Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Designing adaptive interfaces for children : a preliminary study on the effect of age and gender on children’s… Rajamanickam, Mohan Raj 2011

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2012_spring_rajamanickam_mohanraj.pdf [ 4.94MB ]
Metadata
JSON: 24-1.0052139.json
JSON-LD: 24-1.0052139-ld.json
RDF/XML (Pretty): 24-1.0052139-rdf.xml
RDF/JSON: 24-1.0052139-rdf.json
Turtle: 24-1.0052139-turtle.txt
N-Triples: 24-1.0052139-rdf-ntriples.txt
Original Record: 24-1.0052139-source.json
Full Text
24-1.0052139-fulltext.txt
Citation
24-1.0052139.ris

Full Text

Designing Adaptive Interfaces for Children: A Preliminary Study on the Effect of Age and Gender on Children’s Interaction in the Context of Dialoguing with Computers  by Mohan Raj Rajamanickam  B.E., CSE, Coimbatore Institute of Technology, 2002  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Computer Science)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) December 2011  © Mohan Raj Rajamanickam, 2011  Abstract Software targeted at children does not typically take into consideration the significant variation in skills and capabilities across age and gender. The overall goal of our research was to design adaptive interfaces that change to accommodate the inherent age and gender differences among children. We conducted two studies towards this goal at Science World with 195 children between ages 3 to 12. In the first exploratory study, we observed how 111 children interacted with Tux Paint, a painting application designed for children, and the difficulties they encountered in general. We were also interested in the possibility of accelerating children‟s learning of the interface with the least help from adults. Hence, we observed how they used the help system and how they learned by watching their peers. We found that designing an effective help system for children was a tricky proposition fraught with challenges. As for our inquiry into the general difficulties, we identified that dialogs were a significant source of problems for children. We classified the problems with dialogs by age groups and set out to solve them with potential design solutions targeted at three different age groups. In the second observational study, we observed how 84 children interacted with our various dialog box designs embodying 8 design factors. The dialog boxes were designed with the goal of enabling efficient communication of information; children need to understand the information that is communicated and make informed decisions. We found that while some design factors helped achieve effective communication, some did not. We present our results and an analysis of children‟s information consumption behavior, especially with respect to age and gender differences, in the context of their interaction with dialog boxes. We put forth theories and present models on how children of different age and gender consumed ii  information differently from different information channels (textual and non-textual). We discuss the design implications of our findings that could help designers in constructing adaptive interfaces that improve the interaction by taking the age and gender into account.  iii  Preface The studies described in this thesis were conducted with the approval of the UBC Behavioural Research Ethics Board (BREB) certificate number: H10-01618.  iv  Table of Contents Abstract .................................................................................................................................... ii Preface ..................................................................................................................................... iv Table of Contents .................................................................................................................... v List of Tables ......................................................................................................................... xii List of Figures ...................................................................................................................... xvii List of Abbreviations ........................................................................................................... xxi Acknowledgements ............................................................................................................. xxii Dedication ........................................................................................................................... xxiv Chapter 1: Introduction ........................................................................................................ 1 1.1  Research goals ...................................................................................................... 3  1.2  Overview ............................................................................................................... 3  1.3  Thesis contributions .............................................................................................. 4  Chapter 2: Related work ....................................................................................................... 6 2.1  Understanding how children learn ........................................................................ 6  2.1.1  Why gender matters ...................................................................................... 8  2.1.2  Designing help information systems ............................................................. 8  2.2  Dialoguing with the computer ............................................................................ 10  2.2.1  How users consume information ................................................................ 12  2.2.1.1 Effects of literacy level and age ............................................................. 13 2.2.2  Designing for effective communication...................................................... 15  2.2.2.1 Highlighting ........................................................................................... 16 2.2.2.2 Contrast .................................................................................................. 17 v  2.2.2.3 Comics and icons ................................................................................... 19 2.3  Summary ............................................................................................................. 22  Chapter 3: Study 1 - Observing children’s interaction with computers......................... 24 3.1  Methodology ....................................................................................................... 24  3.1.1  Participants .................................................................................................. 25  3.1.2  Setting ......................................................................................................... 25  3.1.3  Methods....................................................................................................... 26  3.1.3.1 Data analysis .......................................................................................... 27 3.2  Findings............................................................................................................... 27  3.2.1  Dialog boxes present numerous problems for children .............................. 28  3.2.1.1 Design by age ......................................................................................... 29 3.2.1.1.1 Pre-literate: Ages 3 to 5 .................................................................. 29 3.2.1.1.2 Semi-literate: Ages 6 to 7 ............................................................... 30 3.2.1.1.3 Literate: Ages 8 to 12 ...................................................................... 32 3.2.1.2 Problems with dialogs ............................................................................ 32 3.2.2  Watching peer interactions accelerated learning ........................................ 34  3.2.3  Gender differences and others observations ............................................... 34  3.3  Discussion ........................................................................................................... 38  3.3.1  Challenges of conducting studies in a public space .................................... 38  3.3.2  Challenges in communicating information ................................................. 39  3.3.2.1 Desired independence of interaction ...................................................... 39 3.3.2.2 Patience declines with experience ......................................................... 40 3.3.2.3 Consuming information passively ......................................................... 41 vi  3.3.2.4 Summary of challenges .......................................................................... 42 3.4  Design implications ............................................................................................ 43  3.4.1  Minimizing hindrance ................................................................................. 43  3.4.2  Improving affordances ................................................................................ 44  3.4.3  Improving communication .......................................................................... 44  3.4.4  Addressing impatience ................................................................................ 45  3.4.5  Safer arbitrary choices ................................................................................ 46  3.4.6  Audio........................................................................................................... 46  3.4.7  Is this a problem worth solving? ................................................................. 47  3.5  Summary ............................................................................................................. 48  Chapter 4: Designing effective dialogs ............................................................................... 50 4.1.1  Improving affordances ................................................................................ 52  4.1.1.1 Split structure ......................................................................................... 52 4.1.2  Facilitating safer arbitrary choices .............................................................. 52  4.1.2.1 “I don‟t know” button ............................................................................ 53 4.1.2.2 Delayed-click ......................................................................................... 53 4.1.2.3 Switching button positions ..................................................................... 54 4.1.3  Improving communication .......................................................................... 55  4.1.3.1 Visibility of title and body text .............................................................. 55 4.1.3.2 Visibility of title ..................................................................................... 56 4.1.4  Indicating consequences ............................................................................. 56  4.1.4.1 Color coding of buttons ......................................................................... 57 4.1.4.2 Highlighting the safest button ................................................................ 58 vii  4.1.5  Addressing impatience ................................................................................ 59  Chapter 5: Study 2 - Investigating the impact of different design factors on children’s interaction with dialog boxes........................................................................................... 62 5.1  Methodology ....................................................................................................... 62  5.1.1  Software ...................................................................................................... 63  5.1.2  Participants .................................................................................................. 64  5.1.3  Methods....................................................................................................... 64  5.1.4  Reading test ................................................................................................. 65  5.2  Findings............................................................................................................... 65  5.2.1  Improving affordances ................................................................................ 66  5.2.1.1 Split structures did not improve affordances ......................................... 66 5.2.2  Improving communication .......................................................................... 66  5.2.2.1 Visibility of body text and title superfluous to communication ............ 66 5.2.2.2 Level of difficulty of dialogs – Not all dialogs were created equal ....... 67 5.2.3  Indicating consequences to improve decision making................................ 68  5.2.3.1 Color coding was partly successful at indicating consequences ............ 68 5.2.3.2 Highlighting – the runner up to color coding......................................... 69 5.2.4  Making arbitrary clicks safer ...................................................................... 70  5.2.4.1 Delayed-click made it safer at a cost ..................................................... 70 5.2.4.2 Switching position provides mixed results on children‟s interaction .... 71 5.2.5  Addressing impatience ................................................................................ 72  5.2.5.1 Impatience could not be addressed by skimming .................................. 72 5.2.6  Other results ................................................................................................ 73 viii  5.2.6.1 Children exhibited symptoms of learning as the session progressed ..... 73 5.2.6.2 Differences in reaction times ................................................................. 74 5.2.6.3 Interaction with split structure ............................................................... 74 5.2.7 5.3  Summary of findings................................................................................... 75  Discussion ........................................................................................................... 76  5.3.1  Challenges with conducting field studies with children in the wild ........... 77  5.3.1.1 Intention: an elusive measurement......................................................... 77 5.3.1.2 Parents‟ involvement ............................................................................. 78 5.3.1.3 Delivering help....................................................................................... 78 5.3.1.4 Fitts‟s law ............................................................................................... 79 5.3.2  Differences in interaction across age and gender ....................................... 79  5.3.3  Two phases of information processing ....................................................... 80  5.3.4  Towards a model for information consumption by age .............................. 82  5.3.4.1 Role of position and content .................................................................. 82 5.3.4.2 Mechanisms for information consumption across age groups ............... 83 5.3.5  Interaction of information channels ............................................................ 