Using Overview Style Tables on Small Devices
by
Rui Zhang
Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy
at
Dalhousie University Halifax, Nova Scotia November 2008
© Copyright by Rui Zhang, 2008 Library and Bibliotheque et 1*1 Archives Canada Archives Canada Published Heritage Direction du Branch Patrimoine de I'edition
395 Wellington Street 395, rue Wellington Ottawa ON K1A0N4 Ottawa ON K1A0N4 Canada Canada
Your file Votre reference ISBN: 978-0-494-50083-5 Our file Notre reference ISBN: 978-0-494-50083-5
NOTICE: AVIS: The author has granted a non L'auteur a accorde une licence non exclusive exclusive license allowing Library permettant a la Bibliotheque et Archives and Archives Canada to reproduce, Canada de reproduire, publier, archiver, publish, archive, preserve, conserve, sauvegarder, conserver, transmettre au public communicate to the public by par telecommunication ou par Plntemet, prefer, telecommunication or on the Internet, distribuer et vendre des theses partout dans loan, distribute and sell theses le monde, a des fins commerciales ou autres, worldwide, for commercial or non sur support microforme, papier, electronique commercial purposes, in microform, et/ou autres formats. paper, electronic and/or any other formats.
The author retains copyright L'auteur conserve la propriete du droit d'auteur ownership and moral rights in et des droits moraux qui protege cette these. this thesis. Neither the thesis Ni la these ni des extraits substantiels de nor substantial extracts from it celle-ci ne doivent etre imprimes ou autrement may be printed or otherwise reproduits sans son autorisation. reproduced without the author's permission.
In compliance with the Canadian Conformement a la loi canadienne Privacy Act some supporting sur la protection de la vie privee, forms may have been removed quelques formulaires secondaires from this thesis. ont ete enleves de cette these.
While these forms may be included Bien que ces formulaires in the document page count, aient inclus dans la pagination, their removal does not represent il n'y aura aucun contenu manquant. any loss of content from the thesis. Canada DALHOUSIE UNIVERSITY
To comply with the Canadian Privacy Act the National Library of Canada has requested that the following pages be removed from this copy of the thesis:
Preliminary Pages Examiners Signature Page (pii) Dalhousie Library Copyright Agreement (piii)
Appendices Copyright Releases (if applicable) DEDICATION PAGE
To my parents,
for preparing me for this journey
To my husband,
for holding my hand during it
And to my son,
for supporting me in such a sweet way
IV TABLE OF CONTENTS
LIST OF TABLES ix
LIST OF FIGURES xiii
ABSTRACT xvi
LIST OF ABBREVIATIONS USED xvii
ACKNOWLEDGEMENTS . ...xviii
CHAPTER 1 Introduction 1
1.1 Research Summary 2
1.2 Organizational Overview 5
CHAPTER 2 Background 6
2.1 Table Structure 6
2.2 Table Use 10
2.3 Tables for Use on Multiple Devices 14
2.3.1 Transformation 14
2.4 Display Problems on Small Screen Devices 17
2.5 Table Views 19
2.6 Summary : , 23
v CHAPTER 3 Research Questions 24
3.1 Research Questions 24
3.2 Tasks and Current Features Analysis 27
3.3 New Feature Analysis 30
3.3.1 Prototype..... 30
3.3.2 Table Overview (Zoom-in/Zoom-out) 34
3.3.3 Subtables 35
3.3.4 Cell Expansion 38
3.3.5 Landmark 43
3.3.6 Other Functions 44
3.4 Summary 44
CHAPTER 4 Sequence of Studies 46
4.1 Table Views Studies: Study 1.1 and Study 1.2 46
4.1.1 Study 1.1 - Table Views Study 49
4.1.2 Study 1.2 - Table Views Study with Task Factor 52
4.1.3 Discussion of Both Table View Studies 57
4.1.4 Conclusions for Table View Studies 1.1 and 1.2 59
4.2 Study 2: Browsing Method Study - Column Expansion vs. Cascade 60
4.2.1 Study 2 Experimental Design 62 vi 4.2.2 Study 2 Results 64
4.2.3 Discussion for Study 2 67
4.2.4 Conclusions of Study 2 70
4.3 Study 3: Expansion Method Study - Column Expansion Comparison 70
4.3.1 Column Expansion Comparision Experimental Design 72
4.3.2 Column Expansion Study Results 78
4.3.3 Discussion for Column Expansion Comparison 83
4.3.4 Conclusions of Column Expansion Comparison Study.... 85
4.4 Study 4: Landmarks 85
4.4.1 Landmarks Experimental Design 86
4.4.2 Landmarks Study Results 90
4.4.3 Discussion for Landmark Study 93
4.4.4 Conclusions of Landmark Study 97
4.5 Study 5: Search Study 97
4.5.1 Search Study Experimental Design 100
4.5.2 Search Study Results 106
4.5.3 Discussion for Search 112
4.5.4 Conclusions of Search Study 115
vii CHAPTER 5 Conclusions 116
5.1 Contributions 120
5.1.1 Theoretical Contributions 120
5.1.2 Applied Contributions 121
5.2 Study Limitations 122
5.3 Future Work 124
5.3.1 Over All Examination using Our Design 124
5.3.2 Extending Study for Various Table Styles and Other Data Types 124
5.3.3 Supporting Efficient Web Use 125
5.3.4 Providing a Framework for Equivalent Task Complexity Design 125
BIBLIOGRAPHY 126
APPENDIX A. TABLE VIEWS STUDY 1.1 QUESTIONNAIRE 132
APPENDIX B. TABLE VIEWS STUDY 1.2 QUESTIONNAIRE - WITH TASK FACTOR 135
APPENDIX C. STUDY 2: BROWSING METHOD STUDY - COLUMN EXPANSION VS. CASCADE QUESTIONNAIRE 139
APPENDIX D. STUDY 3: EXPANSION METHOD STUDY - COLUMN EXPANSION COMPARISON QUESTIONNAIRE 141
APPENDIX E. STUDY 4: LANDMARKS STUDY QUESTIONNAIRE 145
APPENDIX F. STUDY 5: SEARCH STUDY QUESTIONNAIRE 148
viii LIST OF TABLES
Table 2.3.1 Bickmore and Schilit transcoding matrix 15
Table 2.3.2 Extended Bickmore and Schilit transcoding matrix 16
Table 3.3.1 Relationship of tasks and features 30
Table 4.1.1 Task types and sample queries 49
Table 4.1.2 Mixed factorial table 50
Table 4.1.3 Accuracy comparison between three techniques 50
Table 4.1.4 Efficiency comparison between three techniques 50
Table 4.1.5 Univariate linear results - efficiency (study 1.1) 51
Table 4.1.6 Mixed factorial table 53
Table 4.1.7 Accuracy comparison between three techniques 53
Table 4.1.8 Efficiency comparison for efficiency between three applications 54
Table 4.1.9 Univariate linear results - efficiency (study 1.2) 54
Table 4.1.10 Hypotheses summary 58
Table 4.1.11 Summarization of significant results 59
Table 4.2.1 Task types and sample queries 63
Table 4.2.2 Mixed factorial table 64
Table 4.2.3 Accuracy 64
ix Table 4.2.4 Comparison for efficiency between two methods 65
Table 4.2.5 Univariate linear results - efficiency (study 2) 65
Table 4.2.6 Efficiency between two methods in different task complexity 66
Table 4.2.7 Valuable comments examples 66
Table 4.2.8 Hypotheses summary 67
Table 4.2.9 Summarization of significant results 69
Table 4.3.1 Task complexity and examples 77
Table 4.3.2 Mixed factorial table 78
Table 4.3.3 Accuracy 79
Table 4.3.4 Univariate linear results - accuracy (study 3) 79
Table 4.3.5 Overall efficiency 80
Table 4.3.6 Univariate linear results - efficiency (study 3) 80
Table 4.3.7 Comparison for short tables 81
Table 4.3.8 Comparison for efficiency for long tables 82
Table 4.3.9 Hypotheses summary 84
Table 4.3.10 Summarization of significant results 85
Table 4.4.1 Task complexity and examples 89
Table 4.4.2 Mixed factorial table 90
Table 4.4.3 Accuracy based on data pattern 90
x Table 4.4.4 Accuracy based on task complexity 91
Table 4.4.5 Efficiency based on data pattern 92
Table 4.4.6 Efficiency based on task complexity 92
Table 4.4.7 Hypotheses summary 94
Table 4.4.8 Matrix of accuracy 94
Table 4.4.9 Summarization of significant results 97
Table 4.5.1 Sample table A 102
Table 4.5.2 Sample table B 102
Table 4.5.3 Sample table 103
Table 4.5.4 Task type 104
Table 4.5.5 Sample questions 104
Table 4.5.6 Mixed factorial table 106
Table 4.5.7 Effectiveness based on methods 106
Table 4.5.8 Univariate linear results - accuracy (study 5) 107
Table 4.5.9 Effectiveness based on table complexity 108
Table 4.5.10 Univariate linear results - efficiency (study 5) 109
Table 4.5.11 Efficiency 110
Table 4.5.12 Method order 112
Table 4.5.13 Hypotheses summary 113 xi Table 4.5.14 Summarization of significant results 115
Table 5.1 Conclusion summary for five studies 116
XII LIST OF FIGURES
Figure 2.1.1 Table sample 8
Figure 2.1.2 Calendar sample 9
Figure 2.1.3 Time table sample 9
Figure 2.1.4 Football league table sample 10
Figure 2.1.5 Content table sample 10
Figure 2.2.1 Table reading and navigation modes 11
Figure 2.3.1 The sample web table displayed on mScope [43] 17
Figure 2.5.1 The sample page displayed with IE on PDA 20
Figure 2.5.2 Sample table linearized 21
Figure 2.5.3 The sample web table displaying on Palmscape 22
Figure 2.5.4 Sample using Excel on PDA 22
Figure 3.3.1 University enrollment sample table 31
Figure 3.3.2 Numeric sample table with Overview model on simulator prototype 32
Figure 3.3.3 Full table view of hotel check-in sample 33
Figure 3.3.4 Compressed view of hotel check-in table sample 34
Figure 3.3.5 Table Overview sample with Zoom-in/Zoom-out feature 35
Figure 3.3.6 An example of table block selection & generation 36
xiii Figure 3.3.7 Non sequential table multiple rows selection & generation 36
Figure 3.3.8 Non sequential table multiple columns selection & generation 37
Figure 3.3.9 An example of table special cells selection & generation 38
Figure 3.3.10 Tooltips sample 39
Figure 3.3.11 Cascade sample 40
Figure 3.3.12 Column, row, column/row expansion sample 41
Figure 3.3.13 Single column expansion vs. multiple columns expansion 42
Figure 3.3.14 Landmark sample 43
Figure 3.3.15 Menu and context menu sample 44
Figure 4.1.1 University enrollment sample table 47
Figure 4.1.2 Efficiency graph for user study 1.1 51
Figure 4.1.3 Graph of efficiency for user study 1.2 55
Figure 4.1.4 Complex task sample using LV and OV 56
Figure 4.2.1 Sample table with row Cascade 61
Figure 4.2.2 Sample table with column Cascade 61
Figure 4.2.3 Cascade cell outside the screen 61
Figure 4.2.4 Sample table with 2 columns and 1 row expanded 61
Figure 4.2.5 Landmarking feature 68
Figure 4.2.6 Single click automatic column expansion 69
xiv Figure 4.3.1 Single column expansion 71
Figure 4.3.2 Multiple column expansion 71
Figure 4.3.3 Single column 72
Figure 4.3.4 Multiple columns 72
Figure 4.3.5 Numeric table sample - University enrollment 73
Figure 4.3.6 Mixed data table sample - Hotel check-in (a) Original version 74
Figure 4.3.6 Mixed data table sample - Hotel check-in (b) Short version 75
Figure 4.4.1 Data patterns 87
Figure 4.4.2 Accuracy 95
Figure 4.5.1 Feature menu 100
Figure 4.5.2 Data patterns 101
Figure 4.5.3 Effectiveness 108
Figure 4.5.4 Task complexity vs table complexity on efficiency 111
Figure 4.5.5 Efficiency 111
XV ABSTRACT
Users increasingly expect access to Web data from a wide range of devices, both wired and wireless. The overall goal of this research is to address the design of applications that support data access by keeping the consistency and providing reasonably seamless migration of tables among internet-compatible devices with minimal loss of effectiveness and efficiency.
The research concentrates on Overview representations and manipulation of large tables of data on small mobile devices, where the user selectively expands areas as needed to access individual cells. The research purpose is to display table information in a manner that supports readability of the data, accuracy of access, and efficiency of information use on the mobile device (focusing on the PDA size screen). In this way the user can concentrate on the data rather than the device. Plus, there will be less ambiguity while migrating between devices.
The research presented in this thesis is the result of five successive studies conducted to examine the accuracy, efficiency, and preference using table on small device in the context of task. An exploratory field study was first conducted to examine which is the most robust table view model during information seeking tasks on small device, as the result shows that Overview is the model. The study also found significant effect by task complexity. As a next step, a series of user studies was conducted to concentrate on exploring the characteristics of tasks using tables on the small screen device, as well as how the features affect the efficiency, effectiveness and preference of the users across tasks. The results of these analyses can be used to develop new functionalities for large table display and manipulation on small mobile devices in a manner that supports accuracy of access, and efficiency on the mobile device over a range of tasks of varying complexity.
XVI LIST OF ABBREVIATIONS USED
OV: Overview DL: Default View LV: Linear View HTDT: Headings + Text/Digit Data Table TDT: Text/Digit Data Table CWA: Column Width Auto-adjustment CWE: Column Width Expansion TLM: Tooltips with Landmarks WLM: Tooltips without Landmarks UVF: Upper Visual Field LVF: Lower Visual Field IE: Internet Explorer HCI: Human-Computer Interaction HTML: HyperText Markup Language HTTP: HyperText Transfer Protocol PDA: Personal Digital Assistants PC: Personal Computer RSVP: Rapid Serial Visual Presentation SD: Standard Deviation URL: Uniform Resource Locator WAP: Wireless Application Protocol WML: Wireless Markup Language 2-D: Two Dimensional
XVII ACKNOWLEDGEMENTS
I would first like to acknowledge each person who provides support, help, and encouragement to me in many ways. Without them this thesis could not have been completed.
My supervisor, Dr. Carolyn Walters, has been a tremendous mentor throughout both my Master and Ph.D. study. It was she who encouraged me to pursuit my Ph.D. after I finished my Master degree under her instruction. Dr. Watters provided me with important advices and very helpful suggestions, as well as inspired and challenged me during our innumerable discussions and meetings. She taught me to keep asking myself "WHY?" during the research and how to step back and look at the big picture. Her constant support, encouragement and understanding have been always with me intellectually, emotionally, and financially.
I would like to express my sincere thanks to Dr. Jack Duffy and Dr. Michael Shepherd, who were always there to give me insightful comments, essential guidance and very helpful advices. Dr. Duffy taught me how to think about problems in statistic way and gave me very important suggestions of analysis methods. Dr. Shepherd's insightful comments, feed back on my work, and unconditional encouragement were always my assets. Finally, I would also like to thank my external examiner for his valuable comments and feedback on my work and time taken from his busy schedule to attend my defense.
I cannot imagine how different and difficult my Ph.D. experience would have been without the support and friendship I received from my classmates, friends, members of the WIFL research lab, and the staff of Faculty of Computer Science, Dalhousie University. Thank you for being there and generally putting up with me.
Special thanks are given to my son, Christopher, who was born during my Ph.D. study. I would never forget the picture that he was sitting on the meeting table playing by himself with his little bear, while I was meeting with Dr. Watters. His smile, love, and happiness are always my strongest backbone and driving force to my study.
