .36’” .06 1 .00
GEFT 99 13.73 3.94 .18 .26“ .03 .07 1.00

 

*p<.05,**p<.01,***p<.001

Table 9
Multiple Regression for Hypotheses 3 and 4

 

 

Intercept REATKS AAKS KC GEFT
Parameter -57.58 1 20*“ 27.92 9.88" 0.88
estimate
Standard (114.99) (0.26) (18.28) (5.77) (1.71)
error '
Standardized [0] [0.44] [0.1 3] [0.1 5] [0.04]
estimate

 

* Q < _05, ** p < .01, *** p < .001 (one-tailed test)

Table 10
Sensitivity Analysis of Differently Weighted Dependent Variable
parameter estimate; (standard error); and [standardized estimate]

 

 

 

PERFORMANCE Intercept REATKS AAKS KC GEFT
a. Rank Ordered 69.35 083*“ 18.92 668* 0.52
(77.25) (0.18) (12.28) (3.88) (1.15)
[0] [0.45] [0.13] [0.15] [0.04]
b. Equally Weighted 245.48” 0.28*** 6.85 239* 0.02
(26.58) (0.06) (4.23) (1.33) (0.39)
(0] [0.451 [0.14] [0.16] [0.00]

 

 

* p < .05, ** p < .01, *** p < .001 (one-tailed test)

Table 11
Multiple Regression (with interaction term) for Hypotheses 3 and 4

 

 

Intercept REATKS AAKS KC GEFT REATKS
xAAKS

Parameter -76.64 1.43’“ 51.62 10.13" 1.31 -0.55
estimate
Standard (117.82) (0.39) (35.61) (5.79) (1.80) (0.71)
error
Standardized
estimate [0] [0.53] [0.25] [0.16] [0.07] [- 0.18]

 

* p < .05, ** p < .01, *** p < .001 (one-tailed test)

85

compared to knowledge content of .15 (p < .05). These results are consistent with
Hypothesis 3. The standardized coefﬁcient for attribute aggregation knowledge structure
was marginally signiﬁcant (p = .07). This result is weakly consistent with Hypothesis 4.
Additionally, the standardized coefﬁcient for GEFT was not signiﬁcant (p = .30).

In order to determine whether or not the results in Table 9 are sensitive to the
weighting43 of the dependent variable (System Design Performance), two additional
regression models were estimated. The “rank ordered” model in Table 10 was estimated
using a dependent variable which was determined by a rank ordering of design errors.
The original scoring method (described in Chapter Three) weighted different classes of
design errors as either - 1, - 2, - 4, - 6, or - 10. This scoring equated to a rank ordered
scoring of - 1, - 2, - 3, - 4, or - 5. The results of the regression analysis indicate no
signiﬁcant difference between the original and rank ordered scoring methods. However,
an even more stringent test was performed weighting all instances of design errors as -1.
The results of this regression (the “equally weighted” model) are shown in Table 10.
Once again, the results were not signiﬁcantly diﬁ'erent.“4

A second regression was calculated adding an interaction term (i.e., REATKS x
AAKS) to control for any nonlinearities between REA template knowledge structure and
attribute aggregation knowledge structure. These results are presented in Table 11. The

ﬁve independent variables accounted for 34% of the variance in system design

 

‘3 Refer to the discussion in Chapter Three for the speciﬁc weighting of design errors.
“ The intercept was signiﬁcantly positive, but it is of no interest in this dissertation.

86

performance, E (5, 93) = 9.55, p < .001, adjusted R2 = .30. The standardized coefﬁcient
on the interaction term was not signiﬁcant (p = .44, two-tailed test)“
4.4.1 Specification Test of Hypotheses 3 and 4

A speciﬁcation test was performed to determine whether or not any extreme
observations were inﬂuencing the results. All observations from studentized residuals
greater than 2 or - 2 were deleted. This resulted in six deleted observations“ (all were
negative residuals, and all were from the experienced group). The new regression results
are in Table 12. The results appeared to be robust. The four independent variables
accounted for 33% of the variance in system design performance, E (4, 89) = 10.81,

p < .001, adjusted R2 = .30. The REA template knowledge structure coefﬁcient remained

Table 12
Multiple Regression (Specification Test) for Hypotheses 3 and 4

 

Intercept REATKS AAKS KC GEFT

 

Parameter -57.90 126*” 25.12 9.71 0.97
Estimate

Standard (117.93) (0.28) (19.13) (5.93) (1.74)
error

Standardized [0] [0.44] [0.12] [0.15] [0.05]
estimate

 

* p < .05, ** p < .01, *** p < .001 (one-tailed test)

 

‘5 As mentioned above, there is no evidence that the obtained scores are not randomly sampled from the
population of interest. It is reasonable to assume that the individual treatment populations and the
dependent variable were drawn from normal distributions. Using the Wilk-Shapiro test on the univariate
variables of interest, the null hypothesis (L19: the sample was drawn from a normally distributed population)
could not be rejected. A review of the histograms conﬁrmed that the normality assumption is reasonable.
There is no reason to suspect that the observations are not independent. A review of scatterplots suggested
that the relationship between the independent variables and the dependent variable was linear.