85  5.3.6  Text as contextual cue................................................................................. 86  5.3.7  Formulating a level of difficulty for a dialog.............................................. 87  5.4  Design implications ............................................................................................ 88  5.4.1  Designing for gender................................................................................... 88  5.4.2  Designing for age ........................................................................................ 88  5.4.3  Designing for children in general ............................................................... 90  5.5  Summary ............................................................................................................. 91 ix  Chapter 6: Conclusions and future work .......................................................................... 93 6.1  Contributions....................................................................................................... 93  6.2  Limitations .......................................................................................................... 94  6.3  Future work ......................................................................................................... 95  6.4  Closing comments ............................................................................................... 97  Bibliography .......................................................................................................................... 99 Appendices ........................................................................................................................... 109 Appendix A Mockups of potential design solutions for effective dialogs ................. 110 A.1  Improving affordances .............................................................................. 110  A.2  Illustrating consequences .......................................................................... 110  A.3  Safer arbitrary choices .............................................................................. 112  A.4  Addressing impatience .............................................................................. 112  Appendix B Additional details of the second study ................................................... 113 B.1  Narrowing down of design choices........................................................... 113  B.2  Why these three dialogs? .......................................................................... 114  B.3  Notes for interpreting the data .................................................................. 115  B.4  Iterative evolution of design factors ......................................................... 116  B.5  Getting to know the data ........................................................................... 118  B.6  Data pre-processing .................................................................................. 119  B.7  Reaction times across genders and age group........................................... 120  B.8  Support for skimming ............................................................................... 123  B.9  Split structure ............................................................................................ 127  B.10  Controlling user‟s actions – delayed-click ................................................ 128 x  B.11  Color coding of buttons ............................................................................ 129  B.12  Highlighting the safe option...................................................................... 133  B.13  Switching the button position ................................................................... 137  B.14  Preference for a choice in terms of its position and how it is worded ...... 143  B.15  Visibility of body text and title ................................................................. 147  B.16  Reading level ............................................................................................ 151  B.17  Reading level analysis on dialog text........................................................ 153  B.18  Learning effect .......................................................................................... 156  Appendix C Call for participation poster displayed at Science World....................... 163 Appendix D Consent form for parents and legal guardians ....................................... 164 Appendix E Assent form for children aged 3 – 6 years.............................................. 167 Appendix F Assent form for children aged 7 – 12 years ............................................ 168  xi  List of Tables Table 1 Three groups of children categorized according to their age. The categorization was based on observations made in the current study and supported by literature from developmental psychology. ................................................................................................ 29 Table 2 Summary of the various problems with dialog boxes by age groups. The two shades of green denote varying intensity levels of the problem while gray denotes that it is not a significant problem area. .................................................................................................... 33 Table 3 Summary of the observed interaction differences between girls and boys. ............... 35 Table 4 Design factors manipulated in the study, their respective rationale and problems addressed. ........................................................................................................................... 51 Table 5 Distribution of the participants of the study across genders and age groups. ............ 64 Table 6 Readability scores from the three reading tests performed on the three dialogs. ...... 67 Table 7 Summary of how different age groups and gender interacted with the design factors. ............................................................................................................................................ 75 Table 8 List of the choices considered during prototyping and reasons as to why we chose some over others. ............................................................................................................. 114 Table 9 Details of the three dialogs used in the study along with their intended disruption level. ................................................................................................................................. 115 Table 10 Significance levels of the differences in RT for the purpose of the discussion of results. .............................................................................................................................. 115 Table 11 Summary of the evolution of design factors through the course of the study ....... 117 Table 12 Break up of the number of participants by gender across all ages. ....................... 118 Table 13 Distribution of gender among participants in the study. ........................................ 119 xii  Table 14 Distribution of the participants of the study across genders and age groups. ........ 119 Table 15 Activity of participants as measured by the number of dialogs faced and the average length of the session during the study, split across genders. ............................................ 120 Table 16 Activity of participants across age groups. The most active group is highlighted. 121 Table 17 Activity of participants across age groups and gender. The most active group is highlighted........................................................................................................................ 121 Table 18 Average RTs of the children across genders. ........................................................ 121 Table 19 Average RTs across age groups. ............................................................................ 122 Table 20 Average RT across age groups and gender. ........................................................... 122 Table 21 Effect of factors to support skimming across age groups and gender. .................. 124 Table 22 Interaction of factors to support skimming and delayed-click. ............................. 125 Table 23 Interaction of factors to support skimming and dialog structure. .......................... 125 Table 24 Interaction of skimming factors and dialog structure across age groups. .............. 125 Table 25 Interaction of factors to support skimming, delayed-click and dialog structure. .. 126 Table 26 Effect of the presence of icons in buttons on RT. .................................................. 126 Table 27 Effect of dialog structure on RT across age groups and gender. ........................... 127 Table 28 Interaction between dialog structure and delayed-click. ....................................... 127 Table 29 Interaction between dialog structure and delayed-click across genders. ............... 128 Table 30 Effect of delayed-click on RT. ............................................................................... 128 Table 31 Effect of delayed-click on RT across age groups. ................................................. 128 Table 32 Interaction between delayed-click and type of response clicked. .......................... 129 Table 33 Effect of button color coding on RT across genders. ............................................ 130 Table 34 Effect of button color coding on RT across age groups. ....................................... 130 xiii  Table 35 Interaction between button color coding and response type. ................................. 131 Table 36 Interaction between button color coding and type of dialog. ................................ 131 Table 37 Interaction between button color coding and type of dialog across age groups and gender. .............................................................................................................................. 131 Table 38 Effect of button color on RT across genders. ........................................................ 132 Table 39 Effect of button color on RT across age groups and genders. ............................... 132 Table 40 Interaction between button color coding and button highlighting. ........................ 132 Table 41 Effect of highlighting on RT. ................................................................................. 133 Table 42 Effect of highlighting on RT across genders. ........................................................ 134 Table 43 Effect of highlighting across age groups. .............................................................. 134 Table 44 Interaction between highlighting and type of response. ........................................ 135 Table 45 Interaction between highlighting and response text clicked. ................................. 136 Table 46 Interaction between the different colors used for highlighting and color coding of buttons. ............................................................................................................................. 136 Table 47 Effect of switching the button position on RT. ...................................................... 137 Table 48 Effect of switching the button position on RT across age groups and genders. .... 137 Table 49 Interaction between switching of button positions and the type of response across genders. ............................................................................................................................ 139 Table 50 Interaction between switching of button positions and the type of response across age groups. ....................................................................................................................... 139 Table 51 ................................................................................................................................ 140 Table 52 Interaction between switching of button positions and response text clicked. ...... 140 Table 53 Interaction between switching of button positions and button color coding. ........ 141 xiv  Table 54 Interaction between switching of button positions and highlighting. .................... 141 Table 55 Interaction between switching of button positions and button color coding across genders. ............................................................................................................................ 141 Table 56 Interaction between switching of button positions and button color coding across age groups. ....................................................................................................................... 142 Table 57 Interaction between switching of button positions and visibility of body text and title.................................................................................................................................... 142 Table 58 Count of the buttons clicked per side of the dialog box from B3, B4, and B5. ..... 143 Table 59 Count of the buttons clicked per side of the dialog box from B2. ......................... 143 Table 60 Effect of switching button positions on the count of buttons clicked per side. ..... 144 Table 61 Effect of the interaction between dialog type and switching of the button positions on the count of the responses per side of the dialog......................................................... 144 Table 62 The different dialogs and their corresponding response texts divided according to the side in which the responses were placed on the dialog. ............................................. 144 Table 63 Count of the responses sorted in descending order. The safer responses are highlighted in green. ......................................................................................................... 145 Table 64 Count and percentage of the responses clicked across age group. ........................ 145 Table 65 Interaction between the switching of button position and response text across age groups. .............................................................................................................................. 146 Table 66 Interaction between the switching of button position and response text across genders. ............................................................................................................................ 147 Table 67 Information on the text contained in the three dialogs along with their title ......... 148 Table 68 Effect of the visibility of body text and title on RT. .............................................. 148 xv  Table 69 Effect of the visibility of body text and title on RT across genders. ..................... 148 Table 70 Effect of the visibility of body text and title on RT across age groups. ................ 149 Table 71 Interaction between visibility of body text, title and button color coding. ............ 150 Table 72 Interaction between visibility of body text, title and button color coding across age groups. .............................................................................................................................. 151 Table 73 Reading level scores of the children against their corresponding average RTs. ... 152 Table 74 Readability scores of three tests on the text of the three dialogs. .......................... 153 Table 75 Average RTs across dialogs. .................................................................................. 