My husband, Lei Dong, was an endless source of support during my Ph.D. Thank you for the encouragement, patience, tolerance, and everything...
Finally, I would like to thank all of the participants who took part in my studies.
XVIII CHAPTER 1 INTRODUCTION
The growth in interest in mobile devices is substantial [11]. As wireless technology provides a more substantial backbone for mobile mail and Internet access, users expect an integration of mobile and fixed access for applications. Many web applications, in business, personal finances, medicine, sales, and education, are ideal scenarios for the integrated use of different devices with different screen sizes.
As a result, it is not necessary for users to be bound to their office or desk, and the requirements for the portable devices, such as Personal Digital Assistants (PDAs) and smart phones, are continuously increasing. Groups need the flexibility of collaborating with mixed devices and individuals need the flexibility of migrating between devices in the course of tasks. When users switch back and forth between devices, such as laptops and PDAs, they expect to be able to continue working with the same data. From the user's perspective, accessing data using multiple devices is not enough, there needs to be continuity of task focus and recognition that each device has particular characteristics. This means that access to data is supported in a consistent manner independent of the actual device and that the user can move (reasonably) seamlessly from one device to another without loss of context or content. Furthermore, this means that the user can access data on any device with necessary transformations executed dynamically for the target device. This transparency of device means that software is required to track the task parameters from device to device, be knowledgeable about the device characteristics, and interpret the requirements for data within the context of each device so that the user is free to migrate between devices.
Tables have been identified as one of the most difficult areas of web accessibility through current browsers [48], and this is even more challenging for small screen devices because of the limitation of the display space and the table's inherently two dimensional structure, and in particular, the spatial layout of a table is an essential component of the semantics encoded in it. The overall goal of this research is to examine features in the design of applications that keep the consistency of the table structure and support for access to data
1 in such tables reasonably seamless throughout the migration of tables, which were designed for use on personal computers (PC) or laptops, to other internet-compatible devices, such as PDA handheld devices, with minimal loss of effectiveness and efficiency. Our research focuses on PDA size devices. The small device's characteristics have an effect on both how information should be presented and how users interact with the device [31]. This study concentrates on the display and manipulation of large tables on PDA sized mobile devices. The objective is explore design features for use with table information on small screens that support readability of the data, accuracy of access, and efficiency on the mobile device (focusing on the PDA size screen) for information seeking tasks.
1.1 RESEARCH SUMMARY
A series of five studies, which are summarized below, were designed to meet this objective.
Study 1 - Table Views Study. A Pilot Study Comparing Three Views to Display a Table on Small Screen Device
As a first step, a pilot study was conducted to examine the choice of the model of table display for information seeking tasks on small device as well as how users interact with a table on a small screen device to complete these tasks. The study examined effectiveness, efficiency and preference of three models: Default View, Linear View, and Overview. The analysis showed that although both the Linear View and Overview techniques were better on the basis of efficiency for the user, as the task became more complex the effect of view became more pronounced. The Overview technique gave the least variation over the five tasks while the Default View and Linear View both exhibited irregular patterns related to the individual task. Consequently, the Overview model was chosen for use in the follow-up studies.
2 Study 2 - Browsing Method Study. A User Study Comparing Column Expansion and Cascade Methods
The findings from the pilot study indicated that Overview was the most suitable way to display a table on small screen devices, but that expanding columns created difficulties for users. Consequently a new technique, Cascade was developed to allow users to quickly scan rows or columns in the compressed table. As a next step, a study was conducted specifically to compare the difference between normal Column Expansion and the Cascade method. The results indicated that little difference was detected for simple tasks, but as the task became more complex, an advantage was found for the scanning potential of the Cascade method. Users may benefit from using both methods.
Study 3 - Expansion Method Study. A User Study Comparing Two Column Expansion Methods
The findings from the previous study indicated that it often was necessary to expand entire columns for table tasks on the small screen. Therefore, the next step was to examine two column expansion methods, a drag option and a single click automatic option. The results indicated that although the single click automatic column expansion method always performed at least as well as the manual drag method, users found benefit from both methods.
Study 4 - Landmark Study. A User Study Examining Landmarks
One of the difficulty users experienced using the Overview model was in refinding cells. A landmark feature was developed to assist users in refinding cells model was in the condensed table. This study was conducted to evaluate the effect of highlighted landmark method for information refinding in a table using the Overview model. The results of this study indicate that using landmarks significantly improved effectiveness as task complexity increased. Therefore, the landmark feature is strongly recommended for more complex tasks for information refinding and comparison tasks using small screen devices.
3 Study 5 - Search Study. A User Study Comparing Manually Next Method and Automatically Cascade Methods on Search Results Scanning
Users indicated a preference for a "Find" function, and a new method using the Cascade model was developed to allow fast scanning of matches in the compressed table. A user study was conducted to evaluate two scanning methods, the traditional Next and over new Cascade, to access the results. Text search is a widely used feature in commercial applications, and it is a very efficient way to narrow down the focus in large information sets like the table. This study did not show overall significant difference between two methods. The scanning potential of the Cascade method was shown to be somewhat more effective.
Summary of Studies
Overall, this series of studies indicates that the Overview model is suitable for table displays on PDA devices. First, Overview mode presents a consistent view of the entire table data set, and this concept has been widely applied in the interface design for small screens in recent years. Second, the Overview model was approved as the most robust model over various levels of tasks compared with more commonly used Linear View and Default View. Third, users preferred having options which reduced the use of scroll bar and other stylus based inputs. Our design presented a set of new features based on this interaction difficulty concern. According to our study result, the Cascade method is more suitable for general browsing or scanning tasks, therefore, the combination of using Cascade for initially browsing and column expansion on subsequent tasks is recommended; in order to minimize the interaction between users and devices, using the Column Width Auto-adjustment method to expand a column by a single click is strongly recommended; the landmark feature is recommended when task complexity and cognitive load increase.
4 1.2 ORGANIZATIONAL OVERVIEW
Chapter 2 provides background literature relevant to the research presented in this thesis. The chapter begins with a general table structure analysis. This is followed by an overview of how tables are used and an introduction of features used in commercial applications, as well as current products and research for table use on small screen devices. The chapter then introduces some display problems and table views using small screen devices, followed by a summary of the related research.
Chapter 3 describes the analysis of tasks and features current commercial products provide for table use on small screen devices, presents a summarization of the research questions which motivate the development of new features and introduces the user study methodologies used, tasks analysis, the data collection tools employed, as well as the new features we developed to meet our research goal.
Chapter 4 describes the series of user studies designed to test our hypotheses. For each study, a corresponding result analysis was conducted using data collected during the study. These analyses explore the characteristics of tasks using tables on the small screen device, as well as how the features affect the efficiency, effectiveness and preference of the users across tasks. The results of these analyses can be used to develop new functionalities for large table display and manipulation on small mobile devices in a manner that supports accuracy of access, and efficiency on the mobile device over a range of tasks of varying complexity.
Chapter 5 offers overall conclusion based on the series of studies, the limitation of this research, as well, a series of recommendations for the future design and evaluation of the data readability, access accuracy and efficiency improvement using tables on the mobile device. These analyses also raise the question of the impact of task complexity and data pattern on user performance. Finally, a summary of the research presented in this thesis, is followed by a description of the major research contributions. The chapter concludes with an overview of the planned future work.
5 CHAPTER 2 BACKGROUND
A table can be defined as a structure which consists of an ordered arrangement of rows and columns [66]. Tables, with their inherent two (or more) dimensional structures, have visual multidimensional layouts, and, in most cases, the visual layout is an essential component of the semantics of the table. In addition, in a variety of situations the visual layout is actually used to provide the semantics of a more complex structuring of the information within the table, e.g., the spatial layout can be used to describe a 3- dimensional structure or even a tree structure, by properly combining the table structure with the use of different colours, header elements, etc. As the result, it may be necessary to keep constant the layout of the table when using devices with the different screen sizes to maintain any visual semantics.
Suitable design features may be used to improve the readability of tables, accuracy of access, and efficiency of information look up. The consistency of table views may become important in situations where users shift or migrate from one device to another. This chapter will discuss table structures; models of tables for migration between devices; how users use tables; and task analysis for table use.
2.1 TABLE STRUCTURE
The first step of our research was to consider the structure of tables. Yesilada [68] organized tables into 11 classes based on their logical organization, as follows: Periodic table, Time table, Content table, Football league tables, Calendar, Statistical tables, Truth tables, Nested tables, Row tables, Column tables, and Row and Column tables.
1. Periodic table: A tabular arrangement of the elements according to their atomic numbers. 2. Time tables: A tabular statement of the time at which or within which, several things is to take place, for example; train, flight, lecture, bus, etc timetables.
6 3. Content tables: A tabular representation of subject matter of a written work, such as book or magazine. 4. Football league tables: Tabular representation of performances of football teams for a period. 5. Calendar: A table showing the months, weeks and days in at least one specific year. 6. Statistical tables: Tables that show statistical data. 7. Truth tables: A table that displays the truth-value of a compound sentence as a function of the varying truth-values of its components. 8. Nested tables: Tables that contain another table in at least one of its cells. 9. Row tables: General purpose tables that are visually organized as rows. 10. Column tables: General purpose tables that are visually organized as columns. 11. Rows and Columns tables: A table whose cell contents are logically accessed by intersection of row and column.
The table classification in Yesilada's study is not, however, entirely relevant to our use as her classification is based on both table content and table structure. For example, some table types can be treated as the same, if content is not taken into consideration, such as Periodic table and Content tables, or Football league tables and Statistical tables.
For the purposes of this thesis, we concentrate on the table structure as it relates to the presentation of a table. Furthermore, we restrict our work to tables based on the rows and columns structure, in which the cell contents are logically defined by the intersection of a row and a column. For each column, there is at most one column header and the data of all cells in a column refer to the same attribute. For each row, there is at most one row header and the data in the cells of that row are related to the same entity, but do not refer to the same attribute. An example is shown in Figure 2.1.1, where each cell represents the enrollment number of one department in one year.
7 Department 1993 1994 1995 1996 1997 1998 1999 2000 2001 Body Composition 2,335 2,430 2.425 2.440 2.615 2.540 2,1 V) 2.435 2,720 Cancer Diagnostic? l.H'O 1.245 1.220 1.215 12265 1,290 1.255 1,665 1/90 Cancer therapy 4.MO 4,175 4,120 4,115 4,035 4,060 4,135 4,140 4.070 Dental specialties 2,475 2.470 2,455 2,465 2,460 2,445 2,430 2,475 2,435 Dentistry 2,435 2,435 2.415 2,430 2,435 2.470 2,490 2.510 2.540 Food and Nutrition 1,630 1,575 1,645 1,710 1.730 1,710 1,690 1,740 1.630 Medical specializations 1.5«X' 1.43s"' 1.550 1.390 1,425 1,335 1,335 1,395 US'"' Medical technology 4,445 4,460 4,470 4.490 4.490 4.490 4.450 4.430 4.440 Medicine 1,745 1.355 1.635 1.600 1,660 1,615 1,475 1.315 1.250 Nut sing 4.055 4,070 4,445 4,745 4,305 4.280 4.040 4.010 3.S40 Occupational therapy 1.500 1.570 1.610 1,650 1,555 1,595 1/15 1.575 1,620 Optometry 1,115 1.110 1,105 1,100 1,110 1,115 1.105 1.120 1.100 Pharmacy 4,955 4,975 4,975 4,305 4,335 4,945 4,950 4,975 4,990 Physical therapy 1,590 1,635 1,660 1.645 1,635 1.655 1,675 1.665,1,690 Psychology 1,225 1,060 1.230 1,125 1,050 1.675 1,455 1.175 1,330 Radiation Biology 4.560 4,135 4.645 4.345.4.430 4,420 4.745 4.650 4.330 Social work and welfare 2,840 2,935 2.750 2.990 2.995 3.110 2.955 2,310 3,155 Space Research 1,190 1,330 1.165 1,245'1,135 1,150 1,215 1.260 1.145 Surgery and specialties 2.490 2.4S0 2.430 2,470 2,4w't 2.4in 2.4^5 2,420 2.415
Figure 2.1.1 Table sample
Our studies include two table styles: with headers (HTDT) and without headers (TDT). Text/Digit Data Table (TDT) refers to tables in which the cells contain text and/or digits data and for which the column and row headings are implied but not presented, such as Periodic table, or the Table of Contents shown in Figure 2.1.5. Headings + Text/Digit Data Table (HTDT) refers to tables which have explicit column and/or row headings and in which the cells contain text and/or digits, such as Time tables, Football league tables, Calendars, Statistical tables, and Truth tables, as defined in Yesilada's study, and shown in Figure 2.1.2, Figure 2.1.3, and Figure 2.1.4.