“5 In essence, only ﬁve observations were deleted. The sixth observation was not included in the original
regression due to a missing GEFT score. The participant did complete the GEFT. However, the booklet
was missing some pages. Therefore no score could be calculated.

87

signiﬁcantly positive (p < .001). The only real difference is that the coefﬁcient for
knowledge content became marginally signiﬁcant (p = .053). These results are consistent
with Hypothesis 3, but not Hypothesis 4.
4.4.2 Ex post Analysis of Hypotheses 3 and. 4

I The ability construct in the accounting literature has proven difﬁcult to measure
and theory is under-developed.47 Empirically, Dunn and Grabski (1998) provide evidence
that the GEFT is positively and signiﬁcantly associated with conceptual database design
performance using undergraduate participants. The data collected in this dissertation
suggest that GEF T is not signiﬁcantly associated with conceptual database design
performance.48 This raises the question of whether or not the GEFT is a good proxy for
ability. However, direct comparisons between this dissertation and Dunn and Grabski
(1998) are difﬁcult due to task differences. Therefore, more research needs to be done to
determine whether GEF T is a reasonable proxy for ability.

As part of an ex post analysis, participants’ ACT scores49 are added to the

regression model. Table 13 presents descriptive statistics and the Pearson correlation

 

‘7 1n the accounting literature the ability construct has been measured by asking a few questions from the
GRE exam (Bonner and Lewis 1990). While this probably measures some innate ability, there has been
little attempt to theoretically specify the precise abilities requisite for a speciﬁc task.

" The correlation and regression analysis was based on all participants. Separating the data into
undergraduates and graduates revealed correlations of (r = .09, p = .51, E = 53 versus 5 = .l l, p = .45, ﬂ =
46, respectively) between GEFT and system design performance.

‘9 Grade point average and individual class grades were not used as proxies of ability since they are likely
proxies for other constructs such as motivation, knowledge, and associated interactions.

8.3.2.5338.

vme

 

 

 

2: cm. 8. 2. :8. :8. 88 :8 NR 5<

2: 8. 8.- 8. z. 8.” 8.2 me two

8.. 8.- :8. :8. 8.0 5.8 we 9.

2: .8. :8. 88 Be .2 8:2

2: .38. 8.8 8.3 Q. 8.2mm

63 8.8 8.85 we 86

5< two ox 8.5. 8.2mm age. a .2. a 838.;
2.26.280

 

Aeecgoa S_Emﬁv 5.5a: .8229..an 5.3 3:335 93.3.685—

m— 035—.

5.88.7.6 emu—8.8.85.8. .cotm Emacs». 8.9.5.8 8.0.5.8.
62.96;. m. 5...; 88.2... .8. .883 .8. 8.9.88. .88. v m ..8. v m .. .8. v a .

 

 

.88 -. .28. .88. .38. .88. .8.
.88. .88. .88. .88. .88. .88..

8. 8 .88- .8; :88 .88 :8... .888- 8
.88. .88. .88. .88. .8.
.88. .8.-8. .88. .88. 6......

8. 8 88 .88. :88 :8. 8.88- 8
.88-. .88. .88. :88. .88. .8.
.88. .38. .88. .88. .88. .88..

.8. N. m...- 88 :28 .88.. :8. 88.8. 8..
.88 -. .88. .88. .88. .88. .8.
...8. .88... .88. :88. .88. .8.»...

8. 8 88- .8.. .88. 8.8 in... 8.8- 8..
.88. .88. .88. .88. .8.
:88. .888. .88. .88. 6.88..
.88. .88. .88. .38. .8.
.E... Em. _ .88.. .88. 8......

.8. 8 88 .88 8.8 58.. 8.8- 8
.88-. .88. .88. .88. .88. .38. .8.
.88. .88. .38. ...-8. .888... .88. .88..

8. N. 8...- 88 8.. :88 :88. :8. .887 a
.88. .88. .88. .88. .88. .8.
.38. .88. .88. .88. .88. .88....

8. N. 88 8. .38. :8... :88 888- .

M 8...):

.8 z. 8.2mm .o< two 8. 8.5. 8.28. 38.6....

 

9 :5. m 8852:: .5.— =e_8~..we~_ 8353). 3.2. Rm

3 833—.