153 Table 76 Average RTs of dialogs across age groups. ........................................................... 154 Table 77 Maximum and average number of dialogs across genders. ................................... 156 Table 78 Maximum and average number of dialogs across age groups. .............................. 156 Table 79 Effect of the sequence with which the dialogs appeared on RT across genders ... 157 Table 80 Effect of sequence on RT across age groups. ........................................................ 159  xvi  List of Figures Figure 1 Articles displayed with varying contrast levels reflecting their importance scores in a news reader using the PostRank plug-in. ........................................................................ 18 Figure 2 Our study setup at Science World showing the laptops and printer we used for the study along with the poster (resting on an easel) that was used to invite participants to the study. .................................................................................................................................. 25 Figure 3 The sticker that was handed out to children at the end of the session. ..................... 27 Figure 4 Examples of dialog boxes from Tux Paint that pose several problems to children. 28 Figure 5 Dialog boxes in Tux Paint with just one choice. ...................................................... 37 Figure 6 Interface of Tux Paint consisting of the prominently displayed help section. ......... 41 Figure 7 A split dialog without the factors to support skimming. .......................................... 52 Figure 8 Dialog with the third “I don‟t know” button. ........................................................... 53 Figure 9 A split dialog revealing its options in sequence. The buttons are revealed in the clockwise direction............................................................................................................. 54 Figure 10 The Quit dialog with the button positions switched (shown both with and without button color coding). .......................................................................................................... 55 Figure 11 The Quit dialog with the button positions switched and without title or body text (shown both with and without button color coding). ......................................................... 55 Figure 12 The Erase, Quit and Save dialog with title and body text. ..................................... 56 Figure 13 The Quit dialog being displayed without title or body text. ................................... 56 Figure 14 The Save dialog without the body text but with just the title. ................................ 56 Figure 15 The Quit dialog without button color coding (shown both with and without title and body text). .................................................................................................................... 57 xvii  Figure 16 The Erase dialog with the safer option highlighted in yellow, disruptive option coded in red, without body text, with the button positions switched. ................................ 58 Figure 17 The Quit and Erase dialogs with the safer options highlighted in green. Quit dialog is also color coded while Erase is not. ............................................................................... 58 Figure 18 The Quit dialog with the safer option highlighted in yellow (shown both with and without color coding). ........................................................................................................ 58 Figure 19 The Erase and Quit dialogs with the safer option highlighted in yellow. .............. 59 Figure 20 Dialog without any of the factors to support skimming. ........................................ 60 Figure 21 Dialogs containing factors to support skimming - emphasizing key words using varying levels of contrast in the body text, with icons inside buttons, and a clear title with an accompanying icon. ....................................................................................................... 60 Figure 22 Children participating in the study, using our instrumented version of Tux Paint software. In the picture on the right, a research assistant can be seen at the far end assisting a child. ................................................................................................................. 63 Figure 23 Reliance on different sources of information to varying degrees by different age groups. ................................................................................................................................ 84 Figure 24 Call-out style dialog box originating from the relevant icon. .............................. 110 Figure 25 Two way conversational dialog between the computer (represented by Penguin, Tux paint‟s mascot) and the user. .................................................................................... 110 Figure 26 Visual illustration of the consequences of the choices with the mouse pointer hovering over “Yes, Save it”. ........................................................................................... 111 Figure 27. Visual illustration of the consequences of the choices with mouse pointer hovering over “No, Don‟t bother” ................................................................................................... 111 xviii  Figure 28 Mock up of button icons conveying potential consequences using images and typography........................................................................................................................ 111 Figure 29. Gradually revealing information and choices to solve some problems with dialogs. Dialogs also have the persistent „Back‟ button. ............................................................... 112 Figure 30 Mockup of a dialog with multiple layers to support skimming such as the title with an icon and body text with varying contrast levels. ......................................................... 112 Figure 31 Histogram showing the distribution of the age of the children who participated in the study. .......................................................................................................................... 118 Figure 32 Average RTs across age groups and gender. ........................................................ 122 Figure 33 Effect of factors to support skimming across gender. .......................................... 123 Figure 34 Effect of factors to support skimming across age groups..................................... 124 Figure 35 Interaction between button color coding and button highlighting. ....................... 133 Figure 36 Effect of highlighting across age groups. ............................................................. 134 Figure 37 Effect of switching the button position on RT across age groups. ....................... 138 Figure 38 Effect of the visibility of body text on RT across genders and age groups. ......... 150 Figure 39 Interaction between reading level and button color coding (left), highlighting (right)................................................................................................................................ 152 Figure 40 Interaction between reading level of children and the different dialogs. ............. 155 Figure 41 Effect of the sequence with which the dialogs appeared on RT across genders. . 158 Figure 42 Effect of sequence on RT across age groups. ....................................................... 160 Figure 43 Interaction between the sequences with which the dialog appeared and button color coding. .............................................................................................................................. 161 Figure 44 Interaction between the sequence with which the dialog appeared and visibility of xix  title.................................................................................................................................... 161 Figure 45 Interaction between the sequence with which the dialog appeared and visibility of body text. .......................................................................................................................... 162  xx  List of Abbreviations RT. Reaction Time, the time elapsed between the dialog appearing on the screen and a choice on the dialog being clicked upon.  xxi  Acknowledgements I thank my supervisor Professor Joanna McGrenere for guiding me over these two years. I am grateful for her kindness, understanding and support. Many thanks to Dr. Charlotte Tang for bringing this thesis to its current form, emphasizing readability and structure, patiently through a number of reviews. I thank Professor Gail Murphy for being the second reader. Special thanks to Science World for allowing us to conduct our studies on their premises and Mila Cotic for being our friendly point of contact in Science World. I thank the following UBC graduate students who volunteered and helped me run the studies at Science World: Yaojie Chan, Rongrong Wang, Asad Kiyani, Pearl Siganporia and Swiya Nath. I thank members of the HCI grad forum for providing feedback on initial prototypes. I thank Alex Garnett, Vincent Levesque, Noreen Kamal, and Zoltan Foley-Fisher for participating in peer design evaluations and interviews and helping me with their frank feedback. I thank Matthew Brehmer for providing me bits of feedback all along the course of my research that turned out to be both useful and important. I thank Professor Ron Rensink for providing feedback and pointers on some of the design prototypes, Professor Linda Siegel for helping me in choosing a reading assessment test for the study, Professor Janet C Read for advising me about my research, Professor Andrea Bunt for helping me during data analysis and Professor Andrew Baron for sharing his experience in conducting studies at Science World. I thank the following graduate students for sharing their knowledge about regression analysis: Hind Sbihi, Weina Zhou, Oliver Schneider, Emtiyaz Khan, Pearl Siganporia and Rongrong Wang. xxii  I am thankful to the Tux Paint developers who answered my questions on the mailing list especially when I had problems instrumenting the source code of Tux Paint. Thanks to all the innumerable open source projects, whose labor I cherish and enjoy every working day. Thanks to Jen Fernquist, Rock Leung, Oliver Schneider and Vincent Levesque for lending a patient ear. Thanks to all my peers in MUX lab. This two year journey started with the recommendation letters from Krishnakumar Gopalakrishnan (Manager at Google), Dr. Arunkumar Balakrishnan and Dr. R Prabhakar (Professors at my undergraduate university). I owe them my gratitude. Thanks to all my friends and family who helped me hang through this ride - it‟s been a tough one.  xxiii  Dedication  To the Divine love which surrounds me in the form of family, friends and others.  xxiv  We as designers of new technologies for children, sometimes forget that young people are not “just short adults” but an entirely different user population with their own culture, norms, and complexities. - Allison Druin, 2002 xxv  Chapter 1: Introduction Computers are playing an important role in almost all aspects of our modern society; parents believe that it is important to provide children with access to computers and governments are spending increasing portions of their educational budgets on technology (Shields & Behrman, 2000). Children have embraced technology and are spending increasingly more time on computers and at increasingly earlier ages (Rideout, Foehr, & Roberts, 2010). More parents are using computers as free baby sitters (Budiu & Nielsen, 2010) and many children are therefore left to use computers on their own, without any guidance from adults. This however has been found to affect children‟s learning (Klein, NirGal, & Darom, 2000). Hence interfaces that help a child to interact independently without adult assistance should be valuable to both the child and the parent. In addition, children do not always benefit from or prefer adult guidance. Children, especially those under age 10, have a strong desire to perform well and accomplish tasks on their own (Lueder & Rice, 2008) and this sense of mastery is very important for them from the developmental perspective (Miserandino, 1996; Holt, 1991). Thus different learning strategies might be required depending upon children‟s age. Computers are increasingly being used to educate children, given their potential to improve learning by tailoring the content according to the capabilities of the children. However, in practice software designed for children pay limited attention to individual differences (Lepper & Gurtner, 1989). By contrast, teachers of classrooms that include multiple age groups effectively facilitate children‟s learning by taking into account the individual developmental needs, even though it is acknowledged as a challenging task 1  (Stone, 1995). Designing usable software and hardware systems for children is more challenging than designing for adults, as children constitute a moving target in terms of their capabilities (Hourcade, 2008; Gelderblom & Kotzé, 2008). Hence, despite the tremendous differences between children of age 3 and 12 in terms of physical, cognitive and other abilities (Lueder & Rice, 2008), designers often create software for children considering them as a single homogenous group. It is thus not surprising that children, especially the younger ones, encounter difficulties when they interact with software by themselves. There is a dearth of empirical and theoretical research on the area of underlying processes, such as learning strategies employed by children and how they direct their attention to different aspects of the software (Lepper & Gurtner, 1989). A review of literature reveals that most of the interaction design research with children is of an applied nature, highlighting the need for conducting more basic research aimed at deeper understanding of children‟s evolving abilities as they age, and development of guidelines based on how this evolution affects children‟s interaction with technologies (Hourcade, 2008). It is therefore important to study children‟s use of computers from a developmental perspective as it provides us with knowledge to make computers more developmentally appropriate rather than merely user friendly(Yan & Fischer, 2004). Thus in order to design usable software for children, several characteristics that govern children‟s behavior such as their age and gender should be taken into consideration (Shneiderman, 1986; Sax, 2005; Druin, 2002). Therefore in this research we set out to empirically investigate and establish guidelines on the design of developmentally appropriate and usable interfaces, taking factors such as age and gender into account. This research entails investigation in the fields of developmental psychology and human computer interaction, more specifically, child 2  computer interaction.  1.1 Research goals Our long-term research goal is to design usable interfaces that can adapt to children‟s age and gender. We seek to address the research question of how to design adaptive interfaces that will change and adapt according to a child‟s age and gender in order to make it more usable. As a first step towards achieving this goal, we strive to learn about the differences in interaction patterns among children of different ages and gender in this research. We conducted two studies with children towards this goal. The goal of the first study was to observe how children of different ages and gender interact with software and the general difficulties they face. The goal of the second study was to evaluate the design solutions aimed at addressing the difficulties identified in the first study, in order to come up with age and gender specific design implications.  