8 Sunday Monday Tuesday Wednesday Thursday Friday Saturday 27 28 29 30 31 1 2
Recurring every Font Size Bold Tuesday Ilalic Color
4 5 6 7 e 9 Recurring every Tuesday leaves room for more visible tent
10 11 12 13 14 15 16 [-recurring every Valentine's Day Tuesday is an example of a built in holiday
17 18 19 20 21 22 23 Recurring every T uesday
24 25 2E 27 28 1 2 Recurring every Tuesday
Figure 2.1.2 Calendar sample
Time Table to and from Tokyo ilaneria (.lul.2UU5)
All flights .11 •• il.im I'Vi'i-pi MIUU- v>illi )»iii.nl<. \im:ift t\pi> is rli,ui 4. Figure 2.1.3 Time table sample 9 Team H W ill i [Chelsea ?3 ."• 12 ;95 : Arsenal 136 [25 5 [87 J3B isi [Man Utd 138 122 [11 Is ise"' ]2S" ";77 Evertan :3B [18 If 13!45 ..."s i Liverpool S [if J7 ' 14 [52 Bolton J38 7 MB fici' 12 \ia" !58 Middlesbrough [38 il4 J13 in i53 46 [55 8 Man City !3B :13 12 147 39 :S2 9 Tottenham 138 [10 14 47 41 [52 10 Aston Villa !38 in ;i5 46 52 [47 T 11 | Charlton [38 fib' Vo 42 58 |46" 12 [Birmingham .'M..' |12 15 40 [46 [45 13 Fulham ]8 18 [52 [60 [44 14 Newcastle "J3B " [14 [14 [47 157 ' I-44"" 15 Blackburn |3B ;15 ~j14 [32 '\A3 J42 ; 16 [Portsmouth [10 Isi is lis" [59 W' 17 [West Brom [38 i flE ;16 [36 fel m 18 ; Crystal Palace ;38 :7 ''l2'"jig J41 "62" [33 19 [Norwich |38 |7 [12 19 I42 [77 [33 20 ; Southampton 38 1B ]14 18 [45 '|6B "" "!32 " Figure 2.1.4 Football league table sample 1.0 INTRODUCTION^ 1.1 Content and Exclusion *-' 1.2 Therapeutic Equivalenc e-Related Terms *> 1.3 Statistical Criteria for Bio equivalenc e *•' 1.4 Reference Listed Drug *•' 1.5 General Policies and Le pi Status *-' 1.6 Practitioner/User Responsibilities *' 1.7 Therapeutic Equivalence Evaluations Codes Figure 2.1.5 Content table sample 2.2 TABLE USE Tables provide users access to data, including text, numeric data, images, links, and other tables using both navigation and manipulation features. Table Navigation Modes The following seven types of navigation can be identified as commonly used on 2-D tables, as shown in Figure 2.2.1. © By Row: access the cells horizontally across a row. © By Column: access the cells vertically up or down a column. 10 ® Browsing: scan through the table of information in a casual way (horizontally and/or vertically). ® Link: follow hyperlinks between cells. © Group: expand or hide rows and/or columns to access a small part of the table (manually and/or automatically, such as Figure 3.3.6). © Refind: return to designated cells which the user has accessed earlier. © Direct access: use a search method to go directly to cells. jMaiRiFotiBi $* <4i*SM % Figure 2.2.1 Table reading and navigation modes Table Manipulation Features As the result of screen size limitation, the ease of use of features available for manipulating the table and the data in the table become important. A number of such features often used in displays on larger screens may be useful on smaller screens. We category them in three types: emphasis of cells, zooming, and option selection. 11 Emphasis of cells Colour [24] is an important and frequently used feature to emphasize data elements during visualization. A study by Healey [24] showed that seven isoluminant colours is the maximum number that can be displayed at one time that still allows rapid and accurate identification of any one of the colours for users [24]. When using five different colours, the visual system can identify the colours in parallel, that is, users could accurately identify the target in all cases. Boldness is another method that is often used to emphasize cells. Zooming Fontsize, many current applications allow variable font sizes, including browsers, Word, Excel, etc. This feature lets users select different font sizes for their comfort level when reading. In the case of the small screen, font size can be used directly to expand or reduce the footprint of tables. Tooltips, is a method that provides enlarged insets and detail-in-context presentations [10] to expand and magnify the content of cells beside the cell so that the entire content can be read. A study by Carpendale and Montagnese [10] examined several different presentation methods including distortion presentations, magnified insets and detail-in- context presentations (offset) to promote the ease of exploration and experimentation into the use of varied presentation combinations. Focus + Context [52] techniques work on the principle of displaying both the data of current focus at full size and detail, and the area around the focal point (the context) in reduced size and detail to help the user make sense of how the focus information relates to the entire data structure. Regions outside the focal point may be distorted smaller (as in fisheye views [21]) or selectively omitted. Tooltips have two drawbacks. First, they necessarily cover adjoining data and second, they may affect user performance [46]. Po et al. [47] suggest that the magnified area of Tooltips should be displayed above the selected cell. RSVP - rapid serial visual presentation, is a method of showing text one word at a time in a fixed focal position on the screen ([50], [67]). This technique, introduced by Forster 12 in 1970 [18], was developed for studying language processing and comprehension, and generalized for information navigation by de Bruijn and Spence [13]. de Bruijn and Spence concluded that RSVP is a valuable technique for searching and browsing information on small screen displays. Oquist's research [45] indicated that when screen space is so limited that traditional text presentation becomes ineffective, RSVP appears to offer an improvement in readability. A prototype designed by Wobbrock et al. [67] provided a RSVP tool to support web page reading on an iPaq screen. Grouping (subtable) refers to the creation of subtables [7] by the selection of cells, rows and/or columns. MS Excel, for example, allows users to re-organize the table by hiding rows/columns or to cut and paste cells manually to create new tables. The subtable concept provides flexibility to users by letting them create a new fully functioning subtable, which can be more easily managed. Expansion of columns/rows is the manual expansion of individual columns by dragging or by double clicking on a column header. This may be problematic on small screen devices because of the inaccuracy of stylus usage on touch screen, especially in mobile situation. Furthermore, in tables with cells that have multiple data values (text or numeric), the user may need to be able to expand the row height to display the full contents of the cells in that row. Option selection Menus, are used in almost all applications, allowing users to select from a list of available options. Menus save screen space where not all the functions or features can be shown on a dedicated tool panel, and, in the case of small screens, allow additional rows of the table to be visible, but require fine motor skill with stylus. A study by Po et al. [47] indicated that the upper visual field (UVF) is specialized to support perceptual tasks in the distance, while the lower visual field (LVF) is specialized to support visually-guided motor tasks, such as pointing. 13 2.3 TABLES FOR USE ON MULTIPLE DEVICES For applications using tables of data on multiple devices, there are three basic methods: the design and implementation of device specific applications such as those of Gaedke [20], the creation of specific tables for each destination device, and the automatic transformation of the data for the targeted device. The first method is to design applications that are dependent on the characteristics of each target device. The second method is to design the table using a specific language for use on each target device. Both of these methods are labor intensive and costly to maintain. 2.3.1 Transformation Transformation refers to the process of automatically altering the original data or display into one suited to the target device. Little work has been done explicitly on tables, and consequently we review related work on automatic transform of web pages. The content of the web page may be changed through the transformation and/or the presentation may be changed. There are several approaches to the migration of web pages across multiple platforms. Dynamic compression of web content has been used ([19], [51]) to improve the transmission rates of these full web pages to wireless devices. The major advantages of automated conversion methods are their lower maintenance costs, and independence from individual sites. One approach, the most simplistic and most frequent, is to just let users of small screen devices do their best with large pages on their small screens. Many approaches, as described below, alter either the content or data structure. For our purposes, this violates our principle of maintaining consistency of both table structure and data content. Transcoding ([26], [57]) performs the transformation of presentation components of web pages algorithmically, at run time. Eisenstein, et al., [17] proposed a general hierarchical architecture that models the dynamic composition of components for display at run time, although generally the designer would provide the mappings for each device for each application. 14 Bickmore and Schilit [5] provided a useful matrix, shown in Table 2.3.1, which summarized general transcoding approaches. Table 2.3.1 Bickmore and Schilit transcoding matrix Elide Transform Syntactic Section Outlining Images Reduction Semantic Remove "irrelevant content" Text summarization Syntactic techniques operate on the structure of the page, while semantic techniques operate on some level with the content. Elision techniques basically remove some information, leaving everything else untouched. For example, the contents of a section can be removed from the document and converted into a hypertext link which, when selected, loads the elided content into the browser. Transformation techniques may involve modifying some aspect of the page's presentation or content. For example, the IBM transcoding service ([22], [26], [27]), provides a server-side approach to modify web content or structure by choosing components for presentation on target devices based on both user profiles and device profiles. For example, tables are converted to lists, and large texts summarized with a link to entire content. All of these actions, however, result in changes in the original data and loss in fidelity. The Spyglass Prism system [61] provided dynamic conversions of HTML to specific target markup languages, such as WAP and WML, and dealt with image and HTML transformation at run time. The system traded fidelity for presentation compactness on the target device ([27], [56]), using key-term extraction, text summarization, document heading extraction, spatial size reduction, colour depth reduction, and lossy compression. These trade offs, however, are not always appropriate especially in sessions where collaborative or data sensitivity are issues. Digestor [5] was implemented as an HTTP proxy to dynamically re-create web pages using a heuristic algorithm and a set of structural page transformations to achieve the best looking document for a given display size. Digester adopted section header outlining 15 techniques, first sentence elision, and image reduction to reduce the target display size. It replaced the original text and images with hyperlinks so that the user could access to them. For our purposes, we extend the Bickmore-Schilit Matrix [65] to include an important additional transformation type, which we call retention. Retention refers to transformations that reduce size but not data fidelity, so that no data is removed or summarized. Retention as a transformation technique does not trade fidelity of data for compactness but rather trades navigation ease and display for compactness. In earlier work [65], a large form was logically separated into several small forms and displayed in several pages to fit the small screen1, and web pages [65] were compressed in size. This means that the original data is not altered. Table 2.3.2 Extended Bickmore and Schilit transcoding matrix Elide Transform Retention Syntactic Section Outlining Image Reduction Table size compression Semantic Remove "irrelevant Content" Text summarization No data is removed There are tasks for which changing or removing data from the original so that the result fits on a small screen are not suitable. For example, a user who migrates from one device to another may need to have consistency and completeness in the data displayed. Collaborative users on a variety of devices may need to have access to exactly the same data, and the same format. Finally, some tables do not have the same semantics when the content has been altered. For example, if the table structure that is logically row based as shown in Figure 2.1.1, may become problematic to the user when converted to a list as displayed on mScope [43] shown in Figure 2.3.1. In this case, none of the row components of the table have a context separated from the table itself. 1 Rui Zhang Master Thesis (2002), Transformation of Forms for Displaying on Small Screen Devices 16 '{Internet Explorer •* HI? 4:5? %ff http.7/gatew3y.oceanlakexorn/rc Uiw«ersi»v and colters «j»sdijst« 122?m?-m* w^*- 5S« i«3-:ooo:o(io-:0ij i»o L_ 5»3 Social wort arid welfare 2.S40 2.750 View Tools 4 ^ £& Figure 2.3.1 The sample web table displayed on mScope [43] Related work has been done on the display and use of large amounts of data on large screens. In particular Focus + Context ([21], [52]) designs have been used with success on regular size screens. Focus + Context designs present a consistent overview of the entire data set, which may be too small to be legible especially on small screens, accompanied by functionality that allows the user to focus on parts of the data within the context of the overall view. The trade off is be an increase in interactions and potentially an increase in difficulties in executing the interactions. 2.4 DISPLAY PROBLEMS ON SMALL SCREEN DEVICES A design for use on a small screen must be informed by consideration of those characteristics that have been shown to affect the performance of users on small screens. An early study by Reseil and Shneiderman [53] showed that the smaller screen size does slow down the reading time. Jones et al. [29] and Walters et al. [62] found that the smaller screen size impedes search task performance. However, Jones et al. [30] also suggested providing direct access, e.g. search mechanism, because the small screen users seem to choose and prefer direct access strategies over less directed, browsing 17 approaches. Duchnicky and Kolers [15] showed that even for very small displays of only a few lines of text, the ability of the user to read and understand information was not adversely affected. In addition, Nielson [40] found that, most users are willing to scroll if necessary. This means that a reasonable amount of scrolling on small screens will not lead to usability disasters. A study of task performance on tables using small screens by Watters et al. [62] indicated that for both simple and complex tasks the allocation of screen space for context information improved efficiency and effectiveness, where the context included headings and relative position within the table. Han and Kwahk [23] found that searching for menu items on single line displays (such as those commonly found on consumer electronic devices) was three times slower than when a conventional display was used. However, Swierenga [60] found that with larger than single line displays, there was no significant effect on hierarchical menu search time with the smaller display. These studies suggest that unless the handheld screen display is very small, the impact of the simple menu selection tasks on small screens will not be catastrophic. A further constraint for using tables on small devices is the physical limitation imposed by the screen size. Using a small screen device to view a table designed for use on a PC or laptop is like trying to use a small window to read the newspaper. Users can only ever see a small part of the text. The challenge is to balance the need for context with the need to display as much data as possible. This can be done by tailoring the data presentation and/or navigation functionalities to minimize negative performance effects on both efficiency and effectiveness. The constraints of interaction with tables on small screen devices include physical difficulties such as imprecise target selection and button clicking, inefficient text input, etc. For example, it becomes problematic to select shrunken cells or modify the cell content when the table size is compressed or ellipsed to fit the width of the small screen. A study by Mizobuchi, et al., [38] showed that both accuracy and efficiency were affected when the input box was 2mm wide, but not for the wider size. The effect of size may be compounded based on whether participants are walking or standing. The small 18 screen area may make it more difficult for users when they try to click functional button on the tool bar or select hierarchical menu using the stylus. In addition, it may be more difficult for the user to expand rows or columns by dragging or double clicking the edge of row or column header, which are quite small. 