(listwise deletion) matrix. The correlations indicated four variables associated with
system design performance: REA template knowledge structure (; = .49), attribute
aggregation knowledge structure (1 = .34), knowledge content (I = .35) and ACT

(; = .32); these correlations were in the predicted direction and signiﬁcant at p < .001,

p < .01, p < .01, and p < .01, respectively. These results are consistent with Hypotheses 3
and 4. The correlation between the Group Embedded Figures Test and system design
performance is not signiﬁcant (; = .11, p = .34).

Table 14 presents the multiple regression analysis.‘2 The different equations
include/exclude the interaction term and the two ability proxies (GEFT and ACT). The
sign of REA template knowledge structure is in the predicted direction. Additionally, the
coefficient is signiﬁcant across all speciﬁcations of the regression equation. This is
consistent with Hypothesis 3. The coefﬁcient for attribute aggregation knowledge
structure is signiﬁcant in equations 1, 2, 5, and 6. However, the fact that so many
participants omitted their ACT scores prevents any strong support for Hypothesis 4.

4.4.3 Ex post Analysis of Hypotheses 3 and 4 Based on Audit Class

As a ex post test, regression analysis was performed including a dummy variable
for whether or not an audit class was taken. Since transaction cycle knowledge structures
have been exhibited by auditors, it is necessary to determine if this is an alternative
explanation for the obtained results, or if it can be ruled out as an explanation. Using a
dummy variable as a proxy for transaction cycle knowledge structure is admittedly crude,

but time constraints did not allow for any other data gathering. The following analysis is

91

based on the experienced group only (since only one inexperienced participant had taken
auditing). Since the experienced group had both accounting and non-accounting students
there was adequate variation in auditing as a dummy variable."

Table 15 presents descriptive statistics and the Pearson correlation (listwise
deletion) matrix. The correlations indicated one variable associated with system design
performance: REA template knowledge structure (r = .37) this correlation was in the
predicted direction and signiﬁcant at p < .05. This result is consistent with Hypothesis 3.

Table 16 presents the multiple regression analysis. The different equations
include/exclude the two ability proxies (GEF T and ACT) and the AUDIT dummy
variable-52 The sign of REA template knowledge structure is in the predicted direction.
Additionally, the coefﬁcient is signiﬁcant across all speciﬁcations of the regression

Table 15
Descriptive Statistics and Correlation Matrix for Hypotheses 3 and 4

 

 

 

Correlations
Van'able u M _S_D SDP REATKS AAKS KC GEF ACT
' T
SDP 31 278.30 22.39 1.00
REATK 31 61.65 27.61 .37' 1.00
S
AAKS 31 0.50 0.35 .17 -.16 1.00
KC 31 20.32 0.75 .18 .42‘ .28 1.00
GEFT 31 14.13 3.88 .17 .06 .08 .08 1.00
ACT 31 26.55 2.72 .35 .07 .01 -.07 .08 1.00
AUDIT 31 0.61 0.50 -.15 -.26 -.05 -.10 -.16 -.46”

 

'p<.05,“p<.01,“'p<.001

 

’° Models 3b and 4b in Table I4 are based on a reduced sample (i.e., the observation was dropped if there
was no ACT or GEFT score) from that used in models 38 and 4a This is an attempt to make analysis of
models I through 6 more comparable

5' 22 experienced participants had taken an auditing class, 24 had not, and one did not respond.

’1 The interaction term is omitted from this analysis since it did not reveal any signiﬁcant relationships in
the prior regressions.

92

Ex post Multiple Regression for Hypotheses 3 and 4 (with Audit)

Table 16

 

 

Intercept REATKS AAKS KC GEFT ACT AUDIT 3 pg],
8.?

1 170.92 0.34: 15.01 -0.87 .73 3.09: 6.74 31 .15
(130.56) (0.16) ' (11.88) (6.08) (0.99) (1.58) (8.96)
[0] [0.42] [0.24] [-0.03] [0.13] [0.38] [0.15]

2 180.01 .45" 0.43 3.13 -0.04 7.59 45 .13
(115.51) (0.16) (12.94) (5.84) (1.12) (8.38)
[0] [0.45) [0.00] [0.08] [-0.01] [0.13]

3 177.12 0.34: 15.53 -0.71* 312* 5.97 32 .18
(124.50) (0.16) (11.51) (5.87) (1.50) (8.36)
[0] [0.43) [0.24] [-0.02] [0.38] [0.13]

4 193.50 031* 14.37 -0.92 0.63 2.56' 31 .16
(125.94) (0.16) (11.75) (6.03) (0.97) (1.40)
[0] [0.39] [0.23] [-0.03] [0.11] [0.31]

5 190.06 .41** 0.41 2.91 -0.04 46 .13
(113.75) (0.15) (12.79) (5.77) (1.10)
[0) [0.42] [0.00) [0.08] [001)

6 193.81 0.32: 14.77 -0.66 2.66“ 32 .20
(121.18) (0.15) (11.36) (5.82) (1.35)
(0] [0.40) (0.23] 1.0.02] [032]

7 179.80 0.45" 0.33 3.12 7.58 46 16
(112.04) (0.15) (12.22) (5.69) (8.01)
101 [0.451 [0.00] [0.08] [0.13)

 

‘ p < .05, “ p < .01, m p < .001 (one-tailed test, except for intercept which is two- tailed)
parameter estimate, (standard error), [standardized estimate]

93

equation. Taken as a whole, the correlation and regression analysis weakly rules out the
competing explanation that a knowledge structure, acquired from the experience of taking
an audit class, predicts system design performance.