1.2 Overview Chapter 2 presents prior work that is related to this research bringing together literature from developmental psychology on how children learn and how learning is affected when adult guidance is scarce, and the need to present information effectively. Work on designing user interfaces for effective communication of information is presented followed by a discussion on how we consume information and how it is affected by age, literacy and information abundance. The importance of understanding the crucial role of gender in designing usable systems for both genders is also touched upon. Chapter 3 presents an exploratory study conducted with 111 children between ages 3 to 12 in Science World aimed at providing preliminary knowledge about designing adaptive 3  interfaces and intelligent help systems for children. We identify the challenges in designing an effective help system. We identify several problems that children faced with a specific UI component - dialog boxes. We classify the problems by age groups and present the design implications. Chapter 4 presents our various design solutions motivated by the findings of the exploratory study (Chapter 3). In total we explore eight different design factors aimed at solving the different problems that children faced with dialog boxes. Chapter 5 presents a follow-up observational study conducted with 84 children between ages 3 to 12 at Science World, in which we studied children of different age groups interacting with dialogs containing our various design factors. We report how these design factors impact children‟s interactions by age and gender. Further, we propose theories on how children process information from dialogs, the existence of different channels of information (textual and non-textual), how children of different age groups and gender consume information differently from these channels and the effect of interaction of these channels on the behavior of children. We discuss the design implications of our findings. Chapter 6 concludes this thesis with an elaboration of our contributions, limitations of this research, and identifies avenues for future research.  1.3 Thesis contributions The research presented in this thesis makes the following contributions: 1. Identifies dialog boxes as one of the significant interaction challenges faced by children and classifies the challenges according to different age groups (Chapter 3). 2. Identifies design factors (Chapter 3) and presents design solutions (Chapter 4) to enable effective dialogue between children and computers taking children‟s age into  4  account. 3. Realizes the effects of different design factors on different age groups and gender (Chapter 5). 4. Proposes preliminary theories about how children consume information on dialogs differently depending on their age and gender (Chapter 5).  We expect that our contributions will provide the foundation for more controlled studies which will verify or challenge our observations and will provide more elaborate insights into the impact of age and gender on children‟s interactions with computers. Collectively, such research should enable us to design software that adapts itself taking into account the inherent differences among children across age, gender, and other such factors.  5  Chapter 2: Related work The overarching goal of this research is to design adaptive user interfaces which change to match children‟s competency levels, as characterized by age, thus making the software more usable. A secondarily goal is to design interfaces that help children interact independently, with the least amount of adult help. Specifically, this research focuses on designing dialogs to enable effective exchange of information between the computer and children of diverse age groups and gender. This requires bringing together knowledge from several areas. First, we review the literature gathered from developmental psychology on how children learn, the role of adults in children‟s learning, the scarcity of adult guidance and how it affects their learning. We then discuss ways to make up for the lack of adult guidance by designing software that effectively communicates information. We also touch upon the importance of understanding the role of gender research in designing usable software for both genders. Second, we present existing limited work on designing dialog boxes to facilitate exchange of information followed by a discussion on designing user interfaces for effective communication of information. This discussion involves research about how we consume information and how the problem of information abundance has changed our information consumption behaviors and the effects of age and literacy on interaction and information consumption. We conclude this chapter by presenting different methods to increase the communicability of information such as highlighting, using contrast and alternate representations such as icons.  2.1 Understanding how children learn Children learn by tweaking and playing with objects in their environment, primarily by 6  mimicking adults and other (older) children who seem to be doing it right (Tizard & Hughes, 1984; Holt, 1991). They prefer to perform tasks without intervention. In fact, letting children perform the tasks they undertake with as little adult help as possible is crucial to the development of their sense of mastery of the world which in turn is paramount for learning (Holt, 1991). In addition, children often strive for perfection in performing even the seemingly trivial tasks (the need for which adults do not usually understand) and experience pride filled moments after accomplishing a task. This behavior is important for the learning process (Holt, 1991). Ability is necessary but not sufficient for learning. More importantly, the basic psychological needs of competence1 and autonomy need to be satisfied for positive affect, engagement and subsequent learning. Competence is being effective in one‟s interactions with the environment and autonomy is being self-determined and having choice and control over an activity (Miserandino, 1996). Autonomy as described by Miserandino can be likened to the sense of mastery as defined by Holt. Recognizing the role of competence and autonomy in children‟s development (Miserandino, 1996), our research set out to examine the relation between children‟s competence, as characterized by their age, and their interaction patterns. Thus our research goal is to design user interfaces which change to match children‟s competence levels and also help them interact autonomously with as little help from others as possible. As a first step towards achieving this goal, we examine the effect of some factors, such as age and reading level that may determine competency of children‟s interactions with a computer.  1  While ability is the capacity or skill to perform a task, competency is the ability to perform it effectively.  7  2.1.1 Why gender matters The developmental curve varies among genders. Brain areas relating to language and fine motor skills mature earlier in girls (by six years) while brain areas involving targeting and spatial memory mature earlier for boys (by four years) (Hanlon, Thatcher, & Cline, 1999). These gender differences are larger and hence more important for children than for adults for whom the brain has attained full maturity. By ignoring the inherent gender differences and putting both girls and boys through uniform learning systems we are rendering disadvantage to both and thereby helping to reinforce the gender stereotypes rather than breaking them down (e.g., stereotypes in math and computer education) (Sax, 2005). Hence it is of crucial importance to understand the inherent differences among genders and design our software for children based on those understandings. But society in general and the design community in particular has failed largely to take into account the inherent gender differences when designing for children. In our studies and subsequent analysis, we have given importance to understanding the role of gender in the observed interaction patterns which should enable us to design better software for both genders.  2.1.2 Designing help information systems Children‟s abstract reasoning with respect to the technological domain is improved when their interactions with computers are mediated by adults (Klein, Nir-Gal, & Darom, 2000). But adults do not typically sit with and help their children when they are using computers. They instead use computers as free baby-sitters (Rideout, Foehr, & Roberts, 2010). But children are not better off for lack of adult guidance; their competence suffers. For example, a study showed that children who did not usually receive mediation from adults with 8  computers at home had internalized a trial-and-error approach to solving problems without conceptualization; thereby forfeiting the full advantage of technology (Klein, Nir-Gal, & Darom, 2000). This internalization without conceptualization could affect other spheres of children‟s learning. Thus although children prefer performing tasks on their own and value their independence (Holt, 1991), the absence of adult help and guidance when they are using computers affects their learning. It is thus important that help be available to children when they need assistance with the computer. Perhaps digital technology can be designed to make up for the lack of human guidance. By modeling on dialogs and interactions that occur in the real world during learning (and thereby building on the innate tendency among humans to interact socially), digital artifacts such as animated puppets could be used to interact with users and enhance learning curiosity (Shaw, LaBore, Chiu, & Johnson, 2004). Well-designed software that effectively communicates helpful information could help children gain autonomy by reducing their reliance on adult help, which is not readily available in many instances, and eventually achieve competence. Complexity in software is ever increasing (McGrenere & Moore, 2000) and hence software that can accelerate children‟s learning of the interface and discovery of features becomes very important, especially given the general lack of adult help (Budiu & Nielsen, 2010). While there is good potential for a well-designed help system, in practice, children do not read help pages voluntarily. This behavior is not for lack of need; rather children are looking for precise solutions to the problems that they are currently facing, but are provided with general explanations instead (Large & Beheshti, 2005). Adults exhibit similar behavior (Nielsen, 2000; Krug, 2000). In addition, young children find it difficult to comprehend help information in the way it is presented, which is predominantly text. It has been suggested that 9  help should instead be delivered using a movie that shows someone performing the related action. Therefore, help should ideally be both context sensitive and presented in a format that is appropriate for the children (Large & Beheshti, 2005). Displaying context-sensitive inline help in both text and video formats has been found to enable adult users to overcome their learning difficulties, explore more features of the software and demonstrate better retention of the help information (Grossman & Fitzmaurice, 2010). Alternatively, software that allows one to learn from their peers has been found to improve learning. A community based recommender system technology that helped adult users discover new functions based on their interaction history yielded better learning of the software (Matejka, Li, Grossman, & Fitzmaurice, 2009). Given the potential learning benefits offered by software that communicates help information effectively, specifically a help system that provides a video replay of peers‟ interaction history, we examined the role of such a help system in our exploratory study presented in the next chapter.  2.2 Dialoguing with the computer Though the previous section focused on help systems, the discussion can be generalized to a class of systems designed to communicate helpful information to the user, such as dialog boxes. Dialog boxes are GUI widgets that allow exchange of information between the user and the computer. Interactive software relies on dialog boxes to communicate information to and receive inputs from the user. They present information that demands the user‟s attention along with various choices that lead to different outcomes. But it has been observed that interacting with the computer via dialog boxes is an experience prone to fatigue and 10  frustration, even for adults as it disrupts their attention and breaks their workflow (Cooper, Reimann, & Cronin, 2007). In addition, dialogs can often be ineffective. For example, error messages are typically delivered using dialog boxes. But users dislike error messages and so they often simply click on a button to dismiss the dialog. Thus error messages do not often prevent users from making further mistakes (Cooper, Reimann, & Cronin, 2007). Over time, users risk becoming inured to dialog boxes and routinely dismissing them without sufficiently processing the information on them (Cooper, Reimann, & Cronin, 2007). Users do not typically wait till their mental models are well formed before they can interact comfortably with a system. Instead they tend to focus on words/phrases that match the task at hand and current context (Krug, 2000). They take the satisficing approach by choosing the first reasonable option, without weighing in all the options and deciding on an optimal one (Klein G. , 1998) and most users tend to go with default choices (Johnson, 2003). Users also prefer to interact by trial and error, viewing their approach as good enough as long as they get some useful results, resorting to available help information only after they have failed in their trial and error approach (Krug, 2000). Users also tend to get confused if the distinctions between the various choices are not clear enough (Krug, 2000). A dialog‟s design could be improved in many ways to mitigate these problems. A dialog should have a clear title – preferably a verb summarizing the function of the dialog (Cooper, Reimann, & Cronin, 2007), thus making it easier to identify the purpose of the dialog. There should be a consistently labeled button that dismisses the dialog without performing any action (Cooper, Reimann, & Cronin, 2007). While there is paucity of research on designing dialog boxes for adult users, to our knowledge there has been no research on designing dialog boxes for children. Therefore in 11  the following section we will review literature covering the broader area of information consumption patterns among users and how to design user interfaces that effectively communicate information taking these into account. From these we will identify the design features and principles that can be applied to designing better dialog boxes. However, since the research in this area has been conducted predominantly with adult users, we are not sure if the identified design features and principles are also applicable to children. We expect that our research will contribute towards filling this gap.  2.2.1 How users consume information We live in an age of information explosion (Davenport & Beck, May 2001; Berghel, 1997) and our information consumption behaviors have evolved accordingly. Consuming information has become a rapid activity and it has shifted from intensive consumption from few sources to extensive consumption from many sources (Krug, 2000; Carmody, 2010). In fact, the amount of information consumed is reported to be inversely proportional to the amount of information available from the source (Nielsen, 2003). Thus the way in which the information is presented should facilitate extensive consumption, for example by facilitating cursory reading (Weinreich, Obendorf, Herder, & Mayer, 2008; Krug, 2000; Nielsen, 1997). Users also tend to develop a variety of mechanisms to optimize their information consumption. For instance, users have been found to start reading a page thoroughly but resorting to more scanning as time passes (Nielsen, 2006) and to ignore noisy objects such as pop-up ads and banner advertisements (Porter, 2010). Several personalized mechanisms have been developed to deal with the information overload problem by making important and relevant information more accessible to users such as social filters (Hill & Terveen, 1996), aggregators (Gulli, 2005) and personalized 12  recommender engines (Shardanand & Maes, 1995). Children also seem to be afflicted by the problem of information overload. The amount of time that children spend online has tripled in the last decade (Rideout, Foehr, & Roberts, 2010). Subsequently, while many children were willing to read instructions a decade ago, they have now become reluctant to read text including help instructions, like adults, besides wanting instant gratification when consuming information (Budiu & Nielsen, 2010). Thus we believe that research on presenting information more effectively to children is important, particularly since children often interact with computers on their own. Interfaces that help children to receive more of the communicated information and make informed decisions independently should be valuable. 