2.5 TABLE VIEWS There are several ways to display a large table on a small screen that parallel the work for web pages. For example, Mackay [34] compared three methods: Direct Migration in which the data is simply sent to the small device and the user navigates by scrolling or paging; Data Modification techniques which modify the data for use on small screens, for example, by reducing images and text summarization [35]; Data Suppression techniques, which remove original data such as selecting keywords, first n words or sentences or Z- thru mapping [58]; Data Overview, based on the Focus + Context model [58] used on larger screens, which provides an overview of the entire data set and users can focus in to get finer grained detail on some selected part of the Overview. Following the recommendations of the research by Watters and Mackay [63], we consider here Default View, Linear View, and Overview. Although the Default and Linear Views are frequently used in the commercial world, the Overview was found to be preferred by user in the Watters and Mackay study [63]. Default View - The most straightforward method to present tabular data on the small screen is to simply display as much of the source as fits the screen with scrolling function to navigate the rest of the data. The result can be seen in Figure 2.5.1, using Internet Explorer (IE). This imposes a burden on the user as they now have to scroll both vertically and horizontally to view the whole table. In addition, unless the headers of the columns and the rows are locked on the screen, it is very easy to get lost or disoriented [62]. For the Default View, the readability of large tables is reduced because of the limitation of the screen size but the cell size remains constant. This means using Default View model may increase difficulty in understanding the data in large tables on small 19 devices. Browsers like NetFront v3.0 [1] and ThunderHawk [6] use similar approaches, although the ThunderHawk browser also reduces the font and flips the table so that users use the PDA horizontally to view the table. This may result in more difficult use of the control buttons of the PDA. Internet &ptorer <^4:5» 0 http://toTe,cs,da[.ca/~iifiang/Tabl »•[ ^» University and college graduates 1993 1995- University 199-4- 1994 1996 P.-vchologvl 9,22^. 10,060: 10.280 iq Social worl- 2,840 arei welfare Food arid! nutiition, 6S0 575: 6451 dietetics: Dentistry 435 435: 415 View Toots ^ 4*} Q Figure 2.5.1 The sample page displayed with IE on PDA Linear View - Another general approach used for PDAs has been to transform tables from their original 2-D structure into a linearized version. AvantGo [3], for example, and OceanLake's mScope [43] use this technique to produce Linear Views of tables, as shown in Figure 2.3.1. A difficulty for the user with this kind of transformed table is the increase in cognitive overhead needed to maintain the context of the column headers. Figure 2.5.2 [36] shows an augmented version of the mScope scheme in which the column headers are repeated for each row. This provides context, although it is problematic for long headers, and it makes row-wise access much easier for the user for some types of tasks. For example, it is reasonably easy to find the number of graduates for a given department in a given year. At the same time, linearization in this fashion makes it harder to complete tasks that require column-wise access. Any deconstruction of tables into lists of rows or columns creates more complexity for classes of tasks, such as which department has the most students in a given year. 20 ItateroetEKplorer Uraw*rei>v mi wikae qfjdua**: (Ur Overview - The third technique, reminiscent of the Focus + Context techniques ([21], [52]) used on larger screens, shrinks the table so that it fits in its entirety on the PDA screen. Information in the table may be shrunk but an overview is always available. Focus + Context techniques follows a principle of information visualization to display the most important data at the focal point at full size and detail, and display the area around the focal point (the context) to help make sense of how the important information relates to the entire data structure. Regions outside the focal point may be displayed smaller (as in fisheye views [21]) or selectively omitted. Data cells are displayed within the context of the Overview to provide a consistent cognitive model of the information space. On larger screens, this consistency has been shown to improve performance and lessen navigational disorientation ([2], [58]). Data Overview navigation techniques include fisheye views [21], context maps, Zoom and Pan, brushing techniques, such as riffling techniques [58], and image representation of the layout [29] on larger screens. Word wrap, unfortunately, reduces readability for the user on the small screen and may be problematic. The Palmscape [28] browser in Figure 2.5.3, for example, keeps the two-dimensional structure of the table by shrinking the column width, word wrapping within the columns, to reduce the width of the table to fit the device screen. 21 Internet Explorer *» <4$ 4:5© |( ittp'Z/ftjme c:.dJ ij/- -r:l MI «g/T •» | f* University and college graduates 1993- 1994- 1995 lUniwrsity 1994 1995 199 6 PJ ethology 9,JJb IU.LOC: lu.jy 0 Social »"otl and "veffare 2.840' Z750: Fojd and 680 64= CbetetKi- Dentiftt * 4^'rf 435 4151 View Took •$ <*i J3 /j Figure 2.5.3 The sample web table displaying on Palmscape Microsoft's Pocket PC version of Excel provides a model of functionalities for spreadsheets that can be applied to generic tables, including easy column width adjustment, row and column selection, frozen headings, and spreadsheet overviews. -IP ocket_Pt - C«, > "' "wo Toc's Hefe ;> •*•<. -^ ® 1 | _A. J e i c l 1 «.••'-. .- 5 ^; «..- :•- -•£. SC / J i . .»,.-• , ' " . '-•-. < \ ;~"?-1 I -f " ~ ' * ^~~ V as i> •6'lSf.. :-j»o. :->3i. : ^'t Figure 2.5.4 Sample using Excel on PDA The Overview model keeps the consistency of table's content and layout, which is one of the principles of Human-Computer Interface Design known as Shneiderman's Principle [55]. This principle calls for consistency in sequences of actions, terminology, colour, layout, capitalization, and fonts. Norman [14] also notes that an overview map that matches the users' mental map of how things work is very important. In addition, both 22 Shneiderman and Norman emphasis reduction of short-term memory load as a key to good interface design. Norman indicates that the rule for the number of items that a person can keep in short-term memory is 5±2 [14]. Miller's "chunking" concept describes the capacity of short term memory: 7±2 [49]. People will have trouble remembering information that exceeds seven items. When users use tables on different devices, they construct a mental model [14] and it is important to minimize the effect of adaptation of that existing mental model when migrating between devices. A variety of factors, such as layout, affect the integrity of the existing mental model when switching between devices ([12], [58]). The Overview is used in this thesis as the presentation model for use on small screen devices. 2.6 SUMMARY The chapter provides an overview of related work in the area of table structure analysis, table use and table view over different devices and device sizes. This literature review has highlighted the gaps in the literature that motivated the research presented in this thesis, specifically related to the study of the table use and displays suitable for small screen devices. The next chapter presents a summarization of the research questions which motivate the new features introduced in this thesis, user studies, the methodologies used, tasks analysis and the data collection tools employed. 23 CHAPTER 3 RESEARCH QUESTIONS 3.1 RESEARCH QUESTIONS In recent years, the mobile device market has shown dramatic growth, despite very limited screen space which makes it difficult for users to read and use information. This presents a serious challenge for interface designers. In this work, we focus on design issues for two dimensional tables. Table users may also require reasonably seamless migration of tables of data, when they need to move between devices using the same table or discuss or collaborate using a shared table with a person using a different device. Our research is focused on display, manipulation, navigation and usability of tables on small screen devices, such as the screens on PDAs. The research may also benefit the designers for other two dimensional data modes, such as pictures, web pages, or multi dimensional data. This research may also benefit application design of tables for larger screens. Other researchers, as introduced in chapter 2, have applied automatic transformations to display tables on mobile devices by either content elision or structure change. The purpose of this research is to allow users to use the same mental model when migrating among various devices. The first assumption we test is the suitability of the Overview model for tables on the PDA, in terms of effectiveness and efficiency. We selected the Overview model for this work because it keeps the same two dimensional table structure that the user would see on a normal personal computer. Although the trade off of the Overview model is the reduction of font size and obscuring of much of the table data for each cell, it provides a fidelity to the table structure, provides the users a big picture of the table, and gives them a sense of context within which to locate their focus. These factors are especially relevant when the user reuses a table which has been used previously on a larger screen device. The table content displayed in a cell in the Overview model may be hidden because of the space limitation, but it is not elided. Consequently, users can use a tool, such as tooltips or column 24 expansion, to read the entire content of a cell without giving up the overall table structure. In this research we examine several such tools for a range of tasks to assist the use of the table, both efficiently and effectively. Hypothesisi = The Overview model is a more efficient, effective and user preferred model than the more commonly used Default View and Linear View models. Following the initial study we chose Overview as the table display model. The next phase of the research explores how to alleviate the effect of the compression the data in cells for the users to perform table related tasks, in terms of efficiency and effectiveness. There are many features or functionalities available for various tasks as we discussed in previous session, and our research is focusing on the following four tasks: browsing/scanning, Comparison, refinding, and search. For Browsing or Scanning Tasks When the user tries to browse a compressed table, he or she normally needs to adjust the width and/or height of columns, rows and cells. This means, the interaction between users and the device will be increased significantly over use of an uncompressed table. Our goal is to minimize the number of interactions for users to browse down the columns, across the rows or over the entire table. We applied the RSVP theory [67] to design a new method, the Cascade method. The Cascade tool provides an automatic sequential expansion of cells with tool-tip popup boxes either down a column or across a row using a single click on the header of a row or column. Hypothesis2 = Cascade is a more efficient, effective and user preferred method for browsing or scanning tasks compared with the more commonly used column expansion method. 25 For Comparison Tasks The Cascade method dynamically displays the cell content in a designated area based on an assigned display time, therefore, short term memory is required for information gathering for comparison tasks. In this case, it may become difficult for users to compare two or more pieces of information dynamically by relying on their memorization. Within a column: If the information required for the comparison is located within a column, which is the simplest case, users need only expand that column and use the scroll bar to move back and forth, if necessary, to do the comparison. Across columns: If the information required for comparison is located within a row, users normally would have to expand every column to read the information for the comparison. In the case of a compressed table with many columns this opening (and closing) of multiple columns becomes problematic. If the comparison information is located randomly in a table, both column and row expansion as well as scrolling may be required to finished the comparison task. The problem remains how to reduce the amount of interaction between the user and the device while doing such comparison tasks. We propose that simplifying the column expansion operation from click and drag to single click and introducing an expansion method based on the Cascade method may be of benefit for the comparison tasks. Hypothesis3 = The Column width auto-adjustment (CWA) method is a more efficient, effective and user preferred method for comparison tasks compared with the more common used click and drag Column width expansion (CWE) method. Hypothesiszo = The Cascade method is a more efficient, effective and user preferred method for comparison tasks compared with the more common used click and drag Column width expansion (CWE) method. 26 For Refinding Tasks For tasks that require refinding information already seen, it is very easy for the users to lose focus on the cells of interest during the operation on a small screen. Users scroll back and forth to get previously seen data in the small screen view, or re-expanded previously expanded columns. We introduce a landmark feature to assist in information refinding. Hypothesis4 = Landmarks are an efficient, effective and user preferred method for refinding tasks. For Searching Tasks A common strategy for users of large tables is to use a text search feature. Earlier work has shown that the common search feature used on larger screens has a negative impact on performance on the small screen [30]. Therefore, we focus our concern to test a new search result browsing feature, the auto-Cascade, for use with large tables on the small screen. Hypothesiss = The Cascade mode will be more efficient, more effective, and preferred by the users to scan search return results than the Next mode for various tasks. 3.2 TASKS AND CURRENT FEATURES ANALYSIS In this research, we have identified four types of tasks that users do using tables on their small devices: Browsing: to browse through the information in the table and get a general mental model of the table, for example, the first time using a table. 27 Comparison: - Comparison of values within a row or column: for example, to find the maximum or minimum value within a column, or a row. This type of task requires the user to hold in memory some values while scanning a single row or column. - Using data involving different rows and/or columns: for example, to compare two or more cell values, in different rows and/or columns. The user needs to scan a row or column, find one value, memorize the value, and scan a different row or column based on the first value to compute the task. Refinding cells: to refind some particular cell values in the table. Finding cells: to look up or search for specific cell values in the table, for example. Current Available Features The following list interprets current available and commonly used features which are for tasks involving tables on small screens. Font: users can adjust the font size using zoom-in/zoom-out for browsing tasks on mobile devices. The size of the table itself will be adjusted accordingly, and may not automatically fit into the small screen. It may require several attempts for a user to adjust the font size to display an entire table on the small screen. It may be difficult for users to make the table readable by font size adjustment. Colour: colour is frequently used to emphasize data elements during visualization. Different colours can help users find and refind information, and can also be useful for comparison tasks. One problem could arise if users may use as many colours as they want to finish a task, which may, have bad effect on the task performance. The study by Healey [24] shows that seven isoluminant colours is the maximum number to be displayed at one time that still allows rapid and accurate identification of any one of the colours for users [24]. 28 Preview: for large tables, users may use a preview function to see the entire table. SubTable selection/generation: sometimes, users do not need to use the entire data table and it would be useful for users to generate a subtable based on their needs. Currently, products (such as Pocket PC Excel) provide a cut and paste function that allows users to create new tables. This may require considerable effort and may require users to migrate back and forth between two tables. MS Excel provides row and column hiding features which may be helpful, however, these are difficult to operate on the small screens. Column/row expansion: is a well known feature available in MS Excel. There are two ways to expand columns and rows. One is to manually drag the right edge of column/row header; another one is to double click the right edge of column/row header to automatically expand the column/row width based on the inside content. One concern for this feature on the small screen is that most small screen devices use touch screen as their display screen. It is relatively more difficult and more inaccurate to select small spaces using a stylus than using a mouse, especially when a double click of a header is required. Search: is a common method to match certain text strings in a table. When a user searches in a table, the cells with content matching the search string are selected and users can navigate through matching cells. Current applications especially for tables, such as Pocket MS Excel, have problems for users. They are not flexible enough to re-organize the content of the table data in order to let users focus on the part they really interested in, for example, if users need to frequently focus on some parts of one large table, they either have to re-create many subtables or use some way, such as highlight, to help them focus. Using too many different colours for highlighting in MS Excel sometimes cause confusion as well [24], if the task requires drilling in the table in several steps. Second, frequent stylus entry on small screen is difficult especially for expanding columns or rows, because it requires the user to accurately click and hold on the separating line of a column or a row and drag to expand. 