4.5 Summary of Chapter Four .

This chapter described the statistical techniques used to test the hypotheses
developed in Chapter Two of this dissertation. Hypothesis I predicted that experience
with REA conceptual modeling would facilitate the organization of REA template
knowledge structuring in long-term memory. This hypothesis was supported.
Hypothesis 2 predicted that experience with REA conceptual modeling would lead to a
better developed schema (i.e., a knowledge structure) for attribute aggregation. This
hypothesis was supported. Hypothesis 3 predicted that REA knowledge structure would
be correlated with successful conceptual modeling performance, while controlling for
knowledge content and ability. This hypothesis was supported. Hypothesis 4 predicted
that attribute aggregation knowledge structure would be correlated with successful
conceptual modeling performance. This hypothesis was weakly supported, although ex
post tests revealed stronger support particularly when ACT score was included in the
regression.

Chapter Five will conclude the dissertation with a summary, and a discussion of

contributions, limitations, and future research directions.

94

Chapter Five
CONCLUSION

This chapter concludes the dissertation with a summary of research questions,
theory, and hypotheses in Section 5.1, a summary of the research method and results in
Section 5.2, a discussion of contributions in Section 5.3, an evaluation of limitations in
Section 5.4, and consideration of future research directions in Section 5.5.

5.1 Summary of Research Questions, Theory, and Hypotheses

The objective of this dissertation was to answer the following research questions:
(1) does experience with conceptual modeling of REA systems result in the organization
of knowledge on an REA basis in long-term memory; (2) does this experience help
designers aggregate attributes to entities; and (3) how do these two types of knowledge
(REA and attribute aggregation knowledge) affect REA design performance while
controlling for knowledge content and ability?

These research questions are important because information system design is a
complex and costly activity, and the implementation of an incorrect design will have dire
consequences. Since conceptual design is the ﬁrst phase of database design, and each
phase builds upon it in succession, it can be argued that it is the most important phase.
Siau et al. (1997, 155) indicates “information modeling is the most difﬁcult, yet most
important phase, in systems development.” Any undetected errors made in this phase
will be implemented in the ﬁnal DBMS and costly to undo. Boehm (1976) suggests that
post-implementation changes cost 75 times more (on average) than changes made during

the analysis and conceptual design stage.

95

To aid in the development of theory and understand the relationships between
constructs, Libby and Luft’s (1993) model of Antecedents and Consequences of
Knowledge was employed. In this model experience and ability are antecedents of
knowledge and performance is a consequence of knowledge and ability. For the purposes
of this dissertation it was necessary to further deﬁne the knowledge construct as having a
content (information stored in memory) component and a structure (a speciﬁc
organization of the knowledge content) components. Since the research questions relate
speciﬁcally to the structure of knowledge, the speciﬁcation of these variables allowed for
a measure of whether or not the participants had sufﬁcient knowledge. This content
knowledge could then be partitioned out allowing for inferences regarding knowledge
structure.

Prior accounting literature has provided insights into the relationships between
these variables.” Studies have shown that knowledge structures are acquired with
experience (e.g., Choo and Trotman 1991), that different knowledge structures exist (e.g.,
Frederick 1991), and that knowledge structure and task structure can interact to affect
task performance (e.g., Nelson et al. 1995). These ﬁndings document different
determinants of task performance and have implications for task assignment and training.

Prior to this dissertation, no attempts had been made to empirically test the
relationships between experience, knowledge, ability, and performance in the discipline
of conceptual database design. Since the REA template has a general form which can

map to (be instantiated in) transaction cycles, it was proposed that the template could

 

’3 The majority of studies have related to audit settings.

96

 

assist the organization (structuring) of REA knowledge in long-term memory. However,
this process does not happen immediately - it happens with experience. Typically, when
individuals learn the REA framework they have already been exposed to an alternative
framework: the journal entry framework underlying the double entry accounting system.
Knowledge acquisition under this framework tends to result in journal entry knowledge
structuring (such as that proposed in Frederick and Libby 1986).

In order to measure REA template knowledge structure, participants were asked to
recall an REA diagram; participants also recalled a random diagram as a control. Based
on the above theory, Hypothesis I predicted that, “There will be an ordinal interaction
between experience and information format organization on knowledge structure.”