2.2.1.1  Effects of literacy level and age  Information consumption and interaction behaviors have been known to change depending on various factors such as the age, literacy level and technological expertise. The literacy levels of children vary greatly within the same age. A five-year range in terms of variation of literacy related skills and corresponding functioning in a classroom is not uncommon (Riley, 1996). The information consumption patterns of adult users with low literacy levels differ dramatically from that of literate users (Nielsen, 2005). For example, lower-literacy users typically do not scan text – they read through it and when they cannot they skip sections instead of skimming them. They also tend to satisfice, i.e. interact based on very little information considering it to be good-enough. Adult users with low literacy levels were able to appreciate software agreements enhanced with graphic design and typography better than others with higher literacy levels (Campbell, Goldman, Boccia, & Skinner, 2004). Children‟s prior experience with technology has been reported to be the main predictor 13  for their ability to interact with software (Budiu & Nielsen, 2010). The technological expertise of adult users also plays an important role in their success at interpreting alternative representations of information such as icons (Baecker, Small, & Mander, 1991). Certain interface elements such as the „Back‟ button in the browser were not used by younger children, but were relied upon by older children (Budiu & Nielsen, 2010), suggesting differences in mental models based on increasing technological expertise. Further, technological expertise has been known to influence learning patterns (Yan & Fischer, 2004). Age related differences in interaction patterns have been observed (Druin, 2002). Using images has been found to accelerate learning to a greater extent in older children (9 – 10 years) than in younger children (4 – 8 years) (Pressley, 1977), suggesting differences in learning methods. A study of children between ages 3 to 12 showed that the youngest children (3 to 5) did not read for lack of ability, children of the middle age group (6 to 8) read tentatively, while older children (9 to 12) scanned the text (Budiu & Nielsen, 2010), suggesting development of adaptive strategies as children‟s capabilities change and are exposed to more information. Individual differences in terms of preferences for icons, characters, colors, animation, etc. are greater among children than adults (Large & Beheshti, 2005). Teenagers (age 13 to 17) have different needs from both adults and younger children. They are characterized by dramatically lower patience levels when compared to adults, tend to mark down sites that look glitzy and tend to dislike anything that seems to be designed for kids. Hence separate designs are needed for teens and younger children (Nielsen, 2005). Older children, in general, tend to be acutely aware of age differences and they tend to react unfavorably to designs perceived as being targeted at younger children (Budiu & Nielsen, 2010; 14  Baumgarten, 2003). To accommodate the differences across ages the interface should support personalization. This increases the appeal across age groups and genders as the presentation or stylistic aspects of an interface are the most age and gender sensitive (Large & Beheshti, 2005). A review of literature reveals that robust guidelines, heuristics, rules and usability principles which are present when it comes to designing for adults are generally absent when it comes to designing technology for children. Even among existing research on children, Gelderblom & Kotzé observe that sufficient differentiation between age groups is absent as is a comprehensive solution covering all phases of design (2008). „Designing for children‟ as a single group is hence not appropriate; children should be separated into groups and designs be targeted at them (Budiu & Nielsen, 2010). Children have been grouped broadly into three groups by developmental psychologists (toddlerhood 0-2, early childhood 2-7 and middle childhood 7-14) based on developmental milestones and capabilities (Lueder & Rice, 2008). Although design research with children has been receiving more focus (IDC, 2011), it has not changed substantially from what Gelderblom and Kotzé observed. To address this we classified children into three different age groups (Table 1) based on our observations and support from literature, and tailored our designs at the groups individually to account for the differences across ages (Budiu & Nielsen, 2010; Campbell, Goldman, Boccia, & Skinner, 2004).  2.2.2 Designing for effective communication The explosion of information has rendered human attention as one of the scarcest resources of modern times (Davenport & Beck, May 2001). A good design should manage or guide user‟s attention by presenting the information appropriately (NASA, 2011). While performing repeated visual search tasks (e.g., searching for the relevant button to dismiss a 15  dialog) users can restrict their attention to a subset of items that they have found to be relevant to the search task (e.g., title, keywords in body text and buttons of a dialog) and subsequently learn to guide attention to specific locations and away from irrelevant stimuli which constitutes a performance advantage (Kunar, Flusberg, & Wolfe, 2008). How is a subset of items identified and focused upon? Contextual cueing is a perceptual mechanism driven by incidentally learned associations between spatial configurations (constituting the context) and target locations which can guide attention towards aspects of the scene that are relevant to the task (Chun & Jiang, 1998). Contextual cues have been learned within the narrow time frame of an experimental session and without explicit awareness of the cues. Contextual cues are based on interactions between memory and attention rather than being based on perceptual or attentional processes alone (Chun & Jiang, 1998). From our review of literature we have identified the following ways to build contextual cues and guide children‟s attention to a subset of items, thus improving the effectiveness of communication. 2.2.2.1  Highlighting  Modulations to the luminance channel have been used to guide the user‟s gaze about a scene without disturbing their visual experience (McNamara, Bailey, & Grimm, 2009). One such modulation that can be performed on the luminance channel is highlighting of selective regions. Highlighting of text has been known to attract attention and facilitate learning; consequently highlighting has been recommended as a mechanism to direct user‟s attention towards the relevant information and to aid skimming (Chi, Gumbrecht, & Hong, 2007). Highlighting can also be achieved using motion and by blurring irrelevant information which is termed as semantic field of depth (Ware, 2000).  16  2.2.2.2  Contrast  Contrast is one of the design elements that could impart an organizational hierarchy, add visual interest to a page and aid in skimming the content; where contrast is defined by a difference in size, style of fonts, opposing colors, or other visual elements (Williams, 1994). Designers should strive to simplify and make essential information obvious at a glance. A clear visual hierarchy communicates to the user unambiguously as to what parts of the information are important (Krug, 2000). Noise can be minimized by toning down visual cues that are not important to communicate the information (Krug, 2000) (NASA, 2011). The visual salience of the data layers can be manipulated to match the urgency of the information displayed by adjusting factors such as luminance contrast and typography thereby creating a perceptual hierarchy (NASA, 2011). For example, less important information can be rendered using lower brightness-contrast and important information in higher brightnesscontrast. Users‟ attention is thereby drawn at first to the most important parts of the information while other supporting information is also available (in the background) when the user needs it, instead of presenting all the information at the same perceptual level. As an example, an online tool shown in Figure 1 uses social statistics to determine the importance levels of news articles and assigns corresponding scores (PostRank Labs, 2011). Higher scores are displayed using more saturated shades of orange thus imparting a higher brightness contrast and thereby drawing our attention towards them.  17  Figure 1 Articles displayed with varying contrast levels reflecting their importance scores in a news reader using the PostRank plug-in.  As another example, software agreements have been redesigned in order to enable impatient users to make informed choices by using contrast (e.g., bold typefaces, different font sizes, colors, highlighting) and by summarizing information using symbols, cartoons and quotations (Kay & Terry, 2010). Structuring and limiting information thus, along with highlighting keywords supports the scanning behavior of users (Nielsen, 1997), with the predictability of a regular pattern allowing users to scan ahead with ease (Mullet & Sano, 1995). Users preferred the text based agreements enhanced with typography and graphic design over viewing additional materials such as a video to consume the communicated information (Kay & Terry, 2010). This preference could be attributed to the higher skimmability of the text-enhanced agreements. Skimmability improves the chances of users in finding interesting and relevant information, which subsequently improves the chances of users reading further into the content. This could also be attributed to the increased perception of participatory control over the interaction and increased affordances to graze 18  information (Pirolli, 2007). 2.2.2.3  Comics and icons  In addition to structuring information into layers to make them more accessible, information could also be augmented using alternate representations. For example, cartoons and comic book style representations have been used for making educative non-fictional information more accessible. Comics have been used successfully right from 1951 when they were used to illustrate methods of preventive maintenance of equipment and weapons for training military personnel (Eisner, 2011), to the present day, for example to introduce software (McCloud, 2011). Icons are graphical or symbolic representations of entities. They play an important role in interface design (Mullet & Sano, 1995). Visual images are readily recognized after limited exposure and icons make excellent memory aids helping the user to recall functionality of complex systems (Ware, 2000). Using images accelerates learning in children (Pressley, 1977). Children prefer having icons and they help children to interact with the software (Uden & Dix, 2000). Children prefer having a mascot or character that imparts personality to the interaction (Large & Beheshti, 2005). There is a rich body of literature on how icons work and how to design effective icons for adults. Icons are visual signals and their design should follow from the nature of the message communicated, the intended audience and the context (Massironi, 2001). Icons have been classified along the dimensions of concreteness (Leung, McGrenere, & Graf, 2011; Zammit, 2000; McCloud, 1993) which is a measure of the extent to which the icon resembles the object that it represents. A symbol is an object which stands in to represent another object. In this way icons are 19  visual symbols. To use something as a symbol, children must focus on what the symbol represents and not on the symbol itself. Consequently, objects that are particularly interesting or attractive, such as toys, are difficult for young children to use as symbols as they have insufficient cognitive resources to handle the dual representation (DeLoache, Uttal, & Pierroutsakos, 1998; Zammit, 2000). Furthermore, the abstract nature of many symbols tends to confuse children. Children do not develop the capabilities to process abstract information until they are much older and advanced in their developmental curve (Piaget, 1959). Unlike text, graphical objects such as icons are more likely to be perceived by different users in different ways; hence icons are not always transparent to decode and adding text labels does not always help (due to unfamiliar vocabulary of the text) (Zammit, 2000). Children have found icons easier to understand if they have had direct experience with the object portrayed by the icon and can make the association between the object in the icon and the function in the application (McKnight & Read, 2009; Zammit, 2000). Hence representational icons have been recognized more easily than abstract ones. Designing icons that correspond to the mental models and metaphors of children, which greatly differ from the designers who are usually adults, is a challenging task (Uden & Dix, 2000; McKnight & Read, 2009; Large & Beheshti, 2005). Similar observations have been made with older adults (Leung, McGrenere, & Graf, 2011). Furthermore, there are inherent difficulties in designing representational icons for activities such as recording sound on a computer. In such instances, context plays an important role in helping children decode the meaning of the icons. But contextual interpretation is heavily influenced by culture. For example, the metaphor of an elderly butler assisting children in finding information in Askjeeves for kids failed miserably suggesting that metaphors depend both on age and culture to be useful (Large & Beheshti, 2005). 