29 3.3 NEW FEATURE ANALYSIS According to the analysis of currently available features, the prototype for use in this research was designed to include those features as well as new features. The features included in the test prototype were chosen to provide consistency and reasonably seamless migration of tables, between regular size PCs and PDAs. It is important to minimize the distortion of the existing mental model of the user as they migrating between devices. Table 3.3.1 presents the relationship of tasks and features we explore our research. Table 3.3.1 Relationship of tasks and features Task Comparison Finding Refinding Features ^^---^^^ Browsing Whinin a Within different Cells Cells column/row columns/rows Table Overview (Font • size/preview) o o Rows selection o o o o Columns SubTable • • • selection o selection / Block selection • • generation o o Special cells & • • • • blocks selection Tooltips • o o o o Column expansion • o o • o Row expansion O o o o o Cascade • o o • • Landmark o • • • Search • • Note: •: primary target; O." secondary target; In the following section, we will describe the main features built in to the prototype. 3.3.1 Prototype The first version of the prototype, written in Java, was an emulator running on a regular personal computer mimicing features found in the Pocket PC version of Excel. The font 30 size of the data in the new table was reduced and zoom-in and zoom-out functionalities provided for the user. The table columns were ellipsed to fit the width of the small screen. This worked reasonably well up to 11 columns and 15 rows on our Toshiba PDA. If the table is bigger than 11X15, however, the table appears to be largely empty. In the preliminary study, we chose University enrollment data table as an example, shown in Figure 3.3.1. The reason we choose numeric data table as the first attempt was that: first, we want to show the table content as much as possible and in the mean while providing the most meaning of the table; second, it is easy for users to understand the comparison tasks using numeric data table; at last, we considered the numeric table is one of the most common used two-dimensional table in the market. Department 1993 1994 1995 1996 1997 1998 1999 2000 2001 Body Composition 2,335 2,4 3u 2.425 2,440 2,6 i 5 2,540 2.110 2,435 2.720 Cane er Diagnostic s 1.1 60 1.245 1,220 I ,t~. 1 _' 1,265 1.290 1,255 1.665 1,690 Cancer therapy 4.040 4,175 4.120 4.115 4,1)35 4,060 4,135 4.140 4,070 E'ental specialties 2.475 2,470 -.4 J-1 2,465 2,4o0 2,445 2,430 •-. A 7*- •-• jJO'-, Dentistry- 2.435 2,435 2.415 2,430 2.435 2,470 2.490 2,510 2.540 Food and Nutrition 1.680 1,575 1,645 1,710 1.7S0 1.710 }f/ju 1,740 1.630 Medic id specializations 1,500 1.4 SO 1,550 1,390 1,425 1,335 1,395 1,360 Me die al te c hnology 4.44 5 4.460 4,470 4.490 4,490 4.490 4.4 5 U4,4S O 4,440 Me die me 1,745 1.355 1,685 1,600 1,660 1.615 1.475 1.315 1.250 Nursing 4,055 4.070 4.445 4.745 4.305 4.230 4,040 4.010 3.340 Occupational therapy 1,500 1,570 1.610 1,650 1.555 1.595 1.615 1.575 1.620 Optometry' 1.115 1.110 1.105 1,100 I.HI'I 1.115 1.105 1.120 1.100 Pharmacy 4,955 4.975 4.975 4.305 A 9Cr, 4,945 4.950 4.975 4,990 Physical therapy 1,590 1,635 1,660 1.645 1,635 1.655 1.675 1.665 1,690 Psychology 1,225 1.060 1,230 1,125 1,050 1.675 1,455 1.175 1,330 Radiation Eiology 4.560 4,135 4.645 4,345 4,430 4,420 4,"?45 4,650 4.380 Social work and welfare 2,840 2,935 2.750 2,990 2.995 3,110 2.955 2,81(i 3.155 Space Research 1,190 1,330 1.165 1.245 1,135 1,150 1.215 l,2*-"0 1.145 Surgery and specialties 2,490 2.4S0 2,430 2,470 2.460 2.410 2.405 2.420 2.415 Figure 3.3.1 University enrollment sample table 31 The user can adjust column widths as needed and mouse-over pop-ups can be used to show the contents of individual cells, as seen in Figure 3.3.2. We used this prototype for our first user study comparing table views. t_f MaitiApp Figure 3.3.2 Numeric sample table with Overview model on simulator prototype (a few simple mouse over help widget was shown) In order to further reduce the impact of the prototype itself on the experiment study, we developed a working version of the prototype written in C# and executable on the Pocket PC device, Dell Axim X30. This prototype was used for the user studies. We also extend the sample table type from numeric only to mixed contents in study 3 - expansion method study, and study 5 - search study. A mixed contents sample table, hotel check-in data, was shown in Figure 3.3.3. 32 Name Gender City Check-In Date Check-out Date Balance Credit Telephone Email URL Mary A. Campione Female Toronto Mays, 1999 May 21,1999 $1,233 $2100 (416)929-0019 [email protected] http://Www.carnpione.com Jone B. Helon Male Montreal July 23,1999 August 11,1999 $1,204 $1800 (800)834-843 [email protected] htr.p://www.helori.com Monecia Basney Female Montreal July 25, 1999 July 30,1999 $1,500 $2600 (800)668-9934 [email protected] http://www.basney.com Stephanie Argue Female Mew York Auguest 11, 1999 Auguest16, 1999 $1,480 $2500 (212)522-5868 stephanie@email .com http://www.argue.com Brad Ailing Male Montreal January 14,2000 January 21, 2000 $2,480 $2300 (800)268-8900 [email protected] http://www.alling.com Benjamin Befort Male Montreal March 30, 200D April 07, 2000 $2,460 $2100 (800)736-5923 [email protected] http://www.befort.com Amy Blue Female Vancouver September 22, 2001 September 29, 2001 $1,335 $2100 (877)809-1659 [email protected] http://Www.blue.com James Dennis Male Halifax September 30, 2001 October 05, 2001 $2,405 $1900 (866)220-6045 [email protected] http://Www.dennis.com Lin Varre Female Halifax October 18, 2001 October 27, 2001 $2,420 $2000 (B66)306-4636 [email protected] http://www.va rre .com Justin Ding Male Halifax December 12, 2001 December 22, 2001 £2,415 $2000 (866)722-9226 [email protected] http://www.ding.com Keith Dressier Male Vancouver February 21, 2002 February 27, 2002 $1,395 $2300 (877)426-60D6 [email protected] http://www.dressler.cam Ross Edwards Female Ottawa May 25, 2002 June 02, 2002 11,360 $2200 [888)839-9289 [email protected] http://www.edwards.com Chase Hoozer Male Washington June 11, 2002 fJune 18,2002 $1,425 $1400 [202)647-4000 [email protected] http://www.hoozer.com Oke G. Pamp Male Washington October 21,2002 October 30, 2002 $2,445 $2100 [202)647-5225 [email protected] http://www.pamp.com Errol Gray Male Washington April 14,2003 April 19,2003 $1,105 $15D0 (202)647-6575 [email protected] http:ZAivww.gray.com Linda Grant Female Boston April 17, 2003 April 30, 2003 $1,115 $1600 [617)448-6000 [email protected] http://www.grant.com Jaye Wotherspoon Male Boston August, 31,2003 September, 03, 2003 $1,855 $1700 (617)882-1211 [email protected] http://www.wotherspoon.com Haakon Borde Male Boston September 18,2003 September 29, 2003 $2,460 $2100 [617)619-6523 [email protected] http://www.borde.com Scott Blanchard Male Waltham November 25, 2003 December 23, 2003 $1,595 $2600 (7B1)433-7800 [email protected] http://www.blanchard.com Rich Houseknecht Male Waltham December 24, 2003 January 20, 2004 $1,575 $1600 (781)433-7850 [email protected] http://www.houseknecht.com Figure 3.3.3 Full table view of hotel check-in sample The main interface of this second prototype is shown in Figure 3.3.4 and includes two parts. The upper part is the mode control area and the lower part is the data display. Initially, the table columns are ellipsed to fit the width of the small screen and the rows are compressed to fit the length of the screen, as much as possible. At the same time, the font size of the data in the new table is reduced and zoom-in and zoom-out functionalities are provided for users to control the font size. Users can define sub-tables using the stylus to select rows and columns of interest, expand individual columns and rows, and expand individual cells with tool-tip style pop-ups. 41*30 jof M^ ^t^ ij' A*>- S~» SI* ^^C ;c^ **tt ~ Kc r£ *-c Jul' Jii^ SI- 52* f5C '""C- ~ct m 5c* F~" f.e -u< .i,r S; Sc< ';: SC- rtt; - E'3 Ma Mo Js* is* 22. £3 .5: bra htt ™~ Be' Ma "v '-'s --• •!•;. $" ;j; !v.' nit Data display area ir- Ft- '.;•«• S- "*. -:. £2 ;a~ s- *•,•*. j3-- v3 ~3 Ss: Cc ;i, *2 ;3£ :a- ->tt > ;_,;> Fi.. ,-,!,• £,.-1 £,£< J2. S3 far v.t <;-.-. «a: ~ >*: '"~f to S2. si '^» *8fafc u Select StibTabte Figure 3.3.4 Compressed view of hotel check-in table sample We introduced a new access method, Lookahead Cascade, which provides automatic scanning of rows or columns without opening up the column or row by providing rapid, timed, sequential expansion of cells. 3.3.2 Table Overview (Zoom-in/Zoom-out) Overview is the table view model we choose to introduce in our research following the recommendations of the research by Watters and Mackay [63], where the Overview was found to be preferred by user in their study. The Overview provides the user with the same structure of the table as would be available on regular PCs. This helps the user to get some ideas about what the table is about and keeps consistency of mental model during device migration. Maintaining the mental model of a table may benefit the user when seeking particular information for similar tasks. For example, in Figure 3.3.5, if a user wants to look up the contact information of a person, through the Overview model using font size Zoom-in/Zoom-out feature on the small screen, the user may have a general idea about where to look for telephone numbers or email addresses from previous exposure to the table. 34 I Zoom-in/Zoom-out feature $M{«« © ' 4|M © #• to ~.]^=~--t "J3 v J v^'i D \m Gi a a a B« a T< EI UI > t-%Felf.y,i^slS2{4;u hi - r v Mo ^' ; -ti' .*. SI. %2r ,S^ VT: nt; JoM;M. JijAt$J5l(8 jo ht M< Fe M. Ju Jtt $1 $i (8 ni ht st s^^Ai A< $is;(??f Figure 3.3.5 Table Overview sample with Zoom-in/Zoom-out feature 3.3.3 Subtables Although Overview provides users a general structure of the table, part of the contents will not be displayed entirely in most case because of the space limitation. For users who need to concentrate on sections of the table or need to juxtapose data for some task, the SubTable option provides the functionality to select parts of the table and create a new table (sub-table). A single click enables the user to return to the original table or a previous subtable. We provide four ways to create subtables in our research prototype: Block selection, Row selection, Column selection, and Special cell & Block selection. Block selection allows users to focus on a contiguous part of the table. Users can click the top-left and bottom-right cell of the target block to select a block of a table, and then generate a subtable based on the selection. As an example shows in Figure 3.3.6, users may want to know who are in the same city from the selected customers. This feature may benefit cell value lookup, min/max value finding, and value comparison in a particular area. 35 •lata form & <4$ lk» |p • M.wi Form 4| 1639 0 B "• ncsscsce tgoM -:3 csdtr *3 T 4-' & t D «~s -: '\fi Na Ge Crt Ch Ch 8a €r« Te In; UR A. (tame Gender City Check- Ma ?*< -;. «» "a n.sz: -4! -3 Hrt,: • My~;.3 rfsi *-*-:-":t!r3: .'-jfv 25, - 1 _c- '-'s '-'o ."j- irf si. 5;- ;s; -•- -•?•.•: SMj.-jr. =e"3 t NsV-'C .iU3.;St J ;: 5:? :3: £rs; '-'si* t-'c-r-trss .,3-'.v.a"> SSiSrlZ -' - •*"' "*i ? IISS2E ^:;i- s-- *'2i sta htt- ~#ri3.y.-\ Mate M:"trs3 -'5r:r SSSSSSS-ip' s~. iz: {?". wr ^tr; > gBBB 5e: 51, S2: (3? 3- '»"• v A single click enables the Ja* 3 M3 ss; Oct Si, $21 'Si ;2V ™tt. user to return to the original 1 l L- *=*' -3 C: Or! JO S3; :? I v.; *G '; table or a previous subtable. — *i^i^Li,.^,.,...£z^JZ»zd™Z^^^ s. Select SubTable ^&F Select SubTable Figure 3.3.6 An example of table block selection & generation Row selection allows users to select a single row by clicking the header of the row; to select contiguous rows; and to select non sequential rows by clicking the header of rows with control key button selected (Figure 3.3.7). 'HainFQna <&t «|$ 1*41 {£} £.? Hate Farm £f «<£ «Mi 0 1 3 o an •J ciict - D' >TL" \ m as Na Ge Cit Ch Ch Ba Cr< Te Err- UR\U m Ge Cit Ch Ch Ba Cr»jl« Em UR ' f I ; : , =ffi| Select S*i!?Tafele select SubTable J Figure 3.3.7 Non sequential table multiple rows selection & generation This feature may help data comparison, min/max value look up, and finding cell values in rows when the user can identify a subset related to a given task. Although row selection 36 allows the user to downsize the table by showing fewer rows, users still may not be able to read the entire content of individual cells. Column selection allows users to select a column by clicking the header of the column; to select a sequence of columns; and to select non-sequential columns by clicking the header of columns with control key button selected (Figure 3.3.8). Column selection is useful for data comparison tasks because the user can juxtapose two or more columns for comparison and more cell content will be shown in those columns. This feature reduces the number of columns shown on the small screen, therefore often allowing more space for cell content. yInamfmm £f 4$ UM3 0 Wain Perm 4Cl • cascasr Q Tcc:T.p ® D -OCT:* Na Ge Ch Ch Te Eir UR «r Balam Credit Emaii jk jj?l <"tt • • r-tc'-. -;•;-- r. ::s 32: - si\ S t5£.'.'.-.2 2 ir-s !',:^f* s:.2^ SIS .'-;"t:jT: "VMM 3Z Mai ":"s*c>. S-«:^re si.E" s:c -:-•»: ' 22 •1HP-1SN SKJ+J fe>- si -=: J:F K-J-3- s-ai ;i;rr~ H;.-;: J:: t-ar?.:- is-a »?sa^:s^ &=•;?-• '•'•;"(-» ;:.•»*: 3^1 reqs-:: £.—, •.4<-:CJ J: 3:r fC: a^yS* 2^ -•-•'3121 ."S^tt •35-^5 -s^s< J2,^:S $."?? ;s-esl- ; 3 I'"- *-3i.?B\ s2 4:: s;. - j.*~ ; — i»:-t-" ^miQmnF , , jyp Setect SubTable Select SabTaWe ^f Figure 3.3.8 Non sequential table multiple columns selection & generation Special cell & block selection provides users with the flexibility to focus on selected cells as shown in Figure 3.3.9, where a combination of block, column, and row selection can be seen. Users select target cells using the control key button. Subtables are generated in which the column header of the selected cells will be automatically used in the subtable. This feature is good for data comparison, because it allows users to focus on specific cells of interest. Again, when users have some ideas about which area they want to concentrate on, this feature may be a good choice. 37 Main form g? <4C VMS $ WafciFwrm # « Ra Ge Ctt Ch Ch Ba Cr« Te lm UR * {tame (City |&atan<|credjt Email *"^ s=e- ~;< *"* W3 ii sx -'-*: "3 ~tX * x- '•--•-- r. X4 J:= -1- r>1a ntt IS 22 -"" ** 111! •-< SS i 'Xrec Kcte X XX SZ~ - "=: - ^r^ <^»f *-,» ^ £w f c~ = -'"^ *;^s «^x .' Z :i~ —• •?. 5Z -if EX ;j:t-f Era *-s '•*; -*S" )3' £Z, XX '?C brs r-tt - ™ J g^( ;.;a ;,<,; vj Ap- 3Z. SX f?C ter Ht ^;~~ - - 0 -> ;v;.^ • .»...* ?„;,^,...i*.^il...s;,f;.*...£;.,i,.{.....™.rr. fc.?.."....;.£..~.„..,.r,„a,.,s?|!iw liftS^ iQR/ iyi TV Select SubTabte Seteet SubTable Figure 3.3.9 An example of table special cells selection & generation 3.3.4 Cell Expansion In the Overview, the content of most cells cannot be shown entirely. Cell expansion is accommodated through the following features: Tooltips, Cascade, Column/Row manual expansion and Column width automatical adjustment. Tooltips is a feature that presents the full content of a cell in a popup box overlay when users let the stylus stay on a cell for a brief time. Users can expand any cell of a table while browsing the table. For example, users may want to know which city James comes from (shown in Figure 3.3.10). 38 •I Matoffafiti •^ 4~ Uai3 |JJ i ,i Cascade U cti $ Na Ge Ot Ch Ch &a CnTeEmUR .'•'3 F~* "or Ma' Ma- SI, £2: (41 rg htt Jvf MB Mo M Au: 52. St; fSC jor htt M;- Fe- Ms Jul> :c> £1, «2t (&C "o htt 5t« Fe- Kr- it: iu: SI, £2! (21 ste rtt 6ra M3 MD Jsr Jer S2. S3; J5C brs htt &er Ma Mo Ms Apt $2, Si I (SC ber htt £JJ«l SI, 52: JS? 3 JgjCB"l'v Select Sublahfe Figure 3.3.10 Tooltips sample Cascade, a method based on RSVP [67] technique, provides the automatic sequential expansion of cells with a tool-tip popup boxes either down a column or across a row by a single click on the header of a row or column. This method was designed for use in table browsing, cell finding, and value comparisons when the user is not familiar with the data in some area of the table. It provides a quick scan through data cells without multiple interactions, which may become more important when the user is walking or in an unstable situation. If, for example, the user wants to know the minimum or maximum value in a column or row, one click will let the user check a whole column or a whole row. Short term memory, however, is required for the users. In Figure 3.3.11, one click on the third column header will allow user to scan the city name of hotel tenant, and another click on the row header to scan the information of the chosen tenant. Also when the cascade moves outside the range of cells on the screen, as shown in the middle image of Figure 3.3.11, the popup stays at the end of the column with a different background colour. 39 m !M. Na Ge mt&qrtbd Ch Ch Ba EmU R .*• Na Ge Ctt Te CM Ch Ch 8a Err UR 4k Na Ge Crt Te Cm Ch Ch Ba lir UR * i, ia • Ma Fe-Jfoft^i j2: r-'a r-'a 51. -a »tt > v» |—^" *"^? ',*^1 S2" * ' *'*B *I "5 htt - r-'-j Fa- ~ot C*?i si: !-'a r-'3 S3 -3 htt ;sr "s i'c (SC 51; ,'v* ."s,r ;i. ;;r> H+ .**r f-'3 ?*";• (32 £1; Ji.f JLr SI '%" htt jg* r-"6 r-'c (&•: SJ: J^- 3«r«; ;3" htt H CK Ma ws (:: si- :a .&>.• r„. ch rtt' ,r. ,.tt vVa .,2 c Si- 2J^ £u-. SI Z"l 'tt 3k- t-'s vVa (:: S2; Del Cci 52, <-i rtx.'