USF Libraries
USF Digital Collections

The effect of multidimensional information presentation on the effectiveness and efficiency of a spatial accounting judgment

MISSING IMAGE

Material Information

Title:
The effect of multidimensional information presentation on the effectiveness and efficiency of a spatial accounting judgment
Physical Description:
Book
Language:
English
Creator:
Chan, Chong Ho
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Three-dimensional
Trend analysis
Pattern recognition
Cognitive fit
Dissertations, Academic -- Accounting -- Doctoral -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: This study is the first in a series of planned studies on the application of multidimensional visualization of business information and data within the context of accounting. The study's research question is: When is multidimensional visualization of information a better problem representation, improving both the effectiveness and efficiency of a spatial accounting judgment? To examine when multidimensional visualization can assist auditors in configural cue pattern recognition, the study employs the traditional DuPont analysis as the three pieces of key information to be represented on the X, Y, and Z axes of a single 3-D perspective display. To help determine when use of 3-D perspective display is beneficial in combining pieces of information, I rely on Vessey's (1991) Cognitive Fit Theory, and the Proximity Compatibility Principle (PCP) proposed by Wickens and Carswell (1995). The study has two hypotheses.Hypothesis H1 predicted that participants viewing a set of 2-D displays will be the most effective or most efficient in generating hypotheses for what caused the changes in the trend of accounting data or in estimating values. Hypothesis H2 predicted that participants viewing a single 3-D perspective display will be the most effective or most efficient in recognizing patterns of accounting data or in generating hypotheses for what caused the emerged pattern. To test the hypotheses of the study a 3 x 2 between-subjects design (display format x task) is used. The independent variables are display types and task types. Graphical display was manipulated at three levels: no graphical display (table only), 2-D display, and 3-D perspective display. Task was manipulated at two levels: trend analysis and pattern recognition task.The need for a fit between different types of spatial tasks and display formats is demonstrated by the findings of this study: 1) that 2-D displays appear to be more suitable for spatial tasks involving the generation of hypotheses for causes of trends in accounting data, while 2) 3-D perspective displays appear to be more suitable for spatial tasks involving pattern recognition in accounting data.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2008.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Chong Ho Chan.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 289 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002019013
oclc - 427373955
usfldc doi - E14-SFE0002488
usfldc handle - e14.2488
System ID:
SFS0026805:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 002019013
003 fts
005 20090722091306.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 090722s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002488
035
(OCoLC)427373955
040
FHM
c FHM
049
FHMM
090
HF71 (Online)
1 100
Chan, Chong Ho.
4 245
The effect of multidimensional information presentation on the effectiveness and efficiency of a spatial accounting judgment
h [electronic resource] /
by Chong Ho Chan.
260
[Tampa, Fla] :
b University of South Florida,
2008.
500
Title from PDF of title page.
Document formatted into pages; contains 289 pages.
Includes vita.
502
Dissertation (Ph.D.)--University of South Florida, 2008.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
520
ABSTRACT: This study is the first in a series of planned studies on the application of multidimensional visualization of business information and data within the context of accounting. The study's research question is: When is multidimensional visualization of information a better problem representation, improving both the effectiveness and efficiency of a spatial accounting judgment? To examine when multidimensional visualization can assist auditors in configural cue pattern recognition, the study employs the traditional DuPont analysis as the three pieces of key information to be represented on the X, Y, and Z axes of a single 3-D perspective display. To help determine when use of 3-D perspective display is beneficial in combining pieces of information, I rely on Vessey's (1991) Cognitive Fit Theory, and the Proximity Compatibility Principle (PCP) proposed by Wickens and Carswell (1995). The study has two hypotheses.Hypothesis H1 predicted that participants viewing a set of 2-D displays will be the most effective or most efficient in generating hypotheses for what caused the changes in the trend of accounting data or in estimating values. Hypothesis H2 predicted that participants viewing a single 3-D perspective display will be the most effective or most efficient in recognizing patterns of accounting data or in generating hypotheses for what caused the emerged pattern. To test the hypotheses of the study a 3 x 2 between-subjects design (display format x task) is used. The independent variables are display types and task types. Graphical display was manipulated at three levels: no graphical display (table only), 2-D display, and 3-D perspective display. Task was manipulated at two levels: trend analysis and pattern recognition task.The need for a fit between different types of spatial tasks and display formats is demonstrated by the findings of this study: 1) that 2-D displays appear to be more suitable for spatial tasks involving the generation of hypotheses for causes of trends in accounting data, while 2) 3-D perspective displays appear to be more suitable for spatial tasks involving pattern recognition in accounting data.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Co-advisor: Uday Murthy, Ph.D.
Co-advisor: Jacqueline Reck, Ph.D.
653
Three-dimensional
Trend analysis
Pattern recognition
Cognitive fit.
0 690
Dissertations, Academic
z USF
x Accounting
Doctoral.
773
t USF Electronic Theses and Dissertations.
856
u http://digital.lib.usf.edu/?e14.2488



PAGE 1

The Effect of Multidimensional Information Presentation on the Effectiveness and Efficiency of a Spatial Accounting Judgment John K. Tan A dissertation submitted in partial fulfillment of the requirements for the degree of Doctcir of Philosophy School of Accountancy College of Business Administration University of South Florida Co-Major Professor: Uday Murthy, Ph.D. Co-Major Professor: Jacqueline Reck, Ph.D. Stephanie Bryant, Ph.D. Brad Schafer, Ph.D. Rosann Collins, Ph.D. Date of Approval: July 10,2008 Keywords: three-dimensional, trend analysis, pattern recognition, cognitive fit. by O Copyright 2008, John K. Tan

PAGE 2

Dedication I would like to dedicate this dissertation to my wonderful parents for their unconditional love, support, and encouragement.

PAGE 3

Acknowledgments First and foremost, I would like to thank the co-chairs of my dissertation committee, Dr. Uday Murthy and Dr. Jacqueline Reck, for their guidance, feedback, and support throughout the process. I would like to thank the other members of my dissertation committee, Dr. Stephanie Bryant, Dr. Brad Schafer, and Dr. Rosann Collins for their support and comments. I would also like to express my appreciation for the valuable learning environment, especially the Friday workshops, at the University of South Florida. I acknowledge the support from all the faculty and staff of the College of Business throughout my doctoral program.

PAGE 4

i Table of Contents List of Tables vii List of Figures x Abstract xi Chapter 1: Introduction 1 1.1 Introduction and Significance of the Issue 1 1.2 The Need for Research on the Issue of Visualization 3 1.3 Research Objective and Question 4 1.4 Results and Contribution 5 Chapter 2: Background, Theory, and Hypotheses Development 8 2.1 What is the Problem? 8 2.1.1 Configural Information Processing 8 2.1.2 Analytical Procedures in Auditing 10 2.2 Visualization 14 2.2.1 Accounting Research on Visualization 15 2.2.2 Multidimensional Accounting Information Research 16 2.2.3 Research in Computer Science and Human Factors 17 2.2.4 Conceptual Limitation on Visualization Using 2-D Graphs with Three Components 18 2.2.5 Informationally Equivalent Presentations 19

PAGE 5

ii 2.3 Theory and Hypotheses 27 2.3.1 The Need for a F it between Presentation Format and Task 27 2.3.2 Trend Analysis and Hypotheses 30 2.3.3 Pattern Analysis and Hypotheses 34 Chapter 3: Research Design 39 3.1 Research Model 39 3.2 The Independent Variables 42 3.3 The Dependent Variables 42 3.4 Experimental Setup and Procedures 47 3.5 Training 48 3.5.1 Training for the Trend Analysis Task 49 3.5.2 Training for the Pattern Recognition Task 52 3.6 Measuring the Dependent Variables 55 3.6.1 Measuring the Dependent Variables f or the Trend Analysis Task 55 3.6.2 Measuring the Dependent Variables f or the Pattern Recognition Task 60 3.7 Covariates 65 3.7.1 Practice Questions 65 3.7.2 Mental Rotations Test 65 3.7.3 Gender and Age 67 3.7.4 Mental Workload 67 3.7.5 Demographics Data 69

PAGE 6

iii 3.8 Post Hoc Analysis (Survey Questions) 70 3.9 Student Participants 70 3.10 Manipulation Check 73 Chapter 4: Pilot Study 82 4.1 Research Design of the Pilot Study 82 4.2 Results of the Pilot Study 83 4.2.1 Descriptive Statistics, Test of Assumptions and Outliers 83 4.2.2 Hypotheses Testing of the Trend Analysis Task 94 4.2.2.1 Results of H1a 94 4.2.2.2 Results of H1b 103 4.2.2.3 Results of H1c 110 4.2.2.4 Results of H1d 111 4.2.3 Hypotheses Testing of the Pattern Recognition Task 113 4.2.3.1 Results of H2a 113 4.2.3.2 Results of H2b 118 4.2.3.3 Results of H2c 119 4.2.3.4 Results of H2d 124 4.3 Lessons Learned from the Pilot Results and How to Improve 129 4.3.1 Rank Order Effects 129 4.3.2 Insufficient Training on Display Formats 129 4.3.3 Insufficient Training on ROE 130 4.3.4 Scrolling Up and Down the Screen 130 Chapter 5: Main Experiment 133

PAGE 7

iv 5.1 Sample Size, Inter-Coder Reliability 133 5.1.1 Sample Size 133 5.1.2 Inter-Coder Reliability 133 5.2 Testing of Raw Data 140 5.2.1 Testing of Outliers and Influential Observation 140 5.2.2 Testing of Assumptions 141 5.2.3 Plan of Statistical Analysis 146 5.2.4 146 5.3 Manipulation Check 147 5.3.1 Results of the Manipulation Check Questions 148 5.4 Results of the Trend Analysis Task 152 5.4.1 Descriptive Statistics 152 5.4.2 -Moment Correlation 161 5.4.2.1 Correlation between Dependent Variables 161 5.4.2.2 Multicollinearity 163 5.4.2.3 Correlation between Treatment or Covariate and Dependent Variables 163 5.4.3 Regression Analysis of Possible Covariates 173 5.4.4 Results of H1a 175 5.4.5 Results of H1b 183 5.4.6 Results of H1c 191 5.4.7 Results of H1d 198 5.5 Results of Pattern Recognition Task 201

PAGE 8

v 5.5.1 Descriptive Statistics 201 5.5.2 -Moment Correlation 209 5.5.2.1 Correlation between Dependent Variables 209 5.5.2.2 Multicollinearity 210 5.5.2.3 Correlation between Treatment or Covariate and Dependent Variables 210 5.5.3 Regression Analysis of Possible Covariates 218 5.5.4 Results of H2a 220 5.5.5 Results of H2b 226 5.5.6 Results of H2c 232 5.5.7 Results of H2d 237 5.6 Post Hoc Analysis 246 5.6.1 Trend Analysis Post Hoc Analysis 247 5.6.2 Pattern Recognition Post Hoc Analysis 251 Chapter 6: Discussion 255 6.1 Summary of Hypothesized Results 255 6.2 Discussion of Results of Hypothesis H1 264 6.2.1 Discussion of Results of Hypothesis H1a 264 6.2.2 Discussion of Results of Hypothesis H1b 268 6.2.3 Discussion of Results of Hypothesis H 1c 269 6.2.4 Discussion of Results of Hypothesis H1d 270 6.3 Discussion of Results of Hypothesized H2 271 6.3.1 Discussion of Results of Hypothesis H2a 271

PAGE 9

vi 6.3.2 Discussion of Results of Hypothesis H2b 272 6.3.3 Discussion of Results of Hypothesis H2c 273 6.3.4 Discussion of Results of Hypothesis H2d 274 6.4 Comments on the Implication of Significant Covariates 275 6.5 Discussion on the Responses of the Survey Questions 276 6.6 Contribution 277 6.7 Limitations 282 6.8 Future Research 283 References 284 About the Author End Page

PAGE 10

vii List of Tables Table 1: Tabular Display of a Co a Period of Five Years (Case I) 20 Table 2: Tabular Display of Companies with Same Size ROE in a Year (Case II) 23 Table 3: Overview of Dependent Variables Used in the Trend Analysis Task 44 Table 4: Overview of Dependent Variables Used in the Pattern Analysis Task 46 Table 5: Training Questions and Grading Schema of the Trend Analysis Task (Pattern Recognition Task) 54 Ta ble 6: Measurement of the Dependent Variables for the Trend Analysis Task 58 Table 7: Measurement of the Dependent Variables for the Pattern Recognition Task 63 Table 8: Demographic Questions 69 Table 9: Mental Workload Questions and Survey Questions 72 Table 10: Manipulation Checks Questions 75 Table 11: Definition of Variables Used in the Statistical Analysis 76 Table 12: Pilot Study Descriptive Statistics 87 Table 13: Multivariate Tests of H1a 95 Table 14: Pilot Test Results of H1a 97 Table 15: Pilot Test Results of H1a 99 Table 16: Pilot Test Results of H1a 101

PAGE 11

viii Table 17: Multivariate Tests of H1b 104 Table 18: Pilot Test Results of H1b 107 Table 19: Pilot Test Results of H1b 109 Table 20: Multivariate Tests of H 1c 111 Table 21: Multivariate Tests of H 1d 112 Table 22: Multivariate Tests of H2a 114 Table 23: Pilot Test Results of H2a 117 Table 24: Multivariate Tests of H2b 118 Table 25: Multivariate Tests of H2c 120 Table 26: Pilot Test Results of H2c 122 Table 27: Multivariate Tests of H2d 125 Table 28: -Moment Correlation Coefficient of the Scores Assigned by Coder One and Coder Two to the Trend Analysis Task 136 Table 29: -Moment Correlation Coefficient of the Scores Assigned by Coder One and Coder Two to the Pattern Recognition Task 138 Table 30: Results of Testing of Normality, Testing For Homogeneity of Variance 143 Table 31: Results of the Manipulation Check Questions of the Trend Analysis Task 151 Table 32: Results of the Manipulation Check Questions of the Pattern Recognition Task 151 Table 33: Descriptive Statistics for the Trend Analysis Task 156

PAGE 12

ix Table 34: -Moment Correlation Coefficient of the Variables of the Trend Analysis Task 165 Table 35: Test Results of H1a 180 Table 36: Te st Results of H1b 188 Table 37: Te st Results of H1c 195 Table 38: Test Results of H1d 200 Table 39: Descriptive Statistics for the Pattern Recognition Task 205 Table 40: -Moment Correlation Coefficient of the Variables of the Pattern Recognition Task 212 Table 41: Test Results of H2a 224 Table 42: Te st Results of H2b 229 Table 43: Test Results of H2c 235 Table 44: Te st Results of H2d 242 Table 45: Mean, Standard Deviation, Range and ANOVA Pairwise Comparison (Mean) of the Survey Questions Assessing Display Usefulness and Ease of Use, for the Trend Analysis Task 250 Table 46: Mean, Standard Deviation, Range and ANOVA Pairwise Comparison (Mean) of the Survey Questions Assessing Display Usefulness and Ease of Use, for the Pattern Recognition Task 254 Table 47: Summary of the Results of the Trend Analysis Task 262 Table 48: Summary of the Results of the Pattern Recognition Task 263

PAGE 13

x List of Figures Figure 1: 2(Case I) 21 Figure 2: 3D Perspective of Five Years (Case I) 22 Figure 3: 2D Display of Companies with Same Size ROE in a Year (Case II) 25 Figure 4: 3D Perspective Display of Companies with Same Size ROE in a Year (Case II) 26 Figure 5: Independent and Dependent Variables at Conceptual and Operation Level 40 Figure 6: Research Model 41

PAGE 14

xi The Effect of Multidimensional Information Presentation on the Effectiveness and Efficiency of a Spatial Accounting Judgment John K. Tan ABSTRACT Th is study is the first in a series of planned studies on the application of multidimensional visualization of business information and data within the context of accounting. The research question is: When is multidimensional visualization of information a better problem representation, improving both the effectiveness and efficiency of a spatial accounting judgment? To examine when multidimensional visualization can assist auditors in configural cue pattern recognition, the study employs the traditional DuPont analysis as the three pieces of key information to be represented on the X, Y, and Z axes of a single 3D perspective display. To help determine when use of 3-D perspective display is beneficial in combining pieces of information, I rely on Theory, and the Proximity Compatibility Principle (PCP) proposed by Wickens and Carswell (1995). The study has two hypotheses. Hypothesis H1 predicted that participants viewing a set of 2-D displays will be the most effective or most efficient in generating hypothese s for what caused the changes in the trend of accounting data or in estimating values. Hypothesis H2 predicted that participants viewing a single 3-D perspective display will be the most effective or most efficient in recognizing patterns of accounting data or in generating hypotheses for what caused the emerged pattern.

PAGE 15

xii To test the hypotheses of the study a 3 x 2 between-subjects design (display format x task) is used. The independent variables are display types and task types. Graphical display was manipulated at three levels: no graphical display (table only), 2D display, and 3-D perspective display. Task was manipulated at two levels: trend analysis and pattern recognition task. The need for a fit between different types of spatial tasks and display formats is demonstrated by the findings of this study: 1) that 2-D displays appear to be more suitable for spatial tasks involving the generation of hypotheses for causes of trends in accounting data, while 2) 3-D perspective displays appear to be more suitable for spatial tasks involving pattern recognition in accounting data.

PAGE 16

1 Chapter 1: Introduction 1.1 Introduction and Significance of the Issue In the late 1950s, Miller (1956) reported that a human can enlarge his or her problem solving capabilities by using visualization to chunk information. Information visualization is a process that transforms data into a form that allows viewers to visually perceive the meaning of the information without having to rely on their cognitive powers to perform the necessary transformations (Zhang, 1996). A recent article in the Wall Street Journal by Totty (Sep 24, 2007) described how the corporate world is using visualization techniques, including animated graphics and three-dimensional charts, to reveal underlying patterns in complicated data and to make quicker decisions. Moriarity (1979) and Dull and Tegarden (1999) have also demonstrated the potential benefits of visual representation of multidimensional accounting information. Research in computer science, human factors, and aviation engineering has moved beyond two-dimensional (2-D) analysis and found positive effects using threedimensional (3-D) visualization of objects. Kolata (1982) reported that statisticians and computer scientists, through computer-motion graphical displays of multidimensional data, were able to see patterns in data that would not have been picked up with statistical techniques. As explained by Kolata (1982, p.919), three-dimensional displays of data enable scientists to make use of the uniquely human ability to recognize meaningful patterns in the data.

PAGE 17

2 Humans have distinctive visual abilities, such as the ability to accurately identify twenty-four data positions in a square (Miller, 1956 p.87) or the ability to retrieve information and recognize patterns from visual cues (Kosslyn, 1994). With visual abilities, humans can solve problems through the associative system instead of the rulebased system (Sloman, 1996). According to Sloman (1996), computations based on the rule-based system reflect rules and logical content, while computations based on the associative system reflect similarity structure and relations of temporal contiguity. This suggests that problem solvers have the ability to make a judgment or decision through the use of three-dimensional charts. Three-dimensional charts highlight temporal contiguity load, and allowing for better decision making. Three-dimensional visual representation has interested accountants for several reasons. -D than with 2examples of three-dimensional visualization of business data, ranging from a fixedincome portfolio to a retail analysis tool showing aggregate regional and individual store performance. Th e fact that three-dimensional charts can convey additional information suggests that decision makers can use three-dimensional charts to improve both the ir effectiveness (accuracy) and efficiency (time on task) in decision making.

PAGE 18

3 1.2 The Need for Research on the Issue of Visualization Three-dimensional visual representation has also interested auditors. Representing business data three-dimensionally through the X, Y and Z axes not only transforms the data into an analogy of the problem space, but also encourages the viewer to task requires the combination of cues and configural information processing, a decision aid could assist the auditor by visually displaying cues in a 3-D perspective. Bedard and Biggs (1991) report that, when performing analytical procedures, auditors failed not only to reason with the combination of all crucial cues but also to generate an accurate hypothesis to explain the causal agent underlying the pattern and relationships among pieces of financial information. As Bedard and Biggs (1991) suggest, audit efficiency and effectiveness depend on competency in recognizing patterns in financial data as well as in hypothesizing likely causes of those patterns to serve as a guide for deploying appropriate audit procedures. The ability of auditors to recognize patterns and make hypotheses may be negatively affected given that auditors working under time constraints tend to make inferior judgments, as highlighted by the reports of the American Institute of Certified Public Accountants (1978, 1987). Indeed, Choo and Firth (1998) find that auditors did not invoke configural information processing under high time pressure.

PAGE 19

4 1.3 Research Objective and Question Th is study is the first in a series of planned studies on the application of multidimensional visualization of business information and data within the context of accounting (broadly defined to include financial, auditing, and managerial). It is the objective of the study to explore when the use of a single 3-D perspective display is beneficial in combining pieces of financial information. In the study, I investigate the following research question: When is multidimensional visualization of information a better problem representation, improving both the effectiveness and efficiency of a spatial accounting judgment? As will be discussed, two streams of literature in auditing have called for the development of a decision aid to assist auditors in recognizing configural cue patterns or relationships between pieces of financial information. With a decision aid that combines all crucial cues, auditors should find it easier to generate hypotheses to explain the causes of identified patterns, and make more accurate global judgments or predictions. To examine when multidimensional visualization can assist in configural cue pattern recognition, the study employs the traditional DuPont analysis as the three pieces of key information to be represented on the X, Y, and Z axes of a single 3-D perspective display. Return on Equity (ROE) is the multiplicative function of profitability (income/sales), turnover (sales/average total assets), and leverage (average total assets/average total equity) (Robinson, Munter and Grant 2003). DuPont analysis requires that individuals consider all three components of ROE (i.e., process the information configurally) to generate hypotheses regarding what caused the changes in ROE. When making accounting judgments, individuals also need to discern any emerging pattern or

PAGE 20

5 make accounting judgments. To help determine when use of 3-D perspective display is beneficial in combining pieces of information, I rely on Theory, and the Proximity Compatibility Principle (PCP) proposed by Wickens and Carswell (1995); both of which are discussed in Chapter 2. 1.4 Results and Contribution Rather than being the most effective (accurate), the results suggest that participants viewing the 2-D displays can sometimes be more effective in generating hypotheses for what caused the changes in the trend of accounting data when compared to those participants viewing the 3-D perspective display. Contrary to expectation, results suggest that participants viewing the 2-D displays failed to be more effective (accurate) or more efficient (used less time) than participants viewing the tabular display or participants viewing the 3-D perspective display in estimating values from existing data on hand. In fact, results show that the static threedimensional perspective representation of DuPont analysis can, under certain conditions, equity (ROE). Results of this study empirically demonstrated the benefits of using the 3D perspective display to recognize patterns in accounting data. Benefiting from the emergent features within a three-dimensional space, participants viewing the 3D perspective display were more effective (accurate) than either those participants viewing a tabular display and or those participants viewing the 2-D displays while performing the

PAGE 21

6 task of recognizing patterns in accounting data. Furthermore, results suggest that participants viewing the 3-D perspective display can sometimes be more efficient or use less time while performing the task of recognizing pattern s in accounting data. Contrary to expectations, results suggest that participants viewing the 3D perspective failed to be more effective (accurate) than participants viewing the tabular display or participants viewing the 2-D perspective displays in generating hypotheses for what caused the emerged patterns in the accounting data. In terms of efficiency in generating hypotheses for what caused the emerged patterns in the accounting data, mixed results were found. The study contributes to the literature in several ways. The 3-D perspective display of the DuPont Analysis, newly developed for this study, is the first of its kind in the accounting literature. The study is the first research that technically develops a 3D perspective display to represent a company by a point in a three-dimensional space in terms of the DuPont analysis. As shown by this study, advances in technology have made it feasible to provide and study the impact of 3-D perspective displays on information processing and decision-making. One contribution of the study is the testing of findings from human factors research in an accounting context. This study contributes to the literature on decision aids by demonstrating how a 3-D perspective display can integrate information from the X, Y and Z axes to help decision makers invoke configural information processing. In this study DuPont analysis is used as an example. However, accountants and auditors frequently are called upon to integrate information that could benefit from a display perspective that causes users to invoke configural information processing.

PAGE 22

7 The study contributes to the literature by demonstrating how Cognitive Fit Theory applies to different subtypes of spatial tasks. The need for a fit between different types of spatial tasks and display formats is demonstrated by the findings of this study: 1) that 2D displays appear to be more suitable for spatial tasks involving the generation of hypotheses for causes of trends in accounting data, while 2) 3-D perspective displays appear to be more suitable for spatial tasks involving pattern recognition in accounting data. Thus, when developing decision aids, it is important to align the task and the aid to maximize the effectiveness of the decision process.

PAGE 23

8 Chapter 2 : Background, Theory, and Hypotheses Development 2.1. What is the Problem? As discussed in the next two sections, two streams of auditing research, configural information processing and analytical procedures, have called for the development of a decision aid to help auditors more quickly recognize configural cue patterns, or patterns and relationships between pieces of financial information. Several types of decision aids can potentially assist in the recognition of configural cue patterns and relationships. As explained in later sections of this chapter, certain types of aids may be more appropriate for certain types of tasks. Of major interest in the current study is whether a threedimensional perspective, new in this study, can provide greater benefit than more traditional types of aids in certain decision situations. 2.1.1 Configural Information Processing Prior to the publications of Brown and Solomon (1990, 1991), auditing studies like Ashton infor (1974) documented that the majority of the variance in judgments was due to information cue main effects; the single interaction effect explained less than four percent of the variance Studies being characterized primarily as relying on independent rather than patterned (configural) cue usage (Brown and Solomon, 1990).

PAGE 24

9 However, Brown and Solomon (1990, p. 19) indicate that information processing is cognition in which the pattern (or configuration) of stimuli is Solomon (1990) is in successfully documenting that auditors do invoke configural information processing. In the Brown and Solomon (1990) study, auditors had drawn on their task-specific knowledge in the context of the cash cycle to assess control risk related to cash disbursements procedures. Of the seventy-four auditors, the variance in judgments of thirty (40.50%) was attributable to configural information processing strategies. Both the Cohen Commission (American Institute of Certified Public Accountants, 1978) and the Treadway Commission (American Institute of Certified Public Accountants, 1987) reported that time pressure is a major cause of dysfunctional auditor judgments. Choo and Firth (1998) extended Brown and Solomon (1991) by essing. Choo and Firth (1998) replicated one of the cases from Brown and Solomon (1991) on fifty-one auditors who were randomly assigned to one of three levels of time pressure: high (twenty seconds per case), moderate (one minute per case), and low (two minutes pressure conditions, Choo and Firth (1998) found that the interaction terms were significant only in the low time pressure condition. The interaction terms had little or no explanatory power in the moderate and high time pressure situations; under these constraints auditors did not invoke configural information processing. Choo and Firth (1998) explained that if auditors do not have sufficient time to retrieve specific patterns

PAGE 25

10 of cue interactions from their knowledge structure, auditors will invoke a cognitive simplification strategy instead of employing configural information processing. In light of these findings, citing Sen and Wallace (1991), Choo and Firth (1998, p. 29) suggested developed as a decision aid that helps auditors to more quickly recognize configural cue patterns for expert judgment under time constraint. Choo and recognize configural cue patterns for decision making provides motivation for the current study which seeks to examine whether decision aids in the forms of two dimensional perspective displays and three-dimensional perspective displays improve the effectiveness and efficiency of decisions. 2.1.2 Analytical Procedures in Auditing Bedard and Biggs (1991) investigated whether auditors were able to recognize patterns in accounting data and generate hypothesized causal explanations while performing analytical procedures in accordance with SAS No. 56; the auditing standard requiring the use of analytical procedures in the planning and overall review stages of all audits. While prior accounting studies of pattern recognition have focused on time-series extrapolation of financial trends, Bedard and Biggs (1991) define pattern recognition in analytical procedures to include not only recognizing relationships among pieces of financial information, but also to explaining the concepts or causal agent that underlie such relationships through hypothesis generation. Defining and recognizing such financial trends is critical because the failure to generate a correct hypothesis for

PAGE 26

11 explaining discrepancies in account relationships could result in considerable wasted audit hours. Bedard and Biggs (1991) tested twenty-one auditors with a case involving an error of capitalizing some part of selling, general and administrative expenses to inventory, which caused four discrepancies: increase in inventory, increase in income, decrease in gross margin, and no change in sales. This case was designed in such a way that a participant could not identify an error that caused the discrepancies unless those discrepancies were considered as a pattern. The auditors were asked to view projected (prepared by audit firm) and unaudited (prepared by client) financial information (six ratios and four year-end balances at 1987). Subjects were instructed that only one error caused discrepancies between projected and unaudited financial information. The client explanation for the cause of discrepancies, as included in the case, was a large year-end hypothesis about an error that could have caused the observed discrepancies. Out of the twenty-one auditors participating in the Bedard and Biggs (1991) experiment, three auditors made acquisition errors, while four auditors failed to combine crucial cues into a pattern. Fourteen auditors recognized the pattern, yet only six auditors proposed a hypothesis consistent with the pattern. The most important finding of Bedard and Biggs (1991), in relation to this study was in documenting that, while performing pattern recognition, subjects tended to reason with one or two cues at a time instead of attempting to reason with the combination of all crucial cues. Another important finding of Bedard and Biggs (1991) was in documenting significant differences between seniors and managers in correctly generating hypotheses for the recognized pattern. Managers

PAGE 27

12 had better performance due to their ability to organize diverse pieces of information into meaningful chunks. When performing analytical procedures, Bedard and Biggs (1991) highlighted the fact that auditors faced difficulties not only in combining all crucial cues to recognize patterns and relationships among pieces of financial information, but also in explaining the concept or causal agent that underlies such relationships through analytical procedures as performed in practice do not stress the use of data in combination. Kinney (1987) documented the importance of the ability of auditors to recognize patterns or trends in relation to audit efficiency and effectiveness. One of the decision rules suggested by Kinney (1987) for detecting accounting errors was crude pattern analysis for the monthly cross-sections of three financial ratios: receivables turnover, inventory turnover, and cost of sales ratios. Kinney (1987) explained that if both receivables turnover and the cost of sales ratio indicate anomalies, crude pattern analysis can alert auditors to the possibility of material fictitious credit sales. Given the fact that the auditors in the Bedard and Biggs (1991) study performed poorly despite having sufficient knowledge of accounting principles to determine which cues should be combined, Bierstaker, Bedard and Biggs (1999) extended Bedard and Biggs (1991) to provide possible explanations of why auditors were unable to correctly solve the analytical procedures and generate correct hypotheses for financial statement discrepancies. The reason why auditors were unable to use their accounting knowledge to solve the analytical procedures, as tested in Bierstaker, Bedard and Biggs (1999), was

PAGE 28

13 and they were unable to shift to a productive representation that contained the knowledge relevant to a correct solution. Within the context of analytical procedures, Bierstaker, Bedard and Biggs (1999) defined a problem representation as the knowledge structure of client, double-entry accounting, industry patterns, error frequencies, and search processes that are held in working memory to guide current problem-solving activity. Bierstaker, Bedard, and Biggs (1999) replicated the task of Bedard and Biggs (1991) using twelve unproductive and next whether auditors would shift toward a productive representation after receiving prompts to activate relevant knowledge. The results revealed that all auditors initially formulated an unproductive problem representation. Before prompts were given, only one of twelve auditors shifted to a productive problem representation and correctly solved the task. After prompts were given, nine of the remaining eleven auditors correctly solved the task. These findings suggest that problem representation shifts are necessary to achieve effective decision processes. In relation to this study, the most important finding of Bierstaker, Bedard and Biggs (1999) was the discovery of pattern recognition as an enhancement factor for the shift in problem representation; i.e., if one can see the pattern, one can then form a productive problem representation. In the Bierstaker, Bedard and Biggs (1999) study, the eleven auditors did not shift to a productive problem representation until several key discrepancies in the client financial data had been combined into a single pattern to be explained. It was critical for the eleven auditors to recognize the pattern of discrepancies before prompts could activate the knowledge that would in turn yield shifts toward

PAGE 29

14 productive problem representations. Bierstaker, Bedard and Biggs (1999) suggest that the finding that prompts were not helpful for most auditors until they understood the pattern of financial discrepancies has implications for research on decision aids in auditing. The findings, of Bedard and Biggs (1991) and Bierstaker, Bedard and Biggs (1999), help inform the development of a decision aid for applying analytical procedures. 2.2. Visualization Th is study explores the application of visualization techniques to the development of decision aids that can assist auditors with configural information processing and with performing analytical procedures. The study develops a decision aid that can help auditors (1) reason with the combination of all crucial cues available, (2) recognize the patterns and relationships between pieces of financial information, and (3) generate correct hypotheses to explain the causes of those patterns. The remainder of this section first provides background on visualization and multidimensional presentations by discussing the research that has been conducted. Subsequently, the section illustrates how the study creates the three-dimensional display, representing DuPont analysis that will be used in the experiment.

PAGE 30

15 2.2.1 Accounting Research on Visualization Wickens et al. (1994) defined visualization as both a display technique and a cognitive operation. This cognitive operation is recognized as the forming of a mental image about the relationships and constraints of the data. Tegarden (1999) indicates that the purpose of visualization is not to replace good solid quantitative analysis, but to allow decision makers to use their natural spatial/visual abilities to identify structure, patterns, trends, anomalies, and relationships in data such that decision makers can determine the course of further exploration. Auditors regularly perform ratio analysis to discern the possible risk of misstatement of the financial statements. If auditors can also visualize financial data and identify anomalies, auditors can then plan audit procedures accordingly. In a well-cited study, Wright (1995) studied the effects of 2-D (two-dimensional) line and bar graph past performance in meeting loan obligations. The integration of three measures that have differential weighting makes the loan collectibility judgment a complex task. When information relationships are presented effectively and efficiently. differences between tabular and tabular plus graphical presentation of three financial measures: liquidity, long-term leverage, and profitability. Control subjects received tabular only information while treatment subjects received tabular plus graphical

PAGE 31

16 summarization of financial measures. Wright (1995) provided six sets of line graphs and bar charts two sets for each of the liquidity, long-term leverage and profitability categories to the treatment groups. Wright (1995) found a significant interaction between the complexity of the task and the incremental benefit of 2-D graphs when analyzing a complex task such as the loan collectibility judgment, which demands the integration of several sources of information liquidity, long-term leverage, and profitability. In Wright (1995), the treatment group not only had less judgment bias but also had significantly better judgment accuracy. These findings suggest that when the number of pertinent information relationships is high and the task demands information integration, the availability of graphs improves auditors loan collectibility judgment. The study extends the findings of Wright (1995) by investigating whether, under certain situations, multidimensional visualization technologies, such as a 3-D (threedimensional) representation, can further improve the efficiency and effectiveness of auditors decision making involving three financial ratios, as in the case of loan collectability judgments. 2.2.2. Multidimensional Accounting Information Research Tegarden (1999) described business information as multidimensional and introduced visualization technologies such as 3-D scattergrams, 3-D line graphs, and volume renderings, which require a 3-D data set. Moriarity (1979) and Dull and Tegarden (1999) investigated the potential benefits of multidimensional representation of accounting information. Moriarity (1979) employed the graphic technique of Chernoff

PAGE 32

17 on schematic faces outperformed a statistical bankruptcy prediction model. However, unlike line or bar graphs, subjects cannot retrieve actual quantitative data from the multidimensional Chernoff faces. Dull and Tegarden (1999) studied the relationship between three visual based on the data. In constructing the experiment, Dull and Tegarden relied on momentum accounting, which extended period of time. Using the same of 120 months, Dull and Tegarden (1999) compared the effect of 2-D and 3-D line graph presentations of the data s Tegarden (1999) found that subjects using 3-D line graphs that could be rotated provided the most accurate predictions. This finding implies that as the complexity of variables increases in terms of dimensions, decision-making accuracy can be enhanced by employing a representation with the capability of displaying the relationships among variables. 2.2.3 Research in Computer Science and Human Factors Multidimensional visualization has been ex tensively researched in disciplines such as computer science and human factors. For example, through computer-motion graphical displays of multidimensional data, computer scientists were able to see patterns in data that would not have been detected with statistical techniques (Kolata, 1982). In a human factors study, Wickens et al. (1994) conducted two experiments to compare comprehension effectiveness between 3-D perspective displays and 2-D planar

PAGE 33

18 displays of multidimensional data. During the experimental session, participants were variables price, earnings, and debt. The control group viewed two 2-D orthogonal graphs, one plotting the relationship between price (Y axis) and earnings (X axis) and the other defining points on the price (Y) and debt (X) axes (Wickens et al. 1994). The objective was to determine whether participants could discern the relationships between earnings and debt. Participants in the control group were asked to compare the two 2D graphs in order to integrate related information through the Y axis (price). The treatment group viewed a 3-D perspective representation of a cubic space of price (Y axis), earnings (X axis), and debt (Z axis). Each company was identified by one sphere in the 3D condition and by two circles in the 2-D display (one in each 2-D graph). Wickens et al. (1994) demonstrated that 3-D perspective displays had the advantage of dimensional integrality over 2-D planar displays. The result is that viewers of 3-D representations spend less effort in searching, scanning and comparing information because all information is displayed in a single panel. Viewers of 3D representations also benefit from the emergence of perceptual features such as the surface created by the spatial integration of the dimensions in the 3D perspective display. 2.2.4 Conceptual Limitation on Visualization Using 2-D Graphs with Three Components Kumar and Benbasat (2004) suggested that information with two ratio-interval (continuous) components and one nominal component can be represented as a single 2D line graph because the third nominal component can be represented on a 2-D plane using an appropriate visual variable (size, value, texture, color orientation and shape). Further,

PAGE 34

19 Bertin (1981) suggested that in the case wherein all three components are interval or ratio (continuous) scale, the information can be represented by a pair of 2-D graphs. The information must be split between two 2-D graphs because the third continuous component cannot be represented on a 2-D plane using visual variables such as legends (Kumar and Benbasat, 2004). Instead of representing information with three components, this study involves the plotting of the DuPont analysis 1 which has four continuous components (return on equity (ROE), turnover, profitability, and leverage). Return on equity (ROE) and profitability can be plotted on a single line graph or bar chart, as they can be scaled as percentages. Turnover and leverage can be plotted on another line graph or bar chart, as they can be scaled as multiples. A single 3-D bubble plot, however, can represent turnover (X axis), profitability (Y axis), leverage (Z axis) and ROE (bubble) simultaneously. Using a set of 2-D graphs, problem solvers must mentally combine all variables from different 2D graphs, which invariably increases cognitive load. 2.2.5 Informationally Equivalent Presentations One additional concept that needs to be discussed before the theory and hypothesis development is the concept of informationally equivalent representations. Since the study relies on three different visual displays, it is important that the displays be informationally equivalent. The three informational ly equivalent representations used in a tabular display, a set of 2-D displays, and a single 3D perspective 1Recall from Chapter 1 that DuPont analysis is the analysis of the components of ROE: profitability (income/sales), turnover (sales/average total assets), and leverage (average total assets/average total equity).

PAGE 35

20 display. The next paragraph uses the two spatial tasks of this study (trend analysis and pattern recognition) to illustrate the concept of informational equivalency as applied to ROE, and as used in this study. Representations are informationally equivalent if all of the information in one is also inferable from the other, and vice versa (Larkin and Simon, 1987; Simon, 1978). In this study information is presented equivalently through different display formats using two different cases of DuPont analysis (modified from White, Sondhi, and Fried, 1988) presented in a tabular display, a set of 2-D displays and a 3-D perspective display, years (trend analysis task). Equivalent information can be presented in a set of 2D displays (Figure 1), and a 3-D perspective display (Figure 2). Table 1 T (Case I). Year Turnover Profitability Leverage ROE 1 1.10 5.77% 2.26 14.34% 2 1.04 3.63% 2.24 8.46% 3 1.01 10.37% 2.63 27.55% 4 0.98 1.49% 3.31 4.83% 5 1.06 4.93% 3.07 16.04%

PAGE 36

21 Figure 1: 2(Case I)

PAGE 37

22 Figure 2: 3D Perspective ive Years (Case I)

PAGE 38

23 Table 2 shows the DuPont analysis of companies with the same size ROE in a year (pattern recognition task). Equivalent information can be presented in a set of 2D displays (Figure 3), and a 3D perspective display (Figure 4). Table 2 Tabular Display of Compan ie s with Same Size ROE in a Year (Case II) Company Turnover Profitability Leverage ROE 1 1.16 2.898 2.05 6.891 2 1.56 2.231 1.98 6.891 3 0.91 3.506 2.16 6.891 4 0.79 3.826 2.28 6.891 5 1.47 2.59 1.81 6.891 6 1.04 3.012 2.20 6.891 In this study only a static 3-D graphic display will be tested; the effects of 3D rotation and animation will be left for future research. The three components of return on equity (ROE) will be in a 3D perspective display with the X axis as turnover (a multiple measure), Y axis as profitability (a percentage measure) and Z axis as leverage (a multiple measure). In the 3D perspective display ROE will be shown as a bubble of varying size that indicates the multiplicative function of turnover, profitability and leverage. The actual value of the ROE will be labeled, while actual values of each of the components of ROE can be read by following the drop lines from the bubbles linking to X, Y and Z axes, respectively.

PAGE 39

24 Graphical software from the Golden Software, Inc. ( www.goldensoftware.com ) was used to draw figures 1 to 4. According to Golden Software, Inc, the graphical software will by default draw and show the data in the best viewing position. Figures 1, 2 and 4 do not have their axes start from zero, since the graphical software, by default, does not believe that starting from zero shows the data to the viewers in the best viewing position. If the 3-D perspective displays of Figures 2 and 4 were redrawn with the axes starting from zero, the bubbles would appear much deeper into the back wall or farther away from the viewers, making perception difficult. Although the reason is unknown, the 3-D graph used by Kuamr and Benhasat (2004), also, did not have axes start from zero. Tufte (1983) showed a 2-D line graph that did not have axes start from zero, because the range of the data were distant from the point of zero. Since the aim of the study is to explore when the use of a single 3-D perspective display is beneficial in combining pieces of financial information, it is crucial that graphical displays are drawn to show the data to the viewers in the best viewing position. spatial, which demands integration of information rather than extraction of discrete data points (symbolic task). Therefore, Figures 1, 2, 3 and 4 are drawn to display the data in the best viewing position to facilitate the integration of information.

PAGE 40

25 Figure 3: 2-D Display of Companies with Same Size ROE in a Year (Case II)

PAGE 41

26 Figure 4: 3-D Perspective Display of Companies with Same Size ROE in a Year (Case II)

PAGE 42

27 2.3 Theory and Hypotheses 2.3.1 The Need for a Fit between Presentation Format and Task The Standards of Field Work of General Accepted Auditing Standards (GAAS) includes the following statement, ditor must obtain a sufficient understanding of the entity and its environment, including its internal control, to assess the risk of material misstatement of the financial statements whether due to error or fraud, and to design the nature, timing and exte June 1, 2007, AU Section 150). Following these GAAS guidelines, participants in this study perform ed two tasks one trend analysis task and one pattern recognition task. In this study the trend analysis task is the identification and generation of hypotheses about what is happening between years in terms of the DuPont analysis of a company. The trend analysis task also includes projecting or calculating a new value from a data set on hand. The pattern recognition task is the categorization of companies with the same ROE into different groups and the generation of hypotheses to explain the differences between groups Based on the discussion in the next sections I hypothesize that participants using a set of 2-D displays will be the most effective (accuracy) and the most efficient (less time) when performing the task of trend analysis. Further, I also hypothesize that participants using a single 3-D perspective display will be the most effective (accuracy) and most efficient (less time) when performing the task of pattern recognition. Since th is study is investigating performance on a spatial integration task, it is important to understand the need for a fit between presentation format and task. For this reason the definition and differences of spatial and symbolic tasks are explained. Spatial

PAGE 43

28 tasks require making associations or perceiving relationships in the data (Vessey, 1991). Speier (2006) suggests that spatial tasks assess the problem area holistically and often involve trend analysis or other types of information associations. Symbolic tasks, on the other hand, require a specific amount as the response, which often involves extracting discrete data values. An exampl e 1986) Speier (2006) suggests that symbolic tasks require the recall or identification of precise data values. Based on the preceding definitions of spatial and symbolic tasks, audit analytical procedures, such as trend analysis or pattern recognition, are basically spatial in nature. The association between spatial tasks and analytic procedures is further supported by Bedard and Biggs (1991), who suggest that the objectives of performing analytical procedures are to recognize patterns and relationships among pieces of financial information and to generate hypotheses on what caused the patterns and relationships. It is common for auditors to perform spatial tasks such as: recognizing patterns in accounting data, generating hypotheses for what caused the pattern, identifying trends in accounting data, generating hypotheses for what caused the changes in trend, and calculating a new value from the data on hand. All of these tasks are spatial because they require perceiving relationships in the data and interpolating values (Vessey 1991). Since all of these tasks are spatial, 2-D and 3-D graphic displays should result in better performance than tabular displays. According to Vessey and Galletta (1991) tabular displays are suitable only for symbolic tasks or tasks which involve extracting discrete

PAGE 44

29 data values. Given a spatial task the natural question is whether a single 3-D perspective display (Figures 2 and 4) is a better presentation format than a set of 2-D displays (Figures 1 and 3). Vessey (1991) suggests that performance on a task will be enhanced when there is a cognitive fit or match between the information emphasized in the representation type and the task type. Cognitive fit occurs through mental representation which pertains to the way problem solving elements, including problem representation and task, are being represented in human working memory. The strategies and processes that problem solvers use to solve the problem at hand are the link between problem representation and task. When a mismatch occurs between problem representation and task, Vessey (1991) suggests that problem solvers first need to formulate a mental representation based either on the problem representation or on the task. The problem solvers then need to transform either the mental representation or the data into a form suitable for solving a particular problem. Such a two-step process consumes more cognitive effort, and can potentially result in an incomplete mental representation that will likely lead to a less accurate decision (Vessey 1994 p.107). Accordingly, cognitive fit occurs when graphs support spatial tasks, and when tables support symbolic tasks. The majority of prior studies have focused on the relative effectiveness of tables versus 2-D graphs (line graphs and bar charts). comparing the relative fit between tables or 2-D graphs (line graphs and bar charts) and symbolic or spatial tasks. Cognitive Fit Theory has not been used to examine how fit occurs between different presentation formats (a single 3-D perspective display or a set of

PAGE 45

30 2-D displays) and a spatial task. This study applies Cognitive Fit Theory in the context of a comparison of a set of line graphs or bar charts (2-D displays) with a single 3D perspective display for a spatial task. To help inform the hypotheses presented in the next se ction other theories and principles are used to help explain how the cognitive fit between a presentation format and a spatial task occurs. These other theories and principles help identify when the match between the representation type (presentation format) and the spatial task are optimal resulting in the least cognitive effort or best cognitive fit. 2.3.2 Trend Analysis and Hypotheses Line graphs emphasize the direction and volatility of data over time and improve judgments requiring trend analysis (Wright 1995). According to Jarvenpaa and Dickson (1988), viewers of line graphs can see trends and relationships at a glance, avoiding the steps of reading, comparing, and interpreting that are necessary to spot deviations using tabular data. Viewers of line graphs can discern the trend relationship simply by following the changes in the slope of the line (see Figure 1). For discerning trends, viewers of a 3-D perspective display are essentially in the same position as table users (see Figure 2). In Figure 2, year one to year five are represented through the different sizes of the bubbles and the relative positions of the bubbles the changes in each of the components of ROE are not shown explicitly. Viewers of the 3D perspective display in Figure 2 must first identify the correct bubble that represents year 1 to year 5, and then use the drop lines from the bubbles to

PAGE 46

31 compare the differences between years for each variable of the ROE. In doing so, the cognitive demand for 3-D viewers of Figure 2 is much greater than th at for the 2-D line graph viewers using Figure 1, as the latter can discern changes through the slope of the line. The line graphs in Figure 1 permanently display the slope of a trend relationship to viewers. Thus, there is a better cognitive fit between 2-D line graphs and the task of identifying trends in accounting data and generating hypotheses for what caused the changes in trend. Prior literature such as Triffett and Trafton (2006) suggests that when information is not explicitly shown in a graph, viewers with domain-specific knowledge can infer implicit information through a mental process called spatial transformation. Spatial transformations occur when a spatial object is transformed from one mental state or location to another mental state or location. Triffett and Trafton (2006) provided a number of examples of spatial transformations; among them are creating a mental image, modifying that mental image by adding or deleting features, mental rotation (Shepard & Metzler, 1971), mentally moving an object, animating a static image (Bogacz & Trafton, 2005), making comparisons between different views (Trafton et al., 2005), as well as any other mental operation that transforms a spatial object from one state or location into another. Triffett and Trafton (2006) described a meteorologist spatially transforming the position of a low pressure system toward a certain direction (by hand gesture) even though actual movement of the low pressure system was not explicitly shown in a graph. Similarly, viewers of the 3D perspective display in Figure 2 can spatially create a line of slope in their mental mind, while analyzing trends in accounting data or generating

PAGE 47

32 hypotheses for what caused the changes in trend. Whether auditors do spatially transform implicit information, while performing ratio analysis, is an empirical question. Given that 2-D line graphs explicitly display the trend relationship through the slope of the line (see Figure 1) there is less of a need for data to be transformed, relative to 3-D displays, allowing for a better cognitive fit. Therefore, it is hypothesized that: H1a: Subjects using a set of 2-D displays will be the most effective (accuracy) in generating hypotheses for what caused the changes in the trend of accounting data when compared to subjects using a single 3-D perspective display or subjects using a table. H1b: Subjects using a set of 2-D displays will be the most efficient (less time) in generating hypotheses for what caused the changes in the trend of accounting data when compared to subjects using a single 3-D perspective display or subjects using a table. It is common for accountants or auditors to perform spatial tasks such as estimating or calculating a new v Difficulty Principle can help us to understand why a set of 2-D line graphs is a better representation for a task involving interpolation of values. Lohse (1991) suggested that only a small fraction of the information (about 3 chunks) decoded from a graph can be held in short-term memory at one time. Reorganization and reinterpretation of the information decoded from a graph is subject to capacity and duration limitations in shortterm memory (Lohse, 1991). Pinker (1990) suggested that limits on short-term memory and on processing resources will make specific sorts of information easier or more

PAGE 48

33 extract a particular piece of information from a graph if the graph has attached a specific message flag to that piece of information. Viewers of line graphs can discern the trend relationship by following the changes in the slope to determine whether the data are linear (straight), not changing (flat), or increasing sharply (steep slope). The 3-D perspective display of the study cannot provide message flags with similar richness about the nature of the data. If a graph can attach a rich message flag to a piece of information, viewers can more easily encode that piece of information from the graph. Therefore, a set of 2-D line graphs is a better representation for a task involving estimation or calculation of new values. By following the slope of the line, viewers of the 2-D line graphs can roughly estimate the new value to be larger or smaller than the existing values or perhaps that it remains the same. Pinker (1990) suggested that elements in a 2-D line graph will be seen as being a smooth continuation of one another. Again, the 2-D line graph should result in less cognitive effort or a better cognitive fit. Therefore, it is hypothesized that: H1c: Subjects using a set of 2-D displays will be the most effective (accuracy) in an accounting judgment involving estimation of values when compared to subjects using a single 3D perspective display or subjects using a table. H1d: Subjects using a set of 2-D displays will be the most efficient (less time) in an accounting judgment involving estimation of values when compared to subjects using a single 3D perspective display or subjects using a table.