20  Children like icons but tend to interpret them literally and are prone to criticize them when they do not accurately match the underlying function. A text label is recommended to avoid misinterpretations of icons (Large & Beheshti, 2005; Leung, McGrenere, & Graf, 2011) and care should be taken to tailor the text labels according to children‟s vocabulary (Zammit, 2000). Involving children in the design of the icons is recommended to provide relevant designs (McKnight & Read, 2009). Animated icons have been shown to help adult users decode their meanings better than static icons, but designing animations for abstract functions is a tricky proposition (Baecker, Small, & Mander, 1991). Children have been observed to be critical of animation in general, finding it distracting and superfluous. Furthermore, children have found animation sequences to be age sensitive and to most likely garner criticism as being too childish (Large & Beheshti, 2005). In spite of the challenges associated with designing effective icons in terms of children‟s mental models and the nature of the icons themselves (e.g., concrete and abstract), interfaces based on icons have been a popular recipe for designs targeted at children, especially those who cannot read yet. For example, Netflix, a popular online video streaming service, opened a new section of their website targeted at children under twelve (Netflix, 2011). The interface has minimal text and is driven by icons of characters and images of title covers, and was well received by the media and parents. The following is a quote from a parent (Higginbotham, 2011) which illustrates the potential for such interfaces: “I appreciated this, as does my daughter, who can now watch Hello Kitty or the Backyardigans without intervention [emphasis added] from me or my husband on Saturday mornings.”  21  2.3 Summary In this chapter, we reviewed children‟s learning process and the role of competency and autonomy in learning. We touched upon the importance of designing software with an understanding of the inherent gender differences. We noted how children‟s interaction with computers can be improved by adults‟ guidance but how they are often left on their own. We discussed the potential of a system that could communicate helpful information effectively to children and make up for the absence of adult help. We turned our discussion towards dialogs boxes, which are a specific class of information communication systems, and looked at the general problems with dialogs and briefly touched upon some solutions. We then reviewed literature from the broader area of information consumption including users‟ behavior and its adaptations to tackle information overload, followed by the effects of age and literacy on the behavior. We set out to find recommendations for designing an effective information communication system and identified factors such as highlighting, contrast and use of icons to design systems such as dialog boxes for effectively communicating information. In our work described in Chapter 4, we employed several strategies to communicate the information in dialogs more effectively. We used opposing colors to classify information (Williams, 1994), highlighting to draw attention to important information (Chi, Gumbrecht, & Hong, 2007), bold typefaces in titles (Kay & Terry, 2010) and differences in contrast levels to differentiate important text from background information (NASA, 2011). We used icons to help children learn about the software (Large & Beheshti, 2005). We used mascots, as young users have found them to impart personality (Large & Beheshti, 2005). We also used comic style layouts in interfaces in an attempt to model them on real world interactions 22  to enhance learning curiosity (Shaw, LaBore, Chiu, & Johnson, 2004). We decided not to use animated icons as children have found animations to be generally distracting and superfluous and also criticized as being childish (Large & Beheshti, 2005). We preferred representational icons over abstract ones as children found abstract icons harder to interpret (McKnight & Read, 2009). Involving children in the design of the icons has been recommended to minimize the discrepancy in the mental models between children and designers (McKnight & Read, 2009), but we were unable to carry it out owing to the challenges of our study environment. We aimed at designing adaptive interfaces for children taking age and gender differences into consideration. We conducted two studies (Chapters 3 & 5) towards achieving this goal. Our research falls into the scope of providing preliminary guidelines and heuristics for designing technology for children, more specifically methods to effectively communicate information, taking the age and gender differences into account.  23  Chapter 3: Study 1 - Observing children’s interaction with computers In Chapter 1 we argued for designing adaptive interfaces that change according to a child‟s attributes such as age, in order to facilitate children‟s interactions with computers. We also revealed our interest in designing a help system that accelerates learning and minimizes the requirement for adult intervention. In this chapter we present the methods, findings and discussion of an exploratory study that was aimed as a first step towards designing adaptive interfaces and intelligent help systems. We discuss our observations on the differences in interaction patterns among children by gender, age and the mechanisms children employ to learn the interface and the underlying features, including the help system. We explain why it is challenging to design an effective help system for children. Finally, we discuss the problems that children faced with a specific component of the interface, dialog boxes, classify the problems by age groups and present potential design solutions targeted at different age groups.  3.1 Methodology We conducted an exploratory study with children to investigate how they used a computer application and the difficulties they encountered. More specifically, going into the study we were interested in learning about the differences in interaction patterns among children of different ages, and the mechanisms children employed to learn the interface and system features.  24  3.1.1 Participants 111 children (66 girls, 45 boys) aged 3 to 12 took part in our study. Hereinafter, we use the terms participants and children interchangeably. Each session lasted about 12 minutes on average. The choice of the specified age range was based on two assumptions. First, children less than age 3 would not be expected to interact reliably with a computer for reasonable periods of time, especially in an environment filled with playful distractions. Second, children over the age of 12 would likely exhibit interaction patterns closer to that of adults (who are not our primary interest) and they would not likely be interested in an activity such as the one we used in our study, namely painting using a computer application that was designed specifically for children between ages 3 to 12.  3.1.2 Setting This study was conducted at Science World, a non-profit organization in Vancouver, Canada that hosts science shows, exhibits and events targeted at children. It was conducted in the summer of 2010. A recruitment poster (Appendix C) was put on display to invite children to participate.  Figure 2 Our study setup at Science World showing the laptops and printer we used for the study along with the poster (resting on an easel) that was used to invite participants to the study.  25  3.1.3 Methods Two laptop computers were set up to run an open-source children‟s painting application, Tux Paint, for the study. Participants were invited to use the painting program to create their own drawing. There was no fixed amount of time allocated for each study session; the participants could leave at any time. We had instrumented the source code of Tux Paint to capture children‟s interactions, such as the clicks on various UI elements of the software and the painting canvas, into a log file. A screen capture of the session and the corresponding audio input was also automatically recorded. In addition, one researcher was present by the children‟s side, observing their interactions and taking notes when possible. A UBC graduate student volunteer assisted the researcher in running the study. However, observation notes were taken only sparsely and intermittently, capturing only important out-of-ordinary behavior, due to the demands placed on the researcher and volunteer in terms of study logistics such as obtaining informed parental consent while running the study and having to assist and observe two participants simultaneously. On the last day of the study, we offered to play a video at the beginning of a small number of randomly selected sessions. The video shown consisted of selected video screencaptures from sessions during our first four days of the study. The video stood out in terms of the quality of the demonstrated interaction, scope of features explored and quality of picture created. We explained to parents2 and children that watching the screen-capture video could help the children to better learn the interface and its associated features.  2  Hereinafter, we use parents to denote parents or guardians accompanying the children.  26  At the end of each study session, the children were offered a color print out of their drawing along with a small sticker containing Tux Paint‟s mascot3.  Figure 3 The sticker that was handed out to children at the end of the session.  3.1.3.1  Data analysis  At the end of the study, we went over our observation notes. If we had marked a session as warranting a closer look4, we went through the session‟s screen capture and took additional notes. From our notes we collected and organized any recurring themes by age and gender. For the quantitative analysis we analyzed the log files, which consisted of time stamped interactions with the interface, using scripts and extracted higher level information. We triangulated our qualitative observations and the quantitative data where possible, by trying to find support for qualitative observations in the quantitative analysis and vice versa.  3.2 Findings In this section we identify dialog boxes as a significant challenge faced by children. We divide children into three age groups based on our observations with support from the developmental psychology literature. We classify the challenges experienced with dialogs under six categories and further associate them with individual age groups. Then we report  3  The sticker was inspired by similar creations from New Breed Software led by Bill Kendrick who is also the chief contributor to the open source Tux Paint software. 4 Approximately 1 in 10 sessions in our study had been marked for a closer look.  27  on how watching peer interactions, either via screen-capture videos or live, accelerated children‟s learning of the features, but how most children/parents were not patient enough to watch the videos. We conclude this section by reporting our observations on various gender differences on how children interacted with the software.  3.2.1 Dialog boxes present numerous problems for children A key finding that emerged from the study was that children consistently had problems with dialog boxes. Figure 4 shows some of the default dialogs in Tux Paint and illustrates several problems that children faced, for example: Difficulty in understanding the concepts and underlying abstractions of terms such as „your changes‟, „the old one‟, „new file‟, „to quit‟, „take me back‟, Inability in mapping icons with their intended action or representation, such as o  The icon at the top right corner in Figure 4, which represents the action of replacing an existing picture, is not self-explanatory.  o  Confusion as to whether one should always abstain from the red button with the cross sign (which in the case of the Quit dialog in Figure 4, for example, might not be always desirable),  For children who have difficulty reading text, the dialogs look virtually the same, lacking clear non-textual visual cues that differentiate one dialog from another.  Figure 4 Examples of dialog boxes from Tux Paint that pose several problems to children.  28  3.2.1.1  Design by age  Children have varying needs and demonstrate different interaction patterns according to their age, and hence they could be served better by designs that are targeted at different age groups (Budiu & Nielsen, 2010; Campbell, Goldman, Boccia, & Skinner, 2004; Large & Beheshti, 2005). Based on our observations during the study and post-study data analysis as well as literature in developmental psychology (Lueder & Rice, 2008) we categorized children into three groups: pre-literates, semi-literates and literates, with children in each group sharing common behavioral patterns. These groups could potentially benefit from different targeted designs. Group 1 2 3  Ages  Label  Number of children Boys Girls Total 3, 4, 5 Pre-literate 10 13 23 6, 7 Semi-literate 13 18 31 8, 9, 10, 11, 12 Literate 22 35 57  Table 1 Three groups of children categorized according to their age. The categorization was based on observations made in the current study and supported by literature from developmental psychology.  3.2.1.1.1  Pre-literate: Ages 3 to 5  Based on literature, the children in this group are characterized by their desire to be independent, as well as having the least understanding of the consequences of their actions (Lueder & Rice, 2008). Their attention span is brief, their tolerance for frustration is low and their manual skills are not fine tuned (Baumgarten, 2003). Most of the children in this group have not yet learned to read (Lueder & Rice, 2008). In our observations, we found that they had the least control over the mouse and possibly the least familiarity with computers, and hence relied the most on their parents. Most of the pre-literates were accompanied by their parents during the session and their parents offered assistance ranging from occasionally helping whenever their child faced any difficulty, to 29  assisting completely, including operating the mouse and at times even making decisions on their behalf. This was also the group that contained all of the children who chose to patiently wait and watch the screen-capture videos or live peer interactions on the last day of the study. Over the course of the study, we repeatedly bore witness to the perplexed faces of the children, especially pre-literates, when they were facing dialog boxes. None of the preliterates were able to make a choice independently by themselves, especially with dialog boxes that had multiple buttons to choose from. They were less intimidated by dialogs with just one button (Figure 5). We imagine that a dialog box with a single OK button that is colored in green would be a little easier and less intimidating to dismiss. We found that many of these children could not make the causal link between the dialogs and their own actions that triggered them. A number of the pre-literates would try to continue to interact with the software in spite of the dialog blocking any interaction (modal dialogs). Some seemed to be unaware of the dialog‟s appearance on the screen while others were trying to ignore its presence. Many appeared to be facing a dialog for the first time and hence seemed to be wondering why something had appeared in the center of the screen blocking their view preventing them from using the program. The result was that when a dialog appeared, children got confused and at times even startled. Almost all of them required help from adults to dismiss a dialog especially for the first couple of instances they faced a dialog. We observed that only a small number of children in this age group did not seek adult assistance, but rather randomly clicked on a choice in order to dismiss the dialog. 