. CK* f-'s '.Vs (22 52: Oct Ori 52 r-fc htt 01.'* M3 52: Cd Ocl 52 oh htt ••': ft- i-z \i: $2i. ;ui ;-LI s:, "-: htt; *•'; F?' Vz- ?32 J2: 2 - ;«' 51 S2i J-J* 2u" SI -•;. htt - —|, M "••0 htt ' — Sts Fe< r,~. ^21 £2! At.: Ac: 51. sts Ht St- Ft- F,~- (21 J2!fe A„, SI <•*? htt i>2>tf r^' Ke (21 52* A -^ 51 st-s htt S'S Ha I';. fSC S2: ;sr -V S2, ira ~tt. 3ra Fa *-': (3: 52: ;ar Jar 52 t*3 rttt' 5r» K> S-'D fSC S2: Ja- J3- 52 bra htt i.<": FT' 53'. (61 SI* AC: AP' SI. .<-C htt •. Lir: F*' E>;< :.61 i'J Af: i.p' 51 !T ; *" tt - LI F-e- &c<: (&1 Sit ASi J^?I 51 hrc htt Err vj -As (2C SI! At: Aps SI. e-r htt: Err t-'s >.V2 (21 511 Ap Ac SI ST 'tt Err ?-'a •,*,» (22 SI! Api ftp- SI grr htt 5 ; f a ? S SZ ^^y ^iA. &e< t-'i !<•:• it: S2: '-'a AM 12. z-f rtt. be; i-tt Ber r-*8 i-'z, ft'.: 52: MB A-;- =2 — " '' '""MB' '* "'^ 1 o o= ',•-. Sn„- 'si rn:...i-...... r..r.f ,-* .-.,.• art j •^w-JS^a-tt-JLa-s^^ _^:iw™*3*fefc™. „2,ffl,„,„..t,,:,"i.... C, Column/Row manual expansion is a standard feature, also available in our prototype, that lets users use the stylus to directly manipulate column widths or row heights. Column expansion allows the users to read every value in a column but may distort the current table layout by collapsing other columns. Column expansion makes it easier for the user to compare data and find cells, because the user can read the entire content of each cell in a given column, as shown in Figure 3.3.12 The column expansion feature should also benefit the min/max value look ups within columns, although these may also require a scroll. Row expansion adjusts the row height to provide access to multi-line cells. On its own, the row expansion feature may not very helpful for data comparison, min/max value, or a cell value look up, because users still cannot easily read the whole content of cells in the expanded row, as shown in Figure 3.3.12. The ability to adjust both column width and row height makes table browsing more flexible. For example, users may want to see names and Credit at the same time, as shown in Figure 3.3.12. Column and row expansion features are designed for information comparison, cell value look up, and min/max value in columns. Jtein form *?x 4x 11*5 0- #.•^•41 iM? #1 ]H] C3SC3& Na Ge Cit Ch Ch Ba Cr< Te Eir yRi Nam* GelCtt Ch Ch Cr Tdepr Err *,'» Fe' ""or f-'3 *'"a 51. Si: (41 -rs htt Staph Fe- Ks- AL- AU: S S2 (2:2; st~ J:-r ?»'3 Ms J-t- it,: 5"., «:? fSC pr htt E-33 K3 !•': 33*- Jsr S S3 [iZZ) Jrrs c V.z e Ko J J JJ ~1, sit ;SC '--;- htt 3 E*r:3 Ma Ko I-'? ap; * 53 {SCO} is F-5-' V5f S*; >*s * S2 J3"S MS — TEB| F-' f-3 Cd ^St'r K3 !-3' s s3 (sss;, Select SubTable Select SubTable Select SubTable Figure 3.3.12 Column, row, column/row expansion sample Column Width Auto-adjustment (CWA) replaces the familiar double click on the edge of column header to resize columns in spreadsheets. The CWA feature allows users to expand any column automatically with a single stylus click on a column header. At the same time the other columns will be reduced to accommodate the whole Overview table on the small screens. Of course, all cells remain active and can be expanded as needed. Users are allowed to expand multiple columns, and close an expanded column by a second click. The advantage of this feature is that users can expand an entire column by just one click. In the context of the small device, particularly in motion, users can be frustrated by attempting to double click on a small area twice without error using a stylus. Figure 3.3.13 shows a wide column expanded using CWA method and the effect of a user expanding two wide columns using CWA. One disadvantage of this method is that when multiple columns are expanded simultaneously, the required space may exceed the capacity of the screen. I Wain Form •II K36 ^p ly •• Main ¥mm £»4 42 3.3.5 Landmark Sometimes, users need to find a value in a table that they have already read. For example, a user may look up a maximum Credit value in the Hotel check-in table (Figure 3.3.3), and later want to know to whom that credit belongs. Landmarks let users highlight cells of interest by changing the background colour of the cell. This facilitates refinding, which occurs in many tasks. The prototype also has a table colour feature, which can be used for a variety of purposes. The landmark feature allows the user to focus their attention on a few marked up cells. This feature is designed to alleviate some of the pressure on short term memory, and can be anticipated to be useful for cell comparison, finding minimum or maximum values, or and cell refinding tasks. The use of different isoluminant colours lets the user identify particular cells for different purpose, as shown in Figure 3.3.14. For example, a user wants to find out the maximum credit and balance information within some particular cities of tenants. As shown in the second figure of Figure 3.3.14, the user can use landmark feature to mark the interested cities and/or corresponding figure of credit and balance, which assist the user to focus on particular cells and figure out the overall maximum value by comparing the marked values. Popup context "~\ Maximum value of Minimum value of menu for selecting _L credit within the marked balance within the landmark feature % cities (different colour) » £t <41 marked number Q 63 IB iandHaflrChei^ed ••] "_iJ 1 j Col Width Ad just Oct- ,-. n|-^^ - s •*, Sr Na Ge Crt Te C« Ch Ch 8a Ere UR Jk. Na Ge|c«;JTe few Ch Ch B EirjUR * • ••'£ F-i" ~c: (-1 52; '-'s '-'3 51. ~e htt' '•'a Fs- ~M"4i J;: Ks i-'s * , -i Ht' Zs* !\9 "•'; (-?'. Si- ;^r Jor si. ;ar> htt ;s' *-"s "o ;sc a'. Jvf Jt,- s c*. r-'s :\i-:,z: si- ; J- .~.U, si, e-; htt m c- '•^t~2'-2Z 41" -^ A-f' , ~* rtt : = cs- ;-'3 Aa (;•:: s:: 00 Oct *•;. :•<* M* c< "a iV3 0: sir c* Od s . >••» -n •-'0 =i- *--o (SI S^ .V ;„• =i. -; Hf ;-'c- re- "-•'; fS; skt Z^ -"LI- S . -c -tt: ™ st* ?*• f'.f ;V; si", -v -v. si. stf ftt! >!•« F--L Ju'lsi-... *~ J , 5tf m =t» ;-'s r-<3 iz: £11 Zs' ZM tZ. bra htt „ , t-?-2 htt „>-• p?' 9n< {il Sl> ip; Api St. k»d htt IT ?~- SM JS5 =:; -ici A5i|^;.-: -tt &• Ms :\s [2: SI! Ap' Ap> Si. i'r -tt EM Vs -.Va (2C ;:? .-«• .i;: SI, -p -tt ' 5s: f-'a M; ;S-~ i.1 '-*a Ap C. rs- 4t • y ntt r V &*' sJj"j,~ **' "» -n|3"' ' ^r ."...... ,*...* C...... r.£.S.„..~^.:....*„„....;:....».....~r* ^. hi*. ^, p.* .K.^.r....<.i-....r.;...:...;. ~.,™....^ «**... Select SubTaMe B * Select StibTabte Q ^ Figure 3.3.14 Landmark sample 43 3.3.6 Other Functions The following features were also included in the prototype. •MatoftorBt #m7S8 $ Figure 3.3.15 Menu and context menu sample Menu - Functions are available in a drop down menu which users can choose rather than a dedicated tool panel to save screen space for additional rows of data. Context menu - A list of functions pops up when the user lets the stylus stay on an object's name or icon for a brief time. Some functions for special use are listed in the context menu to save screen space. Font zoom-in/zoom-out - The Zoom-in and zoom-out feature allow users to adjust the font size very easily. An example was shown in Figure 3.3.5 3.4 SUMMARY In this chapter we have outlined the research questions for which we conducted user studies. We analyzed the commonly applied tasks and currently available features related to the table use on small devices. This chapter also provides the definition and description for new features developed in the prototype which we used in the subsequent studies. 44 The next chapter presents the series of user studies designed to test our hypotheses. The result analyses were conducted using data collected during each study. These analyses explore the characteristics of tasks using tables on the small screen device, as well as how the features affect the efficiency, effectiveness and preference of the users across tasks. 45 CHAPTER 4 SEQUENCE OF STUDIES Overall, the purpose of this series of studies is to explore design features for use with table information on small screens that support readability of the data, accuracy of access, and efficiency on the mobile device for information seeking tasks. We specifically test special features to alleviate the effect of the compression of cells using the Overview model on a small screen for scanning, comparison, finding, and refinding based on various levels of task complexity. 4.1 TABLE VIEWS STUDIES: STUDY 1.1 AND STUDY 1.2 Overview. Two user studies were conducted to compare efficiency, accuracy, and preference of the prototype Overview (OV) model with Default (DV) and Linear (LV) Views on the PDA. The first study, Study 1.1, Table Views Study, indicated that task complexity would be a factor affects the performance of users. We added two task complexity levels in the second study, Study 1.2, Table Views Study with Task Factor. For both studies, we measured efficiency by counting the number of actions, button clicks or scrolling actions, taken by the user to complete each task. We measured accuracy as the number of correct results of each task. In addition, we surveyed the users to determine their perception of ease of use and preferences they may have. Study Data. Both studies used the same sample data table, shown in Figure 4.1.1, and the three prototype systems discussed in Section 2.5. The reasons we chose numeric data table in these studies included: first, numbers are succinct and allow us to show as much content as possible without losing the meaning of the table; second, it is easy for users to understand comparison tasks using numeric data; at last, the numeric table is a very commonly used type of two-dimensional table. 46 Department 1993 1994 1995 1996 1997 1998 1999 2000 2001 Body Composition 2,335 2,430 2.425 2.44U 2,615 2,540 2,110 2.435 2.72u Cancer diagnostics 1.160 1.245 1,220 1,215 1.265 1,250 1,255 1,665 1/-90 Cancer therapy 4.040 4.175 4,120 4,115 4,035 4.060 4,135 4,140 4,070 Dental specialties 2,475 2,470 2,455 2,465,2,460 2,445 2,430 2,475 2,485 Dentistry 2,435 2,435 2.415 2,430^2,435 2.470 2,490 2.510 2,54M Food and Nutation 1.6S0 1.575 1.645 1,710 1.730 1.710 1,690 1.740 1.630 Medical specializations 1,500 1,430 1.550 1.390 1.425 1,335 1,335 \395 1.360 Medical technology 4,445 4,460 4.470 4.490 4.490 4,490 4,450 4.430 4,440 Medicine 1.745 1,855 1,635 1.600 1,660 1,615 1,475 1.315 1,250 Nursing 4,055 4.070 4,445 4,745-4,305 4,230 4,040 4.010 3.84U Occupational therapy 1.500 1.570 1,610 1,650 1,555 1,595 1,615 1,575 1,620 Optometry 1.115 1.110 1.105 1.100 1.110 1.115 1,105 1,120 1,100 Pharmacy 4.955 4,975 4,975 4.805 4,835 4,945 4,950 4,975 4.990 Physical therapy 1,590 L635 1.660 1.645 1,635 1.655 1,675 1.665 1.690 Psychology 1,225 1,060 1.230 1.125 1,050 1.675 1.455 1.175 1.330 Radiation Biology 4,560 4,135 4.645 4.345 4.430 4,420 4.745 4.650 4.330 Social work and welfare 2,840 2.935 2,750 2.990.2,995 3.110 2.955 2.310 3.155 Space Research 1,190 1,330 1.165 1,245 1,135 1,150 1,215 1.26U .1,145 Surgery and specialties 2.490 2,430 2,430 2,470 2,460 2,410 2,405 2,420 2.415 Figure 4.1.1 University enrollment sample table Hypotheses. For evaluation purposes, we compared the efficiency, effectiveness, and user preference for three table view models for tasks ranging from simple to complex in each study. The hypotheses were: Hypothesisi-i = The Overview model is the most efficient model (as measured by click times) compared with Default View and Linear View for the tasks from simple lookup to complex comparison. Hypothesis^ = The Overview model is the most effective model (as measured by correctness in completing the task) compared with Default View and Linear View for the tasks from simple lookup to complex comparison. 47 Hypothesisi-3 = The Overview model is the most preferred model (as measured by user's preference scores on a post experiment questionnaire) compared with Default View and Linear View for the tasks from simple lookup to complex comparison. As dependent variables, effectiveness is measured by how accurate the user is, in this case, the number of correctly completed tasks; efficiency is measured by how many actions were taken to complete the task; preference is measured in a post-experiment questionnaire. While users may expect that it may take longer to find information on a smaller display, they, nonetheless, expect accuracy in the results. The experiment was designed as a mixed factorial experiment with independent variables: expansion method, task complexity, and order of expansion methods. Task Types. Five task types were defined to represent a range of table oriented tasks, from simple lookups to more complex comparison tasks. Sample questions are presented in Table 4.1.1. Task Li, the simple lookup task, required the user to locate a cell in the upper left corner of the original table. Task L2, also a lookup, required the user to locate a cell in the bottom right of the original table, which would involve navigation through the table in all cases. The complex tasks involved both lookup and comparison activities on the part of the user. The first of these, task C\, required the user to locate two cells in different columns. Task C2 required the user to locate a cell and then scan that column of the original table to identify the minimum value. Task C3 is similar to C2 with the addition of locating and scanning a second column of the original table. Study 1.1 used only three of the five tasks. 48 Table 4.1.1 Task types and sample queries Task Task Type Sample Query Li Simple Lookup In 1994-1995, how many graduates were enrolled in Dentistry? L2 Scroll Lookup In 2000-2001 how many graduates were enrolled in Pharmacy? ct Complex Grid For the periods 1995-1996 and 1997-1998, which had the larger increase: Optometry or Physical therapy? c2 Complex Column In 1998-1999, which program had the fewest enrollments? c3 Complex Compare In the two periods 1996-1997 and 1999-2000, which program had the fewest students enrolled? 4.1.1 Study 1.1 - Table Views Study As the first study, we were interested in examining the choice of the model of table display for information seeking tasks on small device. The study examined effectiveness, efficiency and preference of three models: Default View (DL), Linear View (LV), and Overview (OV). Participants. Nine participants all familiar with PDAs were used in Study 1.1. This provides adequate statistical power in a repeated measure design [37]. Methodology The users, in blocks of three participants, were randomly assigned to one of three groups and given three tasks; the two lookup tasks (Li and L2) and the complex column task (C2). In this study we were interested in a general level comparison of the three views across straightforward tasks, without any overt training on any of the systems. To control for order of method we used the Latin square technique [37], which provides incomplete counterbalance, with three groups of three participants using different views to deal with different tasks. Each participant used all three views but used each view only once for one task. In this case, we did not consider the effect of task order and sequence. 49 Table 4.1.2 Mixed factorial table Factors Between-Subject Variable WithinSubjects Conditions Methods Methods Method (Mj - Default View, M2 - Task Complexity Task Complexity Linear View, M3 - Overview) Order of Methods Order of Methods Task Complexity (Lb L2, C2) Results Accuracy. The users correctly answered all of the Lj and L2 questions and the C2 questions using the Linear and Overview models but experienced difficulty in identifying the correct answer, for task C2 using the Default View (67%), shown in Table 4.1.3. This difficulty also showed up in the high number of actions taken by the users in trying to answer this question using the Default View. Table 4.1.3 Accuracy comparison between three techniques Average Simple Tasks Complex Tasks Default View 84.00% 100.00% 67.00% Linear View 100.00% 100.00% 100.00% Overview 100.00% 100.00% 100.00% Efficiency. The results, shown in Table 4.1.4, from Study 1.1 show that users were less efficient (measured by number of actions required to complete the task) using the Default View than either the Linear View or the Overview as the task became more complex. Furthermore the standard deviations for these means indicate that individual user abilities to cope varied considerably. As the task difficulty increased, the demands of scrolling in the Default View became problematic for the user. The efficiency of the Linear View and the Overview, however, were not significantly different over the conditions for the users. Table 4.1.4 Efficiency comparison between three techniques Mean number of actions TaskLi (SD) TaskL2 (SD) TaskC2 (SD) Default View 3.33(1.15) 17.00 (9.64) 38.33(18.77) Linear View 3.67(1.53) 3.67 (2.08) 10.00 (6.24) Overview 4.00(1.00) 4.33 (3.21) 10.00 (6.56) 50 Table 4.1.5 Univariate linear results - efficiency (study 1.1) Type III Sum Mean Partial Eta Source of Squares df Square F Sip. Squared Corrected Model 3127.407(a) 8 390.926 6.436 .001 .741 Intercept 2966.259 1 2966.259 48.835 .000 .731 task 1182.519 2 591.259 9.734 .001 .520 view 1112.074 2 556.037 9.154 .002 .504 task * view 832.815 4 208.204 3.428 .030 .432 Error 1093.333 18 60.741 Total 7187.000 27 Corrected Total 4220.741 26 a R Squared = .741 (Adjusted R Squared = .626) Using a univariate general linear model, as shown in Table 4.1.5, we found a strong indication that View, Task, and their interaction explain about 75% of the variation in efficiency (R2 of .74), with significant effect of view alone at the p=.002 level, task alone at the p=.001 level and combined view and task at the p=.030 level. All three contributed about the same amount to predict efficiency (partial eta squared varied from 0.52 to 0.43). Looking at the graph in Figure 4.1.2 we can see that DV is dramatically different from both the LV and OV for efficiency. No significant difference, however, was found for efficiency between the LV and the OV for these tasks. 