PAGE 49

34 Hypothesis H1 relies on different measures related to trend analysis which will be discussed in Chapter Three. 2.3.3 Pattern Recognition and Hypotheses According to Wickens and Carswell (1995) the Proximity Compatibility Principle (PCP) consists of two key concepts: processing proximity and perceptual proximity. Processing, or mental-processing proximity defines the extent to which two or more sources of information are used as part of the same task. If different sources of information must be integrated for a single task, they have close processing proximity. PCP suggests that if there is close processing proximity, then close perceptual proximity, the notion that different sources of information will be perceived as more similar if they are displayed in close proximity, is required. This closeness in space between different sources of information, as illustrated by Wickens and Carswell (1995) principle of perceptual proximity, decreases the cognitive effort of comparison and integration. For example, turnover, profitability, and leverage are different sources of information but they have close processing or mental proximity as they multiplicatively explain what causes the changes in the return on equity. Since turnover, profitability, and leverage have close processing proximity, they should be plotted in a single 3-D perspective display so that their close perceptual proximity enables problem solvers to visualize and integrate them into a single view. Human effort in retrieving information or data through graphical display involves information access costs (IAC) such as movement of attention, the eye and the head (Wickens and Carswell, 1995). Displaying return on equity, turnover, profitability, and

PAGE 50

35 leverage in a set of line graphs or bar charts, instead of a single 3-D perspective display, would increase IAC. An increase in IAC will disrupt performance especially on integration tasks, like the DuPont analysis, because such tasks impose additional load on working memory. To perform an integration task, working memory is used to perform multiple processes of computation as required by integration, retaining information from one source (display) while accessing new information from other sources (displays) (Liu and Wickens, 1992). Displaying different but related information in close proximity will provide a great benefit for integration tasks, particularly those that require sequential retrieval of two or more sources of information. Carswell (1992) suggests that integrative cognitive tasks should be supported by integrated perceptual representations, like 3D perspective displays. An important property of a 3-D perspective display is the function that allows problem solvers to see emergent features between datum. Figure 1 of Wickens et al. (1994, p.48) illustrates an example of an emergent feature, which is the surface in a 3D space from plotting the relationship among earnings (X axis), price (Y axis) and debt (Z axis). The aforementioned surface clearly shows that as price increases, earnings also increase, while debt decreases. Without such an emergent feature, as in the case of problem solvers viewing a set of 2-D graphs, recognition of relationships between plotted variables would require the mental computation or comparison of the individual data values. Bennett and Flach (1992) suggested that a mental combination of values, which puts strain on working memory, can be replaced with an automatic perceptual operation as humans have distinctive visual abilities (Miller 1956).

PAGE 51

36 PCP emphasizes that if the problem-solving task requires integration of information with close processing proximity, a 3-D perspective display has better perceptual proximity than a 2-D planar display. The latter requires mental combination while the former invokes automatic perceptual operations. If each corporation can be defined by three financial ratios, it is possible to represent that corporation by a point in a three-dimensional space (3 -D perspective). Al tman et al.(1974 p.199) suggests that a 3-D perspective display can separate companies into subgroups of good or bad companies in terms of the three financial ratios as represented by the X, Y and Z axes, respectively. While 3-D perspectives may be beneficial in identifying emergent features, 2D perspectives in the form of bar charts are what is commonly used for extracting specific point values. According to Tan and Benbasat (1993), bar charts displaying numeric values at the top of the bar are the best type of graph for answering a wide range of questions. Though the bar charts in Figure 3 display the point values separately for each variable, these variables are not integrated or grouped to show any emergent feature or pattern. In contrast, the 3-D perspective display in Figure 4 is showing an emergent feature. Companies in Figure 4 are automatically categorized into two groups: one group is pursuing high profitability, low turnover and high leverage; the other group is pursuing low profitability, high turnover and low leverage. The 3-D perspective display in Figure 4 is a demonstration of the PCP. The relative position of each bubble or company in Figure D perspective display (Figure 4) has, thus, already categorized the companies in accordance with their relative position within the three-dimensional space. In contrast, bar chart

PAGE 52

37 viewers (Figure 3) information to recognize patterns in the accounting data. In doing so the cognitive load is heavy for viewers of Figure 3. Haskell and Wickens (1993) suggest that humans cannot easily integrate information across separate spatial locations. The resultant implication is that viewers of the bar charts (Figure 3) will have difficulty integrating information across four different bar charts in their effort to recognize the patterns in the accounting data. Thus, there is a better cognitive fit between a 3-D perspective display and the tasks of recognizing patterns of accounting data, and generating hypotheses for what caused the pattern. It is hypothesized that: H2a: Subjects using a single 3-D perspective display will be the most effective (accuracy) in recognizing patterns of accounting data when compared to subjects using a set of 2D displays or subjects using a table. H2b: Subjects using a single 3-D perspective display will be the most efficient (less time) in recognizing patterns of accounting data when compared to subjects using a set of 2D displays or subjects using a table. H2c: Subjects using a single 3D perspective display will be the most effective (accuracy) in generating hypotheses for what caused the emerged patterns when compared to subjects using a set of 2-D displays or subjects using a table. H2d: Subjects using a single 3-D perspective display will be the most efficient (less time) in generating hypotheses of what caused the emerged patterns when compared to subjects using a set of 2-D displays or subjects using a table.

PAGE 53

38 Hypothesis H2 relies on different measures related to pattern recognition which will be discussed in Chapter Three.

PAGE 54

39 Chapter 3 : Research Design 3.1 Research Model To test the hypotheses of the study a 3 x 2 between-subjects design (displa y format x task) is used. Two performance constructs are of interest in the study, effectiveness and efficiency. In this study I defined effectiveness as accuracy. Effectiveness is measured as Efficiency is measured as the response time related to the various tasks the participants are asked to perform. The independent variables are display types and task types. Graphical display was manipulated at three levels: no graphical display (table only), 2D displ ay s, and 3-D perspective display (see Tables 1 and 2, and Figures 1-3 for examples of display formats used). Task was manipulated at two levels: trend analysis and pattern recognition task. A number of covariates measures are used in this study to test for significant correlations with the dependent variables. Figure 5 shows the independent and dependent variables at the conceptual and operational levels. Figure 6 shows the research model.

PAGE 55

40 Figure 5 Independent and Dependent Variables at Conceptual and Operational Level Independent Theory Dependent Cognitive Fit Operationalization Presentation formats Task type Judgment Performance: Eff ectiveness Eff iciency Presentation Formats: Tabular display 2 D displays 3 D display Task Type: Trend analysis task Pattern recognition task Effectiveness: Generate a hypothesis Estimate new values Recognize a pattern Generate a hypothesis Efficiency Time on task Statistic Covariates Practice Questions Mental Rotations Test Gender and Age Mental Workload

PAGE 56

41 Figure 6: Research Model H1a H1c H2a H2c Presentation Format Comparisons 2 D line graphs better than tabular display and 3 D perspective display. 3 D perspective display better than tabular display and 2 D bar charts. Decision Making Effectiveness ( Trend analysis task ) Generating hypotheses for what caused the changes in the trend of accounting data. Decision Maki ng Effectiveness ( Trend analysis task ) Estimation of values. Decision Making Effectiveness ( Patter n recognition task ) R ecogni zing patterns of accounting data. Decision Mak ing Effectiveness ( Pattern recognition task ) Generating hypotheses of what caused the emerged patterns.

PAGE 57

42 3.2 The Independent Variables The independent variables are display types and task types. Graphical display was manipulated at three levels: no graphical display (table only), 2-D displays, and 3-D perspective display (see Tables 1 and 2, and Figures 1-3 for examples of display formats used). Task was manipulated at two levels: trend analysis and pattern recognition task. The trend analysis task is defined as the identification and generating of hypotheses about what is happening between years in terms of the DuPont analysis of a company. The trend analysis task also includes projecting or calculating a new value from a data set on hand. The pattern recognition task is defined as the categorization of companies with the same ROE into different groups and the generation of hypotheses to explain the differences bet 3.3 The Dependent Variables Two performance constructs are of interest in the study, effectiveness and efficiency. In this study I define effectiveness as accuracy in performing the tasks. The tasks include generation of hypotheses regarding what caused the changes in the trend of accounting data; estimation of new values for ROE, profitability, turnover, and leverage; recognizing patterns of financial data between companie s; and generation of hypotheses regarding what caused the patterns in financial data. These same tasks are used to determine the efficiency of participants which is measured as the response time related to the various tasks the participants are asked to perform. To test for the trend analysis task (H1a, H1b, H1c, and H1d), six questions are developed (see Table 3) and each participant answers these six questions in the same

PAGE 58

43 order. The first, second, fifth, and sixth questions are used to test for H1a and H1b. The first question asks the participant to write brief sentences describing the differences between years 1 and 4. The second question asks the participants to write brief sentences describing what they perceive to be occurring in the data going from year 2 to year 3 and year 4. The fifth question asks participants to indicate the differences between years 2 and 4 by selecting choices from a given template. The sixth question asks participants to describe what they perceive to be occurring in the data from year 1 to year 2 to year 3 by selecting choices from a given template. The third and fourth questions are used to test H1c and H1d. The third question asks the participants to estimate what the ROE would be in year 6 if each of the variables comprising ROE in year 5 had doubled. The fourth question asks the participants to estimate the average of turnover, leverage and profitability for the years 1, 2, 4 and 5, and use the estimated average to calculate a new ROE.

PAGE 59

44 Table 3 Overview of Dependent Variables Used in the Trend Analysis Task Dependent variables Tests of Questions developed to test hypotheses Accuracy of hypotheses generated Time (seconds) spent on task H1 a H1 b What are the differences between year s 1 and 4? Write short sentences Accuracy of hypotheses generated Time (seconds) spent on task H1a H1b What is happening as you go from year 2 to year 3 to year 4? (Hint: Please identify the trends from year 2 to year 4).Write short sentences using Accuracy of the new values estimated Time (seconds) spent on task H1c H1d Based on the ROE of year 5 (Bubble 5), what would be year 6 ROE if each of the variables of ROE in year 5 had doubled? Accuracy of the new values estimated Time (seconds) spen t on task H1c H1d Estimate the average of turnover, leverage and profitability for the years 1, 2, 4 and 5, and use them to calculate a new ROE (Hint 1: you do not actually need to calculate the average, please consider the position of ROE as you attempt t o answer. Hint 2: year 3 is not used) Accuracy of hypotheses generated Time (seconds) spent on task H1a H1b What are the differences between year s 2 and 4? Please answer as accurately and as fast as possible.* Accuracy of hypotheses generated Time ( seconds) spent on task H1a H1b What is happening as you go from year 1 to year 2 to year 3? Please answer as accurately and as fast as possible.* Participants selected choices from a given template.

PAGE 60

45 To test for the pattern recognition task (H2a, H2b, H2c and H2d), six questions are developed (see Table 4) and each participant answers these six questions in the same order. The second and fourth questions are used to test H2a and H2b. The second question asks the participants to separate companies 1 through 6 into two groups based on similar characteristics. The fourth question asks the participants to select one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. The first, third, fifth and sixth questions are used to test H2c and H2d. The first question asks the participants to write brief sentences describing the differences between companies 1 and 6. The third question asks the participants to write brief sentences describing the patterns of the financial ratios they perceive in group one. The fifth question asks participants to indicate the differences between companies 4 and 6 by selecting choices from a given template. The sixth question asks participants to describe the patterns of the financial ratios they perceive in group two by selecting choices from a given template. Each of the six questions of the trend (pattern) analysis task has its own page screen, allowing for the time spent by each participant on each question to be recorded in seconds, thus allowing for measurement of efficiency.

PAGE 61

46 Table 4 Overview of Dependent Variables Used in the Pattern Analysis Task Dependent variables Test of Questions developed to test hypotheses Accuracy of hypotheses generated Time (seconds) spent on task H2 c H2 d What are the differences between compan ies 1 and 6? Write short Accuracy in recognizing patterns Time (seconds) spent on task H2a H2b Please separate companies 1 through 6 into 2 groups based on similar characteristics. Note: please assign each company only once to either group one or group two, but the groups need not have the same number of companies Accuracy of hypotheses generated Time (seconds) spent on task H2c H2d Group one includes com panies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group two, what are the patterns of the financial ratios you are seeing in group one? Write short sentences Accuracy in recognizing patterns T ime (seconds) spent on task H2a H2b Assuming you cannot select a company solely because of a single variable, for example higher profitability. Comparatively, if it is better to have a higher profitability, higher turnover but lower leverage at the same t ime, which company you will select? Accuracy of hypotheses generated Time (seconds) spent on task H2c H2d What are the differences between compan ies 4 and 6? Please answer as accurately and as fast as possible Accuracy of hypotheses generated Time (seconds) spent on task H2c H2d Group one includes companies 1, 3, 4 and 6, and group two includes companies 2 and 5. Compared to group one what are the patterns of the financial ratios you are seeing in group two? Please answer as accurately and as fast as possible. * Participants selected choices from a given template.

PAGE 62

47 3.4 Ex perimental Setup and Procedures Two hundred and fifty eight undergraduate business students participated in the main experiment. Each participant was randomly assigned to one of the six treatment conditions: trend analysis task with tabular display, trend analysis task with 2-D display, trend analysis task with 3-D perspective display, pattern recognition task with tabular display, pattern recognition task with 2-D display, and pattern recognition with 3D perspective display. The experiment was conducted in a computer laboratory over a period of two experiment. When the participants arrived at the computer laboratory they were randomly assigned to a computer. The investigator first announced that if any participant had eyesight problems he or she could not participate in the experiment (one participant with color blindness was dismissed). The investigator further announced that calculators, or any external aids were not allowed to be used during the whole experiment. The investigator then distributed and explained the informed consent forms to the participants. After the investigator had collected all of the signed informed consent forms from participants, the investigator invited each participant to draw an envelope containing the web link to one of the six treatment conditions. All participants then typed the web link into the computer to access their randomly assigned treatment conditions. All participants started the experiment at the same time, after the investigator was satisfied that each

PAGE 63

48 participant had access to a randomly assigned treatment condition. The following paragraphs describe the experimental procedures. All participants first completed a training task to familiarize them with their assigned display format. Participants performing the trend analysis task viewed either a table, or a set of 2D line charts, or a single 3-D perspective display showing changes in the desirability rating of apartments rented in the past six years and practiced answering four questions on trend analysis. Participants performing the pattern recognition task viewed either a tabular display, or a set of 2D bar charts, or a single 3-D perspective display showing the differences between apartments available for rent and practiced answering four questions involving the recognition of patterns. After all participants had answered four practice questions with their assigned display format, participants then completed a training task on DuPont analysis. All participants answered practice questions about the concept of DuPont analysis and answered another practice question involving the calculation of ROE. Following the training on DuPont analysis, participants completed the experimental tasks. After completing the trend analysis task (see Table 3) or pattern recognition task (see Table 4), all participants completed post-experiment questionnaire consisting of a manipulation check question (see Table 24), a demographic survey (see Table 22), and the Mental Rotations Test. 3.5 Training Training involved evaluating the desirability of apartments a task to which college students can easily relate. The degree of desirability of an apartment to a potential

PAGE 64

49 renter is represented by a desirability rating, which is a multiplicative function of three variables: income factor, size factor, and distance factor. The training task thus mirrored the experimental task, wherein return on equity is a multiplicative function of three accounting variables (return on assets, profitability, and leverage). 3.5.1 Training for the Trend Analysis Task Participants performing the trend analysis task viewed either a tabular display, or a set of 2D line displays or a single 3-D perspective display showing changes in the desirability rating of the apartments rented by the participants between years one to six. Participants were told to assume that for various reasons (for example roommates graduating or the landlord selling the rental property) they had rented and lived in six different apartments, with different desirability ratings in the last six years. Participants were also told to assume that in the past six years, their fixed disposable income had not changed, and their desire to rent a 1,200 square f oo t apartment with the lowest rent and within 5 miles of campus also had not changed. Each participant answered four practice questions (see Table 5) by reading from either a tabular display, or a set of 2D line displays, or a single 3-D perspective display. Bubbles on the 3while the X, Y and Z axes represented the income factor, size factor and distance factor, respectively. Additionally, participants in the 3-D treatment condition learned how to read values from drop lines. After completing the training on display format, participants completed a training task on DuPont analysis. All participants answered a practice question about the concept

PAGE 65

50 of DuPont analysis and answered another practice question involving the calculation of ROE (see Table 5). Four questions were developed to familiarize participants with their assigned display format. Each participant answer ed these four questions in the same order. The first question asks the participant to write brief sentences describing the values of the factors of the apartment rented in year 2. The correct answer to the first question has four parts: (a) desirability rating = 93.78, (b) size factor = 0.94, (c) income factor = 72.03, and (d) distance factor =1.39. For each correct response to one of the four parts the participant was awarded one point. The score range for the answer to the first question is zero to four points (see Table 5). This score on the first question is one of the covariate measures used to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the first question is also one of the covariate measures used to test for significant correlations with the dependent variables. The second question asks the participant to write brief sentences describing the differences between the apartments rented in year 5 and year 6. The correct answer to the second question has three parts: (a) the year 5 size factor is higher than year 6, (b) year 5 income factor is higher than year 6, and (c) year 5 distance factor is lower than year 6. For each correct response to one of the three parts the participant was awarded one point. The score range for the answer to the second question is zero to three points (see Table 19). This score on the second question is one of the covariate measures used to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the second question is also one of the covariate measures used to test for significant correlations with the dependent variables.

PAGE 66

51 The third question asks the participant to write brief sentences describing the values of the factors of the apartment rented in year 4. The correct answer to the third question has four parts: (a) the desirability rating = 26.03, (b) size factor = 0.97, (c) income factor = 71.38, and (d) distance factor = 0.38. For each correct response to one of the four parts the participant was awarded one point, providing a range for the answers to the third question from zero to four points (see Table 5). The score on the third question is used as a covariate measure when testing for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the third question is also one of the covariate measures used to test for significant correlations with the dependent variables. The fourth question asks the participant to write brief sentences describing the differences between the apartments rented in year 2 and year 4. The correct answer to the fourth question has four parts: (a) year 2 desirability rating is higher than year 4, (b) year 2 size factor is lower than year 4, (c) year 2 income factor is higher than year 4, and (d) year 2 distance factor is higher than year 4. Again, one point was awarded for each correct response. The score range for the fourth question is zero to four points (see Table 5). The score on the fourth question is used as a covariate to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the fourth question is also one of the covariate measures used to test for significant correlations with the dependent variables. The two questions developed to familiarize participants with the concept of DuPont analysis were answered in the same order. The first question asks the participant to calculate

PAGE 67

52 and the leverage ratio is 1.1. The correct answer to the first question is ROE equals 11%. A correct response by the participant was awarded one point, providing a score range of zero to one point (see Table 5). This score is one of the covariate measures used to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the first question is also one of the covariate measures used to test for significant correlations with the dependent variables. The second question asks participants to indicate whether ROE is the sum or multiple of turnover ratio, profitability ratio and leverage ratio. The correct answer to the question is multiple, resulting in the participant receiving one point. The score is either zero or one point (see Table 5). This score on the second question is one of the covariate measures used to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering the second question is also one of the covariate measures used to test for significant correlations with the dependen t variables. 3.5.2 Training for the Pattern Recognition Task Participants were told to assume that they had a fixed disposable income, and want to rent a 1,200 square foot apartment with the lowest rent and within 5 miles of campus. The degree of desirability of an apartment to a renter is represented by a desirability rating that is a multiplicative function of three variables: income factor, size factor, and distance factor. Each participant answered four practice questions (see Table 5) by reading from either a tabular display, or a set of 2D bar displays, or a single 3-D perspective display.

PAGE 68

53 The practice questions were similar to the questions practiced by the participants performing the trend analysis task. For example, instead of describing the values of the factors of the apartment rented in a year, participants performing the pattern recognition task practiced describing the values of the factors of an apartment (see Table 5). Additionally, participants in the 3-D treatment condition learned how to read values from drop lines. Bubbles on the 3-D perspective display rating, while the X, Y and Z axes represented the income factor, size factor and distance factor, respectively. After completing the training on display format, participants completed a training task on DuPont analysis. The same two practice questions about the concept of DuPont analysis and the calculation of ROE were used to train participants performing the trend analysis task, and the pattern recognition task (see Table 5).

PAGE 69

54 Table 5 Training Questions and Grading Schema for the Trend Analysis Task (Pattern Recognition Task) Training Questions Answers to Questions If Answered Correctly Grading Schema What are the values of the factors of (apartment 2) the apartment rented in year 2? Desirability rating 93.7 8 Size factor 0.94 Income factor 72.03 Distance factor 1.39 1 point 1 point 1 point 1 point 0 point to 4 points What are the differences between (apartments 5 and 6) apartments rented in years 5 and 6? Year (apartment) 5 size factor higher tha n year (apartment) 6 Year (apartment) 5 income factor higher than year (apartment) 6 Year (apartment) 5 distance factor lower than year (apartment) 6 1 point 1 point 1 point 0 point to 3 points What are the values of the factors of (ap artment 4) the apartment rented in year 4? Desirability 26.03 Size factor 0.97 Income factor 71.38 Distance factor 0.38 1 point 1 point 1 point 1 point 0 point to 4 points What are the differences between (apartments 2 and 4) apartments rented in years 2 and 4? Year (apartment) 2 desirability rating higher than year (apartment) 4 Year (apartment) 2 size factor lower than year (apartment) 4 Year (apartment) 2 income factor higher than year (apartment) 4 Year (apartment) 2 dist ance factor lower than year (apartment) 4 1 point 1 point 1 point 1 point 0 point to 4 points ratio is 5% and leverage ratio is 1.1? ROE 11% 1 point 0 point to 1 point ROE is the sum or multiple of turnover ratio, profitability ratio and leverage ratio? Multiple 1 point 0 point to 1 point

PAGE 70

55 3.6 Measuring the Dependent Variables 3.6.1 Measuring the Dependent Variables for the Trend Analysis Task To test for the trend analysis task (H1a, H1b, H1c, and H1d), six questions are developed (see Table 3) and each participant answers these six questions in the same order. Questions asked and the scoring of answers is similar to what was used in the training. The first, second, fifth, and sixth questions are used to test for H1a and H1b. The first question asks the participant to write brief sentences describing the differences between years 1 and 4. The correct answer to the first question has four parts: (a) year 1 ROE is higher than year 4, (b) year 1 turnover is higher than year 4, (c) year 1 profitability is higher than year 4, and (d) year 1 leverage is lower than year 4. Each correct response was awarded one point. The score range for the first question is zero to four points (see Table 6). This score on the first question is the dependent measure (accuracy) used to test H1a (see Table 3). The time spent in seconds by each participant when answering the first question is the dependent measure (efficiency) used to test H1b (see Table 3). The second question asks the participants to write brief sentences describing what they perceive to be occurring in the data going from year 2 to year 3 and year 4. The correct answer to the second question has eight parts: (a) year 2 ROE is higher than year 3, (b) year 3 ROE is lower than year 4, (c) year 2 turnover is higher than year 3, (d) year 3 turnover is higher than year 4, (e) year 2 profitability is higher than year 3, (f) year 3 profitability is lower than year 4, (g) year 2 leverage is lower than year 3, and (h) year 3 leverage is lower than year 4. Each correct was awarded one point, providing a score range from zero to eight points (see Table 6). This score on the second question is the

PAGE 71

56 second dependent measure (accuracy) used to test H1a (see Table 3). The time spent in seconds by each participant when answering the second question is the second dependent measure (efficiency) used to test H1b (see Table 3). The fifth question asks participants to indicate the differences between years 2 and 4 by selecting choices from a given template. The correct answer to the fifth question has four parts: (a) year 2 ROE is higher than year 4, (b) year 2 turnover is higher than year 4, (c) year 2 profitability is higher than year 4, and (d) year 2 leverage is lower than year 4. With four parts, the range of possible correct answers is zero to four points (see Table 6). This score on the fifth question is the third dependent measure (accuracy) used to test H1a (see Table 3). The time spent in seconds by each participant when answering the fifth question is the third dependent measure (efficiency) used to test H1b (see Table 3). The sixth question asks participants to describe what they perceive to be occurring in the data from year 1 to year 2 to year 3 by selecting choices from a given template. The correct answer to the sixth question has eight parts: (a) year 1 ROE is higher than year 2, (b) year 2 ROE is higher than year 3, (c) year 1 turnover is higher than year 2, (d) year 2 turnover is higher than year 3, (e) year 1 profitability is higher than year 2, (f) year 2 profitability is higher than year 3, (g) year 1 leverage is higher than year 2, and (h) year 2 leverage is lower than year 3. Each correct response by the participant to each of the eight parts was awarded one point. The score range for the sixth question is zero to eight points (see Table 6). This score on the sixth question is the fourth dependent measure (accuracy) used to test H1a (see table 3). The time spent in seconds by each participant when

PAGE 72

57 answering the sixth question is the fourth dependent measure (efficiency) used to test H1b (see Table 3). H1c and H1d are tested with questions three and four. Question three asks the participants to estimate what the ROE would be in year 6 if each of the variables comprising ROE in year 5 had doubled. The correct answer to the third question is response and the correct answer is the dependent measure (accuracy) used to test H1c (see Table 3). Time spent in seconds by each participant when answering the question is the dependent measure (efficiency) used to test H1d (see Table 3). Question four asks the participants to estimate the average of turnover, leverage and profitability for the years 1, 2, 4 and 5, and use the estimated average to calculate a new ROE. The correct answer to the fourth question has four parts: a) average turnover = 1.045, (b) average profitability = 3.955%, (c) average leverage = 2.72 and the new ROE = 1 correct answers are the dependent measures (accuracy) used to test H1c (see Table 3). Time spent in seconds by each participant when answering the fourth question is the dependent measure (efficiency) used to test H1d (see Table 3). Table 6 shows the grading schema of each of the six questions of the trend analysis task.

PAGE 73

58 Table 6 Measurement of the Dependent Variables for the Trend Analysis Task Questions developed to test hypoth eses Answers to questions If answered Correctly Grading Schema (Score Range) What are the differences between years 1 and 4? Write Year 1 ROE higher than year 4 Year 1 turnover higher than year 4 Year 1 profitability higher than year 4 Year 1 leverage lower than year 4 1 point 1 point 1 point 1 point 0 point to 4 points What is happening as you go from year 2 to year 3 to year 4? (Hint: Please identify the trends for year 2 to year 4).Wri Year 2 ROE higher than year 3 Year 3 ROE lower than year 4 Year 2 turnover higher than year 3 Year 3 turnover higher than year 4 Year 2 profitability higher than year 3 Year 3 profitability l ower than year 4 Yea r 2 leverage lower than year 3 Year 3 leverage lower than year 4 1 point 1 point 1 point 1 point 1 point 1 point 1 point 1 point 0 point to 8 points Based on the ROE of year 5 (Bubble 5), what would be year 6 ROE if each of the variables of ROE in year 5 had doubled? Year 6 ROE 128.34% +/ differences from answer Estimate the average of turnover, leverage and profitability for the years 1, 2, 4 and 5, and use them to calculate a new ROE (Hint 1: you do not actually need to calculate the average, please consider the position of ROE as you attempt to answer. Hint 2: year 3 is not used) Turnover 1.045 Profitability 3.955% Leverage 2.72 New ROE 11.241 +/ differences from answer

PAGE 74

59 Table 6 (Continued) Measurement of the Dependent Variables for the Trend Analysis Task Questions developed to test hypotheses Answers to questions If answered Correctly Grading Schema (Score Range) What are the differences between years 2 and 4? Please answer as accurately and as fast as possible. Year 2 ROE higher than year 4 Year 2 turnover higher than year 4 Year 2 profitability higher than year 4 Year 2 leverage lower than year 4 1 point 1 point 1 point 1 point 0 point to 4 points What is happening as you go from year 1 to year 2 to year 3? Please answer as accurately and as fast as possible. Year 1 ROE higher than year 2 Year 2 ROE higher than year 3 Year 1 turnover higher than year 2 Year 2 turnover higher than year 3 Year 1 profitability higher than year 2 Year 2 profitability higher than year 3 Year 1 leverage higher than year 2 Year 2 leverage lower than year 3 1 point 1 point 1 point 1 point 1 point 1 point 1 point 1 point 0 point to 8 points

PAGE 75

60 3.6.2 Measuring the Dependent Variables for the Pattern Recognition Task To test for the pattern recognition task (H2a, H2b, H2c and H2d), six questions are developed (see Table 4) and each participant answers these six questions in the same order. The second and fourth questions are used to test H2a and H2b. The second question asks the participant to separate companies 1 through 6 into two groups based on similar characteristics. The correct answer to the first question places companies 1, 3, 4, and 6 into group one, and companies 2 and 5 into group two. One point was awarded whenever the participant placed a company into the correct group. The score range for this question is zero to six points (see Table 7). This score on the second question is the dependent measure (accuracy) used to test H2a (see Table 4). Time spent in seconds by each participant when answering the second question is the dependent measure (efficiency) used to test H2b (see Table 4). Question four asks the participant to select one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. Company one is the correct answer to the fourth question. A correct response by the participant was awarded one point. The score range for the fourth question is zero to one point (see Table 7). This score on the fourth question is the second dependent measure (accuracy) used to test H2a (see Table 4). Time spent in seconds by each participant when answering the fourth question is the dependent measure (efficiency) used to test H2b (see Table 4). The first, third, fifth and sixth questions are used to test for H2c and H2d. The first question asks the participant to write brief sentences describing the differences between companies 1 and 6. The correct answer to the first question has three parts: (a) company 1 turnover is higher than company 6, (b) company 1 profitability is lower than

PAGE 76

61 company 6, and (c) company 1 leverage is lower than year 6. Each correct response by the participant to each of the three parts was awarded one point. The score range for the first question is zero to three points (see Table 7). The score on the first question is the dependent measure (accuracy) used to test H2c (see Table 4). Time spent in seconds by each participant when answering the first question is the dependent measure (efficiency) used to test H2d (see Table 4). Question three asks the participant to write brief sentences describing the financial ratio patterns they perceive in group one (companies 1, 3, 4 and 6) relative to group two (companies 2 and 5). Th e correct answer to the third question has three parts: (a) group one turnover is lower than group two, (b) group one profitability is higher than group two, and (c) group one leverage is higher than group two. Each correct response was awarded one point. The score range for the third question is zero to three points (see Table 7). This score on the third question is the second dependent measure (accuracy) used to test H2c (see Table 4). Time spent in seconds by each participant when answering the question is the second dependent measure (efficiency) used to test H2d (see Table 4). The fifth question asks participants to indicate the differences between companies 4 and 6 by selecting choices from a given template. The correct answer to the fifth question has three parts: (a) company 4 turnover is lower than company 6, (b) company 4 profitability is higher than company 6, and (c) company 4 leverage is higher than company 6. Correct responses were awarded one point, resulting in a score range of zero to three points (see Table 7). The score on the fifth question is the third dependent measure (accuracy) used to test H2c (see Table 4). Time spent in seconds by each

PAGE 77

62 participant when answering the fifth question is the third dependent measure (efficiency) used to test H2d (see Table 4). Question six asks participants to describe the financial ratio patterns they perceive in group two relative to group one by selecting choices from a given template. The correct answer to the sixth question has three parts: (a) group two turnover is higher than group one, (b) group two profitability is lower than group one, and (c) group two leverage is lower than group one. Again, correct responses by the participant were awarded one point. The score range is zero to three points (see Table 7). The score on the third question is the fourth dependent measure (accuracy) used to test H2c (see Table 4). Time spent in seconds by each participant when answering the sixth question is the fourth dependent measure (efficiency) used to test H2d (see Table 4).

PAGE 78

63 Table 7 Measurement of the Dependent Variables for the Pattern Recognition Task Questions developed to test hypotheses Answers to questions If answered Correctly Grading Schema (Score Range) What are the differences between companies 1 and 6? Company 1 turnover higher than company 6 Company 1 profitability lower than company 6 Company 1 leverage lower than company 6 1 point 1 point 1 point 0 point to 3 points Please separate companies 1 through 6 into 2 groups based on similar characteristics. Note: please assign each company only once to either group one or group two but the groups need not have the same number of companies A group includes companies 1, 3 4 and 6 Another group includes companies 2 and 5 1 point 1 point 1 point 1 point 1 point 1 point 0 point to 6 points Group one includes companies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group two, what are the patterns of the financial ratios you are seeing in group one? Write short sentences using the Group 1 turnover lower than group 2 Group 1 profitability higher than group 2 Group 1 leverage higher than group 2 1 point 1 point 1 point 0 point to 3 points Assume you cannot select a company solely because of a single variable, for example higher profitability. Comparatively, if it is better to have a higher profitability, higher turnover but lower leverage at the same time which company you will select? Company 1 1 point 0 point to 1 point

PAGE 79

64 Table 7 (Continued) Measurement of the Dependent Variables for the Pattern Recognition Task Questions developed to test hypotheses Answers to questions If answered Cor rectly Grading Schema (Score Range) What are the differences between companies 4 and 6? Please answer as accurately and as fast as possible Company 4 turnover lower than company 6 Company 4 profitability higher than company 6 Company 4 leverage hig her than company 6 1 point 1 point 1 point 0 point to 3 points Group one includes companies 1, 3, 4 and 6, and group two includes companies 2 and 5. Compared to group one what are the patterns of the financial ratios you are seeing in group two? Pl ease answer as accurately and as fast as possible. Group 2 turnover higher than group 1 Group 2 profitability lower than group 2 Group 2 leverage lower than group 1 1 point 1 point 1 point 0 point to 3 points

PAGE 80

65 3.7 Covariates 3.7.1 Practice Questions Tufte (1983 p. 56) suggested that perceptions change with experience; and that is affected by his or her knowledge of the DuPont analysis and his or her familiarity with the (randomly assigned) display format. Compared to the participants viewing the 3D display, it is expected that participants viewing the tabular display or the 2-D displays would have higher scores in each of the four practice questions regarding the display format; as the latter would be more familiar with their assigned display formats. The score on each of the six practice questions (four questions on display format and two questions on ROE) is the covariate measure used to test for significant correlations with the dependent variables. The time spent in seconds by each participant when answering each of the six practice questions is another covariate measure used to test for significant correlations with the dependent variables. 3.7.2 Mental Rotations Test by his or her spatial visualization ability. Using meta-analytic techniques to compare the effect size of the gender difference in fourteen prior studies that administrated the Mental Rotations Test, Masters and Sanders (1993) confirmed that males generally scored higher than females in the Mental Rotations Tests.

PAGE 81

66 Vandenberg and Crawford (1971) developed the Mental Rotations Test to ion abilities. This test uses three-dimensional objects displayed in two-dimensional drawings to measure spatial visualization. In order participants in the study completed th e Mental Rotations Test. This study uses the Mental Rotations Test supplied by the Educational Testing Service (ETS). The Mental Rotations Test presents a three-dimensional drawing on the left hand side and requires participants to indicate which two of the (rotated in three-dimensional space) versions of the drawing on the right hand side (rotated in three-dimensional space) represents the original 3-D drawing, thereby awford 1971). following grading schema: (a) if two drawings were chosen and both choices of drawings are correct, two points are awarded, (b) if two drawings were chosen and one choice of drawings is incorrect, or both choice of drawings are incorrect no points are awarded, (c) if only one drawing was chosen and it is correct, one point is awarded. The score range for the Mental Rotations Test is zero to 40 points. This score on the Mental Rotations Test is the covariate measure used to test for significant correlations with the dependent variables. The total time in seconds spent by each participant when answering the Mental Rotations Test is also used as a covariate measure used to test for significant correlations with the dependent variables.

PAGE 82

67 3.7.3 Gender and Age Hygge and Knez (2001) used gender as an individual variable to test for cognitive performance on four different tasks attention, problem solving, long-term recall and recognition, and short-term recall varying conditions of heat and light. Their results showed that women demonstrated higher performance than men on problem solving tasks, and women also remembered a greater number of words than men. Within the auditing context, Chung and Monroe (1998) confirmed prior findings that females and males process information differently. Accordingly, gender will be one of the covariates in the study (see Table 8). Age will also be collected and tested for significant correlations with the dependent variables (see Table 8). 3.7.4 Mental Workload difference between their capacity for performance and the demands of the task. Benford (2000) further explained that an individual will perceive increases in his/her mental workload as task demands increase. In this study, participants in the 2-D and 3-D treatment conditions had to scroll up and down the screen when answering questions, since the display and the response area did not fit on one screen. Participants viewing the tabular display did not have such a problem. It is expected that participants viewing the 2-D displays and the 3-D perspective display will perceive heavier mental workloads than participants viewing the tabular display.

PAGE 83

68 Participants of this study were undergraduate business students who are familiar with reading data from a tabular display, 2-D line graphs, and 2-D bar charts. The 3D perspective display of DuPont analysis is a newly created presentation format of financial ratios, and had never been seen by the participants. Therefore, it is logical to expect that participants viewing the 3-D perspective display would perceive the highest mental workload (due to unfamiliarity and the need to scroll up and down) when compared to the participants viewing the 2-D displays or the participants viewing the tabular display. Participants viewing the 2-D display could perceive a higher mental workload than the participants viewing the tabular display (due to the need to scroll up and down). This study adopted the four statements developed by Reid and Nygren (1988) to number that indicates the extent to which they agree with each of the four statements. The scale numbers are: 1= strongly disagree, 2 = moderately disagree, 3 = slightly disagree, 4 = neither agree nor disagree, 5 = slightly agree, 6 = moderately agree, and 7= strongly agree (see Table 9). The four statements are: (1) very little mental effort or concentration was required to complete the tasks, (2) the tasks performed were almost automatic, requiring little or no attention, (3) the tasks were very complex and required total attention, and (4) extensive mental effort and concentration was necessary in the tasks. Reid and Nygren The score on mental workload is the covariate measure used to test for significant correlations with the dependent variables.

PAGE 84

69 3.7.5 Demographic Data In addition to the variables already discussed, participants were asked to provide information about their student status, undergraduate GPA, SAT score, number of years of full time working experience, part time working hours per week (if not working full time), working experience in accounting related jobs, and highest level of education (see Table 8). The data on these additional demographic que st ions will be tested as possible covariate measures. Table 8 Demographic Questions Your age is 18 22 23 27 29 32 33 37 38 42 43 47 48 50 Your Gender is Male Female Student Status Freshman Sophomore Junior Senior Graduate Student What was your undergraduate overall GPA? What was your SAT score? Number of years of full time work experience? If you do not work full time now, number of hours you work part time now? Work experience in accounting related jobs (for example: bookkeep ing or auditing)? Full time Part time Both full time and part time None Highest level of education you already achieved? degree

PAGE 85

70 3.8 Post Hoc Analysis (Survey Questions) As indicated earlier, the 3-D perspective display of DuPont analysis is a newly created display format of financial ratios that has never been empirically tested. For this reason, the study adopted six survey questions from Fuller, Murthy, and Schafer (2007) to elic ns on the usefulness and ease of use of their randomly assigned display format. Participants were asked to select the scale number from 1 to 7 that indicates the extent to which they agree with each of the six questions. The six questions are: (1) using th e tables (graphs) was frustrating, (2) the tables (graphs) displayed the task information in a readable format, (3) I found the tables (graphs) useful in how they presented the data for decision making, (4) the tables (graphs) helped me to understand the task data to make a better decision, (5) the tables (graphs) fit the way I needed to view the task information to make better decisions, and (6) overall, I am satisfied with the tables (graphs) in providing the information I needed to complete this task (see Table 9). 3.9 Student Participants Student participants are used in the study for several reasons. DuPont analysis is taught in every principles of accounting class. Wright (2007) demonstrated that students can be good surrogates for real world auditors, as task-specific academic instruction and practice is a good substitute for audit working experience, at least for a task such as evaluating loan collectability. Participants in this study learn and practice DuPont analysis in the classroom, while participants in Wright (2007) also learned about the loan collectability judgment in the classroom. In discussing the use of students as participants

PAGE 86

71 in behavioral auditing research, Peecher and Solomon (2001) contend that it is inappropriate only when theory or prior research indicates that experience interacts with a factor of interest in the study. Given the paucity of research on the effects of multidimensional displays in accounting behavioral research and the lack of widespread use of such displays in practice, it is unknown whether experience would interact with display type. Accordingly, the use of students as participants is deemed appropriate at least in the initial phases of this research stream.

PAGE 87

72 Table 9 Mental Workload Questions and Survey Questions Please use the following scale as the index for your responses. 1 2 3 4 5 6 7 strongly disagree moderately disagree slightly disagree Neither agree nor disagree slightly agree moderately agree strongly agree Please select the scale number which indic ates the extent to which you agree with each of the following statements. There is no right or wrong answer to these statements Using the tables (graphs) was frustrating. Very little mental effort or concentration was required to complete tasks. Tasks p erformed were almost automatic, requiring little or no attention. Tasks were very complex and required total attention. Extensive mental effort and concentration was necessary in tasks. The tables (graphs) displayed the task information in a readable f ormat. I found the tables (graphs) useful in how they presented the data for decision m aking. The tables (graphs) helped me understand the task data to make a better decision. The tables (graphs) fit the way I needed to view the task information to mak e a better decision. Overall, I am satisfied with the tables (graphs) in providing the information I needed to complete the task.

PAGE 88

73 3.10 Manipulation Checks responses were in th conditions. Different manipulation questions were designed for the treatment conditions of display format and task type. Another manipulation question was designed to test ost experiment knowledge of DuPont analysis. Participants were asked to answer three manipulation questions by selecting the choices of true or false. To test the manipulation of display formats, participants viewing the tabular display were asked whether: -D line display were asked whether: f participants viewed the 2Finally participants viewing the 3se To test the manipulation of tasks, participants performing the trend analysis task anies companies with the same ROE also have the same turnover, profitability, and leverage able 10).

PAGE 89

74 Lastly, due to the fact that the study has many variables, it is beneficial to have a table summarizing the definition and symbols of all the variables to be used in statistical analysis (see Table 11).

PAGE 90

75 Table 10 Manipulation Check Questions Tab ular Display Trend Analysis Task Pattern Recognition Task The tables you see in the experiment, have data points of zero The tables you see in the experiment, have data points of zero Return on Equity is the SUM of turnover, profitability, and leverage ratios Return on Equity is the SUM of turnover, profitability, and leverage ratios Within the context of this experiment companies can have negative ROE Within the context of this experiment, companies with the same ROE also have the same turnover, profit ability, and leverage ratios. 2 D L ine (Bar) Display Trend Analysis Task Pattern Recognition Task The line graphs you see in this experiment, have their axis started from zero The bar charts you see in this experiment, have their axis started from zero Return on Equity is the SUM of turnover, profitability, and leverage ratios Return on Equity is the SUM of turnover, profitability, and leverage ratios Within the context of this experiment companies can have negative ROE Within the context of this exper iment, companies with the same ROE also have the same turnover, profitability, and leverage ratios. 3 D Display Trend Analysis Task Pattern Recognition Task The graphs you see in this experiment, have their axis started from zero The graphs you see in t his experiment, have their axis started from zero Return on Equity is the SUM of turnover, profitability, and leverage ratios Return on Equity is the SUM of turnover, profitability, and leverage ratios Within the context of this experiment companies can have negative ROE Within the context of this experiment, companies with the same ROE also have the same turnover, profitability, and leverage ratios.