Successful performance in the task of REA conceptual modeling not only
involves deriving the structure of the diagram (i.e., an REA template or templates),
additionally, it includes aggregating the attributes to the correct entities. It was proposed
that experienced REA designers would have more well-developed schemas for attribute
aggregation. The journal entry knowledge structure that the inexperienced participants
possess is not robust (dealing mainly with accounts, amounts, and possibly dates) in
terms of accounting features. Experience with REA conceptual modeling facilitates
schema development for important information features (i.e., attributes). Therefore,
Hypothesis 2 predicted that, “REA-related entity clustering increases with REA
experience.”

Since successful design performance depends on the structuring of entities,
relationships, cardinalities, and attributes, the structuring of related knowledge should be

positively related to conceptual database design performance. Therefore, Hypotheses 3

97

(and 4) stated that, “Performance on the conceptual modeling task is positively correlated
with REA template (attribute aggregation) knowledge structure.”
5.2 Summary of Research Method and Results

The research method involved three tasks. Tasks 1 and 2 employed the free recall
method to infer knowledge structure. Task 3 involved the conceptual design of an
accounting database system. The inexperienced participants were enrolled in an
undergraduate information systems class; the experienced participants were enrolled in a
graduate information systems class. The experienced participants had all completed a
prior class related to conceptual database design.

The research design for task 1 was a 2 x 2 design with one between-subjects factor
(experience) and one within-subjects factor (diagram type). Participants were given a
diagram to study for three minutes, and then were asked to reconstruct the diagram from
memory. Participants recalled an REA diagram and a random diagram; the diagrams
were counterbalanced and separated by task 2. The obtained results were consistent with
Hypothesis 1. When recalling the REA diagram, the experienced participants recalled
signiﬁcantly more information than the inexperienced participants. When recalling the
random diagram, the amount of information recalled was not signiﬁcantly different
between experienced and inexperienced participants.

The performance on the random diagram suggested that experienced and
inexperienced participants had approximately the same memorization ability. However,
the performance on the REA diagram suggested that some other variable was affecting
recall: knowledge structure. It appears as though the experienced participants were able

to access their long-term memories and use their REA template knowledge structures to

98

help facilitate chunking. Therefore, the ability to chunk according to the REA template
allowed for more information to be stored and later recalled.

The research design for task 2 was a 2 x 1 paired-groups design. Participants
were given a randomized list of attributes to study for four minutes, and then were asked
to reconstruct the list ﬁ‘om memory. As mentioned above, this task was performed
between the two diagram recalls in task 1. An ARC index was calculated to measure the
degree of clustering evident in the attributes recalled. The obtained results were
consistent with Hypothesis 2. Experienced participants exhibited signiﬁcantly greater
levels of clustering (related to entities) than inexperienced participants. However, both
groups had positive average ARC indexes indicating that it is natural to cluster attributes
according to the entities to which they aggregate. Experience with conceptual modeling
appears to enhance the schema development.

The research design for task 3 required the experienced and inexperienced
participants to complete a conceptual database design of an REA system. This task was a
realistic abstraction of what a professional designer could be asked to complete. The
performance on this task was then regressed on the knowledge structure variables
measured in tasks 1 and 2 while controlling for ability (as measured by the Group
Embedded Figures Test) and knowledge content. The obtained results were consistent
with Hypothesis 3; however, Hypothesis 4 was only weakly supported.

REA template knowledge structure is signiﬁcantly associated was conceptual
database design performance while holding attribute aggregation knowledge structure,
knowledge content, and ability constant. This is a distinguishing result -- the way that

knowledge is structured in memory is a determinant of design performance.

99

More research and theory development is needed to understand why Hypothesis 4
was not supported. In supplemental analyses, another ability proxy (ACT score) was
added to the regression equation. This reduced the sample size by 27 participants, and
the coefﬁcient for attribute aggregation knowledge structure became statistically
signiﬁcant. This result held (with the reduced sample) whether or not ACT score was
included in the regression. It is unclear why the data was missing on the ACT scores.
Some participants probably took the SAT test, and some decided not to report for some
reason. Therefore, further research needs to be done to further evaluate the role of
attribute aggregation knowledge structure, and determine better measures of ability.

5.3 Contributions

This dissertation is the ﬁrst study to account for the effects of experience and
knowledge structure on the task of REA conceptual database design. Furthermore, it
suggests that not only is REA useful as a normative theory on enterprise system design, it
also appears to serve as a useful memory structuring device which is important in actually
performing a system design.

On a larger scale, this dissertation also has implications in the area of human
judgment and decision making performance. The theoretical development and related
results indicate that a domain speciﬁc pattern can help humans organize knowledge,
overcome bounded rationality, and manage complex tasks.