3.2.1.1.2  Semi-literate: Ages 6 to 7  With respect to cognitive development, this group has an advantage over the pre-literates in terms of an increase in their attention span (Lueder & Rice, 2008) and a spike in brain 30  development (Herman & Epstein, 1986). They desire autonomy, like to differentiate themselves from the younger pre-literates and fear looking „babyish‟ (Baumgarten, 2003). They have sufficient motor skills and hand-to-eye coordination to enable them to operate a computer without any difficulty (Baumgarten, 2003). They can read at a preliminary level and have better understanding of the consequences of their actions, though they are not yet fully developed (Lueder & Rice, 2008). The increase in attention span and brain growth might explain their curious nature and the fact that we observed this group explore the most number and type of functions in Tux Paint and hence encounter the largest share of dialog boxes5. We observed that this group operated largely without the help of adults. Though children in this age group can read at a preliminary level, we observed that many of them seemed to have conceptual difficulty understanding abstractions in dialogs such as file systems. As a result of their better understanding of the consequences of their actions, we observed that they typically had almost no problem with making the causal link between their actions and appearance of dialog boxes, unlike the pre-literates. Though most of them seemed to be not entirely comfortable with having to deal with dialogs, many seemed to have learned that clicking on a choice, usually arbitrarily, dismisses the dialog. But this behavior of randomly making a choice resulted in loss of work at times, for example when the children chose to overwrite or not save their work, without understanding the implications of their choice.  5  Dialogs in Tux Paint were mostly of confirmatory nature. Hence the number of dialogs encountered was roughly proportional to the extent to which the various functions (especially advanced) were explored in the software.  31  3.2.1.1.3  Literate: Ages 8 to 12  In terms of development, this group is relatively the most likely to understand the consequences of their actions before engaging in them (Lueder & Rice, 2008). This is also the group that tends to be the most self-conscious and hence more susceptible to the Hawthorne effect (Baumgarten, 2003). They are also likely to have the most exposure to technology (Rideout, Foehr, & Roberts, 2010; Ofcom, 2008). We observed that this group had the least problems with dialogs. They did not need help with dialogs, and seemed to process the information on the dialogs effectively. Due to their increased exposure to technology they might have learnt to handle dialogs and become comfortable with them. This however does not mean that this group could not be affected by a redesign of dialogs. The single biggest problem for this group was probably the lack of patience in dealing with dialogs. Furthermore, it has been observed that when it comes to designs, children are very sensitive to the age-factor. Older children tend to treat those designs that they perceive as being designed for younger children unfavorably (Baumgarten, 2003; Budiu & Nielsen, 2010). Hence it remains to be seen how literates will react to the redesigns. 3.2.1.2  Problems with dialogs  From our study, we have identified the following problems that children faced with dialog boxes: 1. Causality (Why did it appear all of a sudden?): Not being able to make the link between their action and the resultant dialog. 2. Hindrance (Why is it stuck?): The modal nature of a dialog box confused many children as to why they cannot  32  continue interacting with the software. 3. Affordance (What is it doing here?): Not being able to understand that the software is trying to have a conversation with them i.e. providing them with some information and asking them to make a choice. 4. Communication (What is it saying?): Not being able to understand the content of the communication. This essentially reduces to not being able to read the text on the dialog and/or not being able to understand the represented abstractions such as file systems. 5. Consequence (What should I do now?): Not understanding the full implications that their choices would have and having to deal with the negative consequences of the choices made in haste. 6. Patience (Whatever…): Not wanting to spend time reading and understanding what the software is trying to communicate to them.  Pre-literates were affected for the most part with problems of causality, hindrance, affordance, communication and consequence; semi-literates with the problem of consequence and to a lesser extent with problems of communication and patience; and literates with the problem of patience and to a lesser extent with the problem of consequence. Problems with dialog box  Age group Pre-literate Semi-literate Literate  Causality (Why did it appear all of a sudden?) Hindrance (Why is it stuck?) Affordance (What is it doing here?) Communication (What is it saying?) Consequence (What should I do now?) Patience (Whatever...) Table 2 Summary of the various problems with dialog boxes by age groups. The two shades of green denote varying intensity levels of the problem while gray denotes that it is not a significant problem area.  33  3.2.2 Watching peer interactions accelerated learning The children who agreed and watched the video of past peer interactions showed clear improvements in their interaction and above-average knowledge of the features when compared to other children of similar age. Similar improvements were also observed in children who watched other children using the software while waiting for their turn to participate in the study. Whenever we saw a child demonstrate such accelerated learning, we enquired about how they knew about the particular interaction or feature in order to verify the source of the knowledge. They reported that they had acquired the knowledge by either watching the video or by observing another child prior to their session. However, most of the parents and children to whom we offered to show the video either declined the offer or would indicate to us shortly after starting the video that they would rather just start painting. We are not sure if the parents or children recognized the value of learning from peers. But prior research has indicated that although users recognized the value of learning from peer interactions, such interactions happened infrequently (Murphy-Hill & Murphy, 2011).  3.2.3 Gender differences and others observations We observed several interesting gender differences in the study. Table 3 summarizes our findings, which are derived from both direct observations and quantitative analysis. We note that we express the findings informally (i.e., less vs. more), which is commensurate with the precision afforded by the study.  34  Observed behaviour Distracted from painting activity Struggled to come up with an idea to paint Tendency to create the painting from scratch Used pre-existing drawing components Inclined to seek help from adults when facing difficulty Time spent on study session Explored features of the software Style of reading text on dialog boxes Speed of interaction with dialogs  Girls Less Less More Less More Less Less Thorough Slower  Boys More More Less More Less More More Casual Faster  Table 3 Summary of the observed interaction differences between girls and boys.  Girls interacted differently than boys. Older children were faster at interacting with the dialogs. Girls were slower than boys on average (by about 3 seconds). The difference between genders was the greatest in semi-literates (about 4 seconds) and lowest in literates. Boys spent more time than girls on the sessions, with the difference being the greatest for semi-literates and lowest for other groups. Boys tend to get more distracted. One of the prominent differences that caught our attention was how easily distracted the boys were when compared to the girls. The environment of Science World replete with attractions could have accentuated the effect. Girls tend to be more creative. Boys thought aloud and pondered on what to draw and seemed to be having difficulty in creating a picture. Boys used more of pre-existing drawing components like Stamps while girls preferred to create their artwork from scratch using basic components such as Paint brushes. This likely explains the differences found in the exploration pattern of functions such as brush clicks and change of colors. Girls, on average, used a 1000 more brush clicks6 than boys per session. Girls changed color more often than boys during a session (13 vs. 8). Older children also changed colors more often than younger children. But the difference among genders persists across age groups with girls changing  6  A brush click is an interaction performed using the mouse on the canvas in the act of painting.  35  colors more often than boys across age groups. Prior research has shown that while girls typically draw objects and people facing the viewer using ten or more colors, while boys typically try to capture an action in their paintings using lesser colors (Iijima, Arisaka, Minamoto, & Arai, 2001). Girls tend to be more result oriented. Boys and girls seemed to interact with the software with different expectations. Boys tended to explore more features of the software. This probably led to them spending more time in the sessions than girls. Girls seemed goal oriented. Girls seemed to have an image in mind that they wanted to replicate onto the digital canvas and would meticulously work towards realizing it. Girls also seemed to have more specific questions enquiring about tools that could help them accomplish a task towards their goal. Boys seemed to be less inclined to be seeking help from adults, especially researchers, than girls, even when they were facing difficulties. Prior research has pointed to the inherent differences between genders in seeking help (Sax, 2005). RT is not proportional to the text on the dialog. The difference in RTs (Reaction times) among dialogs that had one option (Figure 5) vs. the dialogs that had two options (Figure 4, shown earlier) is surprisingly negligible (under a second). But this difference is greater for girls (almost twice) than for boys. The length of the text on the dialog too does not seem to be making a proportional difference in RTs. For different dialogs in Tux Paint whose text length ranges from 27 to 63 characters, the differences in RT remained less than a second.  36  Figure 5 Dialog boxes in Tux Paint with just one choice.  Boys tend to be causal. Boys appeared to be more casual in their interaction with dialog boxes. They seemed to be reading less of the dialog text while girls seemed to be more thorough. This could be attributed to the proportional increase in girls‟ RTs with the amount of information on the dialog, such as the increase in number of options or length of text, while boys showed no such increase. The nature of girls to be less willing to take risks (Sax, 2005) or, as we suspect, be less comfortable in interacting based on limited information might have contributed to this difference. The reading ability of the girls which has been found to be higher on average than boys of the same age (Lueder & Rice, 2008) might also have led the girls in our study to read more text than the boys, thus resulting in slower responses. We suspect that boys and perhaps older children in general tend to skim information rather than process it thoroughly. Usage of different functions. Undo was by far the most frequently used function, followed by Paint and Eraser. Eraser usage seemed to go up with age. But the increase was steeper for girls than boys for whom it was virtually flat. This points to the nature of children to strive for perfection even in seemingly trivial tasks (Holt, 1991) and trying to be competent in one‟s interactions (Miserandino, 1996). It also points to the nature of girls to work meticulously towards a vision. Functions such as Save, Open, New, Quit etc, which can be viewed as meta or house-keeping functions in that they are not directly involved in creating an artwork, were 37  among the least used.  3.3 Discussion In this section we discuss the challenges of conducting studies in a public space such as Science World followed by the challenges we identified in attempting to design an effective help system for children.  3.3.1 Challenges of conducting studies in a public space Science World was replete with attractions that competed with our study, which in turn impacted our recruitment of participants, their session duration, and their attentiveness during the study. Furthermore, to our surprise in many cases it was only either the parent or the child who was interested in participating, but not both. In other cases they would spend only enough time on the software to create an interesting drawing in order to make a printout that they could take home. In spite of the challenges, the distraction filled study environment was perhaps a realistic reflection of many real-world settings such as homes and schools of current generation children who have been identified to immerse themselves in a multitude of simultaneous media exposures and interruptions (Rideout, Foehr, & Roberts, 2010; Ofcom, 2008). Moreover, most of the parents appeared to have a fixed schedule in mind, planning to explore a specific set or number of activities during their visit at Science World. Therefore, many parents would urge their child to finish painting as soon as possible so that they could move on to the next activity. This affected the session length and possibly the nature/style of interaction and the features explored. Increased structuring of children‟s activities by parents has been observed to lead to decrease in children‟s activity levels (Floyd, Bocarro, Smith, & 38  Baran, 2001). Further many parents, as documented in existing reports (Ofcom, 2008), were concerned about the increasing amount of time that children are spending in front of the computer. Thus they generally preferred their children to engage in more physical and cognitive activities, such as those at Science World, rather than participating in our study consisting of a sedentary drawing activity in a computer.  3.3.2 Challenges in communicating information The following are quotes from parents when they were asked about the main difficulties their child faced when interacting with a computer: “Understanding how to use different software. When there are too many choices, not knowing what to do.” 