PAGE 91

76 Table 11 Definition of Variables Used in Statistical Analysis Variable Variable Symbol Definition Refer To Trend Analysis Task Question one and the Time Spent on Task Question one TAQ1 TSTAQ1 Scores and time on Trend Analysis Task between year s Table 3 and 6 Trend Analysis Task Question two and the Time Spe nt on Task Question two TAQ2 TSTAQ2 Scores and time on Trend Analysis Task Table 3 and 6 Trend Analysis Task Question three and the Time Spent on Task Question three TAQ3 TSTAQ3 Scores and time on Trend Analysis Task year 5 (Bubble 5), what would be year 6 ROE if each of the variables of ROE in Table 3 and 6 Trend Analysis Task Question four part a TAQ4a Scores on Trend An alysis Task Question Table 3 and 6 Trend Analysis Task Question four part b TAQ4b Scores on Trend Analysis Task Question leverage for the y Table 3 and 6 Trend Analysis Task Question four part c TAQ4c Scores on Trend Analysis Task Question Table 3 and 6 Trend Analysis Task Question four par t d TAQ4d Scores on Trend Analysis Task Question turnover, leverage and profitability for the years 1, 2, 4 and 5, and use them to Table 3 and 6 Time Spent on Task Question four part a, b, c and d. TSTAQ4 The time spent on answering Task Question four, parts a, b, c and d. Table 3 and 6 Trend Analysis Task Question five and the Time Spent on Task Question five TAQ5 TSTAQ5 Scores and time on Trend Analysis Task fferences Table 3 and 6 Trend Analysis Task Question six and the Time Spent on Task Question six TAQ6 TSTAQ6 Scores and time on Trend Analysis Task Table 3 and 6

PAGE 92

77 Table 11 (Continued) Definition of Variables Used in Statistical Analysis Variable Variable Symbol Definition Refer To Pattern Recognition Task Question one and the Time Spent on Task Question one PRQ1 TSPRQ1 Scores and time on Pattern Recognit ion differences between companies 1 and Table 4 and 7 Pattern Recognition Task Question two and the Time Spent on Task Question two PRQ2 TSPRQ2 Scores and time on Pattern Recognition e companies 1 through 6 into 2 groups Table 4 and 7 Pattern Recognition Task Question three and the Time Spent on Task Question three PRQ3 TSPRQ3 Scores and time on Pattern Recognition i ncludes companies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group two, what are the patterns of the financial ratios you are Table 4 and 7 Pattern Recognition Task Question four and the Time Spent on Tas k Question four PRQ4 TSPRQ4 Scores and time on Pattern Recognition select a company solely because of a single variable, for example higher profitability. Comparatively, if it is better to have higher profitability, h igher turnover but lower leverage at the same Table 4 and 7 Pattern Recognition Task Question five & the Time Spent on Task Question five PRQ5 TSPRQ5 Scores and time on Pattern Recognition re the differences between companies 4 and Table 4 and 7

PAGE 93

78 Table 11 (Continued) Definition of Variables Used in Statistical Analysis Variable Variable Symbol Definition Refer To Pattern Recognition Task Question six and the Time Spent on Task Questio n six PRQ6 TSPRQ6 Scores and time on Pattern Recognition companies 1, 3, 4 and 6, and group two includes companies 2 and 5. Compared to group one what are the patterns of the financial ratios you are seeing in group Table 4 and 7 Practice Question one and the Time Spent on Practice Question one PQ1 TSPQ1 Scores and time on Practice Question of (apartment 2) the apartment rented in Table 5 Practice Question two and the Time Spent on Practice Question two PQ2 TSPQ2 Scores and time on Practice Question (apartments 5 and 6) apartments rented in year s Table 5 Practice Question three and the Time Spent on Practice Qu estion three PQ3 TSPQ3 Scores and time on Practice Question factors of (apartment 4) the apartment Table 5 Practice Question four and the Time Spent on Practice Question four PQ4 TSPQ4 Scores and time on Practice Question (apartments 2 and 4) apartments rented in year s Table 5 Practice Question five and the Time Spent on Practice Question five PQ5 TSPQ5 Scores and time on Practice Question turnover is 2, profitability ratio is 5% Table 5

PAGE 94

79 Table 11 (Continued) Definition of Variables Used in Statistical Analysis Variable Variable Symbol Definition Refer To Practice Question six and the T ime Spent on Practice Question six PQ6 TSPQ6 Scores and time on Practice Question turnover ratio, profitability ratio and Table 5 Score on Mental Rotations Test SMRT Score on Mental Rotations Test, whi spatial ability Section 3.6.2 Time Spent on Mental Rotations Test TSMRT Time in seconds spent on the Mental Rotations Test Section 3.6.2 Age AGE 22, 2 = 23 27, 3 = 28 32, 4 = 33 37, 5 = 38 42, 6 = 43 47, 7 = 48 50). Table 8 Gender GEN Female) Table 8 Student Status SS university ( 1= Freshman, 2 = Sophomore, 3 = Junior, 4 = Senior, 5= Graduate Student) Table 8 GPA GPA Pa Table 8 SAT SAT Table 8 Full Time Working Experience FTWE Number of years of full time working experience Table 8 Part Time Working Hours PTH Number of part time working hours per week Table 8 Acc ounting Related Working Experience ARWE Accounting related working experience (1= full time, 2= part time 3= full time and part time, 4=none) Table 8

PAGE 95

80 Table 11 (Continued) Definition of Variables Used in Statistical Analysis Variable Variable Symbol Def inition Refer To Highest Level of Education HE education already achieved Table 8 Survey Question one S1 Score on the Survey Question one, raphs) was Table 9 Survey Question two S2 Score on the Survey Question two, concentration was required to complete t Table 9 Survey Question three S3 Score on the Survey Question three, were almost automatic, requiring little or no Table 9 Survey Question four S4 Score on the Survey Question four, Table 9 Survey Question five S5 Score on the Survey Question five, xtensive mental effort and concentration was necessary on Table 9 Survey Question six S6 Score on the Survey Question six, task information in a readable Table 9 Survey Question seven S7 Score on th e Survey Question Seven, how they presented the data for Table 9

PAGE 96

81 Table 11 (Continued) Definition of Variables Used in Statistical Analysis Variable Variable Symbol Definition Refer To Survey Que stion eight S8 Score on the Survey Question eight, understand the task data to make a Table 9 Survey Question nine S9 Score on the Survey Question nine, needed to view t he task information Table 9 Survey Question ten S10 Score on the Survey Question ten, tables (graphs) in providing the information I needed to complete the Table 9 Mental Workload MW A verage score on the Survey Question two, three, four and five. (reverse coding on S2 and S3) Table 9 Manipulation Question one M1 The tables (graphs) you see in the experiment, have data points of zero Table 10 Manipulation Question two M2 Return on Eq uity is the SUM of turnover, profitability, and leverage ratios Table 10 Manipulation Question three M3 Within the context of this experiment, companies can have negative ROE or w ithin the context of this experiment, companies with the same ROE also hav e the same turnover, profitability, and leverage ratios. Table 10

PAGE 97

82 Chapter 4 : Pilot Study 4.1 Research Design of the Pilot Study The research design for the pilot study is as is described in Chapter 3, with the following exceptions. For the pilot study, task type is manipulated within-subjects rather than between-subjects. The order of tasks is balanced such that half the participants in each treatment group perform the trend analysis task first and the pattern recognition task second, and vice versa. A total of eighty-six undergraduate business school students participated in a pilot study designed to test the experimental materials and determine whether the experimental manipulations had the intended effect. Each participant was randomly assigned to one of the three treatment conditions. Each participant answered two practice questions by reading from either a table, or a set of 2-D bar charts, or a single 3-D perspective display. After participants answered the two practice questions assessing apartment desirability, DuPont analysis was explained to them using the return on equity of Motorola and Nokia in 1997 as examples. These examples were provided to refresh participants regarding DuPont analysis, as it was expected that participants had already learned the concept of DuPont analysis in the principles of accounting class. Neither a practice question on the calculation of the return on equity nor a practice question about the concept of the DuPont analysis had been provided to the participants. Participants in the 3-D treatment

PAGE 98

83 group viewed an additional page of training demonstrating that bubbles can also be used to display how data changes over time (i.e., from year to year). 4.2 Results of the Pilot Study 4.2.1 Descriptive Statistics, Test of Assumptions and Outliers In terms of effectiveness or accuracy, for both the practice questions one and two, participants viewing the tabular display scored the highest, then followed by those participants viewing the 2-D displays, while those participants viewing the 3D perspective display scored the lowest (see Table 12). In term of effectiveness or accuracy, participants viewing the 3-D perspective display scored the lowest in all the six questions of the trend analysis task. Participants viewing the tabular display scored the highest in the first question (which asked what the data differences were between year 1 and 4), the third question (which asked what the ROE would be in year 6 if each of the variables comprising ROE in year 5 doubled), and average of turnover, profitability, and leverage for the years 1, 2, 4 and 5) of the trend analysis task. While those participants viewing the 2-D display scored the highest in the second question (which asked what was occurring in the data when going from year 2 to average turnover for the years 1, 2, 4, and 5), the fourth qu participants to estimate the average profitability for the years 1, 2, 4, and 5), the fourth

PAGE 99

84 1, 2, 4, and 5), and the sixth question (which asked what was occurring in the data when going from year 1 to 2 to 3) of the trend analysis task (see Table 12). In terms of efficiency or time used (in seconds) by each participant when answering each of the question of the trend analysis task, those participants viewing the 2-D displays were the most efficient or used the least amount of time in answering questions, while those participants viewing the 3-D perspective display used the most amount of time in answering questions (see Table 12). For the pattern recognition task, in term of effectiveness or accuracy, participants viewing the 3-D perspective display scored the highest in the second question (which asked participants to separate companies one through six into two groups), the third question (which participants to describe the pattern of financial ratios in group one), and the fourth question of the (which participants to select one of the six companies if the goal is to have high profitability, high turnover, but low leverage at the same time), while those participants viewing the tabular display scored the lowest in the mentioned above three questions. Those participants viewing the 2-D displays scored the highest in the fifth question (which asked what were the data differences between companies 4 and 6), and question six (which asked participants to describe the pattern of financial ratios in group two) of the pattern recognition task, while those participants viewing the 3D perspective display scored the lowest in the mentioned above two questions. Finally, those participants viewing the tabular display scored the highest in the first question of the pattern recognition task (which asked what data difference were between companies 1 and 6) with those participants viewing the 2-D displays scored the lowest in this question (see Table 12).

PAGE 100

85 In terms of efficiency or time used (in seconds) by each participant when answering each of the question of the pattern recognition task, those participants viewing the 2-D displays were the most efficient or used the least amount of time in answering five out of the six questions of the pattern recognition task. While those participants viewing the tabular display or those participants viewing the 3-D perspective display used the most amount of time in answering three out of the six questions of the pattern recognition task, respectively (see Table 12). The scores on the Mental Rotations Test between tabular, 2-D and 3-D treatment groups were 21, 22, and 23 points, respectively. T-testing shows that there is no significant difference in spatial ability between treatment groups. Participants of each of the three treatment groups had similar GPA and scores of SAT. Participants of the tabular display treatment group had the longest full time working experience, while those participants of the 2-D displays treatment group had the shortest. Participants of the 2D displays treatment group had longest part time working hours while those participants of the tabular display treatment group had the shortest (see Table 12). Two blind coders coded the responses of each participant. Descriptive statistics (see Table 12) reveal that there is considerable variability in the data. Tests of influential observations and outliers, at two standard deviations, were performed on both the coded responses and response time. The time spent by a participant in the 3-D treatment condition in selecting a choice from a given template, which best describes the patterns of tance greater than one. The aforementioned data point was subsequently dropped.

PAGE 101

86 Tests of normality and homogeneity of variance were also performed on both the coded responses and response time. The coded responses and response time were not normally distributed. Since univariate ANCOVA analysis is robust to data that is not normally distributed no adjustments were made to the data. Multivariate MANCOVA w as used first to test whether the manipulation of the presentation formats had significance results. If the manipulation of the presentation formats had significant result, ANCOVA was then used to analysis the univariate results.

PAGE 102

87 Table 12 Pilot Study Descriptive Statistics Panel A: Mean (Standard Deviation) and Range of Practice Questions. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Practice question one: What are the v alues of the factors of apartment one? 3.222 (1.086) 0.00 0 to 4.00 0 3.00 0 (1.389) 0.00 0 to 4.00 0 2.966 (0.680) 0.00 0 to 4.0 0 0 Practice question two: What are the diffe rences between apartments five and six? 3.444 (0.891) 0.000 to 4.000 3.400 (0.932) 0.000 to 4.000 3.034 (0.778) 1.000 to 4.000 Panel B: Mean (Standard Deviation) and Range of Trend Analysis Task. Tabular Displa y (n=27) 2 D Displays (n=30) 3 D Display ( n=29) Question one What are the differences between year 1 and year 4? 3.481 (0.752) 1.000 to 4.000 3.400 (0.968) 0.000 to 4.000 2.793 (0.726) 1.00 0 to 4.00 0 Time (seconds) spent on question one 132 (74) 53 to 331 127 (71) 39 to 390 148 (77) 49 to 3 18 Question two What is happening as you go from year 2 to year 3 to year 4? 5.777 (2.241) 1.000 to 8.000 5.866 (2.849) 0.000 to 8.000 3.862 ( 2.199 ) 0 .00 0 to 8 0 00 Time (seconds) spent on question two 13 7 (91) 35 to 415 12 9 (61) 36 to 258 1 55 (128) 26 to 674

PAGE 103

88 Table 12 (Continued) Pilot Study Descriptive Statistics Panel B: Mean (Standard Deviation) and Range of Trend Analysis Task. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Question three Based on the ROE of year 5, what w ould be year 6 ROE if each of the variables of ROE in year 5 had doubled? 4.148 (2.918) 0.000 to 10.000 3.200 (1.864) 1.000 to 10.000 3.4483 ( 2.30816 ) 0 .00 0 to 10 0 00 Time (seconds) spent on question three 87 (48) 20 to 250 88 (52) 27 to 193 86 (55) 33 to 253 Question four part a Est imate the average of turnover for the years 1, 2, 4, and 5 6.814 (4.582) 0.000 to 12.000 9.100 (3.467) 0.000 to 12.000 8.413 ( 4.939 ) 0 .00 0 to 12 0 00 Question four part b Est imate the average of leverage for the years 1 2, 4, and 5 3.481 (4.846) 0.000 to 12.000 3.500 (4,761) 0.000 to 10.000 2.551 (4.264) 0.00 0 to 10. 0 00 Question four part c Est imate the average of profitability for the years 1, 2, 4, and 5 3.111 (4.870) 0.000 to 12.000 5.400 (3.891) 0.000 to 11.000 3 .448 3.869 0.000 to 11.000 Question four part d Est imate the average of turnover, leverage and profitability for the years 1, 2, 4, and 5, and use them to calculate a new ROE 1 .111 (2.750) 0.000 to 9.000 0.533 (2.029) 0.000 to 8.000 1.000 2.604 0.000 to 9.000 Time (seconds) spent on question four 205 (123) 29 to 595 172 (78) 49 to 364 83 6 (149) 47 to 836

PAGE 104

89 Table 12 (Continued) Pilot Study Descriptive Statistics Panel B: Mean (Standard Deviation) and Range of Trend Analysis Task. Tabular Displa y (n=27) 2 D Displays (n=30) 3 D Display (n=29) Question five What are the differences between years 2 and 4? Please answer as accurately and as fast as possible. 3.407 (0.747) 2.000 to 4.000 3.866 (0.434) 2.000 to 4.000 3.620 ( 0.621 ) 2 .00 0 to 4 0 00 Time (seconds) spent on question five 34 (13) 21 to 79 32 (9) 19 to 54 50 (13) 27 to 86 Question six What is happening as you go from year 1 to year 2 to year 3? Please answer as accurately and as fast as possible. 7.222 (1.368) 3.000 to 8.000 7.666 (0.660) 5.000 to 8.000 6.931 ( 1.412 ) 3 .00 0 to 8 0 00 Time (seconds) spent on q uestion six 69 (27) 31 to 164 54 (17) 29 to 101 100 (38) 15 to 205 Panel C : Mean (Standard Deviation) and Range of Pattern Recognition Task. Tabular Display (n=27) 2 D Displays (n=3 0) 3 D Display (n=29) Question One What are the differences between companies 1 and 6? 2.777 (0.640) 0.000 to 3.000 2.566 (0.897) 0.000 to 3.000 2.689 ( 0.603 ) 1 .00 0 to 3 0 00

PAGE 105

90 Table 12 (Continued) Pilot Study Descriptive Statistics Panel C : Mean (Standard Deviation) and Range of Pattern Recognition Task. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Time (seconds) spent on question one 112 (58) 46 to 317 111 (45) 42 to 192 163 (82) 71 to 365 Question two Please separate companies 1 through 6 into 2 groups based on similar characteristics. 4.555 (0.697) 3.000 to 6.000 4.800 (0.996) 3.000 to 6.000 5.069 ( 0.923 ) 3 .00 0 to 6 0 00 Time (seconds) spent on question two 91 (52) 38 to 232 70 (30) 26 to 183 83 (32) 37 to 164 Question three G roup one includes companies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group two, what are the patterns of the financial ratios you are seeing in group one? 1.777 (1.012) 0.000 to 3.000 2.333 (0.802) 1.000 to 3.000 2.517 ( 0.82 8 ) 0 .00 0 to 3 0 00 Time (seconds) spent on question three 121 (46) 26 to 251 97 (44) 37 to 213 105 (42) 54 to 243

PAGE 106

91 Table 12 (Continued) Pilot Study Descriptive Statistics Panel C : Mean (Standard Deviation) and Range of Pattern Recognition Task. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Question four Assuming you cannot select a company solely because of a single variable, for example higher profitability. Comparatively, if it is better to have a higher profitability, higher turnov er, but lower leverage at the same time, which company will you select? 0.000 (0.000) 0.000 to 0.000 0.200 (0.406) 0.000 to 1.000 0.724 ( 0.454 ) 0 .00 0 to 1 0 00 Time (seconds) spent on question four 67 (22) 20 to 113 65 (28) 28 to 127 55 (20) 21 to 117 Qu estion five What are the differences between companies 4 and 6? Please answer as accurately and as fast as possible. 2.851 (0.362) 2.000 to 3.000 3.00 (0.000 3.000 to 3.000 2.689 ( 0.603 ) 1 .00 0 to 3 0 00 Time (seconds) spent on question five 30 (11) 16 to 65 27 (9) 15 to 57 43 (19) 21 to 122

PAGE 107

92 Table 12 (Continued) Pilot Study Descriptive Statistics Panel C : Mean (Standard Deviation) and Range of Pattern Recognition Task. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Question six Group one includes companies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group one, what are the patterns of the financial ratios you are seeing in group two? Please answer as accurately and as fast as possible. 2.703 (0.724) 0.000 to 3 .000 2.833 (0.592) 0.000 to 3.000 2.655 ( 0.768 ) 0 .00 0 to 3 0 00 Time (seconds) spent on question six 46 (20) 20 to 101 35 (12) 19 to 68 55 (32) 22 to 168 Panel D : Mean (Standard Deviation) and Range of Mental Rotations Test. Tabular Display (n=27) 2 D Displays (n=30) 3 D Display (n=29) Score on Mental Rotations Test 21.592 (10.123) 2.000 to 36.000 22.366 (12.397) 2.000 to 40.000 23.689 ( 9.849 ) 4 .00 0 to 0 .00 0 Time (seconds) spent on Mental Rotation Test 820 (365) 333 to 2031 657 (201) 195 to 1197 771 (263) 306 to 1222

PAGE 108

93 Table 12 (Continued) Pilot Study Descriptive Statistics Panel E : Demographic Mean and Range. Tabular Display ( n=27 ) 2 D Displays ( n=30 ) 3 D Display ( n=29 ) Male/Female n= 9/n = 18 n = 14/n= 16 n = 8/ n = 21 GPA 3.2048 2.27 0 4. 00 0 3.2953 2.50 3.96 3.2583 2.00 3.87 SAT Score 1 194 980 1750 (n=23) 11 58 900 1470 (n=29) 11 49 68 0 1800 (n=28) Full Time Working Experience in Years 3.7778 0.00 0 20.0 0 0 2.2167 0.00 28.00 2.8793 0.00 27.00 Part Time Working Hours Per Week 15.4815 0.0 0 0 40. 0 00 18.667 0.00 0 52. 0 00 16.5862 0.00 0 40. 0 00

PAGE 109

94 4.2.2 Hypotheses Testing of the Trend Analysis Task Results of hypotheses H1a, H1b, H1c and H1d are shown in Tables 13 to 21. Multivariate MANCOVA was first used to test for the significance of the manipulation variable Treatment. If the manipulation variable Treatment was significant, ANCOVA was then conducted to analysis the univariate results. Covariates included in the analysis are the scores on practice questions one and two, gender, age, score on the Mental Rotations Test, time spent on the Mental Rotations Test, and the task order performed by each participant (for example the trend analysis task first and the pattern recognition task second, and vice versa). The following paragraphs report significant results with tables on top of narration. 4.2.2.1 Results of Hypothesis H1a. Hypothesis H1a has four dependent measures between years 1 and 4; the score on the second question, which asked what participants perceived to be occurring in the data when going from year 2 to year 3 and year 4; the score on the fifth question, which asked participants to select from a template to indicate the differences in data between year 2 and 4; and the score on the sixth question, which asked participants to select from a template to indicate changes in data when going from year 1 to year 2 to year 3. The mean results for all dependent variables are hypothesized in H1a. In constructing the models to test H1a, all covariates mentioned in section 4.2.2 were included in the model along with the manipulated variable Treatment.

PAGE 110

95 Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1a, a MANCOVA analysis was conducted. As shown (Table 13), the overall F-statistic for the manipulated variable Treatment is significant (p = 0.003) using significant results allow for analysis of the univariate results which are provided on Table 14 to 16. Table 13 Multivariate Tests of H1a. Variables Multivariate Test Value F stat |p value| Intercept 0.566 23.173 <0.001 PQ1 0.025 0.450 0.772 PQ2 0.343 9.249 <0.001 AGE 0.097 1.907 0.119 GEN 0.087 1.683 0.164 SMRT 0.072 1.373 0.252 TSMRT 0.025 0.460 0.765 Treatment 0.296 3.129 0.003 Rank ace 0.032 0.592 0.667 Treatment*Rank 0.167 1.639 0.119 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks. The first dependent variable used to test hypothesis H1a was the first question in the trend analysis task, which asked the participants which asked what the data differences were between years 1 and 4. Table 14, Panel A indicates that on average those participants viewing the tabular display (mean score 3.481) were the most accurate (had the highest score) on this first trend analysis task. However, those participants

PAGE 111

96 viewing a 3-D perspective display (mean score 2.793) were 18% less accurate than those viewing the 2-D display (mean score 3.400). Table 14, Panel A suggests that score on practice question two is significantly (p < 0.001) associated with the accuracy of the participants in describing what the data difference were between years 1 and 4. As expected, the results also suggest that the manipulation of the presentation formats (Treatment) has a significant (p = 0.003) main effect on the accuracy of the participants in this trend analysis task. A paired comparison test was conducted to determine if the participants receiving the 2-D Treatment were more effective or accurate than those participants receiving the Tabular or 3D Treatment. Results revealed that the participants viewing the 2-D displays were significantly (p = 0.012) more accurate than participants viewing the 3D perspective display (see Table 14, Panel B ). There was no significant difference in the effectiveness or accuracy between participants viewing the 2-D displays and participants viewing the tabular display. Thus, the paired comparison tests provide partial support for hypothesis H1a.

PAGE 112

97 Table 14 Pilot Test Results of H1a ANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Score on the Task and ANCOVA Results Using Score as the Dependent Variable. Actual Mean ANCOVA Adjusted Mean* Tabular Display (n = 27) 3.481 3.404 2 D Displays (n = 30) 3.400 3.368 3 D Perspective Display (n = 29) 2.793 2.891 Source of Variation Type III SS DF Mean Square F stat P value* Corrected Model 35.342 11 3.213 8.070 <0.001 Intercept 1.182 1 1.182 2.968 0.089 PQ1 0.077 1 0.077 0.194 0.661 PQ2 14.099 1 14.099 35.413 <0.001 AGE 0.003 1 0.003 0.00 7 0.931 GEN 0.923 1 0.923 2.317 0.132 SMRT 1.266 1 1.266 3.181 0.079 TSMRT 0.009 1 0.009 0.022 0.884 Treatment 4.233 2 2.116 5.316 0.003 Rank 0.005 1 0.005 0.012 0.915 Treatment*Rank 2.020 2 1.010 2.537 0.086 Error 29.460 74 0.398 Total 957.000 8 6 Corrected Total 64.802 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.478. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H1a (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 2 D Displays Tabular Display 3 D Display 0.036 0.477 0.176 0.176 0.500 0.012 *p -values are one-tail.

PAGE 113

98 The second dependent variable used to test hypothesis H1a was the second question in the trend analysis task, which asked the participants to write short sentences describing what they perceive to be occurring in the data when going from year 2 to year 3 and year 4. Table 15, Panel A indicates that on average those participants viewing the 2-D displays (mean score 5.866) were the most accurate (had the highest score) on this first trend analysis task. Those participants viewing a 3-D perspective display (means score 3.862) were 35% less accurate than those viewing the 2-D display (mean score 5.866). Those participants viewing the tabular display (mean score 5.777) were 2% less accurate than those viewing the 2-D display (mean score 5.866). Table 15, Panel A suggests that, as expected, the results suggest that the manipulation of the presentation formats (Treatment) has a significant (p = 0.003) main effect on the accuracy of the participants in this trend analysis task. A paired comparison test was conducted to determine if the participants receiving the 2-D Treatment were more effective or accurate than those participants receiving the Tabular or 3D Treatment. Results revealed that the participants viewing the 2-D displays were significantly (p = 0.005) more accurate than participants viewing the 3D perspective display (see Table 15, Panel B ). There was no significant difference in the effectiveness or accuracy between participants viewing the 2-D displays and participants viewing the tabular display. Thus, the paired comparison tests provide partial support for hypothesis H1a.

PAGE 114

99 Table 15 Pilot Test Results of H1a ANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Score on the Task and ANCOVA Results Using Score as the Dependent Variable. Actual Mean ANCOVA Adjusted Mean Tabular Display (n = 27) 5.777 5.6 56 2 D Displays (n = 30) 5.866 5. 8 9 7 3 D Perspective Display (n = 29) 3.862 3.895 Sour ce of Variation Type III SS DF Mean Square F stat p value* Corrected Model 152.995 11 13.909 2.435 0. 012 Intercept 4.617 1 4.617 0.808 0. 372 PQ1 0.692 1 0.692 0.121 0. 729 PQ2 14.800 1 14.800 2.591 0. 112 AGE 0.010 1 0.010 0.002 0.967 GEN 11.469 1 11 .469 2.008 0.161 SMRT 1.400 1 1.400 0.245 0.622 TSMRT 0.033 1 0.033 0.006 0.939 Treatment 61.640 2 30.823 5.396 0.00 3 Rank 9.663 1 9.663 1.692 0. 197 Treatment*Rank 22.671 2 11.336 1.984 0. 145 Error 422.726 74 5.713 Total 2868..000 86 Corrected Total 575.721 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0. 157. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H1a (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 2 D Displays Tabular Display 3 D Display 0.2 41 2 00 2 0.6 6 6 0.6 67 0 5 00 0.0 05 *p -values are one-tail.

PAGE 115

100 The third dependent variable used to test hypothesis H1a was the fifth question in the trend analysis task, which asked participants to indicate the differences between years 2 and 4 by selecting choices from a given template. Table16, Panel A shows that those participants viewing the 2-D displays (3.866) were on average the most accurate (had the highest score) on this trend analysis task. Those participants viewing a tabular display (mean score 3.407) were 12% less accurate than those viewing the 2-D display (mean score 3.866). Those participants viewing the 3-D perspective display (mean score 3.620) were 7% less accurate than those viewing the 2-D display (mean score 3.866). Table 16, Panel A suggests that, as expected, the manipulation of the presentation formats (Treatment) has a significant (p = 0.009) main effect on the accuracy of the participants in describing the differences between years 2 and 4. A paired comparison test was conducted to determine if the participants receiving the 2-D Treatment were more effective or accurate than those participants receiving the Tabular or 3-D Treatment. Results revealed that the participants viewing the 2-D displays were significantly (p = 0.007) more accurate than participants viewing the Tabular display (see Table 16, Panel B ). There was no significant difference in the effectiveness or accuracy between participants viewing the 2-D displays and participants viewing the 3-D perspective display. Thus, the paired comparison tests provide partial support for hypothesis H1a.

PAGE 116

101 Table 16 Pilot Test Results of H1a ANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Score on the Task and ANCOVA Results Using Score as the Dependent Variable. Actua l Mean ANCOVA Adjusted Mean* Tabular Display (n = 27) 3.407 3.380 2 D Displays (n = 30) 3.866 3.858 3 D Perspective Display (n = 29) 3.620 3.662 Source of Variation Type III SS DF Mean Square F stat p value* Corrected Model 7.645 11 0.695 1.964 0.044 Intercept 21.669 1 21.669 61.246 <0.001 PQ1 0.643 1 0.643 1.818 0.182 PQ2 0.341 1 0.341 0.963 0.330 AGE 0.729 1 0.729 2.059 0.155 GEN 0.133 1 0.133 0.375 0.542 SMRT 0.014 1 0.014 0.041 0.840 TSMRT 0.450 1 0.450 1.272 0.263 Treatment 2.981 2 1 .490 4.212 0.009 Rank 0.044 1 0.044 0.124 0.726 Treatment*Rank 1.206 2 0.603 1.704 0.189 Error 26.181 74 0.354 Total 1173.00 86 Corrected Total 33.826 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.111. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H1a (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 2 D Displays Tabular Display 3 D Display 0.478 0.196 0.166 0.166 0.007 0.362 *p -values are one-tail.

PAGE 117

102 The fourth dependent variable used to test hypothesis H1a was the sixth question in the trend analysis task, which asked participants to select from a template to indicate changes in data when going from year 1 to year 2 to year 3. Those participants viewing the 2-D displays (mean score 7.666) were on average the most accurate (had the highest score) on this trend analysis task. Those participants viewing the tabular display (mean score 7.222) were 6% less accurate than those viewing the 2-D display (mean score 7.666). Those participants viewing a 3-D perspective display (mean score 6.931) were 10% less accurate than those viewing the 2-D display (mean score 7.666). Results of the ANCOVA analysis show that the covariates score on practice question two (p = 0.010), age of participants (p = 0.015), and gender of participants (p=0.040) are significantly associated with the accuracy of the participants in selecting from a template to indicate changes in data when going from year 1 to year 2 to year 3. As expected, the results also suggest that the manipulation of the presentation formats (Treatment) has a significant (p = 0.040) main effect on the accuracy of the participants in this trend analysis task. A paired comparison test was conducted to determine if the participants receiving the 2-D Treatment were more effective or accurate than those participants receiving the Tabular or 3-D Treatment. Results revealed that the participants viewing the 2-D displays were not significantly (p = 0.051) more accurate than participants viewing the 3D perspective display. There was no significant difference in the effectiveness or accuracy between participants viewing the 2-D displays and participants viewing the tabular display.

PAGE 118

103 4.2.2.2 Results of Hypothesis H1b. Hypothesis H1b has four dependent measures data differences were between years 1 and 4, the time spent in seconds by each participant in describing what was occurring in the data when going from year 2 to year 3 and year 4, the time spent in seconds by each participant when selecting from a template the differences in data between years 2 and 4, and the time spent in seconds by each participant when selecting from a template changes in data when going from year 1 to year 2 to year 3. The mean results for all dependent variables are hypothesized in H1b. In constructing the models to test H1b, all covariates mentioned in section 4.2.2 were included in the model along with the manipulated variable Treatment. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1b, a MANCOVA analysis was conducted. As shown (Table 17), the overall Fsignificant results allow for analysis of the univariate results which are provided on Table 18 to 19. The first dependent variable used to test hypothesis H1b was the time spent in seconds by each participant when answering what the data differences were between years 1 and 4. On average those participants viewing the 2-D displays (mean seconds 127) were the most efficient or used the least amount of time (in seconds) on this first trend analysis task. Those participants viewing a 3-D perspective display (mean seconds 148) used 16% more time (in seconds), than those viewing the 2-D displays (mean

PAGE 119

104 seconds 127). Those participants viewing the tabular display (mean seconds 132) used 4% more time (in seconds), than those participants viewing the 2-D displays (mean seconds 127). Results of the ANCOVA analysis show that age of the participants is significantly (p < 0.001) associated with the time spent by each participant in describing what the data differences were between years 1 and 4. However, the results suggest that the manipulation of the presentation formats (Treatment) do not have a significant (p = 0.118) main effect on the time spent by each participant in this trend analysis task. Since there is not a significant main effect a paired comparison test was not conducted. Table 17 Multivariate Tests of H1b. Variables Multivariate Test Value F stat |p value| Intercept 0.418 12.726 <0.001 PQ1 0.016 0.294 0.881 PQ2 0.102 2.025 0.100 AGE 0.183 3.979 0.006 GEN 0.031 0.575 0.682 SMRT 0.065 1.236 0.304 TSMRT 0.095 1.873 0.125 Treatment 0.431 4.941 <0.001 Rank 0.486 16.803 <0.001 Treatment*Rank 0.150 1.455 0.179 PQ1= the score on practice question 1; PQ2 = th e score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks.

PAGE 120

105 The second dependent variable used to test hypothesis H1b was the time spent in seconds by each participant when describing what was occurring in the data when going from year 2 to year 3 and year 4. On average those participants viewing the 2-D display (mean seconds 129) were the most efficient (used the least amount of time in seconds) on this trend analysis task. Those participants viewing a 3-D perspective display (mean seconds 155) used 20% more time (in seconds), than those viewing the 2-D displays (mean seconds 129). Those participants viewing the tabular display (mean seconds 137) used 6% more time (in seconds), than those participants viewing the 2-D displays (mean seconds 129). Results of the ANCOVA analysis show that age of the participant is significantly (p < 0.001) associated with the time spent by each participant in describing what was occurring in the data when going from year 2 to year 3 and year 4. However, the results suggest that the manipulation of the presentation formats (Treatment) do not have a significant (p = 0.254) main effect on the time spent by each participant in this trend analysis task. Since there is not a significant main effect a paired comparison test was not conducted. The third dependent variable used to test hypothesis H1b was the time spent in seconds by each participant when selecting from a template the differences in data between years 2 and 4. Table 18, Panel A indicates that on average those participants viewing the 2-D display (mean seconds 32) were the most efficient (used the least amount of time in seconds) on this trend analysis task. Those participants viewing a 3D perspective display (mean seconds 50) used 56% more time (in seconds), than those viewing the 2-D displays (mean seconds 32). Those participants viewing the tabular

PAGE 121

106 display (mean score 34) used 6% more time than those participants viewing the 2D displays (mean score 32). Table 18, Panel A suggests that score on the score on the Mental Rotation Test is significantly (p < 0.047) associated with the time spent by each participant when selecting from a template the differences in data between years 2 and 4. The results suggest that the manipulation of the presentation formats (Treatment) has a significant (p <0.001) main effect on the time spent by each participant in this trend analysis task. A paired comparison test was conducted to determine if the participants receiving the 2D Treatment were more efficient than those participants receiving the Tabular or 3D Treatment. Results revealed that the participants viewing the 2-D displays were significantly (p < 0.001) more efficient than participants viewing the 3-D perspective display (see Table 18, Panel B ). There was no significant difference in the efficiency between participants viewing the tabular display and participants viewing the 2D perspective display. Thus, the paired comparison tests provide partial support for hypothesis H1b.

PAGE 122

107 Table 18 Pilot Test Results of H1b ANCOVA Model on Efficiency (Time Spent) in Trend Analysis Task Tests of Between-Subjects Effects on Efficiency Panel A: Mean Time Spent on the Task and ANCOVA Results Using Time Spent as the Dependent Variable Actual Mean (Seconds) ANCOVA Adjusted Mean* (Seconds) Tabular Display (n = 27) 34.407 33.812 2 D Displays (n = 30) 32.000 32.298 3 D Perspective Display (n = 29) 50 .482 50.512 Source of Variation Type III SS DF Mean Square F stat p value* Corrected Model 7867.174 11 715.198 5.214 <0.001 Intercept 5777.610 1 5777.610 42.123 <0.001 PQ1 65.148 1 65.148 0.475 0.493 PQ2 33.177 1 33.177 0.242 0.624 AGE 39.769 1 39. 769 0.290 0.592 GEN 154.097 1 154.097 1.123 0.293 SMRT 557.712 1 557.712 4.066 0.047 TSMRT 130.348 1 130.348 0.950 0.333 Treatment 5274.726 2 2637.363 19.228 <0.001 Rank 756.492 1 756.492 5.515 0.022 Treatment*Rank 19.044 2 9.522 0.069 0.933 Error 1 0149.815 74 137.160 Total 148745.000 86 Corrected Total 18016.988 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.353. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the va riables. Panel B: Bonferroni Pairwise Comparisons for Test H1b (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 2 D Displays Tabular Display 3 D Display 1.514 18.214 3.262 3.267 0.500 <0.001 *p -values are one-tail

PAGE 123

108 The fourth dependent variable used to test hypothesis H1b was the time spent in seconds by each participant when selecting from a template changes in data when going from year 1 to year 2 to year 3. Table 19, Panel A indicates that on average those participants viewing the 2-D display (mean seconds 54) were the most efficient (used the least amount of time in seconds) on this trend analysis task. Those participants viewing a 3-D perspective display (mean seconds 100) used 85% more time (in seconds), than those viewing the 2-D displays (mean seconds 54). Those participants viewing the tabular display (means seconds 69) used 27% more time than those participants viewing the 2D displays (mean seconds 54). Table 19, Panel A suggests that the score on practice question two (p = 0.028) and the time spent on the Mental Rotation Test (p = 0.015) are significantly associated with the time spent by each participant in selecting from a template changes in data when going from year 1 to year 2 to year 3.The results suggest that the manipulation of the presentation formats (Treatment) has a significant (p <0.001) main effect on the time spent by each participant in this trend analysis task. A paired comparison test was conducted. Results revealed that the participants viewing the 2-D displays were significantly (p < 0.001) more efficient than participants viewing the 3-D perspective display (see Table 19, Panel B ). There was no significant difference in the efficiency between participants viewing the tabular display and participants viewing the 2D perspective display. Thus, the paired comparison tests provide partial support for hypothesis H1b.

PAGE 124

109 Table 19 Pilot Test Results of H1b ANCOVA Model on Efficiency (Time Spent) in Trend Analysis Task Tests of Between-Subjects Effects on Efficiency Panel A: Mean Time Spent on the Task and ANCOVA Results Using Time Spent as the Dependent Variable Actual Mean (Seconds) ANCOVA Adjusted Mean* (Seconds) Tabul ar Display (n = 27) 69.296 67.979 2 D Displays (n = 30) 54.466 57.824 3 D Perspective Display (n = 29) 100.206 96.166 Source of Variation Type III SS DF Mean Square F stat p value* Corrected Model 43916.8420 11 3992.440 5.118 <0.001 Intercept 21661 .621 1 21661.621 27.768 <0.001 PQ1 49.906 1 49.906 0.064 0.801 PQ2 3943.532 1 3943.532 5.055 0.028 AGE 54.234 1 54.234 0.070 0.793 GEN 528.119 1 528.119 0.677 0.413 SMRT 994.898 1 994.898 1.275 0.262 TSMRT 4876.167 1 4876.167 6.251 0.015 Treatment 2 2694.485 2 11347.243 14.546 <0.001 Rank 93.118 1 93.118 0.119 0.731 Treatment*Rank 1234.886 2 617.443 0.792 0.457 Error 57726.472 74 780.087 Total 579561.000 86 Corrected Total 101643.314 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.348. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H1b (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 2 D Displays Tabular Display 3 D Display 10.155 40.342 7.780 7.790 0.294 <0.001 *p-values are one-tail.

PAGE 125

110 4.2.2.3 Results of Hypothesis H1c. Hypothesis H1c has five dependent measures rd question, which ask ed the participants to estimate what the ROE would be in year 6 if each of the variables comprising ROE in year 5 double ask ed the participants to estimate the average of turnover, profitability and leverage for the years 1, 2, 4 and 5, and use the estimated average to calculate a new ROE. The mean results for all dependent variables are hypothesized in H1c. In constructing the models to test H1c, all covariates mentioned in section 4.2.2 were included in the model along with the manipulated variable Treatment. A MANCOVA analysis was conducted. As shown (Table 20), the overall statistic for the manipulated variable Treatment is not significant (p = 0.076) usi Trace. Thus, hypothesis H1c was not supported.

PAGE 126

111 Table 20 Multivariate Tests of H1c Variables Multivariate Test Value F stat |p value| Intercept 0.202 3.550 0.006 PQ1 0.107 1.671 0.153 PQ2 0.045 0.653 0.660 AGE 0.116 1.830 0.118 GEN 0.099 1.536 0.190 SMRT 0.124 1.979 0.092 TSMRT 0.036 0.529 0.754 Treatment P 0.219 1.744 0.076 Rank 0.070 1.051 0.395 Treatment*Rank 0.124 0.938 0.500 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks. 4.2.2.4 Result of Hypothesis H1d. Hypothesis H1d has two dependent measures of estimating what the ROE would be in year 6 if each of the variables comprising ROE in year 5 doubles, and the time spent in seconds by each participant when estimating the average of turnover, profitability and leverage for the years 1, 2, 4 and 5, and using the estimated average to calculate a new ROE. The mean results for all dependent variables are hypothesized in H1d. In constructing the models to test H1d, all covariates mentioned in section 4.2.2 were included in the model along with the manipulated variable Treatment. A MANCOVA analysis was conducted. As shown (Table 21), the overall Fstatistic for the manipulated variable Treatment is not significant (p = 0.350) using

PAGE 127

112 Hypothesis H1d was not supported. Table 21 Multivariate Tests of H1d. Variables Multivariate Test Value F stat |p value| Intercept Pillai 0.059 2.285 0.109 PQ1 0.010 0.364 0.696 PQ2 0.021 0.771 0.466 AGE 0.117 4.825 0.011 GEN 0.005 0.182 0.834 SMRT 0.007 0.241 0.766 TSMRT 0.003 0.118 0.8 89 Treatment 0.059 1.118 0.350 Rank 0.023 0.866 0.425 Treatment*Rank 0.073 1.411 0.233 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks.

PAGE 128

113 4.2.3. Hypotheses Testing of the Pattern Analysis Task Results of H2a, H2b, H2c, and H2d are shown in tables 22 to 27. Multivariate MANCOVA was first used to test for the significance of the manipulation variable Treatment. If the manipulation variable Treatment was significant, ANCOVA was then conducted to analysis the univariate results. Covariates included in the analysis are the scores on practice questions one and t wo gender, age, score on the Mental Rotations Test, time spent on the Mental Rotations Test, and the task order performed by each participant (for example the trend analysis task first and the pattern recognition task second, and vice versa). The following paragraphs report significant results, supporting the hypothesis, with tables on top of narration. 4.2.3.1 Results of Hypothesis H2a. Hypothesis H2a has two dependent measures companies one through six into two groups based on similar financial characteristics; and the score on the fourth question which asked the participants to select one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. The mean results for all dependent variables are hypothesized in H2a. In constructing the models to test H2a, all covariates mentioned in section 4.2.3 were included in the model along with the manipulated variable Treatment. Prior to presenting ANCOVA results for the two dependent variables used to test hypothesis H2a, a MANCOVA analysis was conducted. As shown (Table 22), the overall F-statistic for the manipulated variable Treatment is significant (p < 0.001) using

PAGE 129

114 significant results allow for analysis of the univariate results which are provided on Table 23. Table 22 Multivariate Tests of H2a Variables Multivariate Test Value F stat |p value| Intercept 0.419 26.318 <0.001 PQ1 0.017 0.649 0.525 PQ2 0.016 0.579 0.563 AGE Pi 0.087 3.494 0.036 GEN 0.039 1.472 0.236 SMRT 0.172 7.606 <0.001 TSMRT 0.014 0.510 0.602 Treatment 0.470 11.379 <0.001 Rank 0.023 0.867 0.425 Treatment*Rank Pillai 0.139 2.755 0.030 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks. The first dependent variable used to test hypothesis H2a was the second question in the pattern recognition task, which asked participants to separate companies one through six into two groups based on similar financial characteristics. Those participants viewing the 3-D perspective display (mean score 5.069) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (mean score 4.555) were 10% less accurate than those viewing the 3-D perspective display (mean score 5.069). Those participants viewing the 2D

PAGE 130

115 displays (mean score 4.800) were 5% less accurate than those viewing the 3D perspective display (mean score 5.069). Results of the ANCOVA show that the covariates score on Mental Rotations Test (p < 0.001) is significantly associated with the accuracy of the participants in separating companies one through six into two groups based on similar financial characteristics. Results also, contrary to expectation, show that the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.129) main effect on the accuracy of the participants in this pattern recognition task. Since there is not a significant main effect a paired comparison test was not conducted. The second dependent variable used to test hypothesis H2a was the fourth question in the pattern recognition task which asked participants to select one of the six companies given the goal was to have higher profitability, higher turnover but lower leverage. Table 23, Panel A suggests that those participants viewing the 3-D perspective display (mean score 0.724) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (mean score 0.000) were 100% less accurate than those viewing the 3-D perspective display (mean score 0.724). Those participants viewing the 2-D displays (mean score 0.200) were 73% less accurate than those viewing the 3-D perspective display (mean score 0.724). Table 23, Panel A suggests that age of the participants is significantly (p = 0.033) associated with the accuracy of the participants in selecting the company given the goal was to have higher profitability, higher turnover but lower leverage. Panel A also suggests that the interaction between the presentation formats (Treatment) and the Task Order has a significant effect (p = 0.035) for the accuracy of the participants in this

PAGE 131

116 pattern recognition task. A paired comparison test was conducted to determine if the participants receiving the 3-D Treatment were more accurate than those participants receiving the Tabular or 2-D Treatment. Results revealed that the participants viewing the 3D perspective display were significantly (p <0.001) more accurate than participants viewing the Tabular display and participants viewing the 2-D displays (see Table 23, Panel B). Additional analysis on the interaction effect between the presentation format (Treatment) and the Task Order revealed that among the participants performing trend analysis task first and the pattern recognition task later, those participants viewing the 3D perspective display were significantly (p<0.0001) more accurate than participants viewing the Tabular display and participants viewing the 2-D displays. However, among the participants performing pattern recognition task first and the trend analysis task later, those participants viewing the 3-D perspective display were significantly (p = 0.001) more accurate than participants viewing the Tabular display but not significantly more accurate than participants viewing the 2-D displays (p = 0.082). (Results of additional analysis not shown in tables here).

PAGE 132

117 Table 23 Pilot Test Results of H2a ANCOVA Model on Effectiveness (Accuracy) in Pattern Recognition Task annot select a company solely because of a single variable, for example highest profitability. Comparatively, if it is better to have a higher profitability, higher turnover but lower leverage at the same time, which company Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Score on the Task and ANCOVA Results Using Score as the Dependent Variable. Actual Mean ANCOVA Adjusted Mean* Tabular Display (n = 27) 0.000 0.004 2 D Displays (n = 30) 0.200 0.194 3 D Perspectiv e Display (n = 29) 0.724 0.731 Source of Variation Type III SS DF Mean Square F p value* Corrected Model 9.482 11 0.862 7.055 <0.001 Intercept 0.145 1 0.145 1.184 0.280 PQ1 0.000 1 0.000 0.004 0.950 PQ2 0.000 1 0.000 0.003 0.954 AGE 0.574 1 0.574 4 .698 0.033 GEN 0.000 1 0.000 0.003 0.957 SMRT 0.025 1 0.025 0.207 0.650 TSMRT 0.126 1 0.126 1.034 0.312 Treatment 7.363 2 3.681 30.129 <0.001 Rank 0.012 1 0.012 0.098 0.755 Treatment*Rank 0.856 2 0.428 3.505 0.035 Error 9.042 74 0.122 Total 27.00 0 86 Corrected Total 18.523 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.439. *Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H2a (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* 3 D Display Tabular Display 2 D Displays 0.727 0.537 0.097 0.097 <0.001 <0.001 *p -values are one-tail.

PAGE 133

118 4.2.3.2 Results of Hypothesis H2b. Hypothesis H2b has two dependent measures through six into two groups based on similar financial characteristics, and the time spent in seconds by each participant in selecting one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. The mean results for all dependent variables are hypothesized in H2b. In constructing the models to test H2b, all covariates mentioned in section 4.2.3 were included in the model along with the manipulated variable Treatment. A MANCOVA analysis was conducted. As shown (Table 24), the overall Fstatistic for the manipulated variable Treatment is not significant (p = 0.312) using Hypothesis H2b was not supported. Table 24 Multivariate Tests of H2b Variables Multivariate Test Value F stat |p value| Intercept 0.118 4.905 0.010 PQ1 0.035 1.307 0.277 PQ2 0.002 0.065 0.937 AGE 0.024 0.884 0.417 GEN 0.000 0.111 0.989 SMRT 0.076 3.008 0.056 TSMRT 0.293 15.103 <0.001 Treatmen t 0.063 1.202 0.312 Rank 0.017 0.613 0.544 Treatment*Rank 0.113 2.222 0.069 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks.

PAGE 134

119 4.2.3.3 Results of Hypothesis H2c. Hypothesis H2c has four dependent measures differences were between companies one and six; the score on the third question, which asked the participants to describe the pattern of financial ratios they were seeing in group one compared to group two; the score on the fifth question, which asked the participants what the data differences were between companies four and six by selecting choices from a template; and the score on the sixth question, which asked the participants to describe the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template. The mean results for all dependent variables are hypothesized in H2c. In constructing the models to test H2c, all covariates mentioned in section 4.2.3 were included in the model along with the manipulated variable Treatment. Prior to presenting ANCOVA results for the two dependent variables used to test hypothesis H2c, a MANCOVA analysis was conducted. As shown (Table 25), the overall F-statistic for the manipulated variable Treatment is significant (p = 0.010) using significant results allow for analysis of the univariate results which are provided on Table 26. The first dependent variable used to test hypothesis H2c was the first question in the pattern recognition task, which asked participants what the data differences were between companies one and six. Those participants viewing the tabular display (mean score 2.777) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (means score 2.777) were

PAGE 135

120 3% more accurate than those viewing the 3-D perspective display (mean score 2.689). Those participants viewing the 2-D displays (mean score 2.566) were 5% less accurate than those viewing the 3-D perspective display (mean score 2.689). Results of the ANCOVA show that, contrary to expectation, the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.290) main effect on the accuracy of the participants in describing what the data differences were between companies one and six. Since there is not a significant main effect a paired comparison test was not conducted. Table 25 Multivariate Tests of H2c. Variables Mu ltivariate Test Value F stat |p value| Intercept 0.643 31.991 <0.001 PQ1 0.053 0.998 0.414 PQ2 0.065 1.228 0.307 AGE 0.083 1.605 0.182 GEN 0.097 1.903 0.119 SMRT ce 0.029 0.530 0.714 TSMRT 0.094 1.832 0.132 Treatment 0.256 2.645 0.010 Rank 0.014 0.249 0.910 Treatment*Rank 0.116 1.112 0.358 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks.