5.4 Limitations

The choice of a within-subj ects design for the diagram free recall and the

relatively large sample size provided an adequate level of power. Furthermore, since

three of the four hypotheses were supported, it is unlikely that power is a problem. There

100

 

were no major violations of statistical assumptions noted. The administration of each
experimental session followed a script. Since Hypothesis 4 was only weakly supported,
more research should be done to see if the error term can be attenuated.

There are three important issues related to the external validity of this dissertation.
The ﬁrst concerns the representativeness of the participants, the second concerns the
motivation level of the participants, and the third concerns the instruments used in the
experiment.

Caution must always be exercised when using student participants. In this
dissertation, students are not serving as surrogates for practicing database designers.
Rather, they are representative of database designers in their early stages of knowledge
acquisition. It remains an empirical question whether or not the students in AIS classes at
Michigan State University are representative of the population of students in AIS classes
or the population of practicing database designers at the beginning of their careers.

The use of student participants also raises the question of motivation. The
incentive (class credit) used in the study may not have induced a concerted effort.
However, there is no evidence that motivation levels were not evenly distributed across
groups. In fact, with the diagram recall task, if motivation was driving the results, the
ordinal interaction would not have been obtained.

The instruments employed for tasks 1, 2, and 3, related to only the revenue and
acquisition cycles. Whether or not the results hold across other accounting cycles
remains an empirical question. As mentioned above, the conceptual modeling task was a

realistic abstraction.

l0]

5.5 Future Research Directions

There are several possible future research directions extending from this
dissertation. These directions will be discussed using Libby and Luft’s (1993)
Antecedents and Consequences of Knowledge model as a reference point." An obvious
example is examining the effects of actual work experience on knowledge structure.
Libby (1995, 179-180) suggests that there are two types of task-related encounters
(experiences): ﬁrst-hand and second-hand encounters. First-hand encounters involve
completing actual tasks, reviewing/supervising other people’s task completion, and
receiving feedback. Second-hand encounters involve discussing tasks with colleagues,
reading task procedure guides, and other forms of training and education. This
dissertation tested a form of experience which is a cross between ﬁrst- and second-hand
encounters (i.e., the experience obtained by the participants was in an educational setting.
but did involve the completion of the actual task of conceptual modeling). Future
research could examine the effects of these different types of experience.

This dissertation focused on the early stages of experience and knowledge
acquisition. Since this is an initial study, the functional form (e.g., increasing,
decreasing, concave up or down) of the relationship between knowledge structure and
experience is unclear, especially over greater levels of experience. Other types of
knowledge, such as knowledge of other accounting cycles should be explored. The

measurement of knowledge structure could, perhaps, be improved. It would be

 

5‘ The interested reader will ﬁnd many illuminating research suggestions in Libby and Luft ( 1993) and
Libby (1995). These articles are essential resources for researchers interested in accounting knowledge
research.

102

worthwhile to examine different features of chunking behavior such as how much
information is actually being encoded.

Additionally, it is possible that other knowledge structures exist. If so, the
alternative structures could be compared to the ones in this dissertation to see if they
improve performance or interact with different tasks. Task complexity was not varied in
this dissertation. Future studies could take task complexity into account to see if
knowledge structure interacts with it.

Analogous to the audit literature, the role of ability has to be better speciﬁed and
measured, and motivational and environmental considerations should be tested.
Information system design professionals work in a dynamic environment and the
resulting inﬂuence has been largely ignored in the literature.

The opportunities for future research in this area‘are abundant. This dissertation
represents the ﬁrst step toward developing a detailed understanding of the determinants of

successful REA conceptual modeling performance.

103

LIST OF REFERENCES

104

LIST OF REFERENCES

Adelson, B. 1981. Problem solving and the development of abstract categories in
programming languages. Memory & Cognition 9 (4): 422-433.

Ashcraft, M.H. 1994. Human Memory and Cognition, 2nd ed. New York: HarperCollins
College Publishers.

Banker, R.D., S.M. Datar, C.F. Kemerer, and D. Zweig. 1993. Software complexity and
software maintenance costs. Communications of the ACM 36 (11): 81-94.

Batini, C., S. Ceri, and SB. Navathe. 1992. Conceptual Database Design: An Entity
Relationship Approach. The Benjamin/Cummings Publishing Company, Inc.,
Redwood City, California.

Batra, D. and SR. Antony. 1994a. Effects of data model and task characteristics on

designer performance: A laboratory study. International Journal of Human-
Computer Studies 41: 481-508.

Batra, D. and SR. Antony. 1994b. Novice errors in conceptual database design.
European Journal of Information Systems 3: 57-69.

Batra, D. and J .G. Davis. 1989. A study of conceptual data modeling in database design:
Similarities and differences between expert and novice designers. Proceedings of
the Tenth International Conference on Information Systems, Boston, MA: 91 -99.

Batra, D-, J .A. Hoffer, and RP. Bostrom. 1990. Comparing representations with
relational and EER models. Communications of the ACM 33: 126-139.