45 40 35 jg 3D u = 25 10 5 0 L1 L2 C2 Tasks Figure 4.1.2 Efficiency graph for user study 1.1 The Student-Newman-Keuls test, used to compare post hoc pairs of group means for View and Task against efficiency, confirmed that task C2 was more difficult than either of the lookup tasks (Li and L2) for the users and that the efficiency for task C2 was 51 significantly lower than for either task L2 or task Li, at the p<.05 level, and no other differences were found. Preference. Users in this study, in which no training was given and only one question was asked for preference, preferred the Linear View (77.8%). We suggest that this is, at least partly, because it was most familiar to them. Discussion In this study, we found that the performance based on both accuracy and efficiency was very similar using the Linear View and Overview models, and no significant difference was found with the Default View model. Figure 4.1.2 shows the efficiency of all three models. The performance appears to be task dependent, because the effects increase while the task becomes more complex for all three models. Significance was, however, only found between task C2 and other two tasks (Li and L2). The study results indicated that task complexity is an important factor which may impact both effectiveness and efficiency. Therefore, additional complex tasks were introduced to Study 1.2. The results also indicated that a short training session might reduce the effect of view model familiarity. 4.1.2 Study 1.2 - Table Views Study with Task Factor We were interested in following up the differences found in Study 1.1. In particular, we were interested in whether these results would hold for the additional complex tasks, d and C3. Participants. In Study 1.2 we used 9 different participants, again all familiar with PDA use. Methodology In this study, we examined two effects; the effect of adding a short training session on the use of the three prototypes and the effect of varying the complexity of the tasks. Recall 52 that the algorithm used for the Linear View prototype creates sublists based on the values in the rows of the original table. We anticipate, therefore, that users would have more difficulty in answering questions based on comparing values found in columns in the original table as these would now be spread over sublists. Each user was given a short, five minute, introduction to the functionality of the three prototypes. We used all five tasks, i.e., the original three tasks from Study 1.1 plus two additional complex tasks (Q and C3, sample questions shown in Table 4.1.1). In this study, each user completed all fifteen conditions. Table 4.1.6 Mixed factorial table Factor Between-Subject Variable WithinSubjects Conditions Methods Methods Method (Mj - Default View, M2 - Linear Task Complexity Task Complexity View, M3 - Overview) Order Order Task Complexity (Ll5 L2, C{, C2, C3) Results Accuracy. The users performed more accurately, i.e., got more correct answers, using the Default View and the Overview (average accuracies 95.56% and 100%) than with the Linear View (average accuracy 84.46%). This is largely attributed to the difficulty users had with the Linear View for the complex tasks, but no significant was found based strictly on task complexity. The view had no significant effect on accuracy. Table 4.1.7 Accuracy comparison between three techniques Average Taskl Task2 Task3 Task4 Task5 Default View 95.56% 100.00% 88.90% 88.90% 100.00% 100.00% Linear View 84.46% 100.00% 88.90% 66.70% 88.90% 77.80% Overview 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% Efficiency. Table 4.1.8 shows the mean and standard deviations for the number of clicking and scrolling actions for each condition for the different tasks. We see again that 53 the average overall effort to complete the task increases with task complexity for all three views. All three views performed similarly for Task Li, the simple lookup task. For Task L2, we see that the users were the least efficient using the Default View. For complex tasks, Task Q and Task C3, the users performed better with the Default View than with the Linear View. This may be anticipated as the 2D structure of the table makes it easier to make comparisons of two columns across two rows, particularly Ci for which the two cells needed could be seen at the same time on the screen. For Task C2, which required a scan down a whole column, the users struggled with the Default View, compared to the other views. Although the effort to complete the tasks was shown to increase with complexity using the Overview model, the increase is quite small compared to the other views. That is, the user could perform a wider range of tasks using this view with little loss in efficiency. Table 4.1.8 Efficiency comparison for efficiency between three applications Mean Actions TaskLj(SD) TaskL2(SD) TaskCi(SD) TaskC2(SD) TaskC3(SD) Default View 3.33 (1.00) 7.78 (2.33) 5.89 (2.98) 21.22(6.48) 10.67 (4.58) Linear View 2.67(1.12) 3.22 (0.44) 10.11 (3.18) 6.89(2.15) 14.89 (8.59) Overview 3.33 (0.87) 3.67(1.50) 4.00(1.32) 5.33(1.00) 4.78(1.09) Table 4.1.9 Univariate linear results - efficiency (study 1.2) Type III Sum Mean Partial Eta Source of Squares df Square F Sip. Squared Corrected Model 3390.593(a) 14 242.185 20.732 .000 .707 Intercept 6969.630 1 6969.630 596.639 .000 .833 view 703.704 2 351.852 30.120 .000 .334 task 1252.963 4 313.241 26.815 .000 .472 view * task 1433.926 8 179.241 15.344 .000 .506 Error 1401.778 120 11.681 Total 11762.000 135 Corrected Total 4792.370 134 a R Squared = .707 (Adjusted R Squared = .673) The univariate linear results, shown in Table 4.1.9, indicate that the overall interaction of View, Task, and View by Task has an R square of .707. This indicates that two factors, View and Task, and their interaction explain about 70% of the variation in number of 54 clicks used. The view alone, task alone, and task by view taken together were all significant effects at the p=.00 level. All three contributed about the same amount to predict efficiency (partial eta squared varied from 0.334 to 0.506). The post hoc Student- Newman-Keuls test indicates a significant effect of view on efficiency (F=30.12, p=.00). The DV is significantly less efficient than the LV and the LV is significantly less efficient than OV at the p<.05 level. Similarly, task has a significant effect (F=26.82 with df of 4 and 120, p=.00) on efficiency. Over all three views, the efforts required to complete the complex tasks, Q, C2, and C3 were significantly higher than the effort for tasks Li and L2 (F=40.612 with df of 1 and 129, p=.00). The interaction of View and Task on efficiency was also significant (F = 15.34 with df of 8 and 120, p < .001). 25 -1 -•-DV ™*_OV 0 -I 1 1 1 r L1 L2 C1 C2 C3 Tasks Figure 4.1.3 Graph of efficiency for user study 1.2 Preferences. In Study 1.2, the participants were given a short introduction to the navigational features of the three systems. After the study the participants reported on their preferences using a short questionnaire. In this study, all of the participants preferred the Overview. At the same time, the participants reported that the Linear View was easy to use at least partly because of its wide availability and familiarity. Effect of Training. The short training session had a positive effect on accuracy and efficiency and preference between the Linear View to the Overview. 55 Discussion The efficiency of both the LV and the DV are task dependent while the efficiency of the Overview method, when the user is familiar with its functionality, is relatively independent of task (Figure 4.1.3). We speculate that, in the case of the DV, that this is the negative effect using the combination of horizontal and vertical scrolling to locate the relevant data. With the LV, we speculate that users experience different cognitive loads both as the tasks becomes more complex and when the data needed is not found in the same relative order as the original. For example, if the task is "/« the two periods 1993- 1994 and 1995-1996, which program had the fewest students enrolled?", the user had to manipulate multiple sublists to find data that existed in the columns of original table using LV, as shown in Figure 4.1.4 (a). The efficiency of the OV, on the other hand, changed little as the task became more complex, as shown in Figure 4.1.4 (b). There was no significant difference for efficiency for any of the five tasks using the OV. m Internet Explorer $t ^ 4:5? Mawfwin f £M£ **B 0 http://gatew3y.oceantaLe.com/nph •] ^ • cascade •] f^P^' Univera»y arid college gradual; (Uriiv«rt!f\* Pivcholoqy Ki^rmw^ ^ Using OV model, Ii8%-i3s?' 10,125 two columns will (199?. MS: 10.050 Using LV model, users be shown in the 3,6.75 have to scroll the same screen It9«-2ttO0. 9,255 vertical bar all the way which is easier for 12000-2001 8,175 y down to the bottom, users to finish the 12001-200 "3.030 pick up the right comparison. Social went- an* wd figures, and, in a mean (a) (b) Figure 4.1.4 Complex task sample using LV and OV 56 4.1.3 Discussion of Both Table View Studies Including an initial preview of the features of each view improved the performance of all three systems and, furthermore, changed the qualitative assessment of the methods, particularly the Overview. This is likely because the users were already familiar with many of the Default and Linear View features but were not familiar with the features of the Overview and consequently did not exploit those features in Study 1.1. The ability of users to get the correct answer was affected by task complexity and view. For simple lookups, like Task Li, users had no problem getting all of the correct answers. As the task became more complex, however, errors crept in that varied by view used. Users performed accurately (100%) for all tasks using the Overview. Using the Default View, which also maintains the 2-D structure of the original table, users were 100% accurate for Tasks Li, L2, and Ci but less accurate (85%) when comparisons were needed in C2andC3. Users had significant problems with the Linear View, for all of the complex queries. The effort, measured by the number of clicking and scrolling actions, also varied by view and task. Overall, the LV and OV techniques were better on the basis of efficiency for the user. As the task became more complex the effect of view on effort became clearer. The Overview technique gave the least variation over the five tasks while the DV and LV both exhibited irregular patterns related to the individual task. The DV became problematic whenever the user needed to scroll to find data cells and the potential for the user being lost in the table was a real possibility. The LV became problematic when the tasks involved data that were located in a column in the original table and so required multiple finds to complete the task. This could be addressed by allowing the user to switch the transformation algorithm as needed but one could argue that this would then depend on the user understanding the relationship of the table decomposition to the task at hand. The Overview technique, which maintained the 2-D row and column structure of the table was shown to be the most robust of these methods. The results of these two studies provide support for many of the hypotheses presented in Table 4.1.10. Of the six hypotheses, four were confirmed in favor of the Overview model. 57 The first "JVO" in Table 4.1.10 means that there were no significant differences in efficiency found between Overview and the best performing method, Linear View, for the three tasks in Study 1.1. In Study 1.2, the Overview model was found to be the most efficient. We suspected that the reason for this difference is because we added more task complexity level in Study 1.2, and as shown by the study results, Linear View is not as stable as Overview over variations in task complexity. The second "NO" means that Linear View was chosen as the most preferred model in Study 1.1. LV is a commonly used model for table display on small screen devices and so it is understandable for users to prefer the most familiar model. But in Study 1.2, the Overview is the most preferred model. We provided a short training session before the study, which had a positive affect on users' preference. Table 4.1.10 Hypotheses summary Hypothesis Confirmed Study 1.1 - 3 Tasks Hj! Overview is the most effective model for 3 tasks study Yes Hj 2 Overview is the most efficient model for 3 tasks study No H13 Overview is the most preferred model for 3 tasks study No Study 1.2-5 Tasks Hu Overview is the most effective model for 5 tasks study Yes Hj 2 Overview is the most efficient model for 5 tasks study Yes Hj 3 Overview is the most preferred model for 5 tasks study Yes Summary of Results of Studies 1.1 and 1.2 No significant difference was found for effectiveness in either Study 1.1 or Study 1.2. The results on efficiency have variations. In Study 1.1, a strong indication was found that View, Task, and their interaction have significant effects for view alone at the p=.002 level, task alone at the p=.001 level and combined view and task at the p=.030 level. Furthermore considering task, the efficiency for task C2 was significantly lower than for either task L2 or task Li, at the p<.05 level, and no other differences were found. 58 For Study 1.2, the results indicate that the view alone, task alone, and task and view taken together all were significant effects on efficiency at the p<.001 level. A further look into the significant effect of view on efficiency, indicates that the DV is significantly less efficient than the LV and that the LV is significantly less efficient than OV at the p<.05 level. Similarly, over all three views, the effort required to complete the complex tasks, Ci, C2, and C3 was significantly higher than that needed for tasks Li and L2 at p=.00 level. The interaction of View and Task on efficiency was also significant at p < .001 level. Table 4.1.11 Summarization of significant results Factor Effect On Study Factor P Value View 0.001 Task Complexity 0.002 Study 1.1 Task Complexity C2vs(LbL2) <.05 View * Task Complexity 0.030 View 0.000 Efficiency DV vs LV <05 View LV vs OV <.05 Study 1.2 Task Complexity 0.000 Task Complexity (C2,C2andC3)vs(L1,L2) 0.000 View * Task Complexity <.001 4.1.4 Conclusions for Table View Studies 1.1 and 1.2 From these two initial studies, we conclude that the Overview technique is the best of the three models tested for table display on mobile device, and we use this method in the future studies. The poor performance of the common Linear View under some conditions would indicate that designers should be careful in using that transformation technique if the anticipated tasks include tasks that are directly related to the 2D structure of the original table. Alternatively, designers could consider providing the users with an option to select dynamically when and how any linearization should be achieved, i.e., by row or by column. Of course, this would assume that the users could map the requirements of their task to the organization of the lists. Finally, the users in the second study were able to 59 benefit from a very spare instruction and were willing to exploit new features to complete tasks both effectively and efficiently. 4.2 STUDY 2: BROWSING METHOD STUDY - COLUMN EXPANSION VS. CASCADE Overview. According to the result of previous study, Overview was chosen as the most robust model for use in developing table display features for mobile devices. The Overview model, however, presents its own difficulties in finishing tasks efficiently and effectively using the compressed table. Column expansion methods are commonly used features on regular sized computers to quickly expand the width of columns in a table. Simple column expansion, however, may be problematic while using on small screen device because most small screen devices use touch screen as their display screen. It is relatively more difficult and more error prone for a user to drag the small column header separator line using a stylus than using a mouse. In this study, we introduce a new feature, Cascade, and compared it with the standard column expansion feature for various tasks on the small screen device. Lookahead Cascade. The Cascade feature automates a sequential expansion of cells moving either down a column or across a row. It is started with a single click on the header of a row or column. The full content of each cell is displayed in a popup overlay one by one from the first to the last. This feature lets users scan a row or a column without committing to Column expansion and without repeated interaction. Figure 4.2.1 shows a row Cascade and Figure 4.2.2 shows a column Cascade. When the Cascade moves outside the range of cells on the screen, the popup stays at end of the row or column with a different background colour, as shown in Figure 4.2.3. 60 MafitFwm f«H$*l*OI|- ><*»Forra f «? Hi *S» Oll'^ «W»F«o> f j+>&*£» Q 0 >::, •J „ J J ^ „ AKK A^K a t G--— iiS2Sr '_ taf li.i |4I idi IOI ii,i !•;. 1 .. ), U UpJ 1»| IS*| 19l| 19«| 19<| i-»«| 19<|aOC *- f-r -7 kto}; ~ -- i7: •' * a _T t • *» ;-i;->.~>-;-s-;-*;j;4 :,4, i i :•! _v :.-. :- :.4 ;.- : - : J ;• - :- i*..- -i -:-)--: -* • 4,? 4.3: 4,: '<->HHHHB'* •'." -• - -* - •'• - 4 - » 3i < 1 «l 1 f» Figure 4.2.1 Sample table Figure 4.2.2 Sample table Figure 4.2.3 Cascade cell with row Cascade with column Cascade outside the screen Column Expansion. The user uses the stylus to click on a header and drag to adjust column widths or row heights to focus on those units, as shown in Figure 4.2.4. Hi'il. ri'* mfysra c S3 fjc^csce Ml 'J i # C: % • Toe's -,.•'• a? j Un!r»ersitf 19< 1994-1995 191 11' w -5.