PAGE 136

121 The second dependent variable used to test hypothesis H2c was the third question in the pattern recognition task which asked participants to describe the pattern of financial ratios they were seeing in group one compared to group two. Those participants viewing the 3-D perspective display (mean score 2.517) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (mean score 1.777) were 31% less accurate than those viewing the 3-D perspective display (mean score 2.517). Those participants viewing the 2D displays (mean score 2.333) were 8% less accurate than those viewing the 3D perspective display (mean score 2.517) Table 26, Panel A suggests that the manipulation of the presentation formats (Treatment) has a significant (p = 0.001) main effect on the on the accuracy of the participants in describing the pattern of financial ratios they were seeing in group one compared to group two. A paired comparison test was conducted to determine if the participants receiving the 3-D Treatment were more accurate than those participants receiving the Tabular or 2-D Treatment. Results revealed that the participants viewing the 3D perspective display were significantly (p = 0.002) more accurate than participants viewing the Tabular display. There was no significant difference in the efficiency between participants viewing the 2-D displays and participants viewing the 3D perspective display (see Table 26 Panel B). Thus, the paired comparison tests provide partial support for hypothesis H2c.

PAGE 137

122 Table 26 Pilot Test Results of H2c ANCOVA Model on Effectiveness (Accuracy) in Pattern Recognition Task and 5. Compared to group two, what are the patterns of the financial ratios you are seeing in group one? Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Score on the Task and ANCOVA Results Using Score as the Dependent Variable. Actual Mean ANCOVA Adjusted Mean* Tabular Display (n = 27) 1.777 1.727 2 D Displays (n = 30) 2.333 2.387 3 D Perspective Display (n = 29) 2.517 2.501 Source of Variation Type III SS DF Mean Square F p value* Corrected Model 22.149 11 2.014 2.942 0.003 Intercept 1.346 1 1.346 1.966 0.165 PQ1 1.493 1 1.493 2.182 0.144 PQ2 1.825 1 1.825 2.667 0.107 AGE 0.265 1 0.265 0.387 0.536 GEN 1.795 1 1.795 2.623 0.110 SMRT 0.832 1 0.832 1.215 0.274 TSMRT 1.920 1 1.920 2.805 0.098 Treatment 9.071 2 4.535 6.626 0.001 Rank 0.197 1 0.197 0.288 0.593 Treatment*Rank 1.645 2 0.822 1.202 0.307 Error 50.653 74 0.685 Total 497 86 Correct ed Total 72.802 85 *Adjusted Mean is for the effect of the covariate. Adjusted R-Squared = 0.439. Treatment p-values are one-tail, all others are two-tail. See Table 11 for definition of the variables. Panel B: Bonferroni Pairwise Comparisons for Test H2c (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value 3 D Display Tabular Display 2 D Displays 0.775 0.114 0.230 0.231 0.002 0.500 *p -values are one-tail.

PAGE 138

123 The third dependent variable used to test hypothesis H2c was the fifth question in the pattern recognition task, which asked participants to what the data differences were between companies four and six by selecting choices from a template. Those participants viewing the 2-D displays (mean score 3.000) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (2.851) were 6% more accurate than those viewing the 3-D perspective display (mean score 2.689). Those participants viewing the 2-D displays (mean score 3.000) were 11% more accurate than those viewing the 3-D perspective display (mean score 2.689). Results of the ANCOVA show that the manipulation of the presentation formats (Treatment) has a significant (p = 0.028) main effect on the accuracy of the participants in what the data differences were between companies four and six by selecting choices from a template. A paired comparison test was conducted to determine if the participants receiving the 3-D Treatment were more effective or accurate than those participants receiving the Tabular or 2-D Treatment. Contrary to prediction, results revealed that the participants viewing the 2D displays were significantly (p = 0.026) more accurate than participants viewing the 3-D perspective display. There was no significant difference in the efficiency between participants viewing the tabular display and participants viewing the 3-D perspective display. Thus, hypothesis H2c was partially not supported. The fourth dependent variable used to test hypothesis H2c was the sixth question in the pattern recognition task, which asked participants to describe the pattern of financial ratios they were seeing in group two compared to group one. Those participants viewing the 2-D display (mean score 2.833) were on average the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular

PAGE 139

124 display (mean score 2.703) were 2% more accurate than those viewing the 3D perspective display (mean score 2.655). Those participants viewing the 2-D displays (mean score 2.833) were 7% more accurate than those viewing the 3-D perspective display (mean score 2.655). Results of the ANCOVA show that, contrary to expectation, the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.400) main effect on the accuracy of the participants in describing the pattern of financial ratios they were seeing in group two compared to group one. Since there is not a significant main effect a paired comparison test was not conducted. 4.2.3.4 Results of Hypothesis H2d. Hypothesis H2d has four dependent measures differences were between companies one and six; the time spent by each participant in describing the pattern of financial ratios they were seeing in group one compared to group two; the time spent by each participant in describing what the data differences were between companies four and six by selecting choices from a template; and the time spent by each participant in describing the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template The mean results for all dependent variables are hypothesized in H2d. In constructing the models to test H2d, all covariates mentioned in section 4.2.3 were included in the model along with the manipulated variable Treatment. Prior to presenting ANCOVA results for the two dependent variables used to test hypothesis H2d, a MANCOVA analysis was conducted. As shown (Table 27), the overall F-

PAGE 140

125 significant results allow for analysis of the univariate results. Table 27 Multivariate Tests of H2d Variables Multivariate Test Value F stat |p value| Intercept 0.337 9.041 <0.001 PQ1 0.018 0.330 0.857 PQ2 0.099 1.950 0.112 AGE 0.080 1.542 0.199 GEN 0. 059 1.115 0.356 SMRT 0.033 0.602 0.663 TSMRT 0.152 3.172 0.019 Treatment 0.414 4.694 <0.001 Rank 0.497 17.542 <0.001 Treatment*Rank 0.189 1.877 0.068 PQ1= the score on practice question 1; PQ2 = the score on practice question 2; AGE = age of participants; GEN = gender of participants. SMRT = the score on Mental Rotation Test. TSMRT = time spent on Mental Rotations Test. Treatment = manipulation variables. Rank = the order of performing trend analysis task first then pattern recognition task and vise versa. Treatment Rank = interaction between the manipulation variables and the order of performing tasks. The first dependent variable used to test hypothesis H2d was the time spent by each participant in answering what the data differences were between companies one and six. On average those participants viewing the 2-D display (mean seconds 111) were the most efficient (used the least amount of time in seconds) on this pattern recognition task. Those participants viewing the 2-D displays (mean seconds 111) used 46% less time (in seconds) than those participants viewing the 3-D display (mean seconds 163). While those participants viewing the tabular display (mean seconds 112) used 45% less time (in seconds) than those participants viewing the 3-D perspective display (mean seconds163).

PAGE 141

126 Results of the ANCOVA show that, the manipulation of the presentation formats (Treatment) has a significant (p < 0.001) main effect on the efficiency of the participants in describing what the data differences were between companies one and six. A paired comparison test was conducted to determine if the participants receiving the 3D Treatment were more efficient or used less time (in seconds) than those participants receiving the Tabular or 2-D Treatment. Results show that those participants viewing tabular display (p <0.001) and those participants viewing the 2-D displays (p = 0.001) were significantly more efficient or used less time (in seconds) than those participant viewing the 3-D perspective display in this pattern recognition task. The second dependent variable used to test hypothesis H2d was the time spent by each participant in describing the pattern of financial ratios they were seeing in group one compared to group two. On average those participants viewing the 2-D display (mean seconds 97) were the most efficient (used the least amount of time in seconds) on this pattern recognition task. Those participants viewing the 2-D displays (mean seconds 97) used 8% less time (in seconds), than those participants viewing the 3-D display (mean seconds 105). While those participants viewing the tabular display (mean seconds 121) used 15% more time (in seconds), than those participants viewing the 3-D perspective display (mean seconds 105). Results of the ANCOVA show that, the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.221) main effect on the time spent by each participants in describing the pattern of financial ratios they were seeing in group one compared to group two. Since there is not a significant main effect a paired comparison test was not conducted.

PAGE 142

127 The third dependent variable used to test hypothesis H2d was the time spent by each participant in answering what the data differences were between companies four and six by selecting choices from a template. On average those participants viewing the 2D display were the most efficient (used the least amount of time in seconds) on this first pattern recognition task. Those participants viewing the 2-D displays (mean seconds 27) used 38% less time (in seconds) than those participants viewing the 3-D display (mean seconds 43). While those participants viewing the tabular display (mean seconds 30) used 31% less time (in seconds) than those participants viewing the 3-D perspective display (mean seconds 43). Results of the ANCOVA show that, the manipulation of the presentation formats (Treatment) has a significant (p < 0.001) main effect on the efficiency of the participants in describing what the data differences were between companies four and six. A paired comparison test was conducted to determine if the participants receiving the 3D Treatment were more efficient or used less time (in seconds) than those participants receiving the Tabular or 2-D Treatment. Results show that those participants viewing tabular display (p <0.001) and those participants viewing the 2-D displays (p < 0.001) were significantly more efficient or used less time (in seconds) than those participant viewing the 3-D perspective display in this pattern recognition task. The fourth dependent variable used to test hypothesis H2d was the time spent by each participant in describing the pattern of financial ratios they were seeing in group two co mpared to group one by selecting choices from a template. On average those participants viewing the 2-D display (mean seconds 35) were the most efficient (used the least amount of time in seconds) on this pattern recognition task. Those participants

PAGE 143

128 viewing the 2-D displays (mean seconds 35) used 37% less time (in seconds) than those participants viewing the 3-D display (mean seconds 55). While those participants viewing the tabular display (mean seconds 46) used 17% less time (in seconds) than those participants viewing the 3-D perspective display (mean seconds 55). Results of the ANCOVA show that, the manipulation of the presentation formats (Treatment) has a significant (p = 0.017) main effect on the efficiency of the participants in describing the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template. A paired comparison test was conducted to determine if the participants receiving the 3-D Treatment were more efficient or used less time (in seconds) than those participants receiving the Tabular or 2-D Treatment. Results show that those participants viewing the 2-D displays (p = 0.015) were significantly more efficient or used less time (in seconds) than those participant viewing the 3-D perspective display in this pattern recognition task. There is no significant difference in efficiency between those participants viewing the tabular display and those viewing the 3-D perspective display. In conclusion, hypothesis H2d was not supported.

PAGE 144

129 4.3 Lessons Learned from the Pilot Results and How to Improve 4.3.1 Rank Order Effect As shown by Tables 23 the order in which the trend analysis and pattern recognition tasks were presented resulted in a significant interaction with the presentation formats (Treatment). The interaction effect between the Treatment condition and the task order suggests that participant responses depend on whether participants perform the trend analysis task first and the pattern recognition task later or vice versa. To address th e issue of rank order effects, the main experiment employs a full factorial 3x2 betweensubjects design. 4.3.2 Insufficient Training on Display Formats Despite the fact that participants were required to perform two different treatment tasks (trend analysis and pattern recognition tasks), the pilot study employed the same training task to familiarize participants with their assigned display formats. Participants viewing the 3-D perspective display scored, on average, only 2.966 and 3.034 (on a four point scale) on the two practice questions; these scores are lower than that of participants viewing either the tabular display (3.222 and 3.444) or 2D displays (3.000 and 3.400). It seems that participants, especially those viewing the 3D perspective display, need more training in terms of practice questions. The main experiment employs separate training materials for the two treatment tasks: the trend analysis task and the pattern recognition task. The practice questions on display formats are increased from two to four questions, to better train the participants

PAGE 145

130 on their assigned display format, which was particularly important for those receiving the 3-D perspective display. 4.3.3 Insufficient Training on ROE The pilot study did not provide participants an opportunity to review the concepts of the DuPont analysis, or to practice calculating the return on equity. Even though participants had already learned the concepts of DuPont analysis in the principles of accounting class, all treatment groups performed poorly in the trend analysis task involving calculating ROE. The main experiment provides two practice questions to train participants on the concept of DuPont analysis and on the calculation of ROE, respectively. 4.3.4 Scrolling Up and Down the Screen Participants in both the 2-D and 3-D treatment conditions had to scroll up and down the screen when answering questions, since the display and the response area did not fit on one screen. Debriefing with some participants suggest ed that participants felt constrained by having to scroll up and down the screen while answering the questions. It seems that the physical motion of scrolling up and down the screen interfered with performance, an d Sweller (1992). This split-attention effect likely had undesirable consequences on both the time spent and the accuracy of responses from participants viewing either the 2D displays or 3-D perspective display (Chandler and Sweller 1992). Participants viewing the tabular display did not have such a problem.

PAGE 146

131 The 2-D line graphs (see Figure 1) or 2-D bar charts (see Figure 3) and 3D perspective displays (see Figure 2 and 4) cannot be further reduced in area without sacrificing the quality of perception. This prevents fitting the questions and response guidelines for ensuring that the displays exhibit graphical excellence. ce (1983), the study develops figures 1, 2, 3, and 4 to show the data, to induce the viewer to think about the substance and to encourage the viewer to compare different pieces of data. Similar to the 3D perspective display (see Figures 2 and 4), this study places four separate 2-D line graphs (see Figure 1) and four separate 2-D bar graphs (see Figure 3) in a single page area. The total area of each of the Figures 1, 2, 3, and 4 are the same, and viewers can clearly view the scale and description of each figure. experiment has four practice questions to familiarize participants with their assigned display format. Though increased training cannot remove the split-attention effect, increased training can help to familiar participants with the need to scroll up and down the screen while performing the task. Participants viewing the tabular display do not need to scroll up and down the screen. Compared to viewers of the 2-D display and the viewers of the 3-D perspective display, viewers of the tabular display have less mental workload. To measure the differences in mental workload among the three treatment groups four survey questions

PAGE 147

132 main experiment. The four questions adopted from Reid and Nygren (1988) are: 1) Very little mental effort or concentration was required to complete tasks, 2) Tasks performed were almost automatic, requiring little or no attention, 3) Tasks performed were very complex and required total attention, and 4) Extensive mental effort and concentration were necessary in tasks. Measures of mental workload are used as one of the covariates in the main experiment.

PAGE 148

133 Chapter 5 : Main Experiment 5.1 Sample Size, Inter-Coder Reliability 5.1.1 Sample Size According to the analysis two hundred sixteen participants are needed to detect a medium -level of 0.05 and a power of 0.80, if one-way ANOVA statistical analysis of six groups is performed. Cohen (1992) suggests that with a power of 0.80, there is an 80% chance of detecting an effect if that effect genuinely exists. 5.1.2 Inter-Coder Reliability Two hundred fifty eight undergraduate business students participated in the main experiment. Two hypothesis-blind coders worked independently to code the responses of each participant. The same two coders, who had coded the responses from the pilot tests, were used as coders for the main experiments. When assigning score-points to the trend analysis questions, and the pattern recognition questions, the two coders were instructed to follow the grading schema as outlined in Table 5, Table 6, and Table 7, respectively. When assigning score-points to al Rotations Tests, the two coders were instructed to follow the grading schema as discussed in section 3.6.2. Other than the aforementioned assignment of score-points, which required some judgment, all other responses to

PAGE 149

134 questions like the survey or demographic questions were automatically captured and reported by the survey software. Each of the two coders separately coded the responses and then reconciled their differences. Subjectivity of the coder and differences in coding between the two coders should be low, as most of the questions asked were in the form of multiple choices. The survey software used in the study (Select Survey) allowed the experimenter to deploy openended questions such as asking participants to respond by writing short sentences or writing a newly calculated value, or to deploy multiple choices questions such as asking participants to respond by selecting choices from a given template. Measures of the time spent (in seconds) by each participant when answering a question (practice questions, trend analysis questions, and pattern recognition questions) were calculated via a database query based on timestamps captured by the survey software. of multiple choices. Like the dependent variables of the time spent in seconds by each participant when answering a question, the survey software automatically captured the responses of all multiple-choice questions, which could be exported to an Excel spreadsheet. Smith (2000) suggests that when a coding system yields scores, the agreement between the scores assigned by two coders can be tested through use of a correlation coefficient. Since the two blind coders of the study were simply assigning scores rather than actually coding subjective responses, the agreement between the scores assigned by

PAGE 150

135 the two blind coders was tested though a simple correlation between their assigned scores. The correlation between each of the scores assigned by the coders was separately evaluated. For example, the scores assigned by each of the two coders to the TAQ1 were set up as column one and column two in the order of the treatment groups of tabular display, 2-D displays, and a 3-D perspective display to calculate the inter-coder correlation. A similar procedure was performed on all the scores of the Trend Analysis task and the Pattern Recognition task. Smith et al. (1992) suggest that a satisfactory percentage of agreement between two coders will require an inter-coder correlation of 0.85 or more. Table 28 and 29 reports -Moment Correlation Coefficients of the scores (before reconciliation) assigned by coder one and coder two to the trend analysis task and pattern recognition task, respectively. Except for practice question two of the trend analysis task, which has an inter-coder correlation of 0.799, all other inter-coder correlation statistics are well above the benchmark of 0.85 as suggested by Smith et al. (1992). Since practice question two of the trend analysis task is a covariate, rather than a manipulated variable, a 0.799 inter-coder correlation for the scores assigned to practice question two (before reconciliation) is not a major concern. It is concluded that the degree of inter-coder reliability is sufficiently high to proceed with data analysis. Starting from the next section onward, the data set used and reported in the rest of the study was after the reconciliation of differences between the two coders.

PAGE 151

136 Table 28 -Moment Correlation Coefficients of the Scores Assigned by Coder One and Coder Two to the Trend Analysis Task Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 PQ1 TSPQ1 PQ2 TSPQ2 PQ3 TSPQ3 PQ4 Coder 1 PQ1 0.919** Coder 1 TSPQ1 0.959** Coder 1 PQ2 0.799** Coder 1 TSPQ2 0.998** Coder 1 PQ3 0.993** Coder 1 TSPQ3 0.973** Coder 1 PQ4 0.982** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 TSPQ4 PQ5 TSPQ5 PQ6 TSPQ6 TAQ1 TSTAQ1 Coder 1 TSPQ4 0.992** Coder 1 PQ5 0.952** Coder 1 TSPQ5 0.994** Coder 1 PQ6 1.000** C oder 1 TSPQ6 0.989** Coder 1 TAQ1 0.979** Coder 1 TSTAQ1 0.986** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 TAQ2 TSTAQ2 TAQ3 TSTAQ3 TAQ4a TAQ4b TAQ4c Coder 1 TAQ2 0.914** Coder 1 TSTAQ2 0.994** Coder 1 TAQ3 1.000** Coder 1 TSTAQ3 1.000** Coder 1 TAQ4a 1.000** Coder 1 TAQ4b 0.937** Coder 1 TAQ4c 0.987** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 TAQ4d TSTAQ4 TAQ5 TSTAQ5 TAQ6 TSTAQ6 S1 Coder 1 TAQ4d 1.0 00** Coder 1 TSTAQ4 0.999** Coder 1 TAQ5 0.991** Coder 1 TSTAQ5 0.997** Coder 1 TAQ6 0.993** Coder 1 TSTAQ6 0.999** Coder 1 S1 0.995** See Table 11 for definition of the variables

PAGE 152

137 Table 28 (Continued) Pears -Moment Correlation Coefficients of the Scores Assigned by Coder One and Coder Two to the Trend Analysis Task Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 S2 S3 S4 S5 S6 S7 S8 Coder 1 S2 0.998** Coder 1 S3 1.000** Coder 1 S4 1.000** Coder 1 S5 1.000** Coder 1 S6 0.999** Coder 1 S7 1.000** Coder 1 S8 1.000** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 S9 S10 M1 M2 M3 AGE GEN Coder 1 S9 0.974** Coder 1 S10 1 .000** Coder 1 M1 1.000** Coder 1 M2 1.000** Coder 1 M3 0.979** Coder 1 AGE 0.975** Coder 1 GEN 1.000** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 SS GPA SAT FTWE PTH ARWE HE Coder 1 SS 1.000** Coder 1 GPA 1.000** Coder 1 SAT 1.000** Coder 1 FTWE 1.000** Coder 1 PTH 1.000** Coder 1 ARWE 1.000** Coder 1 HE 1.000** Coder 2 Coder 2 SMRT TSMRT Coder 1 SMRT 0.997** Coder 1 TSMRT 0.990** See Table 11 for definition of the variables

PAGE 153

138 Table 29 -Moment Correlation Coefficients of the Scores Assigned by Coder One and Coder Two to the Pattern Recognition Task Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 PQ1 TSPQ1 PQ2 T SPQ2 PQ3 TSPQ3 PQ4 Coder 1 PQ1 0.970** Coder 1 TSPQ1 1.000** Coder 1 PQ2 0.924** Coder 1 TSPQ2 1.000** Coder 1 PQ3 0.982** Coder 1 TSPQ3 1.000** Coder 1 PQ4 0.975** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 TSPQ4 PQ5 TSPQ5 PQ6 TSPQ6 PRQ1 TSPRQ1 Coder 1 TSPQ4 1.000** Coder 1 PQ5 0.910** Coder 1 TSPQ5 1.000** Coder 1 PQ6 1.000** Coder 1 TSPQ6 1.000** Coder 1 PRQ1 0.956** Coder 1 TSPRQ1 1.000** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 PRQ2 TSPRQ2 PRQ3 TSPRQ3 PRQ4 TSPRQ4 PRQ5 Coder 1 PRQ2 0.995** Coder 1 TSPRQ2 1.000** Coder 1 PRQ3 0.993** Coder 1 TSPRQ3 1.000** Coder 1 PRQ4 1.000** Coder 1 TSPRQ4 1.000** Coder 1 PRQ5 0.985** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 TSPRQ5 PRQ6 TSPRQ6 S1 S2 S3 S4 Coder 1 TSPRQ5 1.000** Coder 1 PRQ6 1.000** Coder 1 TSPRQ6 0.975** Coder 1 S1 0.979** Coder 1 S2 1.000** Coder 1 S3 0.974** Coder 1 S4 0.999** See Table 11 for definition of the variables

PAGE 154

139 Table 29 (Continued) -Moment Correlation Coefficients of the Scores Assigned by Coder One and Coder Two to the Pattern Recognition Task Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 S5 S6 S7 S8 S9 S10 M1 Coder 1 S5 1.000** Coder 1 S6 0.895** Coder 1 S7 0.995** Coder 1 S8 0.978** Coder 1 S9 1.000** Coder 1 S10 1.000** Coder 1 M1 1.000** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 Coder 2 M2 M3 AGE GEN SS GPA SAT Coder 1 M2 1.000** Coder 1 M3 1.000** Coder 1 AGE 0.873** Coder 1 GEN 1.000** Coder 1 SS 1.000** Coder 1 GPA 1.000** Coder 1 SAT 0.959** Coder 2 Coder 2 Coder 2 Coder2 Coder 2 Coder 2 FTWE PTH ARWE HE SMRT TSMRT Coder 1 FTWE 0.975** Coder 1 PTH 0.998** Coder 1 ARWE 1.000** Coder 1 HE 1.000** Coder 1 SMRT 0.993** Coder 1 TSMRT 1.000** See Table 11 for definition of the variables

PAGE 155

140 5.2 Testing of Raw Data The study employs a 3 x 2 between-subjects (display format x task) design. The independent variables are display type and task type. Display type was manipulated at three levels: no graphical display (table only), 2-D displays, and 3-D perspective display (see Tables 1-2, and Figures 1-3 for examples of display formats used). Task type was manipulated at two levels: trend analysis and pattern recognition task. The remainder of this section discusses the tests of statistical assumptions. 5.2.1 Testing of Outliers and Influential Observations Field (2005) suggests that both outliers and influential observations have to be considered simultaneously when determining whether to drop an observation that meets the criteria of being either an outlier or an influential observation. For example, an observation can have a small standardized residual (not an outlier), but can be very influential (Field 2005). residual has an absolute value greater than 3, the observation is an outlier. Cook and observation is an influential observation. Tests for influential observations and outliers, at one standard deviation, were performed on each dependent variable (see Tables 3 and 4). ROE of year 5 what would be year 6 ROE if each of the variables of ROE in year 5 had 3, and a standardized residual of -10.93413. This observation, within the 3-D perspective treatment group, is dropped from the statistical an

PAGE 156

141 third question (TAQ3), the scores on parts a, b, c and d of the fourth question (TAQ4a, b, c, d). The aforementioned influential outlier was one of the observations for the dependent variable TAQ3 within the 3-D perspective treatment group, and therefore affects hypothesis H1c. Two separate MANCOVA models were run with and without the influential outlier. Comparison between the multivariate test results of the two different MA NCOVA models showed that the exclusion of the influential outlier improved the significance of the manipulation of the presentation formats, as revealed in the hypothesis H1c had four significant (alpha<0.10) differences (instead of two) in terms of accuracy between the participants viewing the 2-D displays and participants viewing the tabular display or 3-D perspective display. The results of hypothesis H1c are reported without the outlier (see section 5.3.5). Other than the dropped observation, there is no other observation which is either an outlier or an influential observation. 5.2.2 Testing of Assumptions Kolmogorov-Smirnov and Shapiro-Wilk tests were performed on each dependent variable (see Table 3 and 4) to test the assumption of normality. Field (2005) suggests that if the Kolmogrov-Smirnov test or the Shapiro-Wilk test is significant (p < 0.05), then the distribution of the sample data is significantly different from a normal distribution. Table 30 shows that all the dependent variables are significantly different from a normal distribution. Assumptions of the univariate ANOVA analysis include: 1) dependent variable is normally distributed within each group, 2) the variances in each group are roughly equal,

PAGE 157

142 3) each observation should be independent and 4) each dependent variable should be on an interval scale (Field, 2005). Table 30 shows that the dependent variables are nonormed on each of the dependent variables (see Table 3 and 4) to test the assumption of homogeneity of variances. Field (2005) suggests that if been violated. Table 30 shows that about half of the dependent variables violated the assumptions of homogeneity of variance. Field (2005) suggests that ANOVA is fairly robust in terms of violations of the assumption of normality and homogeneity of variance when the cell sizes are relatively equal. This study has relatively equal cell sizes for both the Trend Analysis Task (the treatment group of tabular display has 42 observations, 2D displays has 40 observations, and 3-D perspective display has 42 observations), and the Pattern Recognition Task (the treatment group of tabular display has 43 observations, 2D displays has 41 observations, and 3-D perspective display has 43 observations). Therefore, no transformation of data is necessary. In regard to the third assumption of the univariate ANOVA analysis, each observation of the study is statistically independent. In regard to the fourth assumption of the univariate ANOVA analysis, all dependent variables of the study are on an interval scale (see Tables 6 and Table 7).

PAGE 158

143 Table 30 Results of Testing For Normality, and Testing For Homogeneity of Variance Dependent Variables Results of Testing For Normality Kolmogorov Smirnov Test Shapiro Wilk Test Result of Testing For Homogeneity of Variances Tesr Based on Mean TAQ1 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P < 0.001 TSTQ1 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.273 TAQ2 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.945 TST AQ2 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.124 TAQ3 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P = 0.033 TSTAQ3 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0. 946 TAQ4a Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P = 0.006 TAQ4b Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.157 TAQ4c Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Varianc e P < 0.001 TAQ4d Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P < 0.001 TSTAQ4 Non Normal Distribution P = 0.001 P < 0.001 Homogeneity of Variance P = 0.317 TAQ5 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity o f Variance P = 0.050 TSTAQ5 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P = 0.007 See Table 11 for definition of the variables

PAGE 159

144 Table 30 (Continued) Results of Testing For Normality, and Testing For Homogeneity of Variance Dependent Variables Results of Testing For Normality Kolmogorov Smirnov Test Shapiro Wilk Test Result of Testing For Homogeneity of Variances Tesr Based on Mean TAQ6 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P < 0 .001 TSTAQ6 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.239 PRQ1 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P = 0.007 TSPRQ1 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Varia nce P = 0.011 PRQ2 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.051 TSPRQ2 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.062 PRQ3 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of V ariance P = 0.057 TSPRQ3 Non Normal Distribution P = 0.016 P < 0.001 Heterogeneity of Variance P = 0.035 PRQ4 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P < 0.001 TSPRQ4 Non Normal Distribution P < 0.001 P < 0.001 Homoge neity of Variance P = 0.635 PRQ5 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P = 0.003 TSPRQ5 Non Normal Distribution P < 0.001 P < 0.001 Heterogeneity of Variance P < 0.001 PRQ6 Non Normal Distribution P < 0.001 P < 0.00 1 Homogeneity of Variance P = 0.154 TSPRQ6 Non Normal Distribution P < 0.001 P < 0.001 Homogeneity of Variance P = 0.218 See Table 11 for definition of the variables

PAGE 160

145 MANOVA has similar assumptions to ANOVA but MANOVA further assumes that dependent variables have multivariate normality within groups and that the correlation between any two dependent variables is the same in all groups (homogeneity of covariance matrix). Field (2005) suggests that instead of checking the assumption of multivariate normality, it is more practical and useful to check the assumption of univariate normality for each dependent variable. Field (2005) stipulates that univariate normality is a necessary condition for multivariate normality, though univariate normality does not guarantee multivariate normality. Table30 shows that all the dependent variables are non-normally distributed within each group. Therefore, multivariate normality does an -Bartlett trace, the Hotellingrobust to violations of multivariate normality (Field 2005). -covariance matrices between treatment groups, Box multivariate normality is not tenable. Bray and Maxwell (1985) found that when cell sizes are equal, the Pillai-Bartlett trace is the most robust to potential violations of the homogeneity assumption. Field (2005) suggests that the Pillai-Bartlett trace and Hotelling-Lawley trace are robust to violations of homogeneity when sample sizes are equal. This study has relatively equal cell sizes for both the Trend Analysis Task, and the Pattern Recognition Task. Therefore, no transformation of data is necessary since the Pillai-Bartlett trace and Hotelling-Lawley trace are used in this study.

PAGE 161

146 5.2.3 Plan of Statistical Analysis This study employs MANCOVA analysis to test hypotheses. By including all dependent variables in a single analysis, MANCOVA takes into account the relationships among dependent variables. If the MANCOVA model is significant, then univariate ANCOVA will be used to separately test each dependent variable of a hypothesis. Rather than just putting the theoretical or expected covariates (see section 3.60) into the MANCOVA analysis, a separate regression will be run for each dependent variable with all the theoretical covariates and demographic variables (see section 3.65) to test for significance of all possible covariates to be included in subsequent MANCOVA analyses. Insignificant covariates in MANCOVA or ANCOVA models are dropped from the final analysis. This study used the Reid and Nygren (1988) scale to measure a perceived mental workload. The Reid and Nygren (1988) scale comprises the following four statements: (1) very little mental effort or concentration was required to complete tasks, (2) tasks performed were almost automatic, requiring little or no attention, (3) tasks were very complex and required total attention, and (4) extensive mental effort and concentration was necessary in tasks. Participants were asked to select a scale number, from one (strongly disagree) to seven (strongly agree), indicating the extent to which they agreed with each of the four statements. After reverse coding the first two statements, a reliability analysis was performed to confirm whether the four-statement scale is a dependable measure of the mental workload construct. The reliability test shows a

PAGE 162

147 the four statements (after reverse coding the first two statements). This average reflects a univariate models. 5.3 Manipulation Check As discussed in section 3.90 different manipulation questions were designed for the treatment conditions of display format and task. Another manipulation question was s were asked to answer three manipulation questions by selecting the choices of true or false. The pilot study used the same manipulation questions as described in following paragraphs. To test the manipulation of display formats, participants viewing the tabular -D line displays were asked whether: ted from zero (answer -D bar displays Finally participants viewing the 3-D display were a

PAGE 163

148 To test the manipulation of tasks, participants performing the trend analysis task n the context of this experiment companies companies with the same ROE also have the same turnover, profitability, and leverage turnover, profitability, and lever 5.3.1 Results of the Manipulation Check Questions Table 31 shows that participants in the trend analysis task did not perform well when answering the manipulation check questions. Compared to those participant s viewing the 2-D displays and the 3-D perspective display, those participants viewing the tabular display had higher accuracy when answering the manipulation check questions. Out of 42 participants viewing the tabular display, seven participants wrongly answered one manipulation check question. Out of 40 participants viewing the 2-D displays, 19 participants wrongly answered one manipulation check question, seven participants wrongly answered two manipulation check questions, and one participant wrongly an swered all three manipulation questions. Out of the 42 participants viewing the 3D perspective display, 16 participants wrongly answered one manipulation check question, seven participants wrongly answered two manipulation check questions, and one participant wrongly answered all three manipulation questions.

PAGE 164

149 Table 32 shows that participants in the pattern recognition task also did not perform well when answering the manipulation check questions. Compared to those participants viewing the 2-D displays and the 3-D perspective display, those participants viewing the tabular display had higher accuracy when answering the manipulation check questions. Out of 43 participants viewing the tabular displays, seven participants wrongly answered one manipulation check question, and one participant wrongly answered two manipulation questions. Out of 41 participants viewing the 2-D display, 13 participants wrongly answered one manipulation check question, four participants wrongly answered two manipulation check questions, and one participant wrongly answered all three manipulation questions. Out of the 43 participants viewing the 3-D perspective display, 15 participants wrongly answered one manipulation check question, and two participants wrongly answered two manipulation check questions. It should be noted that the pilot study had a lower error rate in responses to the manipulation check questions. However, the pilot study had a smaller sample size. The reason for such a high error rate in responses to the manipulation check questions was due to the fact that participants had to recall from memory the display when answering the manipulation check. Participants could not look at the tables or graphical displays when answering the manipulation check questions. Another problem is understanding of the characteristics of the display format they were presented, rather than merely assessing their recognition of the display format they were given. As a result, the manipulation check questions did not operationalize as intended. However, the study had significant results partially supporting the hypotheses.

PAGE 165

150 For the trend analysis task, one participant viewing the 2-D displays and one participant viewing the 3-D perspective display answered all three manipulation check questions incorrectly. The results of hypothesis H1 a-d did not change after dropping the se two participants. For the pattern recognition task, one participant viewing the 2D displays answered all three manipulation check questions incorrectly. The results of hypothesis 2a-d did not change after dropping th is participant. Nevertheless, the study will report the high error rate of the response to the manipulation check questions as a limitation.

PAGE 166

151 Table 31 Results of the Manipulation Check Questions of the Trend Analysis Task Trend analysis Task Tabular Display 2 D Display s 3 D Display No of Participants (n=42) (n=40) (n=42) Answered one manipulation question wrongly n=7 n=19 n=16 A nswered two manipulation question s wrongly n=0 n=7 n=7 Answered three manipulation question s wrongly n=0 n=1 n=1 M1 answered wrongly n=1 n=14 n=11 M2 answered wrongly n=2 n=7 n=6 M3 answered w rongly n=4 n=15 n=8 M1 = The tables (graphs) you see in the experiment, have data points of zero M2 = Return on Equity is the SUM of turnover, profitability, and leverage ratios M3 = Within the context of this experiment, companies can have negative ROE Table 32 Results of the Manipulation Check Questions of the Pattern Recognition Task Trend analysis Task Tabular Display 2 D Display s 3 D Display No of Participants (n=43) (n=41) (n=43) Answered one manipulation question wrongly n=7 n=13 n=15 Answered two manipulation question s wrongly n=1 n=4 n=2 Answe red three manipulation question s wrongly n=0 n=1 n=0 M1 answered wrongly n=4 n=13 n=13 M2 answered wrongly n=0 n=4 n=4 M3 answered Wrongly n=5 n=7 n=2 M1 = The tables (graphs) you see in the experiment, have data points of zero M2 = Return on Equity is the SUM of turnover, profitability, and leverage ratios M3 = Within the context of this experiment, companies with the same ROE also have the same turnover, profitability, and leverage ratios

PAGE 167

152 5.4 Results of the Trend Analysis Task The purpose of this section is to report whether there is support for each of the hypotheses H1a, H1b, H1c and H1d. Hypotheses H1a-H1d posit that 2-D displays are more effective and efficient than tabular and 3-D displays for a trend analysis task. This section is organized in the following way: descriptive statistics of the variables, the -Moment Correlation of the variables, results of the regression analyses of possible covariates, and the MANCOVA and ANCOVA results of each of the hypotheses. 5.4.1 Descriptive Statistics Table 33 shows the descriptive statistics for the trend analysis task. Based on the results of one-way ANOVA the following paragraphs briefly describe those covariates and demographic variables that had significant differences in their means among treatment groups. Descriptive information for the dependent variables will be discussed in the results section for each hypothesis. The first covariate evaluated was performance on the practice questions. For the score on practice question one (PQ1) (What are the values of the factors of the apartment rented in year 2?), participants viewing the 2-D displays on average had the highest score. Those participants viewing the 2-D displays (mean score 3.375) were significantly more accurate (p = 0.018) than those participants viewing the 3-D perspective display (mean score 2.619). There was no significant difference in accuracy between those participants viewing the tabular display (mean score 3.238) and those participants viewing the 2D

PAGE 168

153 displays (m ean score 3.375) or those participants viewing the 3-D perspective display (mean score 2.619). For the score on practice question two (PQ2) (What are the differences between the apartments rented in year 5 and 6?), participants viewing the 2-D displays on average had the highest score. Those participants viewing the 2-D displays (mean score 2.825) were significantly more accurate (p = 0.001) than those participants viewing the 3D perspective display (mean score 2.214). Those participants viewing the tabular display (mean score 2.714) were significantly more accurate (p = 0.005) than those participants viewing the 3-D perspective display (mean score 2.214). There was no significant difference in accuracy between those participants viewing the tabular display (mean score 2.714) and those participants viewing the 2-D display (mean score 2.825). For the score on practice question four (PQ4) (What are the differences between the apartments rented in year 2 and 4?), participants viewing the 2-D displays on average had the highest score. Those participants viewing the 2-D displays (mean score 3.825) were significantly more accurate (p = 0.001) than those participants viewing the 3D perspective display (mean score 3.023). Those participants viewing the tabular display (mean score 3.619) were significantly more accurate (p = 0.007) than those participants viewing the 3-D perspective display (mean score 3.023). There was no significant difference in accuracy between those participants viewing the tabular display (mean score 3.619) and those participants viewing the 2-D displays (mean score 3.825). The second covariate evaluated was the time spent on the practice questions. For the time spent by each participant in answering practice question two (TSPRQ2), those participants viewing the tabular display were the most efficient. Those participants

PAGE 169

154 viewing the tabular display (mean seconds 101) were significantly more efficient (p=0.006) or used less time (in seconds) than those participants viewing the 3D perspective display (mean seconds 152). Those participants viewing the 2-D displays (mean seconds 112) were significantly more efficient (p=0.052) or used less time (in seconds) than those participants viewing the 3-D perspective display (mean seconds 152). There was no significant difference in efficiency between those participants viewing the tabular display (mean seconds 101) and those participants viewing the 2-D displays (mean seconds 112). For the time spent by each participant in answering practice question three (TSPRQ3), those participants viewing the tabular display were the most efficient. Those participants viewing the tabular display (mean seconds 60) were significantly more efficient (p = 0.006) or used less time (in seconds) than those participants viewing the 3D perspective display (mean seconds 83). There was no significant difference in efficiency between those participants viewing the 2-D displays (mean seconds 76) and those participants viewing the 3-D perspective display (mean seconds 83). There was also no significant difference in efficiency between those participants viewing the tabular display (mean seconds 60) and those participants viewing the 2-D displays (mean seconds 76). The next covariate evaluated was time spent on the Mental Rotations Test (MRT). On average those participants viewing the 3-D perspective display were the most efficient. Those participants viewing the 3-D perspective display (mean seconds 515) were significantly more efficient (p = 0.022) or used less time (in seconds) than those participants viewing the tabular display (mean seconds 679). There was no significant

PAGE 170

155 difference in efficiency between those participants viewing the 2-D displays (mean seconds 621) and those participants viewing the 3-D perspective display (mean seconds 515). There was also no significant difference in efficiency between those participants viewing the tabular display (mean seconds 679) and those participants viewing the 2D displays (mean seconds 621). The mean MRT scores between tabular, 2-D and 3-D treatment groups were 19, 18, and 17 points, respectively. T-tests indicate no significant difference in spatial ability among treatment groups. The next covariate evaluated was perceived mental workload (MW). As predicted in section 3.6.4, participants viewing the 3-D perspective display had the highest perceived mental workload (MW) (mean 4.803) when compared to participants viewing the 2-D displays (mean 4.343) or participants viewing the tabular display (mean 4.428). There is no significant difference in mental workload among treatment groups.

PAGE 171

156 Table 33 Descriptive Statistics for the Trend Analysis Task Panel A: Mean (Standard Deviation) and Range of Practice Questions. Tabular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) Practice Question One: What are the values of the factors of apartment rented in year 2? 3.238 (1.303) 0.000 to 4.000 3.375 (1.212) 0.000 to 4.000 2.619 (1.146) 0.000 to 4.000 Time (seconds) Spent on Practice Question one? 85 (50) 31 to 263 100 (35) 35 to 215 10 5 (41) 35 to 217 Practice Question Two: What are the differences between the apartments rented in year s 5 and 6? 2.714 (0.596) 1.000 to 3.000 2.825 (0.594) 0.000 to 3.000 2.214 (0.898) 0.000 to 3.000 Time (seconds) Spent on Practice Question Two 101 (37) 50 to 197 112 (46) 29 to 204 152 (112) 29 to 704 Practice Question Three What are the values of the factors of apartment rented in year 4? 3.238 (1.284) 0.000 to 4.000 3.400 (1.215) 0.000 to 4.000 2.833 (1.286) 0.000 to 4.000 Time (seconds) Spent on Pra ctice Question Three 60 (29) 21 to 164 76 (29) 19 to 141 83 (37) 10 to 176 Practice Question Four: What are the differences between the apartments rented in year 2 and 4? 3.619 (0.935) 0.000 to 4.000 3.825 (0.446) 2.000 to 4.000 3.023 (1.092) 0.000 to 4.0 00 Time (seconds) Spent on Practice Question Four 98 (48) 29 to 301 95 (45) 19 to 222 112 (63) 10 to 353 See Table 11 for definition of the variables.

PAGE 172

157 Table 33 (Continued) Descriptive Statistics for the Trend Analysis Task Panel A: Mean (Standard Deviation) and Range of Practice Questions Tabular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) Practice Question Five if its turnover ratio is 2, profitability ratio is 5%, and leverage ratio is 1.1? 0.500 (0.506) 0.000 to 1 .000 0.500 (0.506) 0.000 to 1.000 0.476 (0.505) 0.000 to 1.000 Time (seconds) Spent on Practice Question Five 142 (61) 52 to 340 124 (57) 54 to 389 151 (79) 13 to 405 Practice Question Six ROE is the sum (multiplicative) of turnover, profitability, and l everage? 0.976 (0.154) 0.000 to 1.000 0.925 (0.266) 0.000 to 1.000 0.928 (0.260) 0.000 to 1.000 Time (seconds) Spent on Practice Question Six 30 (15) 10 to 80 28 (12) 8 to 59 30 (12) 5 to 62 Panel B: Mean (Standard Deviation) and Range of Dependent Variables Tabular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) Question One What are the differences between year s 1 and 4? 3.452 (1.130) 0.000 to 4.000 3.650 (0.735) 1.000 to 4.000 2.595 (1.269) 0.000 to 4.000 Time (seconds) Spent on Question One 92 (46) 26 to 236 96 (43) 13 to 211 121 (58) 8 to 331 Question Two What is happening as you go from year 2 to year 3 to year 4? 5.476 (2.370) 2.000 to 8.000 5.925 (2.758) 0.000 to 8.000 4.690 (2.503) 0.000 to 8.000 Time (seconds) Spent on Questio n Two 136 (114) 40 to 762 127 (56) 32 to 247 152 (90) 7 to 342 See Table 11 for definition of the variables.

PAGE 173

158 Table 33 (Continued) Descriptive Statistics for the Trend Analysis Task Panel B: Mean (Standard Deviation) and Range of Dependent Variables T abular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) Question Three Based on the ROE of year 5, what would be year 6 ROE if each of the variables of ROE in year 5 had doubled? 71.242 (53.015) 143.660 to 127.140 87.110 (35.917) 6.660 to 128.340 63.876 (56.813) 173.660 to 120.340 (n=41) Time (seconds) Spent on Question Three 109 (80) 27 to 397 98 (69) 25 to 303 94 (80) 23 to 453 Question Four Part a Estimate the average of turnover for the years 1, 2, 4, and 5 5.001 (22.339) 104.455 to 0.045 1.503 (4.393) 20.295 to 1.045 0.852 (3.508) 19.955 to 0.295 Question Four Part b Estimate the average of profitability for the years 1, 2, 4, and 5 0.796 (1.334) 1.045 to 3.955 0.626 (4.563) 20.045 to 13.955 1.494 (8.845) 56.245 to 3.255 Question Four Part c Estimate the average of leverage for the years 1, 2, 4, and 5 0.032 (0.860) 5.280 to 0.720 1.898 (5.954) 35.280 to 2.720 3.092 (1.196) 6.280 to 0.620 Question Four Part d Estimate the average of turnover, profitability and leverage for the years 1, 2, 4, and 5, and use them to calculate a new ROE 0.359 (12.605) 75.159 to 11.908 16.392 (57.060) 325.759 to 18.241 0.970 (6.957) 32.629 to 9.441 Time (seconds) Spent on Question Four 221 (135) 32 to 581 218 (136) 21 to 55 1 210 (107) 15 to 530 See Table 11 for definition of the variables.

PAGE 174

159 Table 33 (Continued) Descriptive Statistics for the Trend Analysis Task Panel B: Mean (Standard Deviation) and Range of Dependent Variables Tabular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) Question Five What are the differences between year s 2 and 4? Please answer as accurately and as fast as possible. 3.761 (0.691) 1.000 to 4.000 3.850 (0.483) 2.000 to 4.000 3.642 (0.692) 1.000 to 4.000 Time (seconds) Spent on Questi on Five 35 (10) 8 to 59 35 (7) 23 to 55 48 (17) 8 to 113 Question Six What is happening as you go from year 1 to year 2 to year 3? Please answer as accurately and as fast as possible. 7.547 (1.040) 3.000 to 8.000 7.625 (1.147) 3.000 to 8.000 6.357 (1. 818) 3.000 to 8.000 Time (seconds) Spent on Question Six 65 (40) 12 to 280 58 (29) 13 to 174 81 (43) 7 to 274 Panel C: Mean (Standard Deviation) and Range of Mental Rotations Test and Mental Workload Tabular Display (n=42) 2 D Displays (n=40) 3 D D isplay (n=42) Score on Mental Rotations Test 19.880 (11.536) 2.000 to 40.000 18.375 (10.902) 2.000 to 40.000 17.666 (10.309) 2.000 to 38.000 Time (seconds) Spent on Mental Rotation Test 679 (329) 125 to 1645 621 (282) 199 to 1682 515 (201) 40 to 1027 Mental Workload 4.428 (1.3584) 1.750 to 7.000 4.343 (1.3571) 2.000 to 7.000 4.803 (1.3853) 2.000 to 7.000 See Table 11 for definition of the variables

PAGE 175

160 Table 33 (Continued) Descriptive Statistics for the Trend Analysis Task Panel D: Demographic Mean and Range Tabular Display (n=42) 2 D Displays (n=40) 3 D Display (n=42) AGE 18 22 23 27 29 32 33 37 38 42 43 47 48 50 n=32 n=7 n=3 n=29 n=7 n=2 n=1 n=1 n=30 n=10 n=1 n=1 Male/Female n=24/n=18 n=17/n =23 n=15/n=27 Student Status Freshman Sophomore Junior Senior n=14 n=20 n=8 n=8 n=19 n=13 n=12 n=17 n=13 GPA 3.110 (0.682) 2.300 3.960 3.108 (0.431) 2.000 3.880 3.214 (0.379) 2.300 3.800 SAT 1156 (123) (n=33) 950 1410 1135 (151) (n=32) 790 1430 1014 (402) (n=37) 795 1575 Years of Full T ime Working Experience 2.142 (3.227) 0.000 18.000 1.912 (3.123) 0.000 14.000 1.428 (3.742) 0.000 23.000 Part Time Working Hours Per Week 12.938 (13.044) 0.000 36.000 15.750 (13.110) 0.000 45.000 19.142 (13.377) 0.000 45.000 Accounting Related Working Exp erience Full Time Part Time Both Full/Part Time None n=35 n=6 n=1 n=28 n=7 n=3 n=2 n=28 n=10 n=3 n=1 Highest Education High School d egree Master degree n=41 n=1 n=26 n=3 n=1 n=37 n=5

PAGE 176

161 oduct-Moment Correlation 5.4.2.1 Correlation between dependent variables. product-moment correlation coefficients and the significances for the variables of the trend analysis task. Results of two-tailed testing of the significance of the correlation between variables at alpha levels of 0.05 and 0.01 are also shown in Table 34. The following paragraphs discuss in detail the most important correlation results, in terms of whether the dependent variables for each hypothesis are significantly correlated with one another. When reporting significant correlation between two dependent variables, the following paragraphs use the symbols of the dependent variables (see Table 11) to describe each of them and report their correlation coefficient through the notation of r To test for the trend analysis task (H1a, H1b, H1c, and H1d), six questions are developed (see Table 3) and each participant answers these six questions in the same order. The first, second, fifth, and sixth questions are used to test hypotheses H1a and H1b (see Table3). Hypothesis H1a has four dependent measures of accuracy TAQ1, TAQ2, TAQ5, and TAQ6 (see Table 6). TAQ1 is positively related to TAQ2 (r = 0.547), TAQ5 (r = 0.277), and TAQ6 (r = 0.335). TAQ2 is positively related to TAQ5 (r = 0.190) and TAQ6 (r = 0.340). TAQ5 is positively related to TAQ6 (r = 0.467). The preceding discussion highlights the fact that the four dependent measures for hypothesis H1a are intercorrelated and MANCOVA is the appropriate statistical method to test hypothesis H1a. Hypothesis H1b has four dependent measures of efficiency TSTAQ1, TSTAQ2, TSTAQ5, and TSTAQ6 (see Table 6). TSTAQ1 is positively related to TSTAQ2 (r = 0.450), TSTAQ5 (r = 0.362), and TSTAQ6 (r = 0.481). TSTAQ2 is positively related to

PAGE 177

162 TSTAQ6 (r = 0.256). However, there is no significant correlation between TSTAQ2 and TSTAQ5. TSTAQ5 is positively related to the time spent, in seconds, by each participant when answering the sixth question TSTAQ6 (r = 0.579). The preceding discussion highlights the fact that the four dependent measures of hypothesis H1b are intercorrelated and MANCOVA is the appropriate statistical method to test hypothesis H1b. The third and fourth questions are used to test hypotheses H1c and H1d (see Table 3). Hypothesis H1c has five dependent measures of accuracy TAQ3, TAQ4a, TAQ4b, TAQ4c, and TAQ4d (see Table 6). TAQ3 is positively related to TAQ4a (r = 0.256), while TAQ4b is positively related to TAQ4c (r = 0.328) (see Table 34). It seems that the five dependent measures of hypothesis H1c were not highly inter-correlated; therefore MANCOVA is not necessary to test hypothesis H1c. However, for consistency and to avoid any potential for over stating the results of hypothesis H1c the study first employed MANCOVA to test hypothesis H1c (see discussion in results section). Hypothesis H1d has two dependent measures of efficiency TSTAQ3 and TSTAQ4 (see Table 6). TSTAQ3 is positively related to TSTAQ4 (r = 0.316), see Table 34. Since the two dependent measures of hypothesis H1d were correlated, MANCOVA is the appropriate statistical method to test hypothesis H1d.