Bimberg, J. and M. Shields. 1984. The role of attention and memory in accounting
decisions. Accounting, Organizations and Society 9 (June): 365-3 82.

Bodart, F. and R. Weber- 1997. Optional properties versus subtyping in conceptual
modeling: A theory and empirical test. Working Paper, The University of

Queensland.

Boehm, B. 1976. Soﬁware engineering. IEEE Transactions on Computers (December):
1 226-1 241 -

105

Bonner, S.E., J .S. Davis, and B.R. Jackson. 1992. Expertise in corporate tax planning:
The issue identiﬁcation stage. Journal of Accounting Research 30 (Supplement):
1-28.

Bonner, SE. and BL Lewis. 1990. Determinants of auditor expertise. Journal of
Accounting Research 28 (Supplement): -l-20.

Bonner, S.E., R. Libby, and M.W. Nelson. 1996. Using decision aids to improve
auditors’ conditional probability judgments. The Accounting Review 71: 221-240.

Bousﬁeld, WA. 1953. The occurrence of clustering in the recall of randomly arranged
associates. The Journal of General Psychology 49: 229-240.

Brosey, M. and B. Shneiderrnan. 1978. Two experimental comparisons of relational and
hierarchical database models. International Journal of Man-Machine Studies 10:
625-637.

Butt, J. 1988. Frequency judgments in an audit-related task. Journal Of Accounting
Research 26: 315-330. '

Chase, W.G. and HA. Simon. 1973. Perception in chess. Cognitive Psychology 5: 55-81.

Chen, RP. 1976. The entity-relationship model - toward a uniﬁed view of data. ACM
Transactions on Database Systems 1 (March): 9-36.

Chi, M.T.H., P.J. Feltovich, and R. Glaser. 1981. Categorization and representation of
physics problems by experts and novices. Cognitive Science 5: 121-152.

Choo, F. and KT. Trotrnan. 1991. The relationship between knowledge structures and

judgments for experienced and inexperienced auditors. The Accounting Review
66: 464-485.

Cloyd, CB. 1994. Performance in tax research tasks: The joint effects of knowledge and
accountability. The Accounting Review 72: 1 1 1-131.

Codd, E.F. 1972. Further normalization of the database relational model. In R. Rusin
(Ed.), Database Systems. New York: Prentice-Hall.

Cook, TD. and D.T. Campbell. 1979. Quasi-Experimentation: Design & Analysis Issues
for Field Settings. Boston, MA: Houghton Mifflin Company.

Davis, GB. and HM. Olson. 1985. Management Information Systems, Conceptual

Foundations, Structure, and Development, 2nd ed. New York: McGraw-Hill
Book Company.

106

Davis, J. and 1. Solomon. 1989. Experience, expertise, and expert-performance research
in public accounting. Journal of Accounting Literature 8: 150-164.

de Groot, AD. 1965. Thought and Choice in Chess. The Hague: Mouton.

de Groot, AD. 1966. Perception and memory versus thought. In B. Kleinmuntz (Ed.),
Problem Solving. New York: John Wiley.

Dearman, D. and MD. Shields. 1997. Managers’ cost experience, knowledge and
judgment. Working paper.

Denna, E.L., J .O. Cherrington, D.P. Andros, and AS. Hollander. 1993. Event-Driven
Business Solutions: Today 's Revolution in Business and Information Technology.
New York: Irwin.

Dunn, C. 1994. An abstraction hierarchy as a database interface: Does it control
complexity? Unpublished Doctoral Dissertation, Michigan State University.

Dunn, C. and S.V. Grabski. 1997. “Cognitive ﬁt and a comparison of two accounting
models.” Paper presented at the 1997 American Accounting Association Annual
Meeting, Dallas, TX. .

Dunn, C. and S.V. Grabski. 1998. The effect of ﬁeld dependence on conceptual modeling
performance. Paper accepted to F iﬂh AIS Research Symposium, February 1998.

Dunn, CL. and WE. McCarthy. 1997. The REA accounting model: Intellectual
heritage and prospects for progress. Journal of Information Systems 11: 31-51 .

Egan, DE. and 8.1. Schwartz. 1979. Chunking in recall of symbolic drawings. Memory
& Cognition 7 (2): 149-158.

Frederick, D.M. 1991. Auditors’ representation and retrieval of internal control
knowledge. The Accounting Review 66: 240-258.

Frederick, D.M., V. Heiman-Hoffman, and R. Libby. 1994. The structure of auditors’
knowledge of ﬁnancial statement errors. Auditing: A Journal of Practice &
Theory 13: 1-21.

Frederick, D.M. and R Libby. 1986. Expertise and auditors’ judgments of conjunctive
events. Journal of Accounting Research 24: 270-290.