-I ^^5 f 5.; :;,-,:;": :•:. :; > EJ tSS£, m in I = -.-j j.-- : -; ;.;7: : i ; " >:---sv. 1.4 : ~JS : - : - ! 7~-'.i ;.~ ;,-"*: ; - : > '•-;.::•'« :,7 :.?SS :,~ *. ; '•'-3-c*; i5:-3: :.;;.: ;„-=- »,- ;,u ;4j; 2,4 ;,- *..•••"; -4 : -4.:;"; - -*• - :* r i < { « | j * S 1 Figure 4.2.4 Sample table with 2 columns and 1 row expanded 61 4.2.1 Study 2 Experimental Design Study Data. This study used the same "University enrollment table sample " as Study 1.1 and Study 1.2, shown in Figure 4.1.1. We decided to choose the numeric data table to keep the information in each cell consistent. Hypotheses. For evaluation purposes, we compared the efficiency, effectiveness, and user preference for Column expansion and for Cascade for a set of row oriented lookup and comparison tasks on the PDA. Row oriented tasks were chosen because of the inherent difficulty in examining multiple cells in a given row on the small screen. The hypotheses were: Hypothesis2-i = When the user's task requires a comparison of values within one row, the Cascade method is more efficient (as measured by time) than the Column expansion method. Hypothesis2-2 = When the user's task requires a comparison of values in one row, the Cascade method is more effective (as measured by correctness in completing the task) than the Column expansion method. Hypothesis2s = When the user's task requires the comparison of values in one row, the Cascade method is preferred by the user (as measured by user's preference scores on a post experiment questionnaire) than the Column expansion method. As dependent variables, effectiveness is measured by how accurate the user is, in this case, the number of correctly completed tasks; efficiency is measured by the time taken to complete the task (using a stop watch to record the lapsed time); preference is measured in a post-experiment questionnaire. While users may expect that it may take longer to find information on a smaller display, they, nonetheless, expect accuracy in the results. The experiment was designed as a mixed factorial experiment with independent variables: expansion method, task complexity, and order of expansion methods. 62 Task Types. Two task types were defined to represent a range of table-oriented tasks, from simple lookups to more complex comparison tasks. Sample questions are presented in Table 4.2.1. Task T1-3, simple lookup tasks, required the user to locate a cell in the table. The complex tasks T45 involved both lookup and comparisons, that is, users need to locate and scan one row of the original table to identify the minimum or maximum value. In another words, complex tasks were defined to require more steps and more short-term memory load in order to finish a task. Table 4.2.1 Task types and sample queries Task Task Type Sample Query T13 Simple Lookup In which year, were 1,120 students enrolled in Optometry? T4,S Complex Compare In which year, did Dentistry have the lowest enrolment? Participants. A total of ten participants completed the study, which provides adequate statistical power in a repeated measure design [37]. The students were all graduate students in either Computer Science or Management, none of whom were experienced PDA users. Methodology The five tasks for each user generate 50 data points for each method. Each participant completed two blocks of five tasks each, one block per method, using a working prototype on a working PDA. To control for order of method we used the Latin square technique [37], which provides incomplete counterbalance, with two groups of five participants using different methods to deal with different tasks. The participants were randomly assigned to one of two groups. One group completed the first block of tasks using the Cascade method and the second block of tasks using the Column expansion method while the other group did the blocks in the opposite order. Five minutes were allocated for introduction and practice at the beginning of each block so that the participants were familiar and comfortable with each method. In this case, we do not consider the effect of task order and sequence. 63 The order of tasks did not vary. It was felt that simple tasks should precede complex tasks. In this way, if there were a learning effect, it would reduce the effect of the complex tasks, which makes the study conservative in looking for a complex task effect. Table 4.2.2 Mixed factorial table Factor Between-Subject Variable WithinSubjects Conditions Methods Methods Method (Mj - Column Expansion Method Task Complexity Task Complexity and M2 - Cascade Method) Order of Methods Order of Methods Task Complexity (Tb T2, T3, T4, T5) 4.2.2 Study 2 Results Accuracy. Overall both methods resulted in average accuracy of 85% correct (84% accurate rate using Column expansion method and 86% using Look Forward Cascade), shown in Table 4.2.3, with no significant difference (F=0.205, p=.651). There was no significant difference between Cascade and Column expansion on either simple tasks or complex tasks. Table 4.2.3 Accuracy Average Simple Tasks Complex Tasks Column Expansion 84.00% 90.00% 75.00% Lookahead Cascade 86.00% 86.67% 85.00% Efficiency. The results for efficiency are shown in Table 4.2.4, where the standard deviations for these means indicate that the ability of individual users to cope varied considerably. Using a univariate general linear model, as shown in Table 4.2.5, we did not find significant difference either between Cascade and Column expansion method (F=0.895, p=.347), or differences within task complexity (F=3.188, p=.077). There was however a significant effect for method by task complex interaction in efficiency (F=4.677, df 1 and 96, p=.033). The method by task complexity interaction only contributed small amount to predict efficiency (partial eta squared = .046). 64 Table 4.2.4 Comparison for efficiency between two methods Simple Tasks Complex Tasks Mean Time (Sec) Tj (SD) T2(SD) T3(SD) T4(SD) T5(SD) Column Expansion 53.0 (30.5) 32.3 (12.3) 36.9(13.8) 68.4 (30.9) 54.6 (22.6) Lookahead Cascade 59.4 (30.6) 48.0 (30.0) 34.0(15.2) 36.8(14.4) 53.5 (28.9) where SD is Standard Deviation Table 4.2.5 Univariate linear results - efficiency (study 2) Type III Sum Mean Partial Eta Source of Squares df Square F Sip. Squared Corrected Model 5404.507(a) 3 1801.502 2.713 .049 .078 Intercept 227020.402 1 227020.402 341.904 .000 .781 method 594.015 1 594.015 .895 .347 .009 complex 2116.882 1 2116.882 3.188 .077 .032 method * complex 3105.375 1 3105.375 4.677 .033 .046 Error 63742.883 96 663.988 Total 296581.000 100 Corrected Total 69147.390 99 a R Squared = .078 (Adjusted R Squared = .049) As shown in Table 4.2.4, there was a significant practice effect for the simple tasks, within Ti, T2, and T3 (F=3.85 with df 2 and 54, p=.027), but not for the complex tasks. That is, for the simple lookup tasks, the user's performance improved as they gained experience in either method. In another words, the overall time needed to complete the simple tasks decreased with each additional simple task. For the more complex tasks, T4 and T5, participants performed better using the Cascade method than for the Column expansion method (F=4.257 with df 1 and 36, p=.046). We attribute this difference to the time required for the users to repeatedly expand the columns one by one in order to make comparisons of cells within a row. Considering the complexity of tasks and ignoring order effect, there was no significant difference between the Column expansion method and the Cascade method on either the simple tasks or the complex tasks. On the other hand, there was a significant difference in efficiency between the simple and complex tasks using the Column expansion method 65 (F=8.886 with df 1 and 48, p=.005), while no significant difference was found using Cascade method, as shown in Table 4.2.6. Table 4.2.6 Efficiency between two methods in different task complexity Mean Time (Sec) Simple Tasks Complex Tasks Column Expansion 40.73 61.5 Lookahead Cascade 47.13 45.15 Preference. After the study, the participants reported their preferences using a short questionnaire. Participants indicated their preference for method across the following categories: general preference, ease of use, time to find the required information, easier to use while the user stays still, and easier to use while the user is walking or distracted. The internal consistency of this scale had a Cronbach alpha of .728. This level justifies adding the participants' scores across all five questions [42]. The mean preference score across all users and all five questions was in favor of the Cascade method. This rating was significantly above the scale midpoint (t=2.63, p=.025). Valuable comments were provided by the participants, for consideration in future work, several are shown in Table 4.2.7. Table 4.2.7 Valuable comments examples "Double clicking of the column header can expand the whole Example 1 column ... better for column expansion." Example 2 "Clicking on a cell, both headers should be highlighted." Example 3 "For comparing task, cell highlighting would be helpful." Example 4 "Cascade should be able to start from halfway." 66 4.2.3 Discussion for Study 2 The results of Study 2 provide support for the hypotheses presented in Table 4.2.8. Of the five hypotheses, three were confirmed in favor of the cascade method and the other two showed no significant difference between the methods. For effectiveness, from Table 4.2.8, we see that no significant differences were found between the methods for simple tasks. We suspect this reflects that less interaction was required for simple tasks when using the Column expansion method, and so users could more easily focus on the information they needed. The Cascade method did, perform more effectively than the Column expansion method for complex tasks. There was no significant difference found for efficiency for simple tasks using Column expansion or Cascade method. The Cascade method, however, did perform better in terms of efficiency for complex tasks. We attribute this difference to the time required for the users to repeatedly expand the columns one by one in order to make comparisons of cells within a row. Complex tasks may require more column expansions than simple tasks, which means more interactions, therefore, more time. The time cost using the Cascade method, however, was relatively consistent because a timer was set to control the speed of cell content expansion. Finally, we see that the users preferred the Cascade method over the Column Expansion method overall. Table 4.2.8 Hypotheses summary Hypothesis Confirmed Effectiveness H21 Cascade is more effective for simple tasks No H21 Cascade is more effective for complex tasks Yes Efficiency H2 2 Cascade is more efficient for simple tasks No H2 2 Cascade is more efficient for complex tasks Yes Preference H23 Cascade is preferred over Column Expansion Yes 67 From this study we can conclude that for complex tasks, the Cascade method was more accurate, more efficient and more preferred when compared to the more common Column expansion method when accessing an Overview table on the small screen. In a post hoc examination of the data, we found a significant order by method interaction effect for the simple task data (F=7.11, p=.017, r|2= .308). When the participants used the Column expansion technique first they were significantly less efficient using the Column expansion method than the column expansion scores for participants who used the Cascade method first. We speculate that the Cascade technique provides a better initial view of the data in a column for users who are not familiar with the table. This knowledge carried over to subsequent uses of the table allowing the user to derive benefit from the Column expansion functionality. Several suggestions from the users in this study led to improvements in the next version of the prototype including introducing, landmarks, Cascade pause, and single click column resizing. Landmarking, shown in Figure 4.2.5, allows the user to highlight cells especially for comparison purposes. ; ".{Wain form SMffltas # Q Csscs-* *J t: i__ a: * a-**--; "\ * SBr Ha Gejot Chjtt Ba|Cr< Te|trr 08 * '.•'3 Fi- ""'..< r-'s h'i ji, ;2; (4; r s t-tl ;.-.r va r-':- Ji.s Au: si, Si; tec ;;* *•» ¥ '-'o ~f .'•':> Ju» :u<\ |S2' ;'S: i-o rtt Iti =e* M. %; As-. »;. ;;•! ..:: »te r>tt; Bra Ki KD Z3* Jar fi, S3, l>2 bra r-tt> Se> Ki ••'-.- '•'* -xj j;?:;?: b»t "tt' A" =«- vs; i-il 5«: SI, SZ: i?" 2- rtt ;s- :-*s -3: ist c-ci £2, s;> ;as ;s- i-t? ^ Ff- -:ai Ocl Oct s;, S?; {35 h-l rtt - Ji.: Ka rtal De: De| |s3' :'fi ;us t-tt _p . Set ect SuWabte 1 Figure 4.2.5 Landmarking feature Figure 4.2.6 shows the result of the automatic column resizing feature that allows the user with a single stylus click to have a designated column automatically resized so that that 68 column is full size and the others are distorted to accommodate. Of course, all cells remain active and can be expanded as shown. Mainlwiw £f 4i 15b4? 0 O ED 5VJ C?HC3Je i •«MM MM I WW. "tt^ .<•. catena.ix- l&l, ' - ; ^^:' 'V. ** ?"^^^,C^^" s ; > ' ' • i -EC* •. a. -; -;••• ! ' I ' • ; ' ' t*c' ,'.^.,,i>-fo't ..}- . s . ' ' i ! >-iU::''V<*.*.A.vdr!i,c.:- -UiiiiLLJ^^^h;.!:.i.;jr. Select SubTaWe Figure 4.2.6 Single click automatic column expansion Significant Points Summary No overall significant difference was found for effectiveness in this study. There was a significant effect on efficiency of method by task complex interaction at the p=.033 level. A significant practice effect using both methods was found only for the simple tasks at the p=.027 level. For the more complex tasks, a significant difference was found using the Cascade method over using the Column expansion method at p=.046 level. When we consider each method, there was a significant difference in efficiency found between simple and complex tasks using the Column expansion method at the p=.005 level. Table 4.2.9 Summarization of significant results Factor Effect On Factor P Value Method * Task Complexity 0.033 Simple Tasks Practice effect 0.027 Efficiency Complex Tasks Cascade vs Column expansion 0.046 Column expansion Method Simple vs Complex tasks 0.005 69 4.2.4 Conclusions of Study 2 In this repeated measures study, we were able to show significance differences between the Cascade method and Column expansion method for the complex tasks using Overview tables on a small screen. We suggest that users would benefit from both methods of accessing such tables on small screens. For simple tasks, little difference was detected but as the task became more complex, an advantage was found for the scanning potential of the Cascade method. In addition, users may benefit from using the Cascade technique to initially explore a new table and benefit from Column expansion on subsequent tasks. 4.3 STUDY 3: EXPANSION METHOD STUDY - COLUMN EXPANSION COMPARISON Overview The use of column expansion features to provide full access to cell data in compressed tables on small devices, although useful, may be problematic for users for the following reasons. First, the use of column expansion to explore the data values of one column distorts the current Overview by forcing the collapse of other columns to accommodate the expansion. Second, some columns are simply too wide to be viewed without horizontal scrolling. Furthermore, it is often difficult on a small screen for a user to expand more than one column, as often needed for the comparison of values, where the columns are both wide. Finally, the limitations of stylus use on small devices make column expansion frustrating and/or inaccurate at times especially when the user is mobile. At the same time, column expansion is necessary for most complex tasks. Column Expansion Methods In this study, we examine two specific column expansion techniques for use on Overview tables on small screens, shown in Figure 4.3.1; the column width auto-adjustment (CWA) method allows users to expand one column with a single click in the column header and the more familiar manual column width expansion (CWE) method allows the user to 70 adjust the width of a column using the stylus. We anticipate that the CWE technique may cause some frustration with the users on the small screen as dexterity is required to drag the stylus accurately within the confines of the column header to complete the expansion. The CWA method requires only a single click of the stylus in the column header. The CWA, however, adjusts the column to accommodate the widest cell value automatically. Microsoft's Pocket PC version of Excel provides similar functionality for spreadsheets where the user double clicks the column header to expand a column automatically. Automatic column width adjustment may introduce problems related to the limited space confines of the PDA screen if the user opens more than one column at the same time, such as shown in Figure 4.3.2. One could conjecture that the manual CWE method may be preferred for tasks using multiple columns. Furthermore, users may require fine grained adjustments of individual columns for particular tasks. i Wain farm ^4|&SS $ in Main Form #^«C&5f ;c'34-343 SI' „\r .V Si, ;ar bft : 647-522= 52: C-n S-ci 52 :K c^~3« (2C;:.F,47-4:;:- SI- 3u'f iu; Si, Cri W :;6SS'9?;-5 ;2< Zij- ;*•> si "OnC3.£ir' „ (2;S;i47-E2ii 52: C-d Oct S2, z-i rtt :;s22-ss&j S2! Air ^b; SI itetrsrss ;?:c:ss.$-s?3« $2; „v- x^ s:, -.- t-tt :::ss-ss'.c; S2: Jsr „V S2 D"3 ;£.«"• 3 C;;:;s;;-5S£.3 S2! &„; Au Si, sts HI ::44S'S:;c sir Aw Ap si .r 33 .$.?** a ;3c;):6s-53:: SZ: Jar Jar $2, fc'3 >*tt !^4?-=s;'E SH AW At; Si e;TOc.j'c™3s (5r,-45 sc:: ;:; ips ipi«;, J-; f-tt :, 7)i -5.-23 S2: ••',» %-i $2 T> ! (r::;!--»7-'=E"= Si! 4;; Ap« SI, «: Ht _ .:~3i-121i Si: Au: Sci Si : •V r :'i:':;?3s-5523 S2: !- 2 ~p £2, bsf htt ^SDS-1659 12:it;>?:jl 8-, S«*»- " _«i kfe. 4 1yp Sete