PAGE 178

163 5.4.2.2 Multicollinearity. According to Field (2005) when multicollinearity exists, strong correlation between two predictors, in a regression model it is difficult to assess the individual importance of a predictor. Field (2005) further suggests that one way of identifying multicollinearity is to scan a correlation matrix of all the predictor variables and see if any correlate above 0.80. Table 34 shows that age of the participants (AGE) and full time working experience of the participants (FTWE) are highly correlated at 0.825. However, these two variables, AGE and FTWE were not used simultaneously in any of the final MANCOVA analyses reported in the results section. The problem of multicollinearity is not significant in this study. 5.4.2.3. Correlation between treatment or covariates and dependent variables. Table 34 also shows information about the correlations between the treatment variables and the dependent variables, and between the covariates and demographic variables and the dependent variables. As before, significance is shown at alpha levels of 0.05 and 0.01. All significance levels are based on two-tailed tests. A review of the table indicates that there are some significant correlations between the treatment variable and dependent variables. Additionally, several of the dependent measures are correlated with covariates and demographic variables. Field (2005) comments that caution must be taken when interpreting correlation coefficients because the correlation coefficient says nothing about which variable causes the other to change. However, the correlation matrix does provide preliminary evidence that the treatment may be associated with some of the dependent variables. There is also preliminary evidence that some covariate and demographic variables may be important controls in the MANOVA and ANOVA models. Therefore, the study used regression

PAGE 179

164 analysis for each dependent variable to test the significance of the association between the dependent variables and all of the covariates and demographic variables.

PAGE 180

165 Table 34 -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. TREATMENT PQ1 T SPQ1 PQ2 TSPQ2 TREATMENT 1.000 0.203* 0.191* 0.273* 0.274** PQ1 0.203* 1.000 0.072 0.460** 0.089 TSPQ1 0.191* 0.072 1.000 0.019 0.327** PQ2 0.273** 0.460** 0.019 1.000 0.163 TSPQ2 0.274** 0.089 0.327** 0.163 1.000 PQ3 1.310 0.688** 0.016 0.43 9** 0.150 TSPQ3 0.277** 0.042 0.664** 0.214 0.389** PQ4 0.264** 0.518** 0.041 0.648** 0.045 TSPQ4 0.112 0.056 0.307** 0.220* 0.754** PQ5 0.020 0.162 0.098 0.205* 0.133 TSPQ5 0.053 0.064 0.350** 0.104 0.472** PQ6 0.085 0.154 0.013 0.142 0.170 TSP Q6 0.006 0.089 0.260 0.228 0.372** TAQ1 0.305** 0.412** 0.100 0.591** 0.052 TSTAQ1 0.234** 0.085 0.486** 0.128 0.618** TAQ2 0.126 0.489** 0.120 0.382** 0.098 TSTAQ2 0.073 0.221* 0.282** 0.167 0.332** TAQ3 0.059 0.062 0.300 0.045 0.106 TSTAQ3 0.079 0.057 0.197* 0.189* 0.195* TAQ4a 0.128 0.177* 0.052 0.072 0.072 TAQ4b 0.161 0.148 0.200 0.245** 0.074 TAQ4c 0.032 0.086 0.130 0.165 0.150 TAQ4d 0.016 0.086 0.004 0.013 0.063 TSTAQ4 0.037 0.156 0.189* 0.293** 0.251** TAQ5 0.078 0.305 ** 0.138 0.221* 0.161 TSTAQ5 0.395** 0.371** 0.264** 0.239** 0.441** TAQ6 0.330** 0.089 0.025 0.374** 0.159 TSTAQ6 0.167 0.020 0.374** 0.037 0.562** AGE 0.049 0.040 0.004 0.029 0.349** GEN 0.177* 0.007 0.008 0.032 0.183* SS 0.066 0.055 0.165 0.051 0.218* GPA 0.027 0.188 0.061 0.112 0.177 SAT 0.063 0.012 0.229* 0.051 0.197* FTWE 0.088 0.105 0.136 0.112 0.297* PTH 0.193 0.024 0.111 0.036 0.020 ARWE 0.154 0.046 0.118 0.096 0.012 HE 0.126 0.196* 0.046 0.013 0.204 SMRT 0.084 0.104 0.609 0.161 0.102 TSMRT 0.240** 0.104 0.051 0.173 0.050 MW 0.113 0.229* 0.108 0.191* 0.172 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed) See Table 11 for definition of the variables.

PAGE 181

166 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. PQ3 TSPQ3 PQ4 TSPQ4 PQ5 TREATMENT 0.131 0.277** 0.264* 0.112 0.020 PQ1 0.688** 0.042 0.518** 0.056 0.162 TSPQ1 0.166 0.664** 0.041 0.307** 0.098 PQ2 0.439** 0.214* 0.648** 0.220* 0.205* TSPQ2 0.150 0.389** 0.045 0.754** 0.133 PQ3 1.000 0.171 0.552** 0.189* 0.237** TSPQ3 0.171 1.000 0.224* 0.394** 0.204** PQ4 0.552** 0.224** 1.000 0.218* 0.147 TSPQ4 0.189 0.394** 0.218* 1.000 0.086 PQ5 0.237** 0.204* 0.147 0.086 1.000 TSPQ5 0.067 0.448** 0.111 0.458** 0.090 PQ6 0.140 0.105 0.316** 0.175 0.101 TSPQ6 0.095 0.369** 0.081 0.461** 0.162 TAQ1 0.361** 0.093 0.6 64* 0.105 0.101 TSTAQ1 0.109 0.447** 0.190* 0.723** 0.094 TAQ2 0.439 0.186 0.513** 0.159 0.235** TSTAQ2 0.235 0.327* 0.222* 0.480** 0.088 TAQ3 0.142 0.062 0.058 0.011 0.230* TSTAQ3 0.210* 0.143 0.189 0.091 0.367** TAQ4a 0.039 0.077 0.018 0.07 7 0.006 TAQ4b 0.117 0.049 0.072 0.074 0.085 TAQ4c 0.211* 0.139 0.051 0.168 0.098 TAQ4d 0.090 0.029 0.071 0.062 0.139 TSTAQ4 0.206 0.306** 0.228* 0.240** 0.289** TAQ5 0.400** 0.205 0.262* 0.206* 0.237** TSTAQ5 0.124 0.293** 0.209* 0.401** 0.045 TAQ6 0.495** 0.133 0.397** 0.115 0.192* TSTAQ6 0.006 0.402** 0.075 0.479** 0.049 AGE 0.013 0.041 0.013 0.180* 0.008 GEN 0.160 0.073 0.054 0.072 0.177* SS 0.071 0.169 0.211* 0.227 0.109 GPA 0.058 0.150 0.114 0.081 0.029 SAT 0.029 0.019 0.023 0.450** 0.249* FTWE 0.065 0.067 0.018 0.151 0.046 PTH 0.145 0.023 0.082 0.071 0.041 ARWE 0.054 0.119 0.071 0.089 0.153 HE 0.006 0.051 0.009 0.015 0.030 SMRT 0.24 9** 0.036 0.012 0.091 0.294** TSMRT 0.229* 0.059 0.161 0.039 0.292** MW 0 .238 0.057 0.335** 0.064 0.168 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 182

167 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. TSPQ5 PQ6 TSPQ6 TAQ1 TSTAQ1 TREATMENT 0.053 0.085 0.006 0.305** 0.234** PQ1 0.064 0.154 0.089 0.412** 0.085 TSPQ1 0.350** 0.013 0.260** 0.100 0.486** PQ2 0.104 0.142 0.228* 0.591** 0.128 TSPQ2 0.472** 0.170 0.372** 0.052 0.618** PQ3 0.067 0.140 0.095 0.361** 0.109 TSPQ3 0.448** 0.105 0.369** 0.093 0.447** PQ4 0.111 0.310* 0.081 0.664** 0.190* TSPQ4 0.458** 0.175 0.461* 0.105 0.723** PQ5 0.090 0.101 0.162 0.101 0.094 TSPQ5 1.000 0.148 0.526** 0.003 0.439** PQ6 0.148 1.000 0.123 0.199* 0.124 TSPQ6 0.526** 0.123 1.000 0.012 0.394** TAQ1 0.003 0.199* 0.012 1.000 0.146 TSTAQ1 0.439** 0.124 0.394** 0.146 1.000 TAQ 2 0.067 0.143 0.036** 0.547** 0.203* TSTAQ2 0.390** 0.135 0.137 0.184* 0.450** TAQ3 0.107 0.110 0.216* 0.021 0.098 TSTAQ3 0.279** 0.161 0.003 0.154 0.187* TAQ4a 0.035 0.037 0.083 0.009 0.031 TAQ4b 0.006 0.028 0.195 0.192* 0.049 TAQ4c 0.098 0. 052 0.051 0.022 0.181* TAQ4d 0.050 0.050 0.414** 0.020 0.044 TSTAQ4 0.510** 0.126 0.215* 0.085 0.235** TAQ5 0.138 0.069 0.206* 0.277** 0.212* TSTAQ5 0.304** 0.009 0.206* 0.261** 0.362** TAQ6 0.187* 0.193* 0.171 0.335** 0.034 TSTAQ6 0.413** 0.050 0 .409** 0.099 0.481** AGE 0.290 0.013 0.249 0.016 0.172 GEN 0.061 0.059 0.075 0.047 0.002 SS 0.142 0.005 0.126 0.174 0.230** GPA 0.038 0.032 0.124 0.115 0.011 SAT 0.253** 0.051 0.286** 0.014 0.204* FTWE 0.190* 0.002 0.024* 0.092 0.116 PT H 0.090 0.085 0.050 0.182* 0.030 ARWE 0.054 0.080 0.106 0.076 0.169 HE 0.027 0.043 0.058 0.012 0.124 SMRT 0.068 0.076 0.051 0.175 0.026 TSMRT 0.154 0.006 0.222 0.171 0.000 MW 0,086 0.085 0.028 0.252 0.056 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 183

168 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. TAQ2 TSTAQ2 TAQ3 TSTAQ3 TAQ4a TREATMENT 0.126 0.073 0.059 0.079 0.128 PQ1 0.489** 0.221* 0.062 0.197* 0.057 TSPQ1 0.120 0.282** 0.030 0.197* 0.052 PQ2 0.382** 0.167 0.045 0.189** 0.072 T SPQ2 0.098 0.332** 0.106 0.195* 0.072 PQ3 0.439** 0.235** 0.142 0.210* 0.039 TSPQ3 0.186* 0.327** 0.062 0.143 0.077 PQ4 0.513** 0.222* 0.058 0.189* 0.018 TSPQ4 0.159 0.480** 0.011 0.091 0.077 PQ5 0.235** 0.088 0.230* 0.367** 0.006 TSPQ5 0.06 7 0.390** 0.107 0.279** 0.035 PQ6 0.143 0.135 0.110 0.161 0.037 TSPQ6 0.036 0.336** 0.137 0.216* 0.003 TAQ1 0.547** 0.184* 0.021 0.154 0.009 TSTAQ1 0.203* 0.450** 0.098 0.187* 0.031 TAQ2 1.000 0.427** 0.022 0.121 0.065 TSTAQ2 0.427* 1.000 0.1 34 0.204* 0.121 TAQ3 0.022 0.134 1.000 0.408** 0.256** TSTAQ3 0.121 0.204 0.408** 1.000 0.072 TAQ4a 0.065 0.121 0.256** 0.072 1.000 TAQ4b 0.023 0.086 0.055 0.114 0.013 TAQ4c 0.068 0.107 0.061 0.077 0.102 TAQ4d 0.094 0.043 0.033 0.024 0 .101 TSTAQ4 0.107 0.232* 0.071 0.316** 0.043 TAQ5 0.190* 0.267** 0.175 0.182* 0.026 TSTAQ5 0.235** 0.115 0.083 0.077 0.070 TAQ6 0.340** 0.239** 0.132 0.203* 0.036 TSTAQ6 0.050 0.256** 0.052 0.085 0.054 AGE 0.208* 0.086 0.151 0.065 0.004 GEN 0.024 0.060 0.148 0.120 0.016 SS 0.007 0.102 0.035 0.013 0.089 GPA 0.037 0.116 0.085 0.058 0.007 SAT 0.120 0.419** 0.093 0.049 0.021 FTWE 0.219* 0.149 0.096 0.077 0.075 PTH 0.116 0.054 0.122 0.003 0.041 ARWE 0.030 0.090 0.026 0. 053 0.005 HE 0.130 0.132 0.079 0.062 0.008 SMRT 0.235** 0.132 0.236** 0.337** 0.084 TSMRT 0.186* 0.159 0.110 0.235** 0.023 MW 0.353** 0.098 0.064 0.367 0.071 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 184

169 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. TAQ4b TAQ4c TAQ4d TSTAQ4 TAQ5 TREATMENT 0.161 0.032 0.016 0.037 0.078 PQ1 0.177 0.148 0.086 0.086 0.156 TSPQ1 0.020 0.130 0.004 0.189* 0.138 PQ2 0.245** 0.165 0.013 0.293* 0.221* TSPQ2 0.074 0.150 0.063 0.251** 0.161 PQ3 0.117 0.21 1* 0.090 0.206* 0.400* TSPQ3 0.049 0.139 0.029 0.306* 0.205* PQ4 0.072 0.051 0.071 0.228 0.262** TSPQ4 0.074 0.168 0.062 0.240** 0.206* PQ5 0.085 0.088 0.139 0.289** 0.237* TSPQ5 0.006 0.098 0.050 0.560** 0.138 PQ6 0.028 0.052 0.050 0.126 0.06 9 TSPQ6 0.083 0.195* 0.051 0.414** 0.215* TAQ1 0.192 0.022 0.020 0.085 0.277** TSTAQ1 0.049 0.181* 0.044 0.235* 0.212* TAQ2 0.023 0.068 0.094 0.107 0.190* TSTAQ2 0.086 0.107 0.043 0.232** 0.267** TAQ3 0.055 0.061 0.033 0.071 0.175 TSTAQ3 0.11 4 0.077 0.024 0.316** 0.182* TAQ4a 0.013 0.102 0.101 0.043 0.026 TAQ4b 1.000 0.328** 0.002 0.033 0.306** TAQ4c 0.328** 1.000 0.154 0.083 0.311** TAQ4d 0.002 0.154 1.000 0.250** 0.090 TSTAQ4 0.033 0.083* 0.250** 1.000 0.258** TAQ5 0.306* 0.311* 0.090 0.258* 1.000 TSTAQ5 0.171 0.017 0.067 0.069 0.083 TAQ6 0.129 0.349** 0.076 0.274** 0.467** TSTAQ6 0.066 0.144 0.053 0.187* 0.171 AGE 0.052 0.080 0.011 0.036 0.080 GEN 0.061 0.111 0.138 0.004 0.129 SS 0.041 0.055 0.075 0.021 0.095 GPA 0.006 0.069 0.011 0.061 0.029 SAT 0.007 0.175 0.096 0.019 0.061 FTWE 0.059 0.085 0.011 0.045 0.041 PTH 0.044 0.053 0.019 0.052 0.196* ARWE 0.107 0.302** 0.052 0.066 0.173 HE 0.045 0.008 0.025 0.067 0.010 SMRT 0.117 0.056 0.02 4 0.182* 0.190* TSMRT 0.103 0.096 0.116 0.421* 0.309** MW 0.004 0.154 0.068 0.050 0.214* Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 185

170 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. TSTAQ5 TAQ6 TSTAQ6 AGE GEN TREATMENT 0.395** 0.330** 0.167 0.049 0.177* PQ1 0.305** 0.371** 0.089 0.020 0.040 TSPQ1 0.264** 0.025 0.374** 0.004 0.008 PQ2 0.239** 0.374** 0.037 0.029 0.032 TSPQ2 0.441** 0.159 0.562** 0.349** 0.183* PQ3 0.124 0.495** 0.006 0.013 0.160 TSPQ3 0.293** 0.133 0.402** 0.041 0.07 3 PQ4 0.209 0.397** 0.075 0.013 0.054 TSPQ4 0.401** 0.115 0.479** 0.180** 0.072 PQ5 0.045 0.192* 0.049 0.008 0.177* TSPQ5 0.304** 0.181* 0.413** 0.209* 0.061 PQ6 0.009 0.193* 0.050 0.013 0.059 TSPQ6 0.206* 0.171 0.049** 0.249** 0.075 TAQ1 0.261 0.335* 0.099 0.016 0.047 TSTAQ1 0.362** 0.034 0.481** 0.172 0.002 TAQ2 0.235** 0.340** 0.500 0.208* 0.024 TSTAQ2 0.115 0.239** 0.256** 0.086 0.060 TAQ3 0.0083 0.132 0.052 0.151 0.148 TSTAQ3 0.077 0.203* 0.085 0.065 0.120 TAQ4a 0.07 0 0.036 0.054 0.004 0.016 TAQ4b 0.171 0.129 0.066 0.052 0.061 TAQ4c 0.017 0.349** 0.144 0.080 0.111 TAQ4d 0.067 0.076 0.053 0.011 0.138 TSTAQ4 0.069 0.274** 0.187* 0.036 0.004 TAQ5 0.083 0.467** 0.171 0.080 0.129 TSTAQ5 1.000 0.204* 0.579 ** 0.251** 0.163 TAQ6 0.204** 1.000 0.122 0.128 0.005 TSTAQ6 0.579** 0.122 1.000 0.345** 0.204* AGE 0.251** 0.128 0.345** 1.000 0.014 GEN 0.163 0.005 0.204* 0.014 1.000 SS 0.221* 0.017 0.184* 0.473** 0.155 GPA 0.148 0.023 0.171 0.056 0.077 SAT 0.3 43** 0.052 0.225* 0.001 0.025 FTWE 0.195* 0.114 0.325* 0.825** 0.003 PTH 0.005 0.137 0.026 0.005 0.207* ARWE 0.119 0.053 0.155 0.089 0.110 HE 0.241** 0.067 0.082 0.477** 0.102 SMRT 0.228* 0.147 0.218 0.056 0.311** TSMRT 0.053 0.339* 0. 050 0.051 0.077 MW 0.202* 0.270** 0.178* 0.187* 0.072 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 186

171 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. SS GPA SAT FTWE PTH TREATMENT 0.066 0.027 0.063 0.088 0.193* PQ1 0.007 0.055 0.188 0.012 0.105 TS PQ1 0.165 0.061 0.229** 0.136 0.111 PQ2 0.089 0.032 0.051 0.112 0.036 TSPQ2 0.218* 0.177 0.197* 0.297** 0.020 PQ3 0.071 0.058 0.029 0.065 0.145 TSPQ3 0.169 0.150 0.109 0.067 0.023 PQ4 0.211* 0.114 0.023 0.018 0.082 TSPQ4 0.227* 0.081 0.450** 0.151 0.071 PQ5 0.109 0.029 0.249* 0.046 0.041 TSPQ5 0.142 0.038 0.253* 0.190* 0.090 PQ6 0.005 0.032 0.051 0.002 0.085 TSPQ6 0.126 0.124 0.286** 0.224* 0.050 TAQ1 0.174 0.115 0.014 0.092 0.182* TSTAQ1 0.236** 0.011 0.204** 0.116 0 .030 TAQ2 0.007 0.037 0.120 0.219* 0.116 TSTAQ2 0.102 0.116 0.419** 0.149 0.054 TAQ3 0.035 0.085 0.093 0.096 0.122 TSTAQ3 0.013 0.058 0.049 0.077 0.003 TAQ4a 0.089 0.007 0.021 0.075 0.041 TAQ4b 0.041 0.006 0.007 0.059 0.044 TAQ4c 0.0 55 0.069 0.175 0.085 0.053 TAQ4d 0.075 0.011 0.096 0.011** 0.019 TSTAQ4 0.021 0.061 0.019 0.045 0.052 TAQ5 0.095 0.029 0.061 0.041 0.196* TSTAQ5 0.221* 0.148 0.343** 0.195* 0.005 TAQ6 0.017 0.023 0.052 0.114 0.137 TSTAQ6 0.184* 0.171 0.2 25* 0.325** 0.026 AGE 0.473** 0.056 0.001 0.825** 0.005 GEN 0.155 0.077 0.025 0.003 0.207* SS 1.000 0.082 0.181 0.308** 0.027 GPA 0.082 1.000 0.060 0.145 0.165 SAT 0.181 0.060 1.000 0.038 0.060 FTWE 0.308** 0.145 0.038 1.000 0.038 PTH 0.027 0.165 0.060 0.038 1.000 ARWE 0.260** 0.015 0.068 0.099 0.029 HE 0.391** 0.019 0.058 0.347* 0.153 SMRT 0.061 0.080 0.159 0.043 0.163 TSMRT 0.074 0.099 0.053 0.059 0.158 MW 0.068 0.036 0.169 0.176 0.007 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables. The number of observations for the variable of SAT is 10 3.

PAGE 187

172 Table 34 (Continued) -Moment Correlation Coefficients of the Variables of the Trend Analysis Task. ARWE HE SMRT TSMRT MW TREATMENT 0.154 0.126 0.084 0.204 0.113 PQ1 0.024 0.046 0.196* 0.104 0.229* TSPQ1 0.118 0.046 0.069 0.05 1 0.108 PQ2 0.096 0.013 0.161 0.173 0.191* TSPQ2 0.012 0.204* 0.102 0.050 0.172 PQ3 0.054 0.006 0.249** 0.229* 0.238** TSPQ3 0.119 0.051 0.306 0.059 0.057 PQ4 0.071 0.009 0.120 0.161 0.335* TSPQ4 0.089 0.015 0.091 0.039 0.064 PQ5 0.15 3 0.030 0.294** 0.292** 0.168 TSPQ5 0.054 0.027 0.068 0.154 0.086 PQ6 0.080 0.043 0.076 0.006 0.085 TSPQ6 0.106 0.058 0.051 0.222* 0.028 TAQ1 0.076 0.012 0.175 0.171 0.252** TSTAQ1 0.169 0.124 0.026 0.000 0.056 TAQ2 0.030 0.130 0.235** 0.18 6* 0.353 TSTAQ2 0.090 0.132 0.132 0.159 0.098 TAQ3 0.026 0.079 0.236** 0.110 0.064 TSTAQ3 0.053 0.062 0.337** 0.235** 0.037 TAQ4a 0.005 0.008 0.084 0.023 0.071 TAQ4b 0.107 0.045 0.117 0.103 0.004 TAQ4c 0.302** 0.008 0.056 0.096 0.154 TAQ4d 0.052 0.025 0.024 0.116 0.068 TSTAQ4 0.066 0.067 0.182* 0.421** 0.050 TAQ5 0.173 0.010 0.190* 0.309** 0.214* TSTAQ5 0.119 0.241** 0.228** 0.053 0.202* TAQ6 0.053 0.067 0.147 0.339** 0.270** TSTAQ6 0.155 0.082 0.218* 0.050 0.178 AGE 0.089 0.477** 0.056 0.051 0.187* GEN 0.110 0.102 0.311** 0.077 0.072 SS 0.260** 0.391** 0.061 0.074 0.068 GPA 0.015 0.019 0.080 0.099 0.036 SAT 0.068 0.058 0.159 0.053 0.169 FTWE 0.099 0.347** 0.043 0.059 0.176 PTH 0.029 0.153 1. 630 0.158 0.007 ARWE 1.000 0.219* 0.018 0.123 0.010 HE 0.219* 1.000 0.093 0.032 0.098 SMRT 0.018 0.093 1.000 0.592** 0.113 TSMRT 0.123 0.032 0.592* 1.000 0.213* MW 0.090 0.098 0.113 0.213* 1.000 Figures shown in the table are Pearson Correlation Coefficients. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed). See Table 11 for definition of the variables.

PAGE 188

173 5.4.3 Regression Analysis of Possible Covariates Regression models allow researchers to estimate the individual contribution of each predictor to the model. If the t-statistic associated with a predictor has a significant alpha value, then the predictor is making a significant contribution to the model (Field, 2005). This study developed separate regression models for each dependent measure using twenty-four possible predictors. Only those predictors that were common covariates across the models used to test a single hypothesis were retained in the models. A complete dis covariate measures are: the score on each of the six practice questions (PQ1, PQ2, PQ3, PQ4, PQ5 and PQ6) (see Table 5), the time spent in seconds by each participant when answering each of the six practice questions (TSPQ1, TSPQ2, TSPQ3, TSPQ4, TSPQ5, TSPQ6) (see Table 5), the score on the Mental Rotations Test (SMRT), the time spent in seconds by each participant when answering the Mental Rotations Test (TSMRT), mental workload (MW), gender of the participants (GEN), and age of the participants (AGE). The demographic data of the participants that were considered are: student status (SS), undergraduate overall GPA (GPA), SAT score (SAT), amount of full time working experience (FTWE), part time working hours (PTH), accounting related working experience (ARWE), and highest level of education (HE). Hypothesis H1a has four dependent measures of accuracy the second question (TAQ2), the score on the fifth question (TAQ5), and the score on the sixth question (TAQ6). HypothesisH1b uses the time spent on each of the tasks used to test hypothesis H1a to de

PAGE 189

174 seconds answering TAQ1 (TSTAQ1), TAQ2 (TSTAQ2), TAQ5 (TSTAQ5), and TAQ6 (TSTAQ6) question (TAQ3), the score on pa the score on part d of the fourth question (TAQ4d). Two measures of efficiency are used to test hypothesis answering TAQ3 (TSTAQ3), and when answering all four parts of the fourth question; that is, TAQ4a-TAQ4d (TSTAQ4). In total there are fifteen dependent measures for hypotheses H1a, H1b, H1c and H1d (see Table 3 and 17). A separate regression model was run for each of the fifteen dependent measures using the twenty-four possible predictors mentioned above in each regression model. Dependent Measures (TAQ1-6 and TSTAQ1-6) = b 0 + b1 Treatment + b 2 PQ1 + b3 TSPQ1 + b 4 PQ2 + b 5 TSPQ2 + b 6 PQ3 + b 7 TSPQ3 + b8 PQ4 + b 9 TSPQ4 + b10 PQ5 + b 11TSPQ5 + b 12 PQ6 + b 13 TSPQ6 + b 14AGE + b15GEN + b16SS + b17GPA + b 18 SAT + b 19FTWE + b 20 PTH + b 21ARWE + b 22HE + b23 SMT + b24TSMRT + b 25 MW + e. The results of the analysis indicated that for H1a the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H1a: 1) practice question one (PQ1), 2) practice question three (PQ3), 3)

PAGE 190

175 time spent on practice question five (TSPQ5), 4) age (AGE), 5) part time working hours (PTH), 6) highest level of education (HE), and 7) mental workload (MW). For hypothesis H1b the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H1b: 1) practice question one (PQ1), 2) time spent on practice question four (TSPQ4), 3) SAT score (SAT), 4) accounting related working experience (ARWE), 5) highest level of education (HE), 6) the score on Mental Rotations Test (SMRT). For hypothesis H1c the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H1c: 1) highest level of education (HE), 2) time spent on mental rotations test (TSMRT). Finally, for hypothesis H1d the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H1d: 1) practice question five (PQ5), 2) time spent on practice question 5 (TSPQ5), 3) score on the Mental Rotations Test (SMRT), 4) and time spent on Mental Rotations Test (TSMRT). The following sections discuss the results of testing hypotheses H1a, H1b, H1c, and H1d. 5.4.4 Results of H1a Four dependent variables are used to test the trend analysis task h the score on the first question (TAQ1), the score on the second question (TAQ2), the score on the fifth question (TAQ5), and the score on the sixth question (TAQ6). In tice question one (PQ1), practice question three (PQ3), time spent on practice question five (TSPQ5), age (AGE), part time working hours (PTH), highest level of education (HE), and mental

PAGE 191

176 4.3) along with the manipulated variable Treatment. The first dependent variable used to test hypothesis H1a was TAQ1, which asked what the data differences were between years 1 and 4 (see Table 3). Table 35, Panel A indicates that on average those participants viewing the 2-D displays were the most accurate (had the highest score) on this trend analysis task. Those participants viewing a 3-D display (mean score 2.595) were 30% less accurate than those viewing the 2D display (mean score 3.650) and those using the tabular display (mean score 3.452) were 5% less accurate than those viewing the 2-D display (mean score 3.650). The second dependent variable (TAQ2) used to test hypothesis H1a asked what participants perceived to be occurring in the data when going from year 2 to year 3 and year 4 (see Table 3). Table 35, Panel A indicates that on average those participants viewing the 2-D displays were the most accurate (had the highest score) on this trend analysis task. Those participants viewing a 3-D perspective display (mean score 4.690) were 20% less accurate than those viewing the 2-D display (mean score 5.925) and those using the tabular display (mean score 5.476) were 7% less accurate than those viewing the 2-D displays (mean score 5.925). The third dependent variable (TAQ5) used to test hypothesis H1a asked participants to select from a template indicating the differences in data between years 2 and 4 (see Table 3). On average those participants viewing the 2-D displays were again the most accurate. Those participants viewing a 3-D perspective display (mean score 3.642) were 5% less accurate than those viewing the 2-D displays (mean score 3.850)

PAGE 192

177 while those viewing the tabular display (3.761) were 2% less accurate than those viewing the 2-D displays (mean score 3.850) (Table 35, Panel A). Finally, the last dependent variable (TAQ6) used to test H1a asked participants to select from a template indicating changes in data as you go from year 1 to year 2 to year 3 (see Table 3). As indicated (Table 35, Panel A), on average participants viewing the 2D displays were the most accurate. Those participants viewing a 3-D perspective display (mean score 6.357) were 16% less accurate than those viewing the 2-D displays (mean score 7.625) while those viewing the tabular display (7.547) were 1% less accurate than those viewing the 2-D displays (mean score 7.625) (Table 35, Panel A). As discussed in this and the prior three paragraphs, the mean results for all dependent variables are as hypothesized in H1a, which predicted that participants viewing a set of 2-D displays will be the most effective in generating hypotheses for what caused the changes in the trend of accounting data when compared to participants using a single 3-D perspective display or participants using a table. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1a, a MANCOVA analysis was conducted. As shown (Table 35, Panel B), the overall F-statistic for the manipulated variable Treatment is significant (p = 0.001) using Pi These significant results allow for analysis of the univariate results which are provided on Panel C of Table 35. Panel C indicates that the only covariate significantly (p = 0.040) associated with TAQ1 is PQ1. Results indicate that manipulation of the presentation formats (Treatment) is significantly (p = 0.001) associated with the accuracy of the participants in describing

PAGE 193

178 the differences between years 1 and 4 (TAQ1). A paired comparison test (Table 35, Panel D) shows that the participants viewing the 2D displays were significantly (p=0.001) more effective or accurate than those participants viewing the 3-D perspective display in this trend analysis task (TAQ1). There was no significant difference in the effectiveness, or accuracy between participants viewing the 2-D displays and participants viewing the tabular display. Thus, the paired comparison tests provide partial support for H1a, which predicted that participants viewing a set of 2-D displays will be the most effective in generating hypotheses for what caused the changes in the trend of accounting data when compared to participants using a single 3-D perspective display or participants using a table. Panel C indicates that the covariates, PQ1 (p = 0.003), and MW (p = 0.011) are all significantly associated with TAQ2. Contrary to expectation, the results suggest that the manipulation of the presentation formats (Treatment) does not have an significant (p = 0.310) effect on the accuracy of the participants in describing what is happening as you go from year 2 to year 3 to year 4 (TAQ2). Since there is no significant main effect a paired comparison test was not conducted. Panel C indicates that the covariates, PQ1 (p = 0.049), PQ3 (p < 0.001), PTH (p = 0.047), and MW (p = 0.050) are all significantly associated with TAQ5. Again it was found that, contrary to expectations, the results suggest that the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.296) effect on the accuracy of the participants in describing the data differences between years 2 and 4 (TAQ5). Thus, a paired comparison test was not conducted.

PAGE 194

179 Finally, Panel C, indicates that the covariates PQ3 (p < 0.001), TSPQ5 (p = 0.016), AGE (p = 0.027), and MW (p = 0.023) are significantly associated with the of the presentation formats (Treatment) has a significant (p < 0.001) effect on the accuracy of the participants in describing what is happening to the data as you go from year 1 to year 2 to year 3 (TAQ6). A paired comparison test was conducted to determine if the participants viewing the 2-D displays were more effective or accurate than those participants reviewing the tabular or 3-D perspective display. Results revealed that the participants viewing the 2-D display were significantly (p < 0.001) more accurate than participants viewing the 3-D perspective display (see Table 35, Panel D). There was no significant difference in effectiveness or accuracy between participants viewing the 2D displays and participants viewing the tabular display. Thus, the paired comparison tests provide partial support for H1a, which predicted that participants viewing a set of 2D displays will be the most effective in generating hypotheses for what caused the changes in the trend of accounting data when compared to participants using a single 3D perspective display or participants using a table.

PAGE 195

180 Table 35 Test Results of H1a (Participants using a set of 2-D displays will be the most effective in generating hypotheses for what caused the changes in the trend of accounting data) MANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task Tests of BetweenSubjects Effects on Effectiveness Panel A: Mean Scores on the Tasks. Dependent Variable Treatment Actual Mean MANCOVA Adjusted Mean* TAQ1 Tabular Display (n=42) 3.452 3.368 2 D Displays (n=40) 3.650 3.558 3 D Perspective Display (n=42) 2.595 2.767 TAQ 2 Tabular Display (n=42) 5.476 5.180 2 D Displays (n=40) 5.925 5.632 3 D Perspective Display (n=42) 4.690 5.266 TAQ5 Tabular Display (n=42) 3.761 3.723 2 D Displays (n=40) 3.850 3.828 3 D Perspective Display (n=42) 3.642 3.703 TAQ6 Tabular Disp lay (n=42) 7.547 7.435 2 D Displays (n=40) 7.625 7.543 3 D Perspective Display (n=42) 6.357 6.548 *Adjusted Mean is for the effects of the covariates. Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept Pilla 0.615 44.256 < 0.001 PQ1 0.117 3.684 0.007 PQ3 0.174 5.487 < 0.001 TSPQ5 0.059 1.741 0.146 AGE 0.071 2.130 0.082 PTH 0.046 1.344 0.262 HE 0.032 0.931 0.449 MW 0.101 3.112 0.018 Treatment 0.224 3.527 0.001 PQ1= score on practice question 1; PQ2 = score on practice question 2; PQ3 = score on practice question 3 TSPQ5 = time spent on practice question five. PTH = part time working hours. HE = highest level of education. MW = mental workload

PAGE 196

181 Table 35: Test Results of H1a (Continued) Panel C: ANCOVA Results Using Scores as the Dependent Variables Dependent Variables Source of Variation Type III SS DF Mean Square F Stat p value* TAQ1 Corrected Model 49.628 9 5.514 5,417 <0.001 Intercept 22.581 1 22.581 22.182 <0.001 PQ1 4.406 1 4.406 4.328 0.040 PQ3 0.703 1 0.703 0.690 0.408 TSPQ5 0.731 1 0.731 0.718 0.399 AGE 0.003 1 0.003 0.003 0.956 PTH 1.683 1 1.683 1.653 0.201 HE 0.011 1 0.011 0.100 0.919 MW 2.910 1 2.910 2.859 0.094 Treatment 12.488 2 6.244 6.134 0.001 Error 116.049 114 1.018 Total 1456.000 124 Corrected Total 165.677 123 TAQ2 Corrected Model 287.861 9 31.985 6.899 <0.001 Intercept 91.963 1 91.963 19.836 <0.001 PQ1 41.917 1 41.917 9.041 0.003 PQ3 9.017 1 9.017 1.945 0.166 TSPQ5 0.059 1 0.059 0.013 0.910 AGE 10.057 1 10.057 2.169 0.144 PTH 4.340 1 4.340 0.936 0.335 HE 3.972 1 3.972 0.857 0.357 MW 30.812 1 30.812 6.646 0. 011 Treatment 4.436 2 2.218 0.478 0.310 Error 528.526 114 4.636 Total 4372.000 124 Corrected Total 816.387 123 TAQ5 Corrected Model 12.716 9 1.413 4.409 <0.001 Intercept 40.464 1 40.464 126.265 <0.001 PQ1 1.265 1 1.265 3.948 0.049 PQ 3 5.386 1 5.386 16.807 <0.001 TSPQ5 0.579 1 0.579 1.807 0.182 AGE 0.410 1 0.410 1.279 0.260 PTH 1.296 1 1.296 4.044 0.047 HE 0.186 1 0.186 0.580 0.448 MW 1.255 1 1.255 3.916 0.050 Treatment 0.336 2 0.168 0.525 0.296 Error 36.534 114 0.320 Total 1793.000 124 Corrected Total 49.250 123

PAGE 197

182 Table 35: Test Results of H1a (Continued) Panel C: ANCOVA Results Using Scores as the Dependent Variables Dependent Variables Source of Variation Type III SS DF Mean Square F Stat p value* TAQ6 Corr ected Model 116.879 9 12.987 9.456 <0.001 Intercept 120.368 1 120.368 87.644 <0.001 PQ1 0.044 1 0.044 0.032 0.858 PQ3 17.586 1 17.586 12.805 <0.001 TSPQ5 8.244 1 8.244 6.003 0.016 AGE 6.910 1 6.910 5.031 0.027 PTH 1.373 1 1.373 1.000 0.319 HE 4.014 1 4.014 2.923 0.090 MW 7.297 1 7.297 5.313 0.023 Treatment 21.835 2 10.917 7.949 <0.001 Error 156.564 114 1.373 Total 6647.000 124 Corrected Total 273.444 123 TAQ1 Adjusted R Squared = 0.244. TAQ2 Adjusted R Squared = 0.301. TAQ5 Adjusted R Squared = 0.200. TAQ6 Adjusted R Squared = 0.382. PQ1= practice question 1; PQ2 = practice question 2; PQ3 = practice question 3, TSPQ5 = time spent on practice question five. PTH = part time working hours. HE = highest level of education. MW = mental workload *Treatment p-values are one-tail, all others are two-tail. Panel D: Bonferroni Pairwise Comparisons for Test H1a Dependent Variables (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* TAQ1 2 D Displays Tabular Di splay 3 D Display 0.190 0.791 0.229 0.235 0.500 0.001 TAQ6 2 D Displays Tabular Display 3 D Display 0.108 0.995 0.266 0.273 0.500 <0.001 *p -values are one-tail.

PAGE 198

183 5.4.5 Results of H1b Four dependent measures are used to test the trend analysis task hypothesis the time spent in seconds by each participant when answering the first question (TSTAQ1), the time spent in seconds by each participant when answering the second question (TSTAQ2), the time spent in seconds by each participant when answering the fifth question (TSTAQ5), and the time spent in seconds by each participant when answering the sixth question (TSTAQ6). In constructing the models to test hypothesis H1b, six covariates practice question one (PQ1), time spent on practice question four (TSPQ4), accounting related working experience (ARWE), highest level of education (HE) and score on Mental Rotations Test (SMRT) were included in the model (see section 5.4.3) along with the manipulated variable Treatment Section 5.3.1 reports that out of the 42 participants viewing the tabular display, only 33 participants had taken the SAT. Out of the 40 participants viewing the 2D displays, only 32 participants had taken the SAT. Out of the 42 participants viewing the 3-D perspective display, only 37 participants had taken the SAT. By including the SAT score as a covariate in the MANCOVA analysis, the cell sizes not only reduced to 33, 32, and 37 for the tabular display, 2-D displays and 3-D perspective display treatment group, respectively, but the cell sizes also become unequal. With unequal cell sizes among the treatment groups, multivariate tests are no longer robust to the violation of the assumptions of MANCOVA. A separate MANCOVA was run including the SAT variable and the TSTAQ5 results are unchanged and the TSTAQ2 results become is excluded from the reported results.

PAGE 199

184 The first dependent variable used to test H1b was TSTAQ1 the time spent in seconds by each participant when answering what the data differences were between years one and four (see Table 3). Table 36, Panel A indicates that on average those participants viewing the tabular display were the most efficient (used the least time in seconds) on this trend analysis task. Those participants viewing a 3-D perspective display (mean seconds 121) used 26% more time (in seconds), than those viewing the 2D displays (mean seconds 96). But, those participants viewing the 2-D displays (mean seconds 96) used 4% more time (in seconds) than those viewing the tabular display (mean seconds 92). The second dependent variable TSTAQ2 used to test H1b was the time spent in seconds by each participant when answering what he or she perceived to be occurring in the data when going from year 2 to year 3 and year 4 (see Table 3). Table 36, Panel A indicates that on average those participants viewing the 2-D displays were the most efficient (used the least time in seconds) on this trend analysis task. Those participants viewing a 3-D display (mean seconds 152) used 20% more time (in seconds), than those viewing the 2-D displays (mean seconds 127). Those participants viewing a tabular display (mean seconds 136) used 7% more time (in seconds), than those viewing the 2D displays (mean seconds 127). The third dependent variable TSTAQ5 used to test H1b was the time spent in seconds by each participant when selecting choices from a template to indicate the differences in data between years 2 and 4 (see Table 3). Table 36, Panel A indicates on average those participants viewing the 2-D displays and those participants viewing the tabular spent the same amount of time in seconds (mean seconds 35). Those participants

PAGE 200

185 viewing a 3-D perspective display (mean seconds 48) used 37% more time (in seconds), than those viewing the 2-D displays (mean seconds 35). The fourth dependent variable TSTAQ6 used to test H1b was the time spent in seconds by each participant when selecting choices from a template to indicate the changes in data as you go from year 1 to year 2 to year 3 (see Table 3). Table 36, Panel A indicates on average those participants viewing the 2-D displays were the most efficient (used the least time in seconds) on this trend analysis task. Those participants viewing a 3-D display (mean seconds 81) used 39% more time than those viewing the 2D displays (mean seconds 58). Those participants viewing the tabular display (mean se conds 65) used 12% more time (in seconds) than those viewing the 2-D display (mean seconds 58). The mean results for th e four dependent variables from this and the preceding paragraph provide mixed evidence of support for H1b, which predicted that participants viewing a set of 2-D displays will be the most efficient in generating hypotheses for what caused the changes in the trend of accounting data when compared to participants using a single 3-D perspective display or participants viewing a table. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1b, a MANCOVA analysis was conducted. As shown (Table 36, Panel B), the overall F-statistic for the manipulated variable Treatment is significant (p < 0.001) These significant results allow for analysis of the univariate results which are provided on Panel C of Table 32. Panel C indicates that the covariates TSPQ4 (p < 0.001), ARWE (p = 0.010), and HE (p = 0.039) are significantly associated with the time spent in seconds by each

PAGE 201

186 participant (TSTAQ1) when answering the first question in the trend analysis task. Results indicate that manipulation of the presentation formats (Treatment) is significantly (p = 0.005) associated with the time spent in seconds by each participant (TSTAQ1) when describing what the data differences were between years 1 and 4. A paired comparison test (Table 36, Panel D) shows that there was no significant difference in the efficiency (time used in seconds) between participants viewing the tabular display, viewing the 2-D displays, and participants viewing the 3-D perspective display when answering what the data differences were between years 1 and 4. Panel C, indicates that the covariates PQ1 (p = 0.013), and TSPQ4 (p <0.001) are significantly associated with the time spent in seconds by each participant (TSTAQ2) when answering the second question in the trend analysis task. Contrary to expectations, the results suggest that manipulation of the presentation formats (Treatment) does not have a significant (p = 0.182) effect on the time spent in seconds by each participant (TSTAQ2) when describing what he or she perceived to be occurring in the data when going from year 2 to year 3 and year 4. Since there is not significant main effect a paired comparison test was not conducted. Panel C, indicates that the covariates PQ1 (p = 0.003), TSPQ4 (p < 0.001), ARWE (p = 0.019), and HE (p = 0.001) are associated with the time spent in seconds by each participant (TSTAQ5) when answering the fifth question in the trend analysis task. Re sults suggest that the manipulation of the presentation formats (Treatment) is significantly (p <0.001) associated with the time spent by participants (TSTAQ5) when describing the data differences between years 2 and 4. A paired comparison test (Table 36, Panel D) shows that the participants viewing the 2-D displays were significantly (p <

PAGE 202

187 0.001) more efficient or used less time in seconds than participants viewing the 3D perspective display when describing the data differences between years 2 and 4. There was no significant difference in the efficiency (time used in seconds) between participants viewing the 2-D displays and participants viewing the tabular display when answering the fifth question in the trend analysis task. Thus, the paired comparison tests provide partial support for hypothesis H1b, which predicted that participants viewing a set of 2-D displays will be the most efficient in generating hypothesis for what caused the changes in the trend of accounting data when compared to participants using a single 3D perspective display or participants viewing a table. Panel C, indicates that the covariates TSPQ4 (p < 0.001), and SMRT (p = 0.049) are associated with the time spent in seconds by each participant when answering the sixth question in the trend analysis task TSTAQ6. Contrary to expectations, results suggest that manipulation of the presentation formats (Treatment) does not have a significant (p = 0.057) effect on the time spent by the participants in describing what is happening as you go from year 1 to year 2 to year 3 (TSTAQ6). Since, there is no significant main effect a paired comparison test was not conducted.