Geerts, G.L. and WE. McCarthy. 1996. Augmented Intensional Reasoning in

Knowledge-Based Accounting Systems. Working Paper, Michigan State
University.

107

Geerts, G.L. and WE. McCarthy. 1997. Modeling business enterprises as value-added
process hierarchies with Resource-Event-Agent Object Templates. In J.
Sutherland and D. Patel (eds.) Business Object Design and Implementation,
Springer-Verlag.

Geerts, G.L. and WE. McCarthy. 1994. The economic and strategic structure of REA
accounting systems. Paper presented to the 300th Anniversary Program, Martin
Luther University, Halle-Wittenberg, Germany (September).

Geerts, G.L. and WE. McCarthy. 1998. Use of an Accounting Object Infrastructure for
Knowledge-Based Enterprise Models. Working Paper, Michigan State University.

Hollander, A.S., E.L. Denna, and J .O. Cherrington. 1996. Accounting, Information
Technology, and Business Solutions. Chicago: Irwin.

J arvenpaa, S.L. and J .J . Machesky. 1989. Data analysis and learning: an experimental
study of data modeling tools. International Journal of Man-Machine Studies 31:
367-391.

Johnson, H.T. and RS. Kaplan. 1987. Relevance Last: The Rise and Fall of Management
Accounting. Boston: Harvard Business School Press.

J uhn, S. and J .D. Naumann. 1985. The effectiveness of data representation characteristics
on user validation. Proceedings of the Sixth International Conference on
Information Systems, Indianapolis, IN, USA.

Kendall, K. E. and J. E. Kendall. 1995. Systems Analysis and Design, 3rd ed. Englewood
Cliffs, NJ. Prentice-Hall, Inc.

Libby, R. 1995. The role of knowledge and memory in audit judgment. In R. H. Ashton
and A. H. Ashton (Eds. ) Judgment and Decision-Making Research in Accounting
and Auditing. New York: Cambridge University Press.

Libby, R. and D.M. Frederick. 1990. Experience and the ability to explain audit ﬁndings.
Journal of A ccounting Research 28: 348-367.

Libby, R. and J. Luﬁ. 1993. Determinants of judgment performance in accounting
settings: Ability, knowledge, motivation, and environment. Accounting,

Organizations and Society 18: 425-450.

Libby, R., K. Trotrnan, and I. Zimmer. 1987. Member variation, recognition of expertise,
and group performance. Journal of Applied Psychology 72: 81-87.

108

Maines, LA. 1990. The effect of forecast redundancy on judgments of a consensus
forecast’ 5 expected accuracy. Journal of A ccounting Research 28 (Supplement):
29-47.

Malhotra, A., J.C. Thomas, J .M. Carroll, and LA. Miller. 1980. Cognitive processes in
design. International Journal of Man-Machine Studies 12: 119-140.

Mandler, G. 1972. Organization and memory. In KW. Spence and J .T. Spence (Eds.) The
psychology of learning and motivation. New York: Academic Press.

McCarthy, W.E. 1982. The REA accounting model: A generalized framework for
accounting systems in a shared data environment. The Accounting Review (July):
554-578.

McCarthy, W.E. and SR. Rockwell. 1996. REACH: Automated Database Design
Integrating F irst-Order Theories, Reconstructive Expertise, and Implementation
Heuristics for Accounting Information Systems. Working Paper, Michigan State
University.

McKeithen, K.B., J .S. Reitman, H.H- Rueter, and SC. Hirtle. 1981. Knowledge
organization and skill differences in computer programmers. Cognitive
Psychology 13: 307-325.

Miller, GA. 1956. The magical number seven, plus or minus two: Some limits on our
capacity to process information. Psychological Review 63: 81-97.

Moser, D. 1989. The effects of output interference, availability, and accounting
information on investors’ predictive judgments. The Accounting Review 64: 433-
448.

Nelson, M.W., R. Libby, and SE. Bonner. 1995. Knowledge structure and the estimation
of conditional probabilities in audit planning. The Accounting Review (January):
27-47.

Nij ssen, GM. and TA. Halpin. 1989. Conceptual Schema and Relational Database
Design: A Fact Oriented Approach. New York: Prentice-Hall.

Oltrnan, P.K., E. Raskin, and H.A. Witkin. 1971. Group Embedded Figures Test. Palo
Alto, CA.: Psychologists Press, Inc.

Ricchiute, D. 1992. Working-paper order effects and auditors’ going concern decisions.
The Accounting Review 67: 46-58.

Roenker, D.L., C.P. Thompson, and SC. Brown. 1971. Comparison of measures for the
estimation of clustering in free recall. Psychological Bulletin (July): 45-48.

109

Weber, R. 1996. Are attributes entities?: A study of database designers’ memory
structures. Information Systems Research 7: 137-162.

lll

   

"IIIIIIIIIIIIIIIIIES