PAGE 203

188 Table 36 Test Results of H1b (P articipants viewing a set of 2-D displays will be the most efficient in generating hypothesis for what caused the changes in the trend of accounting data) MANCOVA Model on Efficiency (Less Time) in Trend Analysis Task Tests of Between-Subjects Effects on Efficiency Panel A: Mean Time Spent on the Tasks. Dependent Variable Treatment Actual Me an (Seconds) MANCOVA Adjusted Mean* (Seconds) TSTAQ1 Tabular Display (n=42) 92.785 98.230 2 D Displays (n=40) 96.975 100.900 3 D Perspective Display (n=42) 121.904 117.100 TSTAQ2 Tabular Display (n=42) 136.452 132.500 2 D Displays (n=40) 127.525 1 30.400 3 D Perspective Display (n=42) 152.357 153.500 TSTAQ5 Tabular Display (n=42) 35.309 36.341 2 D Displays (n=40) 35.400 36.632 3 D Perspective Display (n=42) 48.452 46.247 TSTAQ6 Tabular Display (n=42) 65.523 66.999 2 D Displays (n=40) 58.5 00 61.243 3 D Perspective Display (n=42) 81.333 77.246 *Adjusted Mean is for the effects of the covariates. Panel B: Multivariae Tests Variables Multivariate Test Value F Stat |p value| Intercept 0.302 12.220 < 0.001 PQ1 ce 0.125 4.030 0.004 TSPQ4 0.594 41.274 < 0.001 ARWE 0.100 3.143 0.017 HE 0.158 5.290 0.001 SMRT 0.067 2.042 0.093 Treatment 0.235 3.791 <0.001 PQ1 = score on practice question 1. HE = highest level of education. TSPQ4 = time spent in seconds by each participant when answering practice question 4. ARWE = accounting related working experience. SMRT = the score on Mental Rotations Test

PAGE 204

189 Table 36: Test Results of H1b (Continued) Panel C: ANCOVA Results Using Time Spent as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSTAQ1 Corrected Model 191337.073 7 27333.868 23.887 <0.001 Intercept 1188.392 1 1188.392 1.039 0. 310 PQ1 1889.213 1 1889.213 1.651 0.201 TSPQ4 143034.257 1 143034.25 7 124.995 <0.001 ARWE 7832.995 1 7832.995 6.845 0.010 HE 5009.286 1 5009.286 4.378 0.039 SMRT 610.019 1 610.019 0.533 0.467 Treatment 10698.636 2 5349.318 4.675 0.005 Error 1 32740.927 116 1144.318 Total 1665262.00 0 124 Corrected Total 324078.000 123 TSTAQ2 Corrected Model 320879.012 7 45839.859 7.740 <0.001 Intercept 7939.585 1 7939.585 1.341 0.249 PQ1 37955.049 1 37955.049 6.409 0.013 TSPQ4 205254.519 1 205 254.51 9 34.657 <0.001 ARWE 819.495 1 819.495 0.138 0.711 HE 17593.043 1 17593.043 2.971 0.087 SMRT 15309.892 1 15309.382 2.585 0.111 Treatment 12083.532 2 6041.766 1.020 0.182 Error 687007.787 116 5922.481 Total 3402301.00 0 124 Corrected Total 1007886.79 8 123 TSTAQ5 Corrected Model 10731.338 7 1533.048 14.160 <0.001 Intercept 5175.406 1 5175.406 47.801 <0.001 PQ1 1004.753 1 1004.753 9.280 0.003 TSPQ4 2516.580 1 2516.580 23.244 <0.001 ARWE 614.737 1 614.737 5.678 0.019 HE 132 5.330 1 1325.330 12.241 0.001 SMRT 272.048 1 272.048 2.513 0.116 Treatment 2368.839 2 1184.419 10.940 <0.001 Error 12559.210 116 108.269 Total 219616.000 124 Corrected Total 23290.548 123

PAGE 205

190 Table 36: Test Results of H1b (Continued) Panel C: ANCOVA Results Using Time Spent as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSTAQ6 Corrected Model 59297.801 7 8471.114 7.593 <0.001 Intercept 11041.823 1 11041.823 9.897 0.002 PQ1 245.762 1 245.762 0.220 0.640 TSPQ4 32975.793 1 32975.793 29.558 <0.001 ARWE 3476.405 1 3476.405 3.116 0.080 HE 1445.381 1 1445.381 1.296 0.257 SMRT 4434.606 1 4434.606 3.975 0.049 Treatment 4909.006 2 2454.503 2.200 0.057 Error 129413.619 116 1115.635 Total 772470.000 124 Corrected Total 188711.419 123 TSTAQ1 Adjusted R Squared = 0.566. TSTAQ2 Adjusted R Squared = 0.277 TSTAQ5 Adjusted R Squared = 0.428. TSTAQ6 Adjusted R Squared = 0.273 PQ1 = score on practice question 1. TSPQ4 = time spent in seconds by each participant when answering practice question 4. ARWE = accounting related working experience. HE = highest Level of Education. SMRT = the score on Mental Rotations Test *Treatment p-values are one-tail, all others are two-tail Panel D: Bonferroni Pairwise Comparisons for Test H1b Dependent Variables (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* TSTAQ1 2 D Displays Tabular Display 3 D Display 7.087 16.188 7.633 7.808 0.500 0.060 TSTAQ5 2 D Displays Tabular Display 3 D Display 0.291 9.614 2.348 2.402 0.500 <0.001 *p -values are one-tail.

PAGE 206

191 5.4.6 Results of H1c question (TAQ4a, b, c, d). In constructing the models to test hypothesis H1c, two 4.3) along with the manipulated variable Treatment. The first dependent variable used to test hypothesis H1c w as TAQ3, which asked the participants to estimate what the ROE would be in year 6 (TAQ3) if each of the variables comprising ROE in year 5 doubles. The correct answer to the third question is response and the correct answer is the dependent measure (accuracy) used to test H1c (TAQ3) (see Table 3). Lower scores on this measure indicate greater accuracy. Table 37, Panel A, shows that all participants had a mean estimation of year 6 ROE (TAQ3) that was less than the correct answer by 71.242 (tabular), 87.110 (2-D), and 63.876 (3-D). The other dependent variable used to test hypothesis H1c was question four, which asked the participants to estimate the average of turnover (TAQ4a), profitability (TAQ4b) and leverage (TAQ4c) for the years 1, 2, 4 and 5, and use the estimated average to calculate a new ROE (TAQ4d). The correct answers to the fourth question are: (a) average turnover = 1.045, (b) average profitability = 3.955%, (c) average leverage = 2.72 responses and the correct answers are the dependent measures (accuracy) used to test H1c (see Table 3). Again, lower scores on this measure indicate greater accuracy.

PAGE 207

192 Ta ble 37, Panel A provides mean results for TAQ4a-d. For TAQ4a participant responses are greater than the correct answer by 5.001 (tabular), 1.503 (2-D), and 0.874 (3 correct answer for tabular display (0.796), and 2-D displays (0.626), while the mean answer for the 3-D perspective display was greater than the correct answer by 1.541. For correct answer by 0.032 (tabular), 1.898 (2-D), and 0.322 (3-D). Finally, for TAQ4d, 2-D (16.392) and 3-D (1.055), while answers were less than the correct answer for the tabular display (0.359) dependent measures (accuracy) used to test hypothesis H1c. The differences between ositive or negative value. The mean results for the dependent variables from the preceding paragraphs provide little support for H1c, which predicted that participants viewing a set of 2D displays will be the most effective in an accounting judgment involving estimation of values when compared to participants using a single 3-D perspective display or participants using a table. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1c, a MANCOVA analysis was conducted. As shown (Table 37, Panel B), the overall F-statistic for the manipulated variable Treatment is significant (p = 0.005)

PAGE 208

193 These significant results allow for analysis of the univariate results, which are provided on Panel B of Table 37. Panel C indicates that manipulation of the presentation formats (Treatment) is significantly (p = 0.032) associated with the accuracy of the participants in estimating what the ROE would be in year 6 (TAQ3) if each of the variables comprising ROE in year 5 doubles. A paired comparison test (Table 37, Panel D) shows that the participants viewing the 3-D perspective display were more (p = 0.031) effective or accurate than those participants viewing the 2-D displays in this trend analysis task (TAQ3). Thus, hypothesis H1c predicting that participants viewing a set of 2-D displays will be the most effective in an accounting judgment involving estimation of values, when compared to participants using a single 3-D perspective display or participants using a table, was not supported. Contrary to expectation, Panel C indicates that the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.170) effect on the accuracy of the participants in estimating the average of turnover for the years 1, 2, 4 and 5 (TAQ4a). Since there is no significant main effect a paired comparison test was not conducted. Thus, hypothesis H1c predicting that participants viewing a set of 2D displays will be the most effective in an accounting judgment involving estimation of values, when compared to participants using a single 3-D perspective display or participants using a table, was not supported. Panel C indicates that the manipulation of the presentation formats (Treatment) does not have a significant (p = 0.094) effect on the accuracy of the participants in

PAGE 209

194 estimating the average of profitability for the years 1, 2, 4 and 5 (TAQ4b). Since there is no significant main effect a paired comparison test was not conducted. Panel C indicates that manipulation of the presentation formats (Treatment) is significantly (p = 0.018) associated with the accuracy of the participants in estimating the average of leverage for the years 1, 2, 4 and 5 (TAQ4c). A paired comparison test (Table 37, Panel D) shows that the participants viewing the tabular display were more (p = 0.032) effective or accurate than those participants viewing the 2-D displays in this trend analysis task (TAQ4c). A paired comparison test (Table 37, Panel D) also shows that the participants viewing the 3-D perspective display were more (p = 0.047) effective or accurate than those participants viewing the 2-D displays in this trend analysis task (TAQ4c). Thus, H1c was not supported. Panel C indicates that manipulation of the presentation formats (Treatment) is significantly (p = 0.022) associated with the accuracy of the participants in estimating the average of turnover, profitability and leverage for the years 1, 2, 4 and 5, and using the estimated average to calculate a new ROE (TAQ4d). A paired comparison test (Table 37, Panel D) shows that the participants viewing the tabular display were more (p = 0.026) effective or accurate than those participants viewing the 2-D displays in this trend analysis task (TAQ4d). Thus, hypothesis H1c was not supported. In summary, H1c is not supported. While the MANCOVA result showed a significant treatment effect, the significant pairwise comparisons at the ANCOVA level show that TAQ3, TAQ4c and TAQ4d, are opposite what was predicted, indicating that use of 2-D displays did not result in greater accuracy by participants.

PAGE 210

195 Table 37 Test Results of H1c ( Participants viewing a set of 2-D displays will be the most effective in an accounting judgment involving estimation of values), MANOVA Model on Effectiveness (Accuracy) in Trend Analysis Task the profitability and leverage for the years 1, 2, (TAQ4d) Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Scores on the Tasks. Dependent Variable Treatment Actual M ean MANCOVA Adjusted Mean* TAQ3 Tabular Display (n=42) 71.242 72.184 2 D Displays (n=40) 87.110 88.030 3 D Perspective Display (n=41) 63.876 62.015 TAQ4a Tabular Display (n=42) 5.001 5.054 2 D Displays (n=40) 1.503 1.490 3 D Perspective Dis play (n=41) 0.874 0.834 TAQ4b Tabular Display (n=42) 0.796 0.777 2 D Displays (n=40) 0.626 0.563 3 D Perspective Display (n=41) 1.541 1.457 TAQ4c Tabular Display (n=42) 0.032 0.105 2 D Displays (n=40) 1.898 1.933 3 D Perspective Displa y (n=41) 0.322 0.213 TAQ4d Tabular Display (n=42) 0.359 1.756 2 D Displays (n=40) 16.392 16.329 3 D Perspective Display (n=41) 1.055 -2.547 The number of observations of 3-D Perspective Display is 41 instead of 42 as an outlier was dropped *Adjusted Mean is for the effects of the covariates. Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept 0.229 6.789 < 0.001 HE 0.013 0.306 0.908 TSMRT 0.048 1.148 0.33 9 Treatment 0.205 2.627 0.005 HE = highest level of education. TSMRT = time spent on mental rotations test.

PAGE 211

196 Table 37: Test Results of H1c (Continued) Panel C: ANOVA Results Using Scores as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TAQ3 Corrected Model 19410.708 4 4852.677 1.998 0.099 Intercept 73061.681 1 73061.681 30.077 0.000 HE 2331.893 1 2331.892 0.960 0.329 TSMRT 5704.231 1 5704.231 2.348 0.128 Treatment 13 641.724 2 68200.862 2.808 0.032 Error 286643.074 118 2429.179 Total 978643.709 123 Corrected Total 306053.782 122 TAQ4a Corrected Model 414.336 4 103.584 0.563 0.690 Intercept 28.689 1 28.689 0.156 0.694 HE 2.950 1 2.950 0.016 0.899 TS MRT 0.622 1 0.622 0.003 0.954 Treatment 400.509 2 200.255 1.088 0.170 Error 21714.821 118 184.024 Total 22890.690 123 Corrected Total 22129.157 122 TAQ4b Corrected Model 171.343 4 42.836 1.246 0.295 Intercept 29.571 1 29.571 0.860 0.356 HE 15.855 1 15.855 0.461 0.498 TSMRT 15.943 1 15.943 0.464 0.497 Treatment 116.794 2 58.397 1.698 0.094 Error 4057.884 118 34.389 Total 4229.407 123 Corrected Total 4229.227 122 TAQ4c Corrected Model 98.817 4 24.704 2.004 0.098 Inter cept 24.965 1 24.965 2.025 0.157 HE 1.571 1 1.571 0.127 0.722 TSMRT 15.390 1 15.390 1.248 0.266 Treatment 84.540 2 42.470 3.429 0.018 Error 1454.674 118 12.328 Total 1620.097 123 Corrected 1553..492 122

PAGE 212

197 Table 37: Test Results of H1c (Continued) Panel C: ANOVA Results Using Scores as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TAQ4d Corrected Model 9167.503 4 2291.876 2.029 0.095 Intercept 41.362 1 41.362 0.037 0.849 HE 290.027 1 290.027 0.257 0.613 TSMRT 1875.519 1 1875.519 1.660 0.200 Treatment 7175.704 2 3587.852 3.176 0.022 Error 133298.654 118 1129.650 Total 146268.140 123 Corrected Total 142466.157 122 TAQ3 Adjusted R Squared = 0.032. TA Q4a Adjusted R Squared = -0.015 TAQ4b Adjusted R Squared = 0.008. TAQ4c Adjusted R Squared = 0.037 TAQ4d Adjusted R Squared = 0.033 HE = highest level of education. TSMRT = time spent on mental rotations test. *Treatment p-values are one-tail, all others are two-tail. Panel D: Bonferroni Pairwise Comparisons for Test H1c Dependent Variables (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* TAQ3 2 D Displays Tabular Display 3 D Display 15.846 26.015 11 .026 11.098 0.230 0.031 TAQ4c 2 D Displays Tabular Display 3 D Display 1.828 1.720 0.785 0.791 0.032 0.047 TAQ4d 2 D Displays Tabular Display 3 D Display 18.085 13.782 7.519 7.568 0.026 0.106 *p -values are one-tail.

PAGE 213

198 5.4.7. Results of H1d Two dependent variables were used to test the trend analysis task for hypothesis (TSTAQ3), and the time spent in seconds by each participant when answering all four parts of the fourth question (TSTAQ4). In constructing the models to test hypothesis (TSPQ5), score on the mental rotations test (SMRT), and time spent on the mental 4.3) along with the manipulated variable Treatment. The first dependent variable used to test hypothesis H1d was the time spent in seconds by each participant when answering the third question (TSTAQ3) which asked what the ROE in year six would be if each of the variables of ROE in year 5 doubled (see Table 3). Table 38, Panel A indicates that those participants viewing a tabular display (mean seconds 109) used 11% more time (in seconds) than those viewing the 2D displays (mean seconds 98) .Those participants viewing the 3-D perspective display (mean seconds 94) were as efficient as those participants viewing the 2-D displays when answering the third question. The second dependent variable used to test hypothesis H1d was the time spent in seconds by each participant when answering the fourth question in the trend analysis task (TSTAQ4), which asked the participants to estimate the average of turnover, profitability, and leverage for the years 1, 2, 4, and 5, and use them to calculate a new ROE (see Table 3). Table 38, Panel A indicates on average participantsviewing the tabular display (mean seconds 221), participants viewing the 2-D displays (mean seconds 218), and

PAGE 214

199 participants viewing a 3-D perspective display (mean second 210) of different treatment groups spent roughly the same amount of time in seconds when answering the fourth question of the trend analysis task (TSTAQ4). The mean results for the dependent variables from the preceding paragraphs provide little support for H1d, which predicted that participants viewing a set of 2D displays will be the most efficient in an accounting judgment involving estimation of values when compared to participants viewing a single 3-D perspective display or subjects viewing a table. A MANCOVA analysis was conducted. As shown (Table 38, Panel B), the overall F-statistic for the manipulated variable Treatment is not significant (p = 0.751) Trace. Analysis of the univariate results is not necessary. Thus, hypothesis H1d was not supported.

PAGE 215

200 Table 38 Test Results of H1d (Participants viewing a set of 2-D displays will be the most efficient in an accounting judgment involving estimation of values ) MANCOVA Model on Efficiency (Less Time) in Trend Analysis Task and 5, and use th Tests of Between-Subjects Effects on Efficiency Panel A: Mean Time Spent on the Tasks. Dependent Variable Treatment Actual Mean (Seconds) MANCOVA Adjusted Mean* (Seconds) TSTAQ3 Tabular Display (n=42) 109.500 107.300 2 D Displays (n=40) 98.925 103.800 3 D Perspective Display (n=42) 94.809 92.387 TSTAQ4 Tabular Display (n=42) 221.333 207.700 2 D Displays (n=40) 218.325 227.900 3 D Perspective Display (n=42) 210.095 214.600 *Adjusted Mean is for the effects of the covariates. Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept 0.006 0.329 0.720 PQ5 0.113 7.369 0.001 TSPQ5 0.274 21.855 < 0.001 SMRT 0.078 4.92 1 0.009 TSMRT 0.130 8.696 <0.001 Treatment 0.016 0.479 0.751 PQ1 =score on practice question 5. TSPQ5 = time spent in seconds by each participant when answering practice question 5. SMRT = Score on Mental Rotations Test. TSMRT = Time Spent on Mental Rotations Test.

PAGE 216

201 5.5 Results of the Pattern Recognition Task The purpose of this section is to test whether there is support for each of the hypothesis H2a, H2b, H2c and H2d. For the pattern recognition task, H2a-d posit that a 3-D display will result in greater effectiveness and efficiency than tabular or 2D displays. This section is organized in the following way: descriptive statistics of the -Moment Correlation of the variables, results of the regression analyses of possible covariates, and the MANCOVA results of each of the hypothese s. 5.5.1 Descriptive Statistics Table 39 shows the descriptive statistics for the pattern recognition task. Based on the results of one-way ANOVA, the following paragraphs briefly describe those covariates and demographic variables that had significant differences between their means among treatment groups. Descriptive information concerning the dependent variables will be reported separately in the results section of each of the hypotheses. The first covariate evaluated was performance on the practice questions. For the score on practice question two (PQ2) (What are the differences between the apartments 5 and 6?), participants viewing the tabular display on average had the highest score. Those participants viewing the tabular display (mean score 2.860) were significantly more accurate (p = 0.014) than those participants viewing the 3-D perspective display (mean score 2.441). There was no significant difference in accuracy between those participants viewing the tabular display (mean score 2.860) and those participants viewing the 2D displays (mean score 2.682). There was also no significant differences in accuracy

PAGE 217

202 between those participants viewing the 2-D displays (mean score 2.682) and those participants viewing the 3-D perspective display (mean score 2.441). For practice question three (PQ3) (What are the values of the factors of apartment 4?), participants viewing the tabular display on average had the highest score. Those participants viewing the tabular display (mean score 3.488) were significantly more accurate (p< 0.001) than those participants viewing the 2-D displays (mean score 1.926). Those participants viewing the 3-D perspective display (mean score 3.139) were significantly more accurate (p< 0.001) than those participants viewing the 2-D displays (mean score 1.926). There was no significant difference in accuracy between those participants viewing the tabular display (mean score 3.488) and those participants viewing the 2-D displays (mean score 3.139). For practice question four (PQ4) (What are the differences between the apartments 2 and 4?), participants viewing the tabular display on average had the highest score. Those participants viewing the tabular displays (mean score 3.488) were significantly more accurate (p = 0.039) than those participants viewing the 3D perspective display (mean score 2.976).There was no significant difference in accuracy between those participants viewing the tabular display (mean score 3.488) and those participants viewing the 2-D displays (mean score 3.414). There was no significant difference in accuracy between those participants viewing the 2-D display (mean score 3.414) and those participants viewing the 3-D perspective display (mean score 2.976). The second covariate evaluated was the time spent on the practice questions. For the time spent by participants on practice question one (TSPQ1), participants viewing the tabular display were the most efficient. Participant viewing the tabular display (mean

PAGE 218

203 seconds 68) were significantly more efficient (p <0.001) or used less time (in seconds) than those participants viewing the 2-D displays (mean seconds 119) and those participants viewing the 3-D perspective display (p = 102). There was no significant difference in terms of efficiency between participants viewing 2-D displays (mean seconds 119) and those participants viewing the 3-D perspective display (mean seconds 102). For the time spent by participants on practice question two (TSPQ2), participants viewing the tabular display were the most efficient. Participants viewing the tabular display (mean seconds 95) were significantly more efficient (p <0.001) or used less time (in seconds) than those participants viewing the 3-D displays (mean seconds 137). Those participants viewing the 2-D displays (mean seconds 105) were significantly (p = 0.001) more efficient than those participants viewing the 3-D perspective display (mean seconds 137). There was no significant difference in terms of efficiency between participants viewing tabular display (mean seconds 95) and those participants viewing the 2D displays (mean seconds 105). For the time spent by participants on practice question three (TSPQ3), participants viewing the tabular display were the most efficient. Participant viewing the tabular display (mean seconds 58) were significantly more efficient (p <0.001) or used less time (in seconds) than those participants viewing the 2-D displays (mean seconds 101). There was no significant difference in terms of efficiency between participants viewing tabular display (mean seconds 58) and those participants viewing the 3-D perspective display (mean seconds 85). There was no significant difference in terms of efficiency between

PAGE 219

204 participants viewing 2-D displays (mean seconds 101) and those participants viewing the 3-D perspective display (mean seconds 85). As expected (section 3.6.4), participants viewing the 3-D perspective display had the highest perceived mental workload (MW) (mean 4.447) when compared to participants viewing the 2-D displays (mean 4.164) or participants viewing the tabular display (mean 3.720). Comparison of mean mental workload (MW) using one-way ANOVA shows that there is a significant difference on mental workload between participants viewing the tabular display and participants viewing the 3-D perspective display (p = 0.053). The scores on the Mental Rotations Test (SMRT) between tabular, 2-D and 3D treatment groups were 24, 20, and 21 points, respectively. T-tests indicate no significant difference in spatial ability among treatment groups. On the Mental Rotations Test, participants viewing the 2-D displays (mean seconds 632) used less time than participants viewing the tabular display (mean seconds 697) or participants viewing the 3D perspective display (mean seconds 645). Comparison of mean seconds (TSMRT) using one-way ANOVA shows that there is no significant difference on the time spent by participants on the Mental Rotations Test among the treatment groups.

PAGE 220

205 Table 39 Descriptive Statistics for the Pattern Recognition Task Panel A: Mean (Standard Deviation) and Range of Practice Questions. Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=43) Practice Question One: What are the values of the fa ctors of apartment 2? 3.255 (1.236) 0.00 to 4.00 2.658 (1.590) 0.00 to 4.00 2.790 (1.186) 0.00 to 4.00 Time (seconds) Spent on Practice Question one? 68 (27) 33 to 176 119 (44) 47 to 225 102 (35) 47 to 218 Practice Question Two: What are the difference s between the apartments 5 and 6? 2.860 (0.412) 1.00 to 3.00 2.682 (0.819) 0.00 to 3.00 2.441 (0.733) 1.00 to 3.00 Time (seconds) Spent on Practice Question Two 95 (26) 44 to 168 105 (43) 22 to 255 137 (47) 40 to 274 Practice Question Three What are t he values of the factors of apartment 4? 3.488 (0.909) 0.00 to 4.00 1.926 (1.081) 0.00 to 4.00 3.139 (0.804) 1.00 to 4.00 Time (seconds) Spent on Practice Question Three 58 (24) 20 to 119 101 (79) 16 to 416 85 (39) 26 to 290 Practice Question Four: Wh at are the differences between the apartments 2 and 4? 3.488 (0.855) 0.00 to 4.00 3.414 (1.048) 0.00 to 4.00 2.976 (0.912) 1.00 to 4.00 Time (seconds) Spent on Practice Question Four 90 (37) 37 to 193 95 (38) 18 to 211 104 (43) 25 to 251 Practice Ques tion Five turnover ratio is 2, profitability ratio is 5%, and leverage ratio is 1.1? 0.488 (0.505) 0.00 to 1.00 0.463 (0.504) 0.00 to 1.00 0.488 (0.505) 0.00 to 1.00 Time (seconds) Spent on Practice Question Five 127 (51) 6 7 to 315 151 (101) 43 to 617 132 (60) 41 to 258

PAGE 221

206 Table 39 (Continued) Descriptive Statistics for the Pattern Recognition Task Panel A: Mean (Standard Deviation) and Range of Practice Questions. Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=4 3) Practice Question Six ROE is the sum (multiplicative) of turnover, profitability, and l everage? 0.860 (0.350) 0.00 to 1.00 0.853 (0.357) 0.00 to 1.00 0.953 (0.213) 0.00 to 1.00 Time (seconds) Spent on Practice Question Six 28 (11) 8 to 55 28 (13) 8 t o 78 34 (15) 11 to 79 Panel B: Mean (Standard Deviation) and Range of Dependent Variables Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=43) Question One What are the differences between compan ies 1 and 6? 2.906 (0.478) 0.00 to 3.00 2.731 ( 0.671) 0.00 to 3.00 2.697 (0.637) 0.00 to 3.00 Time (seconds) Spent on Question One 83 (32) 34 to 192 89 (35) 23 to 223 115 (61) 22 to 323 Question Two Please separate companies 1 through 6 into 2 groups based on similar characteristics? 4.534 (0.79 7) 3.00 to 6.00 4.560 (0.975) 3.00 to 6.00 5.255 (0.726) 4.00 to 6.00 Time (seconds) Spent on Question Two 76 (37) 30 to 218 96 (52) 18 to 230 82 (52) 31 to 263 Question Three Compared to group two what are the patterns of the financial ratios you are seeing in group one? 1.953 (1.153) 0.00 to 3.00 2.195 (1.054) 0.00 to 3.00 2.116 (0.878) 0.00 to 3.00 Time (seconds) Spent on Question Three 121 (58) 36 to 301 119 (51) 29 to 296 88 (34) 20 to 156 Question Four Comparatively if it is better t o have a higher profitability, higher turnover but lower leverage at the same time, which company you will select? 0.255 (0.441) 0.00 to 1.00 0.219 (0.419) 0.00 to 1.00 0.488 (0.505) 0.00 to 1.00

PAGE 222

207 Table 39 (Continued) Descriptive Statistics for the Pattern Recognition Task Panel B: Mean (Standard Deviation) and Range of Dependent Variables Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=43) Time (seconds) Spent on Question Four 59 (31) 4 to 158 69 (32) 13 to 151 57 (28) 17 to 142 Question Five What are the differences between companies 4 and 6? 2.906 (0.293) 2.00 to 3.00 2.780 (0.612) 1.00 to 3.00 2.720 (0.590) 1.00 to 3.00 Time (seconds) Spent on Question Five 31 (8) 19 to 56 34 (9) 13 to 59 45 (17) 21 to 89 Question Six Compared to group one, what are the patterns of the financial ratios you are seeing in group two? 2.720 (0.629) 0.00 to 3.00 2.780 (0.652) 0.00 to 3.00 2.627 (0.787) 0.00 to 3.00 Time (seconds) Spent on Question Six 44 (26) 16 to 185 41 (13) 14 to 79 43 (17) 19 to 90 Panel C: Mean (Standard Deviation) and Range of Mental Rotations Test and Mental Workload Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=43) Score on Mental Rotations Test 24.302 (9.460) 4.00 to 40.00 20.561 (11.760) 4. 00 to 40.00 21.023 (10.439) 4.00 to 40.00 Time (seconds) Spent on Mental Rotation Test 697 (294) 242 to 2059 632 (308) 64 to 1253 645 (218) 213 to 1141 Mental Workload 3.720 (1.193) 1.50 to 6.00 4.164 (1.490) 1.00 to 7.00 4.447 (1.504) 1.00 to 6.75

PAGE 223

208 Table 39 (Continued) Descriptive Statistics for the Pattern Recognition Task Panel D: Demographic Mean and Range. Tabular Display (n=43) 2 D Displays (n=41) 3 D Display (n=43) AGE 18 22 23 27 29 32 33 37 38 42 43 47 48 50 n=36 n=3 n=3 n=1 n=34 n=5 n=2 n=36 n=3 n=3 n=1 Male/Female n=22/n=21 n=25/n=16 n=11/n=32 Student Status Freshman Sophomore Junior Senior Graduate Student n=9 n=21 n=13 n=2 n=14 n=15 n=8 n=2 n=1 n=15 n=14 n=13 GPA 3.252 (0.355) 2.50 3.80 3.22 (0.450) 2 .00 4.00 3.375 (0.340) 2.80 4.00 SAT 1147 (156) (n=40) 600 1380 1142 (152) (n=38) 800 1600 1163 (137) (n=42) 900 1500 Years of Full Time Working Experience 1.914 (3.055) 0.00 13.00 2.356 (3.874) 0.00 16.00 2.488 (4.817) 0.00 23.00 Part Time Working Hou rs Per Week 13.441 (14.042) 0.00 40.00 14.658 (14.689) 0.00 45.00 17.325 (13.987) 0.00 50.00 Accounting Related Working Experience Full Time Part Time Both Full/Part Time None n=36 n=5 n=2 n=26 n=8 n=4 n=3 n=33 n=5 n=3 n=2 Highest Education High School Master degree n=42 n=1 n=37 n=4 n=39 n=4

PAGE 224

209 -Moment Correlation 5.5.2.1 Correlation between dependent variables. product-moment correlation coefficients and the significances for the variables of the pattern recognition task. Results of two-tailed testing of the significance of the correlation between variables at 0.05 and 0.01 levels are also shown in Table 40. The following paragraphs discuss in detail the most important correlation results, in terms of whether the dependent variables for each hypothesis are significantly correlated with one another. When reporting significant correlation between two dependent variables, the following paragraphs use the symbols of the dependent variables (see Table 11) to describe each of them and report their correlation coefficient through the notation of r To test for the pattern recognition task (H2a, H2b, H2c, and H2d), six questions are developed (see Table 4) and each participant answers these six questions in the same order. The second and fourth questions are used to test hypotheses H2a and H2b (see Table 4). Hypothesis H2a has two dependent measures of accuracy PRQ2 and PRQ4 (see Table 7). PRQ2 is positively related to PRQ4 (r = 0.240) (see Table 40). Since the two dependent measures for hypothesis H2a are positively correlated and MANCOVA is the appropriate statistical method to test hypothesis H2a. Hypothesis H2b has two dependent measures of efficiency TSPRQ2 and TSPRQ4 (see Table 7). TSPRQ2 is positively related to TSPRQ4 (r = 0.441),therefore MANCOVA is the appropriate statistical method to test hypothesis H2b (see Table 40). The first, third, fifth and sixth questions are used to test hypotheses H2c and H2d (see Table 3). Hypothesis H2c has four dependent measures of accuracy PRQ1, PRQ3,

PAGE 225

210 PRQ5 and PRQ6 (see Table 6). PRQ1 is positively related to PRQ5 (r = 0.367) and PRQ6 (r = 0.187) while PRQ3 is positively related to PRQ6 (r = 0.203), and PRQ5 is positively related to PRQ6 (0.281) (see Table 40). It seems that the four dependent measures of hypothesis H2c are inter-correlated; MANCOVA is the appropriate statistical method to test hypothesis H2c. Hypothesis H2d has four dependent measures of accuracy TSPRQ1, TSPRQ3, TSPRQ5 and TSPRQ6 (see Table 6). TSPRQ1 is positively related to TSPRQ3 (r = 0.268) and TSPRQ5 (r = 0.363) while TSPRQ3 is positively related to TSPRQ6 (r = 0.342), and TSPRQ5 is positively related to TS PRQ6 (0.279) (see Table 40). Since the four dependent measures of hypothesis H2d are inter-correlated MANCOVA is the appropriate statistical method to test hypothesis H2d. 5.5.2.2 Multicollinearity. According to Field (2005) when multicollinearity exists, strong correlation between two predictors, in a regression model it is difficult to assess the individual importance of a predictor. Field (2005) further suggests that one way of identifying multicollinearity is to scan a correlation matrix of all the predictor variables and see if any correlate above 0.80. Table 40 shows then non-existence of correlations above 0.80 between any pair of predictor variables. It is concluded that the problem of multicollinearity does not exist and is not a concern here. 5.5.2.3 Correlation between covariates or demographic variables and dependent variables. Table 40 also shows information about the correlations between the treatment variables and the dependent variables, and between the covariates and demographic variables and the dependent variables. As before, significance is shown at alpha levels of 0.05 and 0.01. All significance levels are based on two-tailed tests. A review of the table

PAGE 226

211 indicates that there are some significant correlations between the treatment variables and dependent variables. Additionally, several of the dependent measures are correlated with covariates and demographic variables. Field (2005) comments that caution must be taken when interpreting correlation coefficients because the correlation coefficient says nothing about which variable causes the other to change. However, the correlation matrix does provide preliminary evidence that the treatment may be associated with some of the dependent variables. There is also preliminary evidence that some covariate and demographic variables may be important controls in the MANOVA and ANOVA models. Therefore, the study used regression analysis for each dependent variable to test the significance of the association between the dependent variables and all of the covariates and demographic variables.

PAGE 227

212 Table 40 -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task. TREATMENT PQ1 TSPQ1 PQ2 TSPQ2 PQ3 TREATMENT 1.000 0.141 0.332** 0.250** 0.395** 0.126 PQ1 0.141 1.000 0.023 0.227* 0.166 0.543** TSPQ1 0.332** 0.023 1.000 0.079 0.403** 0.218* PQ2 0.250** 0.227* 0.079 1.000 0.108 0.233** TSPQ2 0.395** 0.166 0.403** 0.108 1.000 0.194* PQ3 0.126 0.543** 0.218* 0.233** 0.194* 1.000 TSPQ3 0.201* 0.086 0.647** 0.059 0.438** 0.063 PQ4 0.220* 0.307** 0.044 0.436** 0.061 0.231** TSPQ4 0.141 0.220* 0.407** 0.205* 0.617** 0.213* PQ5 0.000 0.172 0.010 0.129 0.134 0.099 TSPQ5 0.031 0.129 0.418** 0.022 0.297** 0.058 PQ6 0.122 0.310** 0.063 0.192* 0.009 0.290** TSPQ6 0.158 0.050 0.248** 0.007 0.356** 0.089 PRQ1 0.143 0.216* 0.045 0.504** 0.150 0.176* TSPRQ1 0.280** 0.089 0.360** 0.088 0.497** 0.111 PRQ2 0.332** 0.108 0.168 0.168 0.077 0.043 TSPRQ2 0.052 0.201* 0.269** 0.268** 0.373** 0.091 PRQ3 0.065 0.079 0.054 0.208* 0.170 0.017 TSPRQ3 0.268** 0.188* 0.128 0.180* 0.217* 0.105 PRQ4 0.205* 0.036 0.086 0.003 0.066 0.066 TSPRQ4 0.027 0.155 0.286** 0.048 0.284** 0.056 PRQ5 0.148 0.378** 0.149 0.254** 0.191* 0.302** TSPRQ5 0.417** 0.082 0.227* 0.044 0.392** 0.076 PRQ6 0.056 0.148 0.031 0.256 ** 0.080 0.121 TSPRQ6 0.027 0.086 0.133 0.006 0.101 0.068 AGE 0.051 0.004 0.134 0.007 0.275** 0.009 GEN 0.211* 0.018 0.111 0.083 0.224* 0.017 SS 0.090 0.031 0.007 0.121 0.049 0.063 GPA 0.132 0.082 0.118 0.020 0.066 0.119 SAT 0.096 0.040 0 .109 0.117 0.152 0.015 FTWE 0.057 0.084 0.108 0.019 0.081 0.050 PTH 0.113 0.026 0.057 0.084 0.101 0.073 ARWE 0.064 0.081 0.237** 0.065 0.070 0.171 HE 0.112 0.019 0.173 0.087 0.187* 0.021 SMRT 0.127 0.210 0.029 0.143 0.121 0.237** TSMRT 0. 078 0.257** 0.090 0.216* 0.212* 0.206* MW 0.211* 0.262** 0.161 0.205* 0.123 0.081 Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition of the variables. *Correlation is significant at the 0.05 level (2 tailed).** Correlation is significant at the 0.01 level (2 tailed).

PAGE 228

213 Table 40 (Continued) -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task TSPQ3 PQ4 TSPQ4 PQ5 TSPQ5 PQ6 TREATMENT 0.201* 0220* 0.141 0.000 0.031 0. 122 PQ1 0.086 0.307** 0.220* 0.172 0.129 0.310** TSPQ1 0.647** 0.044 0.407** 0.010 0.418** 0.063 PQ2 0.059 0.436** 0.205* 0.129 0.022 0.192* TSPQ2 0.438** 0.061 0.617** 0.134 0.297** 0.009 PQ3 0.063 0.231** 0.213* 0.099 0.058 0.290** TSPQ3 1.000 0 .131 0.486** 0.036 0.376** 0.195* PQ4 0.131 1.000 0.168 0.013 0.202* 0.024 TSPQ4 0.486** 0.168 1.000 0.061 0.289** 0.029 PQ5 0.036 0.013 0.061 1.000 0.140 0.087 TSPQ5 0.376** 0.202* 0.289** 0.140 1.000 0.024 PQ6 0.195* 0.024 0.029 0.087 0.024 1. 000 TSPQ6 0.090 0.156 0.395** 0.080 0.294** 0.084 PRQ1 0.108 0.317** 0.191* 0.014 0.119 0.248** TSPRQ1 0.327** 0.068 0.570** 0.062 0.295** 0.139 PRQ2 0.026 0.213* 0.037 0.106 0.008 0.001 TSPRQ2 0.304** 0.222* 0.409** 0.048 0.478** 0.147 PRQ3 0.09 8 0.086 0.123 0.026 0.044 0.019 TSPRQ3 0.241** 0.280** 0.348** 0.153 0.231** 0.098 PRQ4 0.014 0.034 0.030 0.124 0.100 0.026 TSPRQ4 0.134 0.122 0.278** 0.098 0.372** 0.191* PRQ5 0.178* 0.291** 0.302** 0.061 0.132 0.157 TSPRQ5 0.161 0.110 0.437** 0.066 0.119 0.090 PRQ6 0.065 0.153 0.165 0.041 0.153 0.070 TSPRQ6 0.140 0.007 0.178* 0.029 0.116 0.123 AGE 0.217* 0.009 0.260** 0.005 0.065 0.064 GEN 0.174 0.114 0.153 0.036 0.126 0.020 SS 0.072 0.109 0.003 0.072 0.032 0.010 GPA 0.145 0.06 0 0.117 0.050 0.038 0.270** SAT 0.091 0.037 0.154 0.063 0.030 0.002 FTWE 0.203* 0.055 0.082 0.000 0.125 0.152 PTH 0.053 0.039 0.093 0.103 0.024 0.030 ARWE 0.183* 0.033 0.003 0.083 0.128 0.124 HE 0.160 0.116 0.090 0.042 0.235** 0.000 SMRT 0.088 0.122 0.219* 0.213* 0.165 0.059 TSMRT 0.144 0.270** 0.262** 0.069 0.367** 0.131 MW 0.137 0.125 0.071 0.116 0.121 0.075 Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition. *Correlation is significant at the 0.05 level (2 tailed). ** Correlation is significant at the 0.01 level (2 tailed).

PAGE 229

214 Table 40 (Continued) -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task TSPQ6 PRQ1 TSPRQ1 PRQ2 TSPRQ2 PRQ3 TREATMENT 0.158 0.143 0.280** 0.332** 0.052 0.065 PQ1 0.050 0.216* 0.089 0.108 0.201* 0.079 TSPQ1 0.248** 0.045 0.360** 0.168 0.269** 0.054 PQ2 0.007 0.504** 0.088 0.168 0.268** 0.208* TSPQ2 0.356** 0.150 0.497** 0.077 0.373** 0.170 PQ3 0.089 0.176* 0.111 0.043 0.091 0.017 TSPQ3 0.090 0.108 0.327** 0.026 0.304** 0.098 PQ4 0.156 0.317** 0.068 0.213* 0.222* 0.086 TSPQ4 0.395** 0.191* 0.570** 0.037 0.409** 0.123 PQ5 0.080 0.014 0.062 0.106 0.048 0.026 TSPQ5 0.294** 0.119 0.295** 0.008 0.478** 0.044 PQ6 0.084 0.248** 0.139 0.001 0.147 0.019 TSPQ6 1.000 0.015 0.290** 0.002 0.256** 0.123 PRQ1 0.015 1.000 0.064 0.014 0.274** 0.146 TSPRQ1 0.290** 0.064 1.000 0.072 0.416** 0.081 PRQ2 0.002 0.014 0.072 1.000 0.045 0.140 TSPRQ2 0.256** 0.274** 0.416 ** 0.045 1.000 0.216* PRQ3 0.123 0.146 0.081 0.140 0.216* 1.000 TSPRQ3 0.256** 0.102 0.268** 0.290** 0.444** 0.149 PRQ4 0.065 0.169 0.054 0.240** 0.088 0.155 TSPRQ4 0.340** 0.134 0.334** 0.022 0.441** 0.151 PRQ5 0.220* 0.367** 0.255** 0.159 0.212* 0.032 TSPRQ5 0.292** 0.146 0.363** 0.085 0.259** 0.059 PRQ6 0.094 0.187* 0.097 0.037 0.169 0.203* TSPRQ6 0.192* 0.163 0.173 0.031 0.297** 0.081 AGE 0.261** 0.055 0.352** 0.069 0.383** 0.028 GEN 0.034 0.111 0.190* 0.047 0.171 0.030 SS 0.248** 0.0 51 0.017 0.024 0.008 0.050 GPA 0.194* 0.054 0.020 0.045 0.076 0.120 SAT 0.119 0.213* 0.284** 0.123 0.156 0.159 FTWE 0.125 0.014 0.104 0.018 0.245** 0.040 PTH 0.034 0.132 0.056 0.057 0.245** 0.100 ARWE 0.005 0.045 0.027 0.009 0.138 0.073 HE 0.113 0.052 0.265** 0.134 0.344** 0.066 SMRT 0.168 0.024 0.135 0.100 0.098 0.163 TSMRT 0.207* 0.185* 0.185* 0.057 0.373** 0.179* MW 0.031 0.020 0.096 0.057 0.021 0.123 Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition. *Correlation is significant at the 0.05 level (2 tailed) ** Correlation is significant at the 0.01 level (2 tailed).

PAGE 230

215 Table 40 (Continued) -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task. TSPRQ3 PRQ4 TSPRQ4 PRQ5 TSPRQ5 PRQ6 TREATMENT 0.268** 0.205* 0.027 0.148 0.417** 0.056 PQ1 0.188* 0.036 0.155 0.378** 0.082 0.148 TSPQ1 0.128 0.086 0.286** 0.149 0.227* 0.031 PQ2 0.180* 0.003 0.048 0.254** 0.044 0.256** TSPQ2 0.217* 0.066 0.28 4** 0.191* 0.392** 0.080 PQ3 0.105 0.066 0.056 0.302** 0.076 0.121 TSPQ3 0.241** 0.014 0.134 0.178* 0.161 0.065 PQ4 0.280** 0.034 0.122 0.291** 0.110 0.153 TSPQ4 0.348** 0.030 0.278** 0.302** 0.437** 0.165 PQ5 0.153 0.124 0.098 0.061 0.066 0.041 TSPQ5 0.231** 0.100 0.372** 0.132 0.119 0.153 PQ6 0.098 0.026 0.191* 0.157 0.090 0.070 TSPQ6 0.256** 0.065 0.340** 0.220* 0.292** 0.094 PRQ1 0.102 0.169 0.134 0.367** 0.146 0.187* TSPRQ1 0.268** 0.054 0.334** 0.255** 0.363** 0.097 PRQ2 0.290** 0. 240** 0.022 0.159 0.085 0.037 TSPRQ2 0.444** 0.088 0.441** 0.212* 0.259** 0.169 PRQ3 0.149 0.155 0.151 0.032 0.059 0.203* TSPRQ3 1.000 0.012 0.355** 0.284** 0.097 0.198* PRQ4 0.012 1.000 0.107 0.165 0.164 0.072 TSPRQ4 0.355** 0.107 1.000 0.307** 0.2 49** 0.195* PRQ5 0.284** 0.165 0.307** 1.000 0.173 0.281** TSPRQ5 0.097 0.164 0.249** 0.173 1.000 0.179* PRQ6 0.198* 0.072 0.195* 0.281** 0.179* 1.000 TSPRQ6 0.342** 0.125 0.305** 0.171 0.279** 0.093 AGE 0.099 0.088 0.125 0.086 0.341** 0.103 GEN 0.0 65 0.160 0.034 0.170 0.248** 0.067 SS 0.031 0.098 0.141 0.093 0.031 0.104 GPA 0.202* 0.021 0.030 0.029 0.057 0.009 SAT 0.197* 0.019 0.094 0.068 0.153 0.081 FTWE 0.053 0.092 0.110 0.060 0.102 0.052 PTH 0.055 0.028 0.010 0.036 0.004 0.082 ARWE 0.022 0.078 0.059 0.049 0.021 0.000 HE 0.058 0.006 0.105 0.014 0.313** 0.017 SMRT 0.058 0.111 0.163 0.229** 0.076 0.150 TSMRT 0.209* 0.196* .462** 0.294** 0.131 0.145 MW 0.028 0.083 0.042 0.244** 0.059 0.181* Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition. *Correlation is significant at the 0.05 level (2 tailed).** Correlation is significant at the 0.01 level (2 tailed).

PAGE 231

216 Table 40 (Continued) -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task. TSPRQ6 AGE GEN SS GPA SAT TREATMENT 0.027 0.051 0.211* 0.090 0.132 0.096 PQ1 0.086 0.004 0.018 0.031 0.082 0.040 TSPQ1 0.133 0.134 0.111 0.007 0.118 0.109 PQ2 0.006 0.007 0.083 0.121 0 .020 0.117 TSPQ2 0.101 0.275** 0.224* 0.049 0.066 0.152 PQ3 0.068 0.009 0.017 0.063 0.119 0.015 TSPQ3 0.140 0.217* 0.174 0.072 0.145 0.091 PQ4 0.007 0.009 0.114 0.109 0.060 0.037 TSPQ4 0.178* 0.260** 0.153 0.003 0.117 0.154 PQ5 0.029 0. 005 0.036 0.072 0.050 0.063 TSPQ5 0.116 0.065 0.126 0.032 0.038 0.030 PQ6 0.123 0.064 0.020 0.010 0.270** 0.002 TSPQ6 0.192* 0.261** 0.034 0.248** 0.194* 0.119 PRQ1 0.163 0.055 0.111 0.051 0.054 0.213* TSPRQ1 0.173 0.352** 0.190* 0.017 0.0 20 0.284** PRQ2 0.031 0.069 0.047 0.024 0.045 0.123 TSPRQ2 0.297** 0.383** 0.171 0.008 0.076 0.156 PRQ3 0.081 0.028 0.030 0.050 0.120 0.159 TSPRQ3 0.342** 0.099 0.065 0.031 0.202* 0.197* PRQ4 0.125 0.088 0.160 0.098 0.021 0.019 TSPRQ4 0.305 ** 0.125 0.034 0.141 0.030 0.094 PRQ5 0.171 0.086 0.170 0.093 0.029 0.068 TSPRQ5 0.279** 0.341** 0.248** 0.031 0.057 0.153 PRQ6 0.093 0.103 0.067 0.104 0.009 0.081 TSPRQ6 1.000 0.078 0.039 0.050 0.069 0.067 AGE 0.078 1.000 0.103 0.157 0 .008 0.326** GEN 0.039 0.103 1.000 0.032 0.180* 0.015 SS 0.050 0.157 0.032 1.000 0.235** 0.057 GPA 0.069 0.008 0.180* 0.235** 1.000 0.010 SAT 0.067 0.326** 0.015 0.057 0.010 1.000 FTWE 0.011 0.695** 0.113 0.255** 0.138 0.105 PTH 0.08 3 0.045 0.053 0.042 0.193* 0.192* ARWE 0.085 0.236** 0.019 0.188* 0.004 0.074 HE 0.024 0.544** 0.130 0.232** 0.099 0.217* SMRT 0.053 0.032 0.203* 0.019 0.037 0.009 TSMRT 0.242** 0.041 0.075 0.063 0.010 0.054 MW 0.126 0.053 0.144 0.079 0.041 0.087 Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition of variables. *Correlation is significant at the 0.05 level (2 tailed).** Correlation is significant at the 0.01 level (2 tailed).

PAGE 232

217 Ta ble 40 (Continued) -Moment Correlation Coefficients of the Variables of the Pattern Recognition Task FTWE PTH ARWE HE SMRT TSMRT MW TREATMENT 0.057 0.113 0.064 0.112 0.127 0.078 0.211* PQ1 0.084 0.026 0.081 0.019 0.210* 0.257** 0.262** TSPQ1 0.108 0.057 0.237** 0.173 0.029 0.090 0.161 PQ2 0.019 0.084 0.065 0.087 0.143 0.216* 0.205* TSPQ2 0.081 0.101 0.070 0.187* 0.121 0.212* 0.123 PQ3 0.050 0.073 0.171 0.021 0.237** 0.206* 0.081 TSPQ3 0.203* 0.053 0.183* 0.160 0.088 0.144 0.137 PQ4 0.05 5 0.039 0.033 0.116 0.122 0.270** 0.125 TSPQ4 0.082 0.093 0.003 0.090 0.219* 0.262** 0.071 PQ5 0.000 0.103 0.083 0.042 0.213* 0.069 0.116 TSPQ5 0.125 0.024 0.128 0.235** 0.165 0.367** 0.121 PQ6 0.152 0.030 0.124 0.000 0.059 0.131 0.075 TSPQ6 0 .125 0.034 0.005 0.113 0.168 0.207* 0.031 PRQ1 0.014 0.132 0.045 0.052 0.024 0.185* 0.020 TSPRQ1 0.104 0.056 0.027 0.265** 0.135 0.185* 0.096 PRQ2 0.018 0.057 0.009 0.134 0.100 0.057 0.057 TSPRQ2 0.245** .245** 0.138 0.344** 0.098 0.373** 0.021 PRQ3 0.040 0.100 0.073 0.066 0.163 0.179* 0.123 TSPRQ3 0.053 0.055 0.022 0.058 0.120 0.209* 0.028 PRQ4 0.092 0.028 0.078 0.006 0.111 0.196* 0.083 TSPRQ4 0.110 0.010 0.059 0.105 0.163 0.462** 0.042 PRQ5 0.060 0.036 0.049 0.014 0.229** 0.2 94** 0.244** TSPRQ5 0.102 0.004 0.021 0.313** 0.076 0.131 0.059 PRQ6 0.052 0.082 0.000 0.017 0.150 0.145 0.181* TSPRQ6 0.011 0.083 0.085 0.024 0.053 0.242** 0.126 AGE 0.695** 0.045 0.236** 0.544** 0.032 0.041 0.053 GEN 0.113 0.053 0.019 0.13 0 0.203* 0.075 0.144 SS 0.255** 0.042 0.188* 0.232** 0.019 0.063 0.079 GPA 0.138 0.193* 0.004 0.099 0.037 0.010 0.041 SAT 0.105 0.192* 0.074 0.217* 0.009 0.054 0.087 FTWE 1.000 0.042 0.418** 0.416** 0.136 0.067 0.123 PTH 0.042 1.000 0.141 0.090 0.033 0.172 0.095 ARWE 0.418** 0.141 1.000 0.392** 0.098 0.020 0.019 HE 0.416** 0.090 0.392** 1.000 0.023 0.065 0.000 SMRT 0.136 0.033 0.098 0.023 1.000 0.531** 0.222 TSMRT 0.067 0.172 0.020 0.065 0.531** 1.000 0.058 MW 0.123 0.09 5 0.019 0.000 0.222* 0.058 1.000 Figures shown in the table are Pearson Correlation Coefficients See Table 11 for definition. *Correlation is significant at the 0.05 level (2 tailed).** Correlation is significant at the 0.01 level (2 tailed).

PAGE 233

218 5.5.3 Regression Analysis of Possible Covariates This study developed separate regression models for each dependent measure used to test hypotheses H2a-H2d, using twenty-four possible predictors. Only those predictors that were common covariates across the models used to test a single hypothesis were retained in the models. Dependent Measures (PRQ1-6 and TSPRQ1-6) = b 0 + b1 Treatment+b 2 PQ1 + b3 TSPQ1 + b 4 PQ2 + b 5 TSPQ2 + b 6 PQ3 + b 7 TSPQ3 + b8 PQ4 + b 9 TSPQ4 + b10 PQ5 + b 11TSPQ5 + b 12 PQ6 + b 13 TSPQ6 + b 14AGE + b15GEN + b16SS + b17GPA + b 18 SAT + b 19FTWE + b 20 PTH + b 21ARWE + b 22HE + b23 SMT + b24TSMRT + b 25 MW + e. second question (PRQ2), and the score on the fourth question (PRQ4). Hypothesis H2b uses the time spent on each of the tasks used to test hypothesis H2a to derive two (TSPRQ2), and PRQ4 (TSPRQ4). the first question (PRQ1), the score on the third question (PRQ3), the score on the fifth question (PRQ5), and the score on the sixth question (PRQ6). Hypothesis H2d uses the time spent on each of the tasks used to test hypothesis H2c to derive four dependent measures of PRQ5 (TSPRQ5), and PRQ6 (TSPRQ6).

PAGE 234

219 The results of the analysis indicated that for hypothesis H2a the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H2a: 1) time spent on practice question five (TSPQ5), 2) practice question six (PQ6), and 3) the time spent in seconds by each participant when answering the Mental Rotations Test (TSMRT). For hypothesis H2b the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H2b: 1) time spent on practice question one (TSPQ1), 2) time spent on practice question five (TSPQ5), 3) time spent on practice question six (TSPQ6), 4) student status (SS), 5) part time working hours (PTH), 6) the score on the Mental Rotations Test (SMRT), 7) and the time spent in seconds by each participant when answering the Mental Rotations Test (TSMRT). Analysis indicated that for hypothesis H2c the following covariates were significant at p-values < 0.05 in one or more of the models, and were therefore retained for testing of H2c: 1) the score on practice question one (PQ1), 2) the score on practice question two (PQ2), 3) the score on practice question six (PQ6), 4) time spent on practice question five (TSPQ5), 5) gender of the participants (GEN), 6) SAT score (SAT), 7) part time working hours (PTH), 8) accounting related working experience (ARWE), and 9) mental workload (MW). For hypothesis H2d the following covariates were significant, and therefore retained for testing of H2d: 1) the score on practice question four (PQ4), 2) the score on practice question six (PQ6), 3) time spent on practice question four (TSPQ4), 4) age of the participants (AGE), 5) SAT score (SAT), 6) the score on the Mental Rotations Test (SMRT), and 7) the time spent in seconds by each participant when answering the Mental Rotations Test (TSMRT).

PAGE 235

220 The next section will discuss the results of testing hypotheses H2a, H2b, H2c, and H2d. 5.5.4 Results of H2a Two dependent variables are used to test the pattern recognition task hypothesis question six (PQ6), time spent on practice question five (TSPQ5), and the time spent in included in the model (see section 5.5.3) along with the manipulated variable Treatment. The first dependent variable used to test hypothesis H2a was PRQ2, which asked (see Table 4) participants to separate companies one through six into two groups based on similar financial characteristics. Table 41, Panel A indicates that on average those participants viewing the 3-D perspective displays were the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (mean score 4.534) were 14% less accurate than those viewing the 3-D perspective display (mean score 5.255) and those using the 2-D displays (mean score 4.560) were also 14% less accurate than those viewing the 3-D perspective display (mean score 5.255). The second dependent variable used to test hypothesis H2a was PRQ4, which asked participants to select one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time (see Table 4). Table 41, Panel A indicates that on average those participants viewing the 3-D perspective display

PAGE 236

221 were the most accurate (had the highest score) on this pattern recognition task. Those participants viewing a tabular display (mean score 0.255) were 48% less accurate than those viewing the 3-D perspective display (mean score 0.488) and those using the 2D displays (mean score 0.219) were 56% less accurate than those viewing the 3D perspective display (mean score 0.488). The mean results from this and the preceding paragraph are as hypothesized in H2a, which predicted that participants viewing a single 3-D perspective display will be the most effective in recognizing patterns of accounting data when compared to participants using a set of 2-D displays or participants using a table. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H1a, a MANCOVA analysis was conducted. As shown (Table 41, Panel B), the overall F-statistic for the manipulated variable Tr eatment is significant (p < 0.001) These significant results allow for analysis of the univariate results, which are provided on Panel C of Table 41. Panel C results indicate that manipulation of the presentation formats (Treatment) is significantly (p < 0.001) associated with the accuracy of the participants in separating companies one through six into two groups based on similar financial characteristics. A paired comparison test (Table 41, Panel D) was conducted to determine if the participants viewing the 3-D perspective display were more effective or accurate than those participants reviewing the tabular or 2-D displays. Results revealed that the participants viewing the 3-D perspective display not only were significantly more effective or accurate than those participants viewing the tabular display (p < 0.001) but also those participants viewing the 2-D display (p < 0.001) on this pattern recognition task (PRQ2).

PAGE 237

222 Thus, the paired comparison tests provide support for H2a, which predicted that participants viewing a single 3-D perspective display will be the most effective in recognizing patterns of accounting data when compared to participants using a set of 2D displays or participants using a table. Panel C indicates that the covariates, TSPQ5 (p = 0.033), and TSMRT (p = 0.002) are significantly associated with PRQ4. Re sults also indicate that manipulation of the presentation formats (Treatment) has a significant (p = 0.004) effect on the accuracy of the participants in selecting one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time (PRQ4). A paired comparison test was conducted to determine if the participants viewing the 3D perspective display were more effective or accurate than those participants reviewing the tabular or 2-D displays. Results revealed that the participants viewing the 3-D perspective display not only were significantly more effective or accurate than those participants viewing the tabular display (p = 0.007) but also those participants viewing the 2D displays (p = 0.016) on this pattern recognition task (PRQ4). Thus, the paired comparison tests provides support for H2a, which predicted that participants viewing a single 3D perspective display will be the most effective in recognizing patterns of accounting data when compared to participants using a set of 2-D displays or participants using a table. Participants answering PRQ4 could either score one point for a correct response or zero for an incorrect response. A logit analysis was performed treating PRQ4 as a dichotomous dependent variable. Results of the logit analysis of the dichotomous PRQ4 variable, with the same covariates, shows that participants viewing the 3-D display had

PAGE 238

223 the highest likelihood of responding correctly compared to participants viewing a tabular display and participants viewing the 2-D displays.

PAGE 239

224 Table 41 Test Results of H2a (Participants viewing a single 3-D perspective display will be the most effective in recognizing patterns of accounting data ) MANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task Pl ease separate companies 1 through 6 into 2 groups based on similar c example higher profitability. Comparatively, if it is better to have a higher profitability, higher turnover but lower leverage at the same time, which company you will select? Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Scores on the Tasks. Dependent Variable Treatment Actual Mean MANCOVA Adjusted Mean* PRQ 2 Tabular Display (n=43) 4.534 4.518 2 D Displays (n=41) 4.560 4.563 3 D Perspective Display (n=43) 5.255 5.271 PRQ4 Tabular Display (n=43) 0.255 0.220 2 D Displays (n=41) 0.219 0.246 3 D Perspective Display (n=43) 0.488 0.499 *Adjusted Mean is for the effect of the covariate. Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept 0.699 139.220 <0.001 TSPQ5 0.037 2.319 0.103 PQ6 0.014 0.844 0.432 TSMRT Trace 0.078 5.095 0.008 Treatment 0.185 6.163 <0.001 PQ6=score on practice question 6. TSPQ5 = time spent on practice question five. PTH = part time working hours. TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test

PAGE 240

225 Table 41: Test Results of H2a (Continued) Panel C: ANCOVA Results Using Scores as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* PRQ2 Corrected Model 15.233 5 3.0477 4 .285 0.001 Intercept 199.370 1 199.370 280.423 <0.001 TSPQ5 0.007 1 0.007 0.010 0.921 PQ6 0.429 1 0.429 0.604 0.439 TSMRT 0.601 1 0.601 0.845 0.360 Treatment 14.877 2 7.439 10.463 <0.001 Error 86.027 121 0.711 Total 3012.000 127 Correc ted Total 101.260 126 PRQ4 Corrected Model 4.090 5 0.818 4.180 0.002 Intercept 0.759 1 0.759 3.879 0.051 TSPQ5 0.906 1 0.906 4.630 0.033 PQ6 0.259 1 0.259 1.322 0.253 TSMRT 1.968 1 1.968 10.058 0.002 Treatment 1.988 2 0.994 5.080 0.004 Erro r 23.764 121 Total 41.000 127 Corrected Total 27.764 126 PRQ2 Adjusted R Squared = 0.244. PRQ4 Adjusted R Squared = 0.301. PQ6= score on practice question 6. TSPQ5 = time spent on practice question five. PTH = part time working hours. TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test *Treatment p-values are one-tail, all others are two-tail. Panel D: Bonferroni Pairwise Comparisons for Test H2a Dependent Variables (I) Treatment (J) Treatment M ean Difference (I J) Std Error p value* PRQ2 3 D Displays Tabular Display 2 D Display 0.754 0.708 0.185 0.187 <0.001 <0.001 PRQ4 3 D Displays Tabular Display 2 D Display 0.279 0.254 0.097 0.098 0 .007 0.016 *p -values are one-tail.

PAGE 241

226 5.5.5 Results of H2b Hypothesi seconds by each participant when answering the second question (TSPRQ2), and the time spent in seconds by each participant when answering the fourth question (TSPRQ4). In constructing the models to test hypothesis H2b, the following seven covariates were included in the model (see section 5.5.3) along with the manipulated variable Treatment: time spent on practice question one (TSPQ1), time spent on practice question five (TSPQ5), time spent on practice question six (TSPQ6), student status (SS), part time working hours (PTH), score on mental rotations test (SMRT), and time spent on mental rotations test (TSMRT). The first dependent variable used to test H2b was TSPRQ2 the time spent in seconds by each participant when separating companies one through six into two groups based on similar financial characteristics (see Table 4). Table 42, Panel A indicates that on average those participants viewing the tabular display were the most efficient (used the least time in seconds) on this pattern recognition task. Those participants viewing a 3D perspective display (mean seconds 82) used 7% more time (in seconds), than those viewing the tabular display (mean seconds 76). But, those participants viewing the 3D perspective display (mean seconds 82) used 17% less time (in seconds) than those viewing the 2-D displays (mean seconds 96). The second dependent variable TSPRQ4 used to test H2b was the time spent in seconds by each participant when selecting one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time (see Table 4). Table 42, Panel A indicates that on average those participants viewing the 3-D perspective

PAGE 242

227 display were the most efficient (used the least time in seconds) on this pattern recognition task. Those participants viewing a tabular display (mean seconds 59) used 4% more time than those viewing the 3-D perspective display (mean seconds 57). Those participants viewing the 2-D displays (mean seconds 69) used 21% more time than those viewing the 3-D perspective display (mean seconds 69). The mean results for the two dependent variables from this and the preceding paragraph provide mixed evidence of support for H2b, which predicted that participants viewing a single 3-D perspective display will be the most efficient in recognizing patterns of accounting data when compared to participants using a set of 2-D displays or participants using a table. Prior to presenting ANCOVA results for the two dependent variables used to test hypothesis H2b, a MANCOVA analysis was conducted. As shown (Table 42, Panel B), the overall F-statistic for the manipulated variable Treatment is weakly significant (p = weakly significant (p = 0.055) using provided on Panel C of Table 42. Panel C indicates that the covariates TSPQ5 (p < 0.001), PTH (p = 0.011), and TSMRT (p = 0.013) are significantly associated with the time spent in seconds by each participant (TSPRQ2) when answering the second question on this pattern recognition task. Contrary to expectations, the results suggest that manipulation of the presentation formats (Treatment) does not have a significant (p = 0.120) effect on the time spent in seconds by each participant (TSTAQ2) when separating companies one through six into two groups based on similar financial characteristics. Since there is not a significant main effect a paired comparison test was not conducted.

PAGE 243

228 Panel C indicates that the covariates TSPQ6 (p < 0.001), SS (p = 0.009), and TSMRT (p < 0.001) are significantly associated with the time spent in seconds by each participant (TSPRQ4) when answering the fourth question of this pattern recognition task. Results indicate that manipulation of the presentation formats (Treatment) is significantly (p = 0.016) associated with the time spent in seconds by each participant when selecting one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. A paired comparison test (Table 42, Panel D) shows that there was no significant difference (p = 0.121) in the efficiency (time used in seconds) between participants viewing the tabular display and participants viewing the 3D perspective display. But those participants viewing the 3-D perspective display were significantly (p = 0.019) more efficient or used less time than those participants viewing the 2-D displays when answering the fourth question on this pattern recognition task. Thus, hypothesis H2b predicting that participants viewing a single 3-D perspective display will be the most efficient in recognizing patterns of accounting data, when compared to participants using a set of 2-D displays or participants using a table, was partially supported.

PAGE 244

229 Table 42 Test Results of H2b (Participants viewing a single 3-D perspective display will be the most efficient in recognizing patterns of accounting data ) MANCOVA Model on Efficiency (Less Time) in Trend Analysis Task Please separate companies 1 through 6 into 2 groups based on similar example higher profitability. Comparatively, if it is better to have a higher profitability, higher turnover but lower leverage at the same time, which company Tests of Between-Subjects Effects on Efficiency Panel A: Mean Time Spent on the Tasks. Dependent Variable Treatment Actual Mean (Seconds) MANCOVA Adjusted Mean* (Seconds) TSPRQ2 Tabular Display (n=43) 76.767 77.413 2 D Displays (n=41) 96.317 94.727 3 D Perspective Display (n=43) 82.744 83.615 TSPRQ4 Tabular Display (n=43) 59.465 64.350 2 D Displays (n =41) 69.847 68.185 3 D Perspective Display (n=43) 57.488 53.871 *Adjusted Mean is for the effect of the covariate Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept 0.042 2.543 0.083 TSPQ1 Pilla 0.022 1.330 0.269 TSPQ5 0.109 7.090 0.001 TSPQ6 0.107 6.933 0.001 SS 0.061 3.756 0.026 PTH 0.093 5.956 0.003 SMRT 0.029 1.709 0.186 TSMRT 0.177 12.438 <0.001 Treatment 0.078 2.374 0.053 TSPQ1 = time spent in seconds by each participant when answering practice question 1. TSPQ5 = time spent in seconds by each participant when answering practice question 5. TSPQ6 = time spent in seconds by each participant when answering practice question 6. SS= student status. PTH = part time working hours. SMRT = the score on Mental Rotations Test TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test

PAGE 245

230 Table 42: Test Results of H2b (Continued) Panel C: ANCOVA Results Using Time Spent as the Dependent Variables. Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSPRQ2 Corrected Model 101971.625 9 11330.181 7.076 <0.001 I ntercept 4722.844 1 4722.844 2.949 0.089 TSPQ1 13.994 1 13.994 0.009 0.926 TSPQ5 22902.634 1 22902.634 14.303 <0.001 TSPQ6 4108.135 1 4108.135 2.566 0.112 SS 0.646 1 0.646 0.000 0.984 PTH 10667.610 1 10667.610 6.662 0.011 SMRT 1394.431 1 1394.4 31 0.871 0.353 TSMRT 10174.204 1 10174.204 6.354 0.013 Treatment 4611.560 2 2305.780 1.440 0.120 Error 187348.044 117 1601.265 Total 1209106.000 127 Corrected Total 289319.669 126 TSPRQ4 Corrected Model 48962.334 9 5440.259 9.012 <0.001 Intercept 2074.348 1 2074.348 3.426 0.066 TSPQ1 1472.688 1 1472.688 2.440 0.121 TSPQ5 530.459 1 530.459 0.879 0.350 TSPQ6 8127.412 1 8127.412 13.463 <0.001 SS 4280.842 1 4280.842 7.091 0.009 PTH 1561.159 1 1561.159 2.586 0.111 SMRT 1924.010 1 1924.010 3.187 0.077 TSMRT 13997.534 1 13997.534 23.187 <0.001 Treatment 4300.291 2 2150.146 3.562 0.016 Error 70629.540 117 603.671 Total 608276.000 127 Corrected Total 119591.874 126 TSPRQ2 Adjusted R Squared = 0.303 TSPRQ4 Adjusted R Squared = 0.364 TSPQ1 = time spent in seconds by each participant when answering practice question 1. TSPQ5 = time spent in seconds by each participant when answering practice question 5. TSPQ6 = time spent in seconds by each participant when answering practice question 6. SS= student status. PTH = part time working hours. SMRT = the score on Mental Rotations Test TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test *Treatment p-values are one-tail, all others are two-tail.

PAGE 246

231 Table 42: Test Results of H2b (Continued) Panel D: Bonferroni Pairwise Comparisons for Test H2b Dependent Variables (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value TSPRQ4 3 D Displays Tabular Display 2 D Display 10.479 14.287 5.952 5.653 0.121 0.019 *p -values are one-tail.

PAGE 247

232 5.5.6 Results of H2c Four dependent variables are used to test the pattern recognition task hypothesis the third question (PRQ3), the score on the fifth question (PRQ5), and the score on the sixth question (PRQ6). In question one (PQ1), the score on practice question two (PQ2), the score on practice question six (PQ6), time spent on practice question five (TSPQ5), gender of the participants (GEN), SAT score (SAT), part time working hours (PTH), accounting ere included in the model (see section 5.5.3) along with the manipulated variable Treatment. The first dependent variable used to test hypothesis H2c was PRQ1, which asked (see Table 4) participants what the data differences were between companies one and six. Table 43, Panel A indicates that on average those participants viewing the tabular display were the most accurate (had the highest score) on this pattern recognition task. Those participants viewing the 3-D perspective display (mean score 2.697) were 7% less accurate than those viewing the tabular display (mean score 2.906). Those participants viewing the 3-D perspective display (mean score 2.697) were also 3% less accurate than those using the 2-D displays (mean score 2.731). The second dependent variable (PRQ3) used to test hypothesis H2c asked participants to describe the pattern of financial ratios they were seeing in group one compared to group two (see Table 4). Table 43, Panel A indicates that on average those participants viewing the 2-D displays were the most accurate (had the highest score) on this trend analysis task. Those participants viewing a 3-D perspective display (mean score

PAGE 248

233 2.116) were 3% less accurate than those viewing the 2-D displays (mean score 2.195). However, those using the tabular display (mean score 1.953) were 8% less accurate than those viewing the 3-D perspective display (mean score 2.116). The third dependent variable used to test hypothesis H2c was PRQ5, which asked (see Table 4) participants what the data differences were between company four and six by selecting choices from a template. Table 43, Panel A indicates that on average those participants viewing the tabular display were the most accurate (had the highest score) on this pattern recognition task. Those participants viewing the 3-D perspective display (mean score 2.720) were 6% less accurate than those viewing the tabular display (mean score 2.906). Those participants viewing the 3-D perspective display (mean score 2.720) were also 2% less accurate than those using the 2-D displays (mean score 2.780). The fourth dependent variable (PRQ6) used to test hypothesis H2c asked participants to describe the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template (see Table 4). Table 43, Panel A indicates that on average those participants viewing the 2-D displays were the most accurate (had the highest score) on this trend analysis task. Those participants viewing a 3-D perspective display (mean score 2.627) were 5% less accurate than those viewing the 2-D displays (mean score 2.780). Those participants viewing the 3D perspective display (mean score 2.627) were also 3% less accurate than those viewing the tabular display (mean score 2.720). The mean results for the dependent variables from this and the preceding paragraph provide little support for hypothesi s H2c, which predicted that participants using a single 3-D perspective display will be the most effective in generating hypotheses

PAGE 249

234 for what caused the emerged patterns when compared to participants using a set of 2D displays or participants using a table. A MANCOVA analysis was conducted. As shown (Table 38, Panel B), the overall F-statistic for the manipulated variable Treatment is not significant (p = 0.506) Trace. Analysis of the univariate results is not necessary. Thus, hypothesis H2c predicting that participants using a single 3-D perspective display will be the most effective in generating hypotheses for what caused the emerged patterns, when compared to participants using a set of 2-D displays or participants using a table, was not supported.

PAGE 250

235 Table 43 Test Results of H2c (Participants using a single 3-D perspective display will be the most effective in generating hypotheses for what caused the emerged patterns.) MANCOVA Model on Effectiveness (Accuracy) in Trend Analysis Task ies 1, 3, 4, and 6, and group two includes companies 2 and 5. Compared to group two, what are the patterns of the financial ratios you are es 1, 3, 4 and 6, and group two includes companies 2 and 5. Compared to group one what are the patterns of the financial ratios you are Tests of Between-Subjects Effects on Effectiveness Panel A: Mean Scores on the Tasks. Dep endent Variable Treatment Actual Mean MANCOVA Adjusted Mean* PRQ1 Tabular Display (n=43) 2.906 2.885 2 D Displays (n=41) 2.731 2.757 3 D Perspective Display (n=43) 2,697 2.695 PRQ3 Tabular Display (n=43) 1.953 1.863 2 D Displays (n=41) 2.195 2.19 4 3 D Perspective Display (n=43) 2.116 2.208 PRQ5 Tabular Display (n=43) 2.906 2.853 2 D Displays (n=41) 2.780 2.820 3 D Perspective Display (n=43) 2.720 2.738 PRQ6 Tabular Display (n=43) 2.720 2.616 2 D Displays (n=41) 2.780 2.803 3 D Perspe ctive Display (n=43) 2.627 2.712 *Adjusted Mean is for the effect of the covariate

PAGE 251

236 Table 43: Test Results of H2c (continued) Panel B: Multivariate Tests Variables Multivariate Test Value F stat |p value| Intercept 0.353 15.243 <0.001 P Q1 0.098 3.055 0.020 PQ2 0.253 9,472 <0.001 PQ6 0.043 1.256 0.292 TSPQ5 0.067 2.025 0.096 GEN 0.066 1.967 0.104 SAT 0.100 3.115 0.018 PTH 0.051 1 .492 0.209 ARWE 0.032 0.923 0.453 MW 0.088 2.685 0.035 Treatment 0.063 0.914 0.506 PQ1 = practice question 1. PQ2 = practice question 2. PQ6 = practice question 6. TSPQ5 = time spent on practice question five. GEN = gender. SAT = scores of SAT. PTH = part time working hours. ARWE = accounting related working experience. MW = mental workload.

PAGE 252

237 5.5.7 Results of H2d Four dependent variables are used to test the pattern recognition task hypothesis (TSPRQ1), the time spent in seconds by each participant when answering the third question (TSPRQ3), the time spent in seconds by each participant when answering the fifth question (TSPRQ5), and the time spent in seconds by each participant when answering the sixth question (TSPRQ6). In constructing the models to test hypothesis H2d, seven practice question four (PQ4), the score on practice question six (PQ6), time spent on practice question four (TSPQ4), age of the participants (AGE), SAT score (SAT), the score on the Mental Rotations Test (SMRT), the time spent in seconds by each model (see section 5.4.3) along with the manipulated variable Treatment. The first dependent variable used to test H2d was TSPRQ1, the time spent in seconds by each participant when describing what the data differences were between companies 1 and 6. Table 44, Panel A indicates that on average those participants viewing the tabular display were the most efficient (used the least time in seconds) on this pattern recognition task. Those participants viewing a 3-D perspective display (mean seconds 115) used 37% more time, than those viewing the tabular display (mean seconds 83).Those participants viewing the 3-D displays (mean seconds 115) used 29% more time than those viewing the 2-D displays (mean seconds 89). The second dependent variable used to test H2d was TSPRQ3 the time spent in seconds by each participant when describing the pattern of financial ratios in group one.

PAGE 253

238 Table 44, Panel A indicates that on average those participants viewing the 3D perspective display were the most efficient on this pattern recognition task. Those participants viewing the tabular display (mean seconds 121) used 37% more time than those viewing a 3-D perspective display (mean seconds 88). Those participants viewing the 2-D displays (mean seconds 119) used 36% more time than those viewing the 3D perspective display (mean seconds 88). The third dependent variable used to test H2d was TSPRQ5 the time spent in seconds by each participant when describing what the data differences were between companies 4 and 6. Table 44, Panel A indicates that on average those participants viewing the tabular display were the most efficient (used the least time in seconds) on this pattern recognition task. Those participants viewing a 3-D perspective display (mean seconds 45) used 43% more time than those viewing the tabular display (mean seconds 31). Those participants viewing the 3-D perspective display (mean seconds 45) used 34% more time than those viewing the 2-D displays (mean seconds 34). The fourth dependent variable used to test H2d was TSPRQ6 the time spent in seconds by each participant when describing the pattern of financial ratios in group two. Table 44, Panel A indicates that on average those participants viewing the 2-D displays were the most efficient on this pattern recognition task. Those participants viewing a 3D perspective display (mean seconds 43) used 3% more time than those viewing the 2D displays (mean seconds 41). Those participants viewing the tabular display (mean seconds 44) used 3% more time than those viewing the 3-D perspective display (mean seconds 43). The mean results for the dependent variables indicate that there may be mixed support for H2d, which predicted that participants viewing a single 3D

PAGE 254

239 perspective display will be the most efficient in generating hypotheses for what caused the emerged patterns when compared to participants viewing a set of 2-D displays or participants using a table. Prior to presenting ANCOVA results for the four dependent variables used to test hypothesis H2d, a MANCOVA analysis was conducted. As shown (Table 44, Panel B), the overall F-statistic for the manipulated variable Treatment is significant (p < 0.001) These significant results allow for analysis of the univariate results, which are provided on Panel C of Table 39. Panel C, indicates that the covariates TSPQ4 (p < 0.001), AGE (p = 0.030), and SAT (p = 0.008) are associated with the time spent in seconds by each participant (TSPRQ1) when answering the first question in the pattern recognition task. Results suggest that the manipulation of the presentation formats (Treatment) is significantly (p = 0.002) associated with the time spent by participants (TSPRQ1) when describing the data differences between companies 1 and 6. A paired comparison test (Table 44, Panel D) shows that both the participants viewing the tabular display (p = 0.002) and the participants viewing the 2-D displays (p = 0.011) were more efficient, using less time than participants viewing the 3-D perspective display when describing the data differences between companies 1 and 6. Thus, hypothesis H2d predicting that participants viewing a single 3-D perspective display will be the most efficient in generating hypotheses for what caused the emerged patterns, when compared to participants viewing a set of 2-D displays or participants using a table, was not supported.

PAGE 255

240 Panel C, indicates that the covariate TSPQ4 (p < 0.001) is associated with the time spent in seconds by each participant (TSPRQ3) when answering the third question in the pattern recognition task. Results suggest that the manipulation of the presentation formats (Treatment) is significantly (p < 0.001) associated with the time spent by participants (TSPRQ3) when describing the pattern of financial ratios in group one. A paired comparison test (Table 44, Panel D) shows that the participants viewing the 3D perspective display were not only significantly more efficient or used less time than participants viewing the tabular display (p = 0.001), but also significantly more efficient than participants viewing the 2-D displays (p = 0.002) when describing the pattern of financial ratios in group one. Thus, hypothesis H2d predicting that participants viewing a single 3-D perspective display will be the most efficient in generating hypotheses for what caused the emerged patterns, when compared to participants viewing a set of 2D displays or participants using a table was supported using TSPRQ3. Panel C indicates that the covariates TSPQ4 (p < 0.001), AGE (p = 0.017), SMRT (p = 0.016), and TSMRT (p = 0.032) are associated with the time spent in seconds by each participant (TSPRQ5) when answering the fifth question in the pattern recognition task. Results suggest that the manipulation of the presentation formats (Treatment) is significantly (p < 0.001) associated with the time spent by participants (TSPRQ4) when describing the data differences between companies 4 and 6. A paired comparison test (Table 44, Panel D) shows that both the participants viewing the tabular display (p < 0.001) and the participants viewing the 2-D displays (p < 0.001) were more efficient or used less time than participants viewing the 3-D perspective display when describing the data differences between companies 4 and 6. Thus, hypothesis H2d was not supported.

PAGE 256

241 Panel C, contrary to expectations, suggests that manipulation of the presentation formats (Treatment) does not have a significant (p = 0.35) effect on the time spent by the participants when describing the pattern of financial ratios in group two (TSPRQ6). Since there is no significant main effect a paired comparison test was not conducted. Thus, TSPRQ6 does not support hypothesis H2d.

PAGE 257

242 Table 44 Test Results of H2d (Participants viewing a single 3-D perspective display will be the most efficient in generating hypotheses for what caused the emerged patterns) MANCOVA Model on Efficiency (Less Time) in Trend Analysis Task and 5. Compared to group two, what are the patterns of the financial ratios you are seeing in and 5. Compared to group one what are the patterns of the financial ratios you are seeing i Tests of Between Subjects Effects on Efficiency Panel A: Mean Time Spent on the Tasks. Dependent Variable Treatment Actual Mean (Seconds) MANCOVA Adjusted Mean* (Seconds) TSPRQ1 Tabular Display (n=43) 83.558 85.458 2 D Displays (n=41) 89.097 90.215 3 D Perspective Display (n=43) 115.279 112.300 TSPRQ3 Tabular Display (n=43) 121.418 122.700 2 D Displays (n=41) 119.902 119.900 3 D Perspective Display (n=43) 88.139 86.876 TSPRQ5 Tabular Display (n=43) 31.976 33.228 2 D Di splays (n=41) 34.097 34.469 3 D Perspective Display (n=43) 45.953 44.348 TSPRQ6 Tabular Display (n=43) 44.604 45.377 2 D Displays (n=41) 41.756 42.462 3 D Perspective Display (n=43) 43.325 41.880 *Adjusted Mean is for the effect of the covariate

PAGE 258

243 Table 44: Test Results of H2d (Continued) Panel B: Multivariae Tests Variables Multivariate Test Value F stat |p value| Intercept 0.181 6.317 <0.001 PQ4 0.057 1.707 0.153 TSPQ4 0.342 14.802 <0.001 PQ6 Pilla 0.035 1.022 0.399 AGE 0.092 2.901 0.025 SAT 0.076 2.356 0.058 SMRT 0.053 1.611 0.176 TSMRT 0.076 2.350 0.058 Treatment 0.356 6.225 <0.001 PQ4 = score on practice question 4. TSPQ4 = time spent in seconds by each participant when answering practice question 4. PQ6 = score on practice question 6. AGE= age. SAT = scores of SAT SMRT = the score on Mental Rotations Test TSMRT =the time spent in seconds by each participant when answering the Mental Rotations Test Panel C: ANCOVA Results Using Time Spent as the Dependent Variables Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSPRQ1 Corrected Model 128262.789 9 14251.421 11.27 4 <0.001 Intercept 3934.007 1 3934.007 3.112 0.080 PQ4 584.685 1 584.655 0.463 0.489 TSPQ4 40991.664 1 40994.664 32.429 <0.001 PQ6 2307.231 1 2307.231 1.825 0.179 AGE 6089.008 1 6089.008 4.817 0.030 SAT 9141.717 1 9141.717 7.232 0.008 SMRT 22 9.995 1 229.995 0.182 0.670 TSMRT 349.556 1 349.556 0.277 0.600 Treatment 14993.777 2 7496.888 5.931 0.002 Error 147895.258 117 1264.062 Total 1448703.000 127 Corrected Total 276158.047 126

PAGE 259

244 Table 44: Test Results of H2d (Continued) Panel C: ANCOVA Results Using Time Spent as the Dependent Variables Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSPRQ3 Corrected Model 99253.737 9 11028.193 5.566 <0.001 Intercept 4437.345 1 4437.345 2.239 0.137 PQ4 5372.637 1 5372.637 2.711 0.102 TSPQ4 29529.177 1 29529.177 14.902 <0.001 PQ6 5252.440 1 5252.440 2.651 0.106 AGE 1.782 1 1.782 0.001 0.976 SAT 3833.227 1 3833.227 1.934 0.167 SMRT 462.676 1 462.667 0.233 0.630 TSMRT 1486.263 1 14862.63 0.7 50 0.388 Treatment 28980.825 2 14490.413 7.313 <0.001 Error 231836.703 117 1981.510 Total 1858345.000 127 Corrected Total 331090.441 126 TSPRQ5 Corrected Model 10610.296 9 1178.922 10.208 <0.001 Intercept 2238.411 1 2238.411 19.382 <0.00 1 PQ4 207.424 1 207.424 1.796 0.183 TSPQ4 2191.325 1 2191.325 18.975 <0.001 PQ6 14.364 1 14.364 0.124 0.725 AGE 679.073 1 679.073 5.880 0.017 SAT 153.426 1 153.426 1.329 0.251 SMRT 685.363 1 685.363 5.935 0.016 TSMRT 543.457 1 543.457 4.706 0 .032 Treatment 2710.509 2 1355.254 11.735 <0.001 Error 13512.019 117 115.487 Total 201705.000 127 Corrected Total 24122.315 126 PQ4 = score on practice question 4. TSPQ4 = time spent in seconds by each participant when answering practice question 4. PQ6 = score on practice question 6. AGE= age. SAT = scores of SAT. SMRT = the score on Mental Rotations Test. TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test. *Treatment p-values are one-tail, all others are two-tail.

PAGE 260

245 Table 44: Test Results of H2d (Continued) Panel C: ANCOVA Results Using Time Spent as the Dependent Variables Dependent Variables Source of Variation Type III SS DF Mean Square F stat p value* TSPRQ6 Corrected Model 5333.715 9 592.635 1.571 0.132 Intercept 2701.757 1 2701.757 7.163 0.009 PQ4 433.582 1 433.582 1.150 0.286 TSPQ4 840.912 1 840.912 2.229 0.138 PQ6 421.258 1 421.258 1.117 0.293 AGE 12.316 1 12.316 0.033 0.857 SAT 78.688 1 78.688 0.209 0.649 SMRT 613.9 89 1 613.989 1.628 0.205 TSMRT 2569.261 1 2569.261 6.812 0.010 Treatment 270.051 2 135.025 0.358 0.350 Error 44130.222 117 377.181 Total 287047.000 127 Corrected Total 49463.937 126 TSPRQ1 Adjusted R Squared = 0.423 TSPRQ1 Adjusted R Squared = 0.246 TSPRQ1 Adjusted R Squared = 0.397 TSPRQ1 Adjusted R Squared = 0.039 PQ4 = score on practice question 4. TSPQ4 = time spent in seconds by each participant when answering practice question 4. PQ6 = score on practice question 6. AGE= age. SAT = scores of SAT. SMRT = the score on Mental Rotations Test. TSMRT = the time spent in seconds by each participant when answering the Mental Rotations Test *Treatment p-values are one-tail, all others are two-tail. Panel D: Bonferroni Pairwise Comparisons for Test H2d Dependent Variables (I) Treatment (J) Treatment Mean Difference (I J) Std Error p value* TSPRQ1 3 D Displays Tabular Display 2 D Display 26.856 22.099 8.260 8.166 0.002 0.011 TSPRQ3 3 D Displays Tab ular Display 2 D Display 35.856 32.976 10.341 10.225 0.001 0.002 TSPRQ5 3 D Displays Tabular Display 2 D Display 11.120 9.879 2.497 2.468 <0.001 <0.001 *p -values are one-tail.

PAGE 261

246 5.6 Post Hoc Analysis As indicated earlier in section 3.7, the 3-D perspective display of DuPont analysis is a newly created display format of financial ratios that has never been empirically tested. For this reason, the study adopted six survey questions from Murthy, Schafer and randomly assigned display format. The six questions are: 1) using the tables (graphs) was frustrating, 2) the tables (graphs) displayed the task information in a readable format, 3) I found the tables (graphs) useful in how they presented the data for decision making, 4) the tables (graphs) helped me to understand the task data to make a better decision, 5) the tables (graphs) fit the way I needed to view the task information to make better decisions, and 6) overall, I am satisfied with the tables (graphs) in providing the information I needed to complete this task (see Table 9). Participants were asked to select the scale number from one to seven that indicates the extent to which they agree with each of the six questions. (1 = highly disagree, 2 = moderately disagree, 3 = slightly disagree, 4 = neither agree nor disagree, 5 = slightly agree, 6 = moderately agree, 7 = highly agree.). Chapter 6 will discuss the implication s of the results of these survey questions or the opinion from participants in regards to the usefulness and ease of use of their randomly assigned display format.

PAGE 262

247 5.6.1 Trend Analysis Task Post Hoc Analysis Table 45 indicates that both the participants viewing the tabular display (mean responses 3.023) and the participants viewing the 2-D displays (mean responses 3.725) disagreed that using the table (graphs) was frustrating. Those participants viewing the 3D perspective display (mean responses 4.690) neither agreed nor disagreed that using the graph was frustrating. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display (p <0.001), and the mean responses of those participants viewing the 2-D displays (p = 0.043), were significantly different from the mean responses of those participants viewing the 3-D perspective display. Table 45 indicates that the participants viewing the tabular display (mean responses 5.785) slightly agreed that the table displayed the task information in a readable format. Both the participants viewing the 2-D displays (means responses 4.700) and the participants viewing the 3-D perspective display (mean responses 4.380) neither agreed nor disagreed that the graphs displayed the task information in a readable format. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 2-D displays (p < 0.001), and from the mean responses of those participants viewing the 3-D perspective display (p < 0.001). Table 45 indicates that both the participants viewing the tabular display (mean responses 5.190) and the participants viewing the 2-D displays (mean responses 5.050) slightly agreed that the table (graphs) was useful in how it presented the data for decision making. Those participants viewing the 3-D perspective display (means responses 4.071) neither agreed nor disagreed that the graph was useful in how it presented the data for

PAGE 263

248 decision making. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display (p = 0.006), and the mean responses of those viewing the 2-D displays (p = 0.022), were significantly different from the mean responses of those participants viewing the 3-D perspective display. Table 45 indicates that both the participants viewing the tabular display (mean responses 5.357) and the participants viewing the 2-D displays (mean responses 5.075) slightly agreed that the table (graphs) helped them understand the task data to make a better decision. Those participants viewing the 3-D perspective display (means responses 4.190) neither agreed nor disagreed that the graph helped them understand the task data to make a better decision. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display (p = 0.001), and the mean response of those participants viewing the 2-D displays (p = 0.018), were significantly different from the mean responses of those participants viewing the 3-D perspective display. Table 45 indicates that the participants viewing the tabular display (mean responses 5.000) slightly agreed that the table fit the way the participants needed to view the task information to make a better decision. Those participants viewing the 2D displays (means responses 4.450) neither agreed nor disagreed that the graphs fit the way the participants needed to view the task information to make a better decision. However, those participants viewing the 3-D perspective display (means responses 3.761) slightly disagreed that the graphs fit the way the participants needed to view the task information to make a better decision. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different

PAGE 264

249 from the mean responses of those participants viewing the 3-D perspective displays (p = 0.001). Table 45 indicates that the participants viewing the tabular display (mean responses 5.214) slightly agreed that they were satisfied with the table in providing the information they needed to complete the task. Both the participants viewing the 2D displays (means responses 4.350) and the participants viewing the 3-D perspective display (mean responses 4.095) neither agreed nor disagreed that they were satisfied with the graphs in providing the information they needed to complete the task. Results of oneway ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 3-D perspective displays (p = 0.006).

PAGE 265

250 Table 45 Mean, Standard Deviation, Range and ANOVA Pairwise Comparison (Mean) of the Survey Questions Assessing Display Usefulness and Ease of Use, for the Trend Analysis Task. Tabular Display (n=42) 2 D Displays (n =40) 3 D Display (n =42) Tabular & 2 D Display Mean Difference Tabular & 3 D Display Mean Difference 2 D & 3 D Display Mean Difference (graphs) was 3.023 (1.569) 1.00 6.00 3.725 (1.782) 1.00 7.00 4.690 (1.960) 1.00 7.00 n/s (p < 0.001) (p = 0.043)* (graphs) displayed the task information in a 5.785 (1.200) 2.00 7.00 4.700 (1.697) 1.00 7.00 4.380 (1.974) 1.00 7.00 (p = 0.001)* (p < 0.001) n/s (graphs) useful in how they presented the data for 5.190 (1.485) 1.00 7.00 5.050 (1.518) 1.00 7.00 4.071 (1.839) 1.00 7 .00 n/s (p = 0.006) (p = 0.022)* (graphs) helped me understand the task data to make a 5.357 (1.077) 2.00 7.00 5.075 (1.327) 2.00 7.00 4.190 (1.783) 1.00 7.00 n/s (p = 0.001) (p = 0.018)* (graphs) fit the w ay I needed to view the task information to m ake a better 5.000 (1.361) 2.00 7.00 4.450 (1.600) 1.00 7.00 3.761 (1.664) 1.00 7.00 n/s (p = 0.001) n/s satisfied with the tables (graphs) in providing the information I needed to complete 5.214 (1.353) 2.00 7.00 4.350 (1.687) 1.00 7.00 4.095 (1.791) 1.00 7.00 n/s (p = 0.006) n/s n/s = non significant results of the pairwise comparison of means at p< =0.05

PAGE 266

251 5.6.2 Pattern Recognition Task Post Hoc Analysis Results of one-way ANOVA testing suggest that there was no significant difference between mean responses of the treatment groups in regard to the question whether using the table (graphs) was frustrating (see Table 46). Table 46 indicates that the participants viewing the tabular display (mean responses 5.860) slightly agreed that the table displayed the task information in a readable format. Both the participants viewing the 2-D displays (means responses 4.829) and the participants viewing the 3-D perspective display (mean responses 4.441) neither agreed nor disagreed that the graphs displayed the task information in a readable format. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 2-D displays (p = 0.005), and from the mean responses of those participants viewing the 3-D perspective display (p < 0.001). Table 46 indicates that the participants viewing the tabular display (mean responses 5.395) slightly agreed that the table (graphs) was useful in how it presented the data for decision making. Both the participants viewing the 2-D displays (mean responses 4.804) and the participants viewing the 3-D perspective display (means responses 4.279) neither agreed nor disagreed that the graph was useful in how it presented the data for decision making. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 3-D perspective display (p = 0.003). Table 46 indicates that both the participants viewing the tabular display (mean responses 5.232) and the participants viewing the 2-D displays (mean responses 5.243)

PAGE 267

252 slightly agreed that the table (graphs) helped them understand the task data to make a better decision. Those participants viewing the 3-D perspective display (means responses 4.472) neither agreed nor disagreed that the graph helped them understand the task data to make a better decision. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display (p = 0.015), and the mean responses of those participants viewing the 2-D displays (p = 0.015), were significantly different from the mean responses of those participants viewing the 3-D perspective display. Table 46 indicates that both the participants viewing the tabular display (mean responses 4.906) and the participants viewing the 2-D displays (means responses 4.634) neither agreed nor disagreed that the graphs fit the way the participants needed to view the task information to make a better decision. However, those participants viewing the 3-D perspective display (means responses 3.953) slightly disagreed that the graphs fit the way the participants needed to view the task information to make a better decision. Results of one-way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 3-D perspective displays (p = 0.008). Table 46 indicates that the participants viewing the tabular display (mean responses 5.209) slightly agreed that they were satisfied with the table in providing the information they needed to complete the task. Both the participants viewing the 2D displays (means responses 4.853) and the participants viewing the 3-D perspective display (mean responses 4.441) neither agreed nor disagreed that they were satisfied with the graphs in providing the information they needed to complete the task. Results of one-

PAGE 268

253 way ANOVA testing suggest that mean responses of those participants viewing the tabular display were significantly different from the mean responses of those participants viewing the 3-D perspective displays (p = 0.027). Chapter Six will discuss the results in detail.

PAGE 269

254 Table 46 Mean, Standard Deviation, Range and ANOVA Pairwise Comparison (Mean) of the Survey Questions on Display Usefulness and Ease of Use, for the Pattern Recognition Task. Tabular Display (n =42) 2 D Displays (n =40) 3 D Display (n =42) Tabular & 2 D Display Mean Difference Tabular & 3 D Display Mean Difference 2 D & 3 D Display Mean Difference (graphs) was 3 .162 (1.462) 1.00 6.00 3.756 (1.894) 1.00 7.00 3.651 (1.837) 1.00 7.00 n/s n/s n/s displayed the task information in a 5.860 (1.245) 2.00 7.00 4.829 (1.610) 1.00 7.00 4.441 (2.015) 1.00 7.00 (p = 0.005)* (p < 0. 001) n/s (graphs) useful in how they presented the data for decision 5.395 (1.311) 1.00 7.00 4.804 (1.720) 1.00 7.00 4.279 (2.027) 1.00 7.00 n/s (p = 0.003) n/s helped me understand the task data to m ake 5.232 (1.377) 2.00 7.00 5.243 (1.462) 1.00 7.00 4.472 (1.964) 1.00 7.00 n/s (p = 0.015) (p = 0.015)* fit the way I needed to view the task information to make 4.906 (1.268) 1.00 7.00 4.63 4 (1.624) 1.00 7.00 3.953 (1.951) 1.00 7.00 n/s (p = 0.008) n/s satisfied with the tables (graphs) in providing the information I needed 5.209 (1.225) 2.00 7.00 4.853 (1.711) 1.00 7.00 4.441 (1.776) 1.00 7.00 n/s (p = 0.027) n/s n/s = non significant results of the pairwise comparison of means at p <=0.05.

PAGE 270

255 Ch apter 6: Discussion 6.1 Summary of Hypothesized Results Hypothesis one was tested in several parts (1a-1d) using several dependent variables, as was hypothesis two (H2a-H2d). After providing a summary of the results for the two hypotheses a discussion of the results will be provided in 6.2. Table 43 summarizes the information presented in this section. Hypothesis H1a predicted that participants using a set of 2-D displays would be the most effective in generating hypotheses for what caused the changes in the trend of accounting data when compared to participants using a single 3-D perspective display or participants using a tabular display. This hypothesis has four dependent measures of were between years 1 and 4; the score on the second question (TAQ2), which asked what participants perceived to be occurring in the data when going from year 2 to year 3 and year 4; the score on the fifth question (TAQ5), which asked participants to select from a template to indicate the differences in data between years 2 and 4; and the score on the sixth question (TAQ6), which asked participants to select from a template to indicate changes in data when going from year 1 to year 2 to year 3. Hypothesis H1a was partially supported by the results for TAQ1 and TAQ6. Participants viewing the 2-D display were significantly (p = 0.001) more effective than participants viewing the 3-D perspective display in describing the differences between

PAGE 271

256 years 1 and 4 (TAQ1). Participants viewing the 2-D display were also significantly (p < 0.001) more accurate than participants viewing the 3-D perspective display in indicating changes in data when going from year 1 to year 2 to year 3 by selecting choices from a template (TAQ6). Hypothesis H1b predicted that participants using a set of 2-D displays would be the most efficient in generating hypotheses for what caused the changes in trend of accounting data when compared to participants using a single 3-D perspective display or participants using a tabular display. This hypothesis has four dependent measures of each participant when answering what the data differences were between years 1 and 4 (TSTAQ1), the time spent in seconds by each participant in describing what was occurring in the data when going from year 2 to year 3 and year 4 (TSTAQ2), the time spent in seconds by each participant when selecting from a template the differences in data between years 2 and 4 (TSTAQ5), and the time spent in seconds by each participant when selecting from a template changes in data when going from year 1 to year 2 to year 3 (TSTAQ6). Hypothesis H1b was not supported since only one of the four dependent variables significantly increased when participants viewed 2-D displays. Participants viewing the 2-D displays were significantly (p < 0.001) more efficient or used less time in seconds than participants viewing the 3-D perspective display when selecting from a template the differences in data between years 2 and 4 (TSTAQ5). Hypothesis H1c predicted that participants using a set of 2-D displays would be the most effective in an accounting judgment involving estimation of values when compared to participants using a single 3-D perspective display or participants using a

PAGE 272

257 the third question (TAQ3), which ask ed the participants to estimate what the ROE would be in year 6 (TAQ3) if each of the variables comprising ROE in year 5 doubles; and the ask ed the participants to estimate the average of turnover (TAQ4a), profitability (TAQ4b) and leverage (TAQ4c) for the years 1, 2, 4 and 5, and use the estimated average to calculate a new ROE (TAQ4d). Hypothesis H1c was not supported. Contrary to what hypothesis H1c predicted, participants viewing the 3-D perspective display were more effective (p = 0.031) or accurate than those participants viewing the 2-D display in estimating the ROE for year 6 (TAQ3) if each of the variables comprising ROE in year 5 doubles. Hypothesis H1c was also not supported in that both the participants viewing the tabular display (p = 0.032) and the participants viewing the 3-D perspective display (p = 0.047) were more effective or accurate than those participants viewing the 2-D displays in estimating the average of leverage for the years 1, 2, 4 and 5 (TAQ4c). On top of the aforementioned result, hypothesis H1c was further not supported by the fact that the participants viewing the tabular display were more (p = 0.026) effective or accurate than those participants viewing the 2-D display in estimating the average of turnover, profitability and leverage for the years 1, 2, 4 and 5, and in using the estimated average to calculate a new ROE (TAQ4d). Hypothesis H1d predicted that participants using a set of 2-D displays would be the most efficient in an accounting judgment involving estimation of values when compared to participants using a single 3-D perspective display or participants using a

PAGE 273

258 tabular display. This hypothesis has two dependent measures o spent in seconds by each participant when estimating what the ROE would be in year 6 if each of the variables comprising ROE in year 5 doubles (TSTAQ3); and the time spent in seconds by each participant when estimating the average of turnover, profitability and leverage for the years 1, 2, 4 and 5, and using the estimated average to calculate a new ROE (TSTAQ4). Hypothesis H1d was not supported as the MANCOVA F-statistic for the manipulated variable Treatment was not significant (p The remainder of this section summarizes the second hypothesis. Hypothesis H2a predicted that participants using a single 3-D perspective display would be the most effective in recognizing patterns of accounting data when compared to participants using a set of 2-D displays or participants using a tabular display. This hypothesis has two the participants to separate companies one through six into two groups based on similar financial characteristics; and the score on the fourth question (PRQ4), which asked the participants to select one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time. Hypothesis H2a was fully supported, in that significant results were found for both dependent variables. Participants viewing the 3-D perspective display not only were significantly more effective or accurate than those participants viewing the tabular display (p < 0.001) but were also more effective than participants viewing the 2D display (p < 0.001) when it came to separating companies one through six into two groups based on similar financial characteristics (PRQ2). Participants viewing the 3D perspective display were also significantly more effective or accurate than those

PAGE 274

259 participants viewing the tabular display (p = 0.007) and those viewing the 2-D displays (p = 0.016) in selecting which of the six companies had high profitability, high turnover but low leverage at the same time (PRQ4). Hypothesis H2b predicted that participants using a single 3-D perspective display would be the most efficient in recognizing patterns of accounting data when compared to participants using a set of 2-D displays or participants using a tabular display. This participant in separating companies one through six into two groups based on similar financial characteristics (TSPRQ2); and the time spent in seconds by each participant in selecting one of the six companies if the goal is to have high profitability, high turnover but low leverage at the same time (TSPRQ4). Hypothesis H2b was partially supported. Participants viewing the 3-D perspective display were only significantly more efficient (p = 0.019) than those participants viewing the 2-D displays when selecting which of six companies had highest profitability, highest turnover but lowest leverage at the same time (TSPRQ4). There was not a significant difference between 3-D display and tabular format, nor was there support for H2b using participant in separating companies one through six into two groups based on similar financial characteristics. Hypothesis H2c predicted that participants using a single 3-D perspective display would be the most effective in generating hypotheses for what caused the emerged patterns when compared to participants using a set of 2-D displays or participants using a

PAGE 275

260 the first question (PRQ1), which asked the participants what the data differences were between companies one and six; the score on the third question (PRQ3), which asked the participants to describe the pattern of financial ratios they were seeing in group one compared to group two; the score on the fifth question (PRQ5), which asked the participants what the data differences were between companies four and six by selecting choices from a template; and the score on the sixth question (PRQ6), which asked the participants to describe the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template. Hypothesis H2c was not supported as the MANCOVA F-statistic for the manipulated variable Treatment was not Hypothesis H2d predicted that participants using a single 3-D perspective display would be the most efficient in generating hypotheses of what caused the emerged patterns when compared to participants using a set of 2-D displays or participants using a tabular display. This hypothesis has four time measures related to the dependent measures identified for H2c; they are TSPRQ1, TSPRQ3, TSPRQ5 and TSPRQ6. Hypothesis H2d had mixed results. No support was found for hypothesis H2d using TSPRQ6 time spent by each participant when describing the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template. Contrary support was found using TSPRQ1 (time spent by each participant in describing what the data differences were between companies one and six) and TSPRQ5 (time spent by each participant in describing what the data differences were between companies four and six by selecting choices from a template). With TSPRQ1 both the participants viewing the tabular display (p = 0.002) and the participants viewing the 2D

PAGE 276

261 displays (p = 0.012) were more efficient, or used less time, than participants viewing the 3-D perspective display when describing the data differences between companies 1 and 6. Again, with TSPRQ5 both the participants viewing the tabular display (p < 0.001) and the participants viewing the 2-D displays (p < 0.001) were more efficient, or used less time, than participants viewing the 3-D perspective display when describing the data differences between companies 4 and 6 by selecting choices from a template. However, the time spent by each participant in describing the pattern of financial ratios they were seeing in group one compared to group two (TSPRQ3) support hypothesis H2d. Participants viewing the 3-D perspective display were not only significantly more efficient than participants viewing the tabular display (p = 0.001), but also significantly more efficient, or used less time, than participants viewing the 2D displays (p = 0.002) when describing the pattern of financial ratios in group one. Table 47 and Table 48 provide a summary of the results of hypotheses H1 and H2, respectively.

PAGE 277

262 Table 47 Summary of the Results of the Trend Analysis Task Hypothesis Dependent Variables Test* Findings H1a Subjects using 2 D displays Score on the 1 st question (TAQ1) S 2 D more accurate than 3 D will be the most effective Score on the 2 nd question (TAQ2) N/S (accuracy) in generating Score on the 5 th question (TAQ5) N/S hypotheses for what caused Score on the 6 th question (TAQ6) S 2 D more accurate than 3 D the chang es in the trend of accounting data H1b Subjects using 2 D displays Time spent on the 1 st question (TSTAQ1) N/S will be the most efficient Time spent on the 2 nd question (TSTAQ2) N/S (less time) in generating Time spent on the 5 th qu estion (TSTAQ5) S 2 D more efficient than 3 D hypotheses for what caused Time spent on the 6 th question (TSTAQ6) N/S the changes in the trend of accounting data H1c Subjects using 2 D displays Score on the 3 rd question (TAQ3) N/S* 3 D more accurate than 2 D will be the most effective Score on the 4 th N/S (accuracy) in the estimation Score on the 4 th N/S of values Score on the 4 th N/S* Table,3 D more accurate Score on the 4 th N/S* Table more accurate than 2 D H1d Subjects using 2 D displays Time spent on the 3 rd question TSTAQ3) N/S will be the most efficient Time spent on the 4 th question (TSTAQ4) N/S (less time) in the estimat ion of values TAQ1 = score on the 1 st question which asked the participants what the data differences were between years 1 and 4. TAQ2 = score on the 2 nd question which asked what the participants perceived to be occurring in the data when going from year 2 to year 3 and year 4. TAQ3 = score on the 3 rd question which asked the participants to estimate what the ROE would be in year 6 if each of the variables comprising ROE in year 5 doubles. icipants to estimate the average of turnover (TAQ4a). profitability (TAQ4b), and leverage (TAQ4c) for the years 1, 2, 4, and 5, and use the estimated average to calculate a new ROE (TAQ4d). TAQ5 = score on the 5 th question which asked the participant to select from a template to indicate the differences between year 2 and 4. TAQ6 = score on the 6 th question which asked the participants to select from a template to indicate changes in data when going from year ` to year 2 to year 3. S = results of testing of the hypothesis indicate that the Mean difference is significant at the 0.05 level and thus supporting the hypothesis in the predicted direction. N/S = results of testing of the hypothesis indicate that the Mean difference is insignificant at the 0.05 level. N/S* = results of testing of the hypothesis indicate that the Mean difference is significant at the 0.05 level but in the opposite direction of the predicted.

PAGE 278

263 Table 48 Summary of the Results of the Pattern Recognition Task Hypothesis Dependent Var iables Test Findings* H2a Subjects using 3 D Score on the 2 nd question (PRQ2) S 3 D more accurate than table, 2 D display will be the most Score on the 4 th question (PRQ4) S 3 D more accurate than table, 2 D effective (accuracy) in recognizi ng patterns of accounting data H2b Subjects using 3 D Time spent on the 2 nd question (TSPRQ2) N/S display will be the most Time spent on the 4 th question (TSPRQ2) S 3 D more efficient than 2 D efficient (less time) in recogni zing patterns of accounting data H2c Subjects using 3 D Score on the 1 st question (PRQ1) N/S display will be the most Score on the 3 th question (PRQ3) N/S effective (accuracy) in Score on the 5 th question (PRQ5) N/S generating hyp otheses for Score on the 6 th question (PRQ6) N/S what caused the emerged patterns H2 d Subjects using 3 D Time spent on the 1 st question (TSPRQ1) N/S* Table, 2 D more efficient than 3 D display will be the most Time spent on the 3 th qu estion (TSPRQ3) S 3 D ,more efficient than 2 D Table efficient (less time) in Time spent on the 5 th question (TSPRQ5) N/S* Table, 2 D more efficient than 3 D generating hypotheses for Time spent on the 6 th question (TSPRQ6) N/S what caused the emerge d patterns PRQ1 = score on the 1 st question which asked the participants what the data differences were between companies one and six. PRQ2 = score on the 2 nd question which asked the participants to separate companies one through six into two groups based on similar financial characteristics. PRQ3= score on the 3 rd question which asked the participants to describe the pattern of financial ratios they were seeing in group one compared to group two. PRQ4 = score on the 4 th question which asked the participants to select one of the six companies if the goal is to have high profitability, high turnover, but low leverage at the same time. PRQ5 = score on the 5 th question which asked the participants what the data differences were between companies four and six by selecting choices from a template. PRQ6 = score on the 6 th question which asked the participants to describe the pattern of financial ratios they were seeing in group two compared to group one by selecting choices from a template. S = results of testing of the hypothesis indicate that the Mean difference is significant at the 0.05 level and thus supporting the hypothesis in the predicted direction. N/S = results of testing of the hypothesis indicate that the Mean difference is insignificant at the 0.05 level. N/S* = results of testing of the hypothesis indicate that the Mean difference is significant at the 0.05 level but in the opposite direction of the predicted.

PAGE 279

264 6.2 Discussion of Results of Hypothesis H1 Hypothesis H1 predicted that participants using a set of 2-D displays would be most effective in two trend analysis tasks generating hypotheses for what caused the changes in the trend of accounting data, and estimating values when compared to participants using a single 3-D perspective display or participants using a tabular display. In relation to the two different trend analysis tasks, hypothesis H1 employed a total of nine dependent measures to test for the differences in accuracy between the responses of participants assigned to different treatment groups. Further, hypothesis H1 predicted that participants using a set of 2-D displays would be most efficient in generating hypotheses for what caused the changes in the trend of accounting data and in estimating values, when compared to participants using a single 3-D perspective display or participants using a tabular display. Hypothesis H1 employed a total of six dependent measures of efficiency to test for the differences in efficiency between participants of different treatment groups. 6.2.1 Discussion of Results of Hypothesis H1a Hypothesis H1a stipulates that those participants viewing the 2-D displays will be the most effective or accurate in generating hypotheses for what caused the changes in the trend of accounting data when compared to those participants viewing the tabular display or those participants viewing a 3-D perspective display. Viewing 2-D displays resulted in significantly more effective or accurate hypotheses by participants for two (TAQ1 and TAQ6) of the four dependent measures. Given the fact that there is no

PAGE 280

265 significant difference in other dependent measures of accuracy between the treatment groups, hypothesis H1a was partially supported. Additional analysis of the results of hypothesis H1a show some facts worthy of discussion. The four dependent measures of accuracy of hypothesis H1a can be divided into two groups: the first group contains the first (TAQ1) and the fifth question (TAQ5), while the second group contains the second (TAQ2) and the sixth question (TAQ6). The first question of the trend analysis task asked the participants to use short sentences to describe the data differences between years 1 and 4. Similarly, the fifth question asked the participants what the data differences were between years 2 and 4 by selecting choices from a template. Both questions involved a time period of two years. Woods (1990) would identify the tasks from TAQ1 and TAQ5 as simple tasks because participants did not need to perform a prior processing step or calculation to create an intermediate value and use that intermediate value to derive the final outcome. A simple comparison between years 1 and 4 or years 2 and 4 is sufficient to answer the first and fifth questions correctly. Though the study is unable to explain why there was no significant differences between those participants viewing the 2-D displays and those viewing the tabular display on all the four dependent measures of accuracy (TAQ1), (TAQ2), (TAQ5), and (TAQ6), the study suggests a possible explanation for why those participants viewing the 2-D displays were significantly (p = 0.001) more accurate than participants viewing a single 3-D perspective display on the first question (TAQ1) but not on the fifth question (TAQ5).

PAGE 281

266 Assuming the template of choices benefited all participants equally when answering the fifth question, those participants viewing the 3-D perspective display possibly became more accurate by being more proficient in reading the 3-D perspective display, which was new to them at the beginning of the experiment. Triffett and Trafton (2006) described how meteorologists spatially transformed the position of a low pressure system toward a certain direction even though the actual movement of the low pressure system was not explicitly shown in a graph. In this study, once participants became more proficient in reading the 3-D perspective display, they were possibly able to mentally create a slope line while generating hypotheses for what the data differences were between years 2 and 4. The aforementioned explanation contributes toward the understanding of why there were no significant differences in the mean score of the fifth question (TAQ5) between the three treatment groups. The next few paragraphs discuss the second group of questions, TAQ2 ant TAQ6. The second question (TAQ2) of the trend analysis task asked the participants to write short sentences about what they perceived to be occurring in the data when going from year 2 to year 3 and year 4; similarly the sixth question (TAQ6) of the trend analysis task asked the participants to select from a template to indicate changes in data when going from year 1 to year 2 to year 3. Both questions involved a time period of sult of a prior process, calculation, and step, then the task is a complex task. Both the second and the sixth question are complex tasks because participants had to discern what was happening between the first two years before correctly hypothesizing what caused the changes in the trend of accounting data across three time periods (from year 1 to year 2 to

PAGE 282

267 year 3 for example). It is interesting to note that the covariate mental workload (MW) was not significantly associated with the dependent measure of TAQ1 (p = 0.094) and only marginally significantly associated with the dependent measure of TAQ5 (p = 0.050), but MW was significantly associated with the dependent measure of TAQ2 (p=0.011) and TAQ6 (p = 0.023). These results support that the second and sixth question of the trend analysis task were both complex tasks. There was no significant difference Benefiting from the message richness of the slope of line, which portrays the trend relationship explicitly, those participants viewing the 2-D displays should have outperformed participants of the other two treatment groups especially on a complex task involving the analysis of multi time period data. While there was no significant difference in accuracy between the three treatment groups on the second question (TAQ2), participants viewing the 2-D displays were significantly more accurate than participants viewing a 3-D perspective display (p < 0.001) in situations such as question six, where a template of choices was provided. Summarizing the above discussion of hypothesis H1a, the following is offered. For simple tasks, those participants viewing the 2-D displays can sometimes be more accurate than those viewing the 3-D perspective display. However, participants of the 3D perspective display were able to improve accuracy enough (possibly due to the ability to spatially transform information) to be on the same footing as participants viewing the 2-D displays when performing a simple task involving two time periods. On the other hand, for the complex task involving three time periods (what was occurring in the data when going from year 1 to year 2 to year 3) viewers of the 2-D displays only out

PAGE 283

268 performed participants viewing the 3-D perspective display when a template of choices was available. Future research should employ methods like verbal protocol analysis to discern what causes the improvement in accuracy in the 3-D perspective display and to understand why participants viewing the tabular display can be more accurate than those viewing the 2-D display when generating hypotheses for what caused the changes in the trend of the data. 6.2.2 Discussion of Results of Hypothesis H1b Hypothesis H1b stipulates that those participants viewing the 2-D displays will be the most efficient or use less time (in seconds) in generating hypotheses for what caused the changes in the trend of accounting data when compared to those participants viewing the tabular display or those participants viewing a 3-D perspective display. Hypothesis H1b uses the time related to answering the four questions asked to test hypothesis H1a. In terms of these four dependent measures of efficiency, those participants viewing the 2D displays were significantly more efficient or used less time than those participants viewing the 3-D display in only one dependent measures of efficiency (TSTAQ5). Given the fact that there are no significant differences in the other dependent measures of efficiency between the treatment groups, hypothesis H1b was not supported. The results of H1b suggest that those participants viewing the 2-D displays were not more efficient than participants viewing the tabular display or participants viewing the 3D perspective display in generating hypotheses for what caused the changes in the trend

PAGE 284

269 of accounting data, the lack of results suggests no one display type is more efficient than the others. 6.2.3 Discussion of Results of Hypothesis H1c Hypothesis H1c stipulated that those participants viewing the 2-D displays would be the most effective or accurate in an accounting judgment involving estimation of values when compared to those participants viewing a tabular display or those participants viewing a 3-D perspective display. There was no support for hypothesis H1c using any of the five dependent measures tested. In fact, for the dependent measure TAQ3, participants viewing the 3-D perspective display were more effective than those participants viewing the 2-D displays in estimating what the ROE would be in year 6 if each of the variables comprising ROE in year 5 doubles. This same result was also found for the dependent measure TAQ4c, which asked participants to estimate the average of leverage for the years 1, 2, 4, and 5, both the participants viewing the tabular display and the participants viewing the 3-D perspective display were more effective than those participants viewing the 2-D displays. Lastly, for the dependent variables TAQ4d participants viewing the tabular display were more effective than those participants viewing the 2-D displays. Though hypothesis H1c was not supported, it is interesting to notice that when performing a trend analysis task that involves the estimation of values, those participants viewing a static three dimensional perspective display of DuPont analysis, which does not have a line of slope connecting data points, can be more effective, or accurate, than those participants viewing the 2-D displays with the line of slope explicitly shown. Dull

PAGE 285

270 and Tegardan (1999) found that participants using three-dimensional line graphs of accounting wealth, momentum and impulse that could be rotated provided the most -dimensional or three dimensional fixed line graphs of equivalent information. It seems that neither the richness of message flags provided by the slope of line (Pinker 1990) nor the smooth continuation of elements as portrayed in a 2-D line displays (Pinker 1990) benefited those participants viewing the 2-D displays in an accounting judgment involving estimation of values. The results of this study indicate that 3-D displays may provide a better fit for such complex tasks, providing a basis for future research investigating whether other types of static or rotatable three-dimensional displays of accounting information that do not have the slope of line connecting data points estimation of values. 6.2.4 Discussion of Results of Hypothesis H1d Hypothesis H1d stipulated that those participants viewing the 2-D displays would be the most efficient or use less time in an accounting judgment involving the estimation of values when compared to those participants viewing a tabular display or those participants viewing a 3-D perspective display. Hypothesis H1d has two dependent measures of efficiency (TSTAQ3 and TSTAQ4). Contrary to prediction, there was no significant difference between treatment groups in terms of time used in a trend analysis task involving an accounting task that requires an estimation of values. Hypothesis H1d was not supported, indicating that viewing a 2-D display format did not command an

PAGE 286

271 advantage or competitive edge over other participants viewing a different type of display format when performing a trend analysis task involving the estimation of values. 6.3 Discussion of Results of Hypothesis H2 Hypothesis H2 predicted that participants using a 3-D perspective display would be most effective in pattern recognition tasks recognizing patterns of accounting data and generating hypotheses for what caused the emerged patterns of accounting data when compared to participants using a set of 2-D displays or participants using a tabular display. In relation to the two pattern recognition tasks, hypothesis H2 employed a total of six dependent measures to test for the differences in accuracy of the responses of participants assigned to different treatment groups. Further, hypothesis H2 also predicted that participants using a 3-D perspective display would be most efficient in recognizing patterns of accounting data and generating hypotheses for what caused the emerged patterns of accounting data when compared to participants using a set of 2-D displays or participants using a tabular display. Hypothesis H2 employed a total of six dependent measures to test for the differences in efficiency between participants of different treatment groups. 6.3.1 Discussion of Results of Hypothesis H2a Hypothesis H2a stipulated that participants viewing the 3-D perspective display would be the most effective or accurate in recognizing patterns of accounting data when compared to those participants viewing a tabular display or those participants viewing the 2-D displays. Hypothesis H2a has two dependent measures of accuracy separating

PAGE 287

272 companies one through six into two groups based on similar financial characteristics (PRQ2), and selecting one of the six companies if the goal is to have high profitability, high turnover, but low leverage at the same time (PRQ4). In terms of both the dependent measures of accuracy, those participants viewing the 3-D perspective display were significantly more effective or accurate than those participants viewing the tabular display and those participants viewing the 2-D displays. The implication of having hypothesis H2a fully supported is interesting. Findings of the study have empirically demonstrated that the emergent features portrayed by a 3D perspective display, as proposed by the Proximity Compatibility Principle (PCP), results in greater decision making accuracy for accounting judgments involving the recognition of patterns in accounting data. Based on the findings of hypothesis H2a, it can be concluded that for accounting tasks involving pattern recognition, the 3-D perspective display (newly developed by the study) has the potential to improve decision accuracy. 6.3.2 Discussion of Results of Hypothesis H2b Hypothesis H2b stipulated that participants viewing the 3-D perspective display would be the most efficient or use less time in recognizing patterns of accounting data when compared to those participants viewing a tabular display or those participants viewing the 2-D displays. For the pattern recognition task of selecting which of six companies has the highest profitability, highest turnover, but lowest leverage at the same time (PRQ4), participants viewing the 3-D perspective display were significantly more efficient or used less time (in seconds) than those participants viewing the 2-D displays.

PAGE 288

273 However, support was not found for H2b using the dependent variable PRQ2, thus hypothesis H2b was partially supported. Partial support of hypothesis H2b implies that participants viewing the 3D perspective display can potentially be more efficient or use less time in forming their decision. The findings indicate that there could be circumstances where decision makers will use less time in decision making by using higher dimension display formats such as the 3-D perspective display. The joint implication of having support for hypotheses H2a and H2b for the pattern recognition task is important. Those participants viewing the 3-D perspective display were not only being more effective or accurate, but also more efficient than those participants viewing the 2-D displays in selecting one of the six companies if the goal is to have high profitability, high turnover, but low leverage at the same time. The results suggest that, for accounting tasks involving the recognition of patterns in data, a 3D perspective display not only can improve decision m less time being used by decision makers in forming their conclusions. 6.3.3 Discussion of Results of Hypothesis H2c Hypothesis H2c stipulated that participants viewing the 3-D perspective display would be the most effective (accurate) in generating hypotheses for what caused the emerged patterns when compared to those participants viewing the tabular display or those participants viewing the 2-D displays. Contrary to prediction, there was no significant difference between treatment groups in term of accuracy in generating hypotheses for what caused the emerged pattern.

PAGE 289

274 The implication from the lack of support for hypothesis H2c is that participants viewing a 3-D perspective display do not command an advantage or competitive edge over other participants viewing a different type of display format when performing a pattern recognition task involving the generation of hypotheses for what caused the emerged pattern. 6.3.4 Discussion of Results of Hypothesis H2d Hypothesis H2d stipulated that participants viewing the 3-D perspective display would be the most efficient or use less time in generating hypotheses for what caused the emerged patterns when compared to those participants viewing the tabular display or those participants viewing the 2-D displays. Hypothesis H2d has mixed results. For two of the dependent variables (TSPRQ1 and TSPRQ5) those participants viewing the tabular display and those participants viewing the 2-D displays used significantly less time than those participants viewing the 3-D perspective display. However, when describing the pattern of financial ratios in group one when compared to group two, participants viewing the 3-D perspective display were significantly more efficient than those participants viewing the tabular display or those viewing the 2-D displays. With such mixed results it is difficult to form any conclusions concerning the use of displays to increase the effectiveness of individuals in generating hypotheses for what caused the emerged patterns. As a conclusion, the results of the study provide some support for Cognitive Fit Theory. The results support the need for fit between task and display since it was found that a single method of display was not suitable for all types of tasks.

PAGE 290

275 6.4 Comments on the Implication of Significant Covariates Table 43 shows that the covariates that are significantly associated with the dependent variables of H1a-d and H2a-d include: scores of practice question one (PQ1), scores of practice question three (PQ3), scores of practice question four (PQ4), time spent by each participant when answering practice question four (TSPQ4), time spent by each participant when answering practice question five (TSPQ5), time spent by each participant when answering practice question six (TSPQ6), age of participants (AGE), accounting related working experience (ARWE), highest education (HE), student status (SS), scores of SAT (SAT), mental workload (MW), scores on mental rotations test (SMRT), and the time spent by each participant when answering Mental Rotations Test (TSMRT). The significant covariates can be grouped into two categories: 1) scores of practice questions and time spent in answering the practice questions (PQ1, PQ3, PQ4 TSPQ4, TSPQ5, and TSPQ6), and 2) individual characteristics (AGE, SS, HE, ARWE, SAT, MW, SMRT, and TSMRT). In the current study, the lack of theory and empirical research prevented hypothesizing relationships between the covariates and visualization; however, the significance of the covariates in the various models indicates that these covariates should be considered in future research on visualization.

PAGE 291

276 6.5 Discussion on the Responses of the Survey Questions In this study, participants responded to a set of survey questions regarding their reactions to the display format they were presented both for the trend analysis task and the pattern recognition task (Table 40 and 41 respectively). Despite the differences in the types of tasks, it is interesting to note that the participants of each treatment group responded to the survey questions in a similar fashion. It is also interesting to note that participants viewing the 3-D perspective display in both tasks did not respond negatively towards the survey questions. Tables 47 and 48 show that those participants viewing the 3-D perspective display in both tasks responded to four out of the six survey questions by selecting the scale number four (neither agree nor disagree). Both the participants viewing the 3-D perspective display in the trend analysis task and the pattern recognition task slightly disagreed that the graphs fit the way the participants needed to view the task information to make a better decision. It is understandable that viewers of the 3-D perspective display prefer to use tabular display or 2-D displays, which are more familiar to them. Unlike their counterparts in the trend analysis task who neither agreed nor disagreed that using the graphs was frustrating, participants of the 3-D perspective display in the pattern recognition task actually disagreed that using the graphs was frustrating. The results seem to show that participants viewing the 3-D perspective display did not overwhelmingly dislike this newly created presentation format. The lack of resistance to the 3-D perspective presented in the experiment indicates the potential for researchers to develop three-dimensional presentations of accounting information that are acceptable to users and can be useful to decision making.

PAGE 292

277 6.6 Contribution The study contributes to the literature by extending the Cognitive Fit Theory in the context of a single task type, namely the spatial task. Prior studies focused on the study of the effects of tabular or two-dimensional graphical displays upon the symbolic or spatial tasks. The study extends the literature by using two and three-dimensional displays to demonstrate how Cognitive Fit Theory applies in different types of spatial tasks. Findings of hypothesis H1a suggest that 2-D displays are suitable for the spatial task of generating hypotheses for what caused the changes in the trend of accounting data, while findings of hypothesis H1c and hypothesis H2a suggest that 3-D perspective display is suitable for other types of spatial tasks such as the prediction of values and recognition of patterns in accounting data, respectively. Findings of the study demonstrate that use of Cognitive Fit Theory is not limited to examining the relationships between tabular and two-dimensional formats and spatial tasks. The study contributes to the literature by demonstrating that even without the line of slope connecting data points, multidimensional visual display of complex multidimensional data such as the DuPont analysis can result in greater estimation accuracy of ROE. Prior literature suggests that viewers of 2-D line graphs benefit from richness of message flags provided by the slope of the line (Pinker 1990), which explicitly shows the smooth continuation of elements in a 2-D line display (Pinker 1990). However, findings of hypothesis H1c suggest that those participants viewing a static three-dimensional perspective display of DuPont analysis (without a line of slope connecting data points) can be more effective or accurate than those participants viewing the 2-D displays (with the line of slope explicitly shown) while performing a trend

PAGE 293

278 analysis task that involves the estimation of values of ROE. However, it should be noted that results of hypothesis H1a show that participants viewing the 2-D displays do benefit from the slope of the line (which shows continuation of elements) when performing the spatial task of generating hypotheses for what caused the changes in the trend of accounting data. Results of the study demonstrate the importance of the fit between presentation format and task, consistent with Cognitive Fit Theory. Whenever there is a fit between the presentation format and the task, performance will be enhanced regardless of whether the task-related information is explicitly or implicitly shown in the display. The findings of hypothesis H1a suggest that 2-D displays are suitable for the spatial task of generating hypotheses for what caused the changes in the trend of accounting data, because the slope of line in the 2-D plane explicitly show ed the changes in the accounting data through changes of the slope. Additionally, the findings of hypothesis H2a suggest that the 3-D perspective display is suitable for the spatial tasks of recognition of patterns in accounting data, because the emergent pattern in the 3-D plane explicitly shows the patterns of accounting data through the relative positioning of the data points in the 3-D plane. The most interesting finding of the study is the fact that a presentation format can be a better fit for a task even if the information is not explicitly shown in the presentation format. As the findings of hypothesis H1c indicate, while performing a trend analysis task that involves the estimation of values of ROE, participants viewing the 3D perspective display can be more effective or accurate than those participants viewing the 2-D displays, despite the fact that only the 2-D display has the line of slope explicitly

PAGE 294

279 showing the changes in the accounting data. According to Woods (1990), the spatial task of estimation of values of ROE is a complex task as decision makers need to calculate the values of turnover, profitability, and leverage first before they can estimate the values of ROE. A possible explanation of why participants viewing the 3-D perspective can be more effective in a complex spatial task involving the estimation of values of ROE is the perceptual proximity of having turnover, profitability, leverage and ROE in one single display which facilitates cognitive integrative processing as suggested by Carswell (1992). Tegarden (1999) indicates the purpose of visualization is not to replace quantitative analysis but to allow decision makers to identify trends, patterns anomalies and relationships in data. AICPA Professional Standards (2007) requires the use of analytical procedures in the planning and overall review stages of all audits. (AU section 329). The results of the study clear demonstrate that student participants can benefit from visualization techniques such as the 2-D displays and the 3-D perspective display while performing various spatial tasks involving DuPont analysis. DuPont analysis is a commonly used analytical procedure in the audit and financial world. If student subjects can benefit from the use of visualization techniques, there is no reason to expect auditors would not also benefit from the use of a proper display format to reason with the combination of all cues, properly generate an accurate hypothesis to explain the causal agent underlying the pattern and relationships among pieces of financial information. The results of this study thus have practical implications for external auditors who must perform analytical review procedures at the beginning of the audit and also during the audit. Specifically, for such analytical review procedures, the results of this study

PAGE 295

280 suggest that visualization techniques such as 2-D line graphs can assist auditors in generating hypotheses for what caused the changes in the trend of accounting data, while the 3-D perspective display can help auditors in identifying patterns within accounting data and in estimating new audit-relevant values The 3-D perspective display of the DuPont Analysis, newly developed in this study, is the first of its kind in accounting literature. The concept of representing a company by a point in a three dimensional space, if each company was defined by only three accounting ratios, was first discussed by Altman et al. (1974). The study is the first research that technically develops a 3-D perspective display to represent a company by a point in a three-dimensional space in terms of the DuPont analysis. This study contributes to the accounting literature by applying findings from human factors research in an accounting context. The study applied the Proximity Compatibility Principle (PCP) in an accounting context and demonstrates how a 3D perspective display can integrate information from the X, Y, Z axes to help decision makers invoke configural information processing. Findings of hypothesis H2a demonstrate that the emergent features portrayed in a 3-D perspective display could actually benefit decision makers. Specifically, findings of hypothesis H2a demonstrate that those participants viewing the 3-D perspective display not only were more accurate but could also be more efficient (in some cases), compared to those decision makers viewing the tabular display or those decision makers viewing the 2-D displays, when performing the pattern recognition task. Participants in the Dull and Tegarden (1999) experiment improved their decision accuracy by using the 3-D displays of line graphs but did not use less time in forming

PAGE 296

281 their conclusions when using the higher dimensional display format. A minor contribution of the study is demonstrating that decision makers using higher dimensional displays of accounting data can improve accuracy and efficiency at the same time. Findings of hypothesis H2a and H2b suggest that participants viewing the 3D perspective display were not only being more effective or accurate, but also more efficient than those participants viewing the 2-D displays in selecting one of the six companies if the goal is to have high profitability, high turnover, but low leverage at the same time. Another minor contribution to the literature is the results of the survey questions on the usefulness and satisfaction of the users of the 3-D perspective display. Analysis of the responses to the survey questions show that users of the 3-D perspective display did not negatively or overwhelmingly dislike this newly created presentation format. Participants of the study were able to learn and use a display format that they had never seen before the 3-D perspective display of DuPont analysis.

PAGE 297

282 6.7 Limitations The study has several limitations. Student subjects were used as surrogates for real world auditors. As discussed in section 3.8, student subjects are suitable for the task of the study. However, using real world auditors as participants can increase the external validity of the findings of the study. As discussed in section 4.34, participants in both the 2-D and 3-D treatment conditions had to scroll up and down the screen when answering questions, since the display and the response area did not fit on one screen. Participants viewing the tabular display did not need to scroll up and down the screen when answering questions. The study acknowledges the need to scroll up and down the screens as a limitation. As discussed in section 5.61 the manipulation check did not operationalize as intended. There was a high error rate in participan question. Such a high error rate was due to the fact that participants had to rely on their memory when answering the manipulation check questions about display. Participants did not look at the displays while answering the manipulation check. The study acknowledges this as a limitation. The study employed a single set of data to test the effects of the manipulation of the presentation format in the trend analysis or pattern recognition task. It is beneficial to have several sets of data of similar complexity to test the effects of the manipulation of the presentation format. The study acknowledges the lack of multiple sets of data to test for the effects of the manipulation of the presentation format.

PAGE 298

283 6.8 Future Research There are ample opportunities for future research in visualization of accounting information. The study employed static 3-D perspective display. Future research can investigate the effects of animation and rotation on those decision makers viewing the 3D perspective display. Future research should employ real world auditors and study the effect of 3D perspective display upon novice auditors. An interesting question for future research is whether 3-D perspective display can improve the decision quality of the novice auditors when compared to experienced auditors. Future research can develop other types of 3-D display of accounting information. Future research should also investigate the question of what type of three-dimensional display is suitable for which type of accounting information. During the process of developing new three-dimensional displays of multidimensional accounting data, it is crucial to test the newly developed display format with several sets of data of similar nature. Further, future research can investigate the effect of split attention, since the participants viewing the 2-D displays or 3-D perspective display had to scroll up and down the screen while performing the assigned task, by having a control group (of a display format) without the need to scroll up and down the screen. Future research should employ verbal protocol to record the mental status of participants viewing different types of display formats. It is important to understand how the participants learn, teach themselves and improve in their performance while viewing their randomly assigned display format.

PAGE 299

284 References Altman, E. I., M. Margaine, M. Schlosser, and P. Vernimmen. 1974. Financial and statistical analysis for commercial loan evaluation: A French experience. The Journal of Financial and Quantitative Analysis 9 (2):195-211. American Institute of Certified Public Accountants. Professional Standards (2007, June 1). New York: AICPA. ________. 1978. Conclusions, and Recommendations New York: AICPA. ________. 1987. The Treadway Commission on Fraudulent Financial Reporting New York: AICPA. Ashton, R. H. 1974 An experimental study of internal control judgments Journal of Accounting Research (Spring): 143-157. Bedard, J. C., and S. F.Biggs. 1991. Pattern recognition, hypotheses generation, and auditor performance in an analytical task. The Accounting Review 66 (3): 662642. Bennett, K. B., and J. M. Flach. 1992. Graphical displays: Implications for divided attention, focused attention, and problem solving. Human Factors 34: 513-533. Benford, T.L. 2000. Determinants of Audit Performance: An Investigation of Task/Technology Fit and Mental Workload (Doctoral Dissertation, University of South Florida, 2000). Bertin, J. 1981. Graphics and Graphic Information Processing. Berlin : Walter de Gruyter & Co Bierstaker, J. L., J. C. Bedard, and S. F. Biggs. 1999. The role of problem representation shifts in auditor decision processes in analytical procedures. A uditing: A Journal of Practice and Theory 18 (1): 1836. Bogacz, S., and J. G. Trafton. 2005. Understanding dynamic and static displays: Using images to reason dynamically. Cognitive Systems Research 6 (4): 312319.

PAGE 300

285 Bray, J. H., and S. E. Maxwell. 1985. Mu ltivariate analysis of variance Sage University paper series on quantitative applications in the social science, 07-054. Newbury Park CA: Sage University. Brown, C. E., and I. Solomon. 1990. Auditor configural information processing in control risk assessment. Auditing: A Journal of Practice & Theory 9 (3): 17-37. ________. 1991. Configural information processing in auditing: The role of domainspecific knowledge. The Accounting Review 66 (1): 100-129. Chandler, P., and J. Sweller. 1992. The split attention effect as a factor in the design of instruction. British Journal of Education Psychology 62: 233-246. Carswell, C .M. 1992. Reading graphs: Interactions of processing requirements and stimulus structure. In B. Burns (Ed.), Percepts, Concepts and Categorie s (pp.605647). Amsterdam: North-Holland. Chernoff, H. 1971. The uses of faces to represent points in N-dimensional space graphical technical report No. 71, Department of Statistics, Stanford University. Chung, J., and G. S. Monroe. 1998. Gender differences in information processing: An empirical test of the hypothesis-confirming strategy in an audit context. Accounting and Finance 38: 265-279. information processing. International Journal of Auditing 2: 21-33. Cohen, J. 1992. A power primer. Psychological Bulletin 112 (1): 155-159. Cook, R. D., and S. Weisberg. 1982. Residuals and influence in regression New York: Chapman & Hall. Dickson, G. W., G. DeSanctis, and D. J. McBride. 1986. Understanding the effectiveness of computer graphics for decision support: A cumulative experimental approach. Communications of the ACM 29 (1): 40-47. Dull, R. B., and D. P. Tegarden. 1999. A comparison of three visual representations of complex multidimensional accounting information. Journal of Information Systems 13 (2): 117-131. Faul, F., E. Erdfelder, A. G. Lang, and A. Buchner. 2007. G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical science. Behavior Research Methods 39: 175-191. Field, A. 2005. Discovering Statistics Using SPSS London: Sage Publications.

PAGE 301

286 http://www.goldensoftware.com Haskell, I. D., and C. D. Wickens. 1993. Two and three dimensional displays for aviation: A theoretical and empirical comparison. International Journal of Aviation Psychology 3: 87-109. Hygge, S., and I. Knez. 2001. Effects of noise, heat and indoor lighting on cognitive performance and self-reported affect. Journal of Environmental Psychology 21: 291-299. Jarvenpaa, S. L., and G. W. Dickson. 1988. Graphics and managerial decision making: Research based guidelines. Communications of the ACM 31 (June): 764-774. Kinney, W. 1987. Attention-directing analytical review using accounting ratios: A case study. Auditing: A Journal of Practice & Theory 6 (Spring): 59-73. Kolata, G. 1982.Computer graphics comes to statistics. Science 919-920. Kosslyn, S. M. 1994. Elements of Graph Design New York: W.H. Freeman. Kumar N., and I. Benbasat. 2004. The effect of relationship encoding, task type, and complexity on information representation: An empirical evaluation of 2-D and 3D line graphs. MIS Quarterly (June): 255-281. Larkin, J. H., and A. H. Simon. 1987. Why a diagram is (sometimes) worth ten thousand words. Cognitive Science 11: 65-99. Liu, Y., and C. D. Wickens. 1992. Visual scanning with or without spatial uncertainty and divided and selective attention. Aeta Psychologica 79: 131-153. Lohse, J. 1991. A cognitive model for the perception and understanding of graphs. Communications of the ACM 383: 137-144. Masters, M. S., and B. Sanders. 1993. Is the gender differences in mental rotation disappearing? Behavior Genetics 23 (4): 337-341. Miller, G. A. 1956. The magical number seven, plus or minus two: some limits on our capacity for processing information. The Psychological Review 63 (2): 81-97. Moriarity. S. 1979. Communicating financial information through multi-dimensional graphics. Journal of Accounting Research 17 (Spring): 205-224. Murthy, U., B. Schafer, and R. Fuller. 2007. The impact of model representation on error detection and functional use. Working paper, The University of South Florida.

PAGE 302

287 Nissim, D., and S. H. Penman. 2001. Ratio analysis and equity valuation: From research to practice. Review of Accounting Studies 6: 109-154. Nunnally, J. C. 1978. Psychometric Theory New York: McGraw Hill. Pinker, S. 1990. A theory of graph comprehension. In R. Freedle (Ed.), Artificial intelligence and the future of testing (pp. 73-126). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Reid. G. B., and T. E. Nygren. 1988. The subjective workload assessment technique: A scaling procedure for measuring mental workload. In P. A. Hancock and N. Meshkati (Eds.), Human Mental Workload (pp. 185-218) North Holland: Elsevier Science Publishing Company, Inc. Sen, A., and W. A. Wallace. 1990. An expert systems assistance to internal audit department evaluations. Expert Systems with Applications: An International Journal 3 (1): 51-66. Shepard, R., and J. Metzler. 1971. Mental rotation of three dimensional objects. Science 171: 701-703. Simon, H. A. 1978. On the forms of mental representation. In Savage, C. W. (Ed.), Minnesota Studies in the Philosophy of Science. Vol. IX : Perception and Cognition: Issues in the Foundations of Psychology Minneapolis; University of Minnesota Press. Sloman, S. A. 1996. The empirical case for two systems of reasoning. Psychological Bulletin 119 (1): 3-22. Smith. C. P. 2005. Content analysis and narrative analysis. In T. R. Harry, and C. M. Judd (Eds.), Handbook of Research Method in Social and Personality Psychology (325). New York: Cambridge University Press. Smith, C. P., S. C. Field, and C. E. Franze. 1992. Methodological considerations: Steps in research employing content analysis systems. In C. P. Smith (Ed.). Motivation and Personalit y: Handbook of Thematic Content Analysis (pp 515-516). New York: Cambridge University Press. Speier, C. 2006. The influence of information presentation formats on complex task decision-making performance. International Journal of Human-Computer Studies 64: 1115-1131. Tan. J. K. H. and I. Benbasat. 1993. The effectiveness of graphical presentation for information extraction: A cumulative experimental approach. Decision Science 24 (1): 167-191.

PAGE 303

288 Tegarden D. P. 1999. Business information visualization. Communications of the Association for Information Systems 1 (4): 1-38. Totty, M. 2007, September 24. Making sense of it all: New software aims to put corp understood. Wall Street Journal (Eastern edition) pp. R8. Trafton, J. G., S. B. Trickett, and F. E. Mintz. 2005. Connecting internal and external representation: Spatial transformations of scientific visualizations. Foundations of Science 10: 89-106. Tricket, S. B., and J. G. Trafton. 2006. Toward a comprehensive model of graph comprehension: Making the case for spatial cognition. Lecture Notes in Computer Science 4045: 286-300. Tu fte, E.R. 1983. The Visual Display of Quantitative Information. Graphics Press: Cheshire, Connecticut. Vandenberg, S.G., and A. R. Kuse. 1978. Mental rotations, a group test of threedimensional spatial visualization. Perceptual and Motor Skills 47: 599-604. Vessey, I. 1991. Cognitive fit: A theory-based analysis of the graphs versus tables literature. Decision Science 22 (Spring): 219-241. ________. 1994. The effect of information presentation on decision making: A costbenefit analysis. Information & Management 27: 103-119. Vessey, I., and D. Galletta. 1991. Cognitive fit: An empirical study of information acquisition. Information Systems Research 2 (1): 63-82. Washburne, J. N. 1927. An experimental study of various graphic, tabular and textural methods of presenting quantitative material Journal of Educational Psychology 18 (6): 361-376. White, G. I., A. C. Sondhi, and D. Fried. 1998. The Analysis and Use of Financial Statements. (2nd ed.). Portland, Oregon: Wiley. Wickens, C. D., and C. M. Carswell. 1995. The proximity compatibility principle: Its psychological foundation and relevance to display design. Human Factors 37 (3): 473-494. Wickens, C. D., D. H. Merwin, and E. L. Lin. 1994. Implications of graphics enhancements for the visualization of scientific data: Dimensional integrality, stereopsis, motion, and mesh. Human Factors 36 (1): 44-61.

PAGE 304

289 Human Decision Processes, 1986, Vol. 37, 60-82. Wright, W. F. 1995. Superior audit judgment given graphical displays. Auditing: A Journal of Practice & Theory 14 (Fall): 144-154. Wright, W. F., and J. J. Willingham. 1997. A computational model of loan loss judgments. Auditing: A Journal of Practice & Theory 16 (Spring): 99-113. Wright, W. 1997. Business visualization applications. IEEE Computer Graphics and Applications 17 (4) : 66-70. Zhang P. (Ed.). 1996. Proceeding of the Second American Conference on Information Systems (AIS 1996): The Impact of Information Visualization on Human Problem Solving Performance in a Complex Business Domain

PAGE 305

About the Author I am from Hawaii with a free spirit. Simplicity of life is what I am pursuing. I had disciplinary studies on auditing and accounting information system. Visualization is one of my key future research areas. I have several professional accounting qualifications. My working experience both as an external and internal auditor helps me in teaching the auditing and accounting information system classes. I am confident that I can help my students in passing the examination of the American Institute of Certified Public Accountants on their first attempt. I enjoy surfing, as I am from Hawaii. Aloha.