'Should show how to do something ... demo...'  While the potential for an effective help system, specifically a system that shows past interaction trails from peers, seemed obvious before we started the study, it became clear that designing an effective solution would be much more challenging than we had initially thought. The following are some of the challenges we identified in attempting to design an effective help system. This could generalize to any system that aims to communicate information to children such as dialog boxes. 3.3.2.1  Desired independence of interaction  Children were not enthusiastic in approaching adults for help, unlike we had expected. Children asked adults (their parents and/or us the researchers) for help primarily in the following scenarios When they faced difficulty with the application, for example:  39  o  when they were looking for a feature but could not locate it themselves; or  o  when their work-flow was interrupted by features like dialog boxes or the interface becoming irresponsive.  At the end of the session when they had finished painting and wanted to print/quit.  We observed that pre-literates and girls were more willing to accept help. But most children preferred to use the application on their own and seemed to request help only when it was absolutely necessary. This could be seen as the need for autonomy (Miserandino, 1996). The following are some quotes from the participants which illustrates this behavior: Older brother commenting about his younger brother: 'Doesn't really ask me for help on software. Only on technical problems…' Parent: „usually needs help from mother if computer is slow, but quite independent‟  We suspect that the demonstrated differences in the way children sought adult help could be dependent on how adults in the past in home/school have tended to respond to their requests for help (Klein, Nir-Gal, & Darom, 2000) and how useful that help turned out to be. It could also be related to how the barrier for asking help changes with age and gender (Sax, 2005). We suspect that children could even exhibit this behavior of resisting external help (until absolutely necessary) towards help or information systems in the computer. 3.3.2.2  Patience declines with experience  We found that most children were impatient to read text: either text in help or text in dialog boxes and that they ignored help text even when they needed it, even though it was being displayed in a prominent always-visible section of the Tux Paint interface.  40  Help system  Figure 6 Interface of Tux Paint consisting of the prominently displayed help section.  Prior research has made similar observations. Most children do not read textual help even when they are in need of help (Large & Beheshti, 2005). Very few users read help and most users prefer to forge ahead and interact by trial and error (Krug, 2000; Nielsen, 1997). The more experience a child has, the less likely he/she were to read instructions or help patiently (Budiu & Nielsen, 2010). We observed that for semi-literates and especially literates, that even though they possess the capabilities to read, plain text seemed to fail as an effective information communication medium in both help and dialogs. This suggests that as children age and/or gain experience in technology and get exposed to an ever increasing volume of information, they could be developing a heightened sense of cost-benefit ratio in consuming information. Plain text with its lower information density could be seen as unattractive. 3.3.2.3  Consuming information passively  Children‟s dislike to consume help and other information passively could also be attributed to the improved interaction affordances offered by the computer. Children watch 41  almost 3-5 hours of television daily (Rideout, Foehr, & Roberts, 2010). Online communities have sprung up around people passively watching screen casts of peer interactions (Queeky, 2011). Hence watching passively does not seem to be something that children and adults dislike to do in general. Nielsen reported on how the amount of text that people end up reading online declines continuously, pointing to the possibility that when users sit in front of the computer, they do not want to spend most of their time reading but rather interacting with the computer (Nielsen, 1997). They want to be clicking links and moving from page to page so much so that they end up reading less and less as their experience with technology matures. This might also explain why the only participants who showed some patience to watch the videos in our study were among the youngest. 3.3.2.4  Summary of challenges  Large and Beheshti also acknowledged the challenges in designing an effective help system and suggested that an ideal solution would have to be context sensitive and precise (Large & Beheshti, 2005). Using video to convey information has been noted to perform worse than plain text (Campbell, Goldman, Boccia, & Skinner, 2004). Users have shown disinterest in viewing additional video materials and prefer skimmable text information (Kay & Terry, 2010). This could be attributed to the poor skimmability of video information. The monolithic informational structure of the video probably has lower affordances on several aspects of interaction when compared with other information delivery mechanisms such as text. The problem with using video as a medium of presenting information might not be just related to the patience to sit through it, it might also have to do with the perception of control, ability to navigate around, skimmability and interaction. An effective design solution, should then try to address the problem on these multiple 42  facets. It should present information in a format that has lower cost of consumption to account for the prevailing impatience. In the case of a help system, it should also appear at the right time and context when the child feels the need for help and match the child‟s expertise level in its content. Due to these various challenges and the short timeframe of a graduate degree we decided not to proceed with designing a help system. But some of the challenges that we identified (e.g., with patience), extend beyond the help system to information communication systems such as dialog boxes.  3.4 Design implications Based on the findings from this study, we discuss several design implications that aim at solving the problems faced by children as described in Section 3.2.1.2 - such as minimizing the hindrance caused by modal dialogs, improving affordances of dialoguing, augmenting the information on dialogs to reach a wider audience and presenting information in a format that helps the impatient children. Finally we discuss why we did not consider audio in our proposed solutions and some motivations behind solving this problem.  3.4.1 Minimizing hindrance The modal nature of a dialog box confused many children, especially pre-literates, as to why they cannot continue interacting with the software. Dialogs have been noted to disrupt user‟s attention and break their workflow (Cooper, Reimann, & Cronin, 2007). Making dialogs non-modal and gradually disappearing after a certain number of seconds could make them less of a hindrance and make especially the pre-literates more comfortable with the notion of dialogs until they are ready to deal with them.  43  3.4.2 Improving affordances We observed that a number of pre-literates had difficulty in identifying the purpose of the dialog UI. They did not understand the concept of dialoguing with the computer. We think that infusing a comic book style into the dialogs by splitting it into two could improve the affordances of dialoguing.  3.4.3 Improving communication None of the pre-literates were able to make a choice independently, due to their inability to read text reliably and limited experience in dealing with digital artifacts such as dialogs. They would rely on help from adults to learn to interact with the dialogs for the first few instances and would then interact by themselves, probably relying on non-textual cues (such as color, e.g., green button) on the dialog boxes and remembering the help from the first few interactions (e.g., „click on the green button’). But as the default dialog boxes in Tux Paint (Figure 4) did not provide enough contextual cueing (Chun & Jiang, 1998) to sufficiently differentiate either the dialogs or the choices across dialogs, we found that children‟s interactions based on non-textual cues were uninformed and only marginally better than random. Semi-literates and especially the literates exhibited relatively less confusion in dealing with dialog boxes. But there were instances when children who seemed to be interacting confidently would approach us after losing their work because they had made a choice without completely understanding it. We suspect that this behavior, where children continue interacting in spite of their incomplete understanding, might be a product of their reading ability (which allows them to skim the text for information) and prior computer experience. 44  To solve the problem for pre-literates we need to add sufficient non-textual cues to the dialogs. We could augment dialog text with graphical objects such as symbols or icons. The typography or appearance of words can also be altered to convey more information (McCloud, 1993; Kay & Terry, 2010). This would make the information more accessible to the pre-literates who cannot read text. It would also make it more accessible to those semiliterates and literates who do not prefer reading text, as augmenting text with non-textual cues also increases the information bandwidth and cost-benefit ratio thus increasing the probability of consuming and understanding more of the information (Nielsen, 2003) and thereby making more informed choices. We have to remember however that pre-literates are not yet completely familiar with the concept of symbols, signs and their representative meanings (Lueder & Rice, 2008).  3.4.4 Addressing impatience One of the important traits of the children that we observed during our study was that they tended to be impatient at consuming the information, evaluating it and making an informed choice (with lower thresholds than we had assumed), which led to random choices being clicked. The impatience of the children could be addressed by designing dialogs to support skimming of its content, the goal of which is to reduce the effort required to identify and understand the dialog. We propose to add information using multiple layers (body text with contrast, a title that stands out visually) (NASA, 2011; Krug, 2000) and multiple modalities (text and icons) (McCloud, 1993; Kay & Terry, 2010) that would lower the cost of consuming information. We think that this might reduce the effort required in identifying and understanding the dialog, which might then lead to more informed choices and thereby reduce random clicks made without understanding the dialog. 45  3.4.5 Safer arbitrary choices If the children still fail to understand the consequences, it might lead to negative outcomes. We can reduce negative consequences of choices made in haste by: Having a persistent choice in all dialogs, in a consistent position with a consistent look and feel, that dismisses the dialog without making any change. The choice could be labelled „Take me back‟ or „Do Nothing‟. Revealing the dialog text gradually in parts, where the order of appearance of a word or phrase depends on its importance, could optimize information consumption. Activating the choices (i.e. making them clickable) in a dialog one after another in increasing order of potential disruption of the choice. For example, in a dialog box that asks the child whether to save a new file or overwriting previous file, the order of activation of buttons would be: 1. „Save as a new file‟ 2. „Save over previous file‟  However, mechanisms should be provided to allow children who can make informed decisions to bypass these safety or optimization mechanisms (e.g., with a shortcut key) so that they do not become annoyed unnecessarily. Over time, we believe that this model offers a comfortable learning environment to the child, one which they can grow with as they become more capable and/or comfortable at processing all the presented information.  3.4.6 Audio One could imagine that for pre-literates and semi-literates whose reading skills are limited, having the software read out aloud the text and the choices on a dialog box could serve as a solution. We observed that for a number of children from these groups that this 46  was the kind of help that their parents provided. For example, when their children turned towards them with blank help-seeking stares, some parents would coerce their children instead into reading the text on the dialog by prompting them with a few words or part of the words and prodding them to continue. But we decided to not pursue audio as a design solution owing to its several limitations. Audio only solves the problem of Communicating the content of the dialog and possibly to a certain extent identifying the dialog‟s Purpose. But it affords poor support for skimming of information, similar to video, and hence could worsen the problem of Patience. Users have shown disinterest in consuming such non-skimmable, passive presentation of information and prefer skimmable text information (Kay & Terry, 2010). Further, conveying multiple choices unambiguously using audio modality becomes challenging and hence could worsen the problem of Consequence. Auditory recognition from memory is inferior to recognizing from visual memory (Cohen, 2009) and hence using audio messages with young children is cautioned against owing to their limited short term memory (Gelderblom & Kotzé, 2009). Moreover, audio messages take relatively more time and effort to produce even in a single language and thus would take substantially more effort to translate into other languages7. Hence it would fail as a scalable universal design solution that can be adopted by willing designers with limited resources.  3.4.7 Is this a problem worth solving? What are the ramifications of leaving children to wonder briefly on a couple of dialog boxes? Would they not learn to use them eventually with or without the help of the adults  7  This could be a significant hurdle for projects like Tux Paint which has been translated into 85 languages on last count as of Aug 2011 according to http://tuxpaint.org/help/po/.  47  with trial and error? We argue that when children have to interact with dialogs repeatedly with imperfect understanding of the process and often depend on adults‟ help, it might undermine their confidence (Miserandino, 1996) and hope (Holt, 1991) to master the digital world, both of which are required for effective learning to take place. Dialog boxes, then, might not be just occasional annoyances for children, especially when they do not know how to react to them and are forced to seek external help. Poorly designed dialog boxes might undermine children's confidence in their ability to interact with digital systems and master them, by 1. Presenting information in a format that they do not fully understand, 2. Forcing them to make choices, in many cases often with haste, the consequences of which they do not fully understand, 3. Seeking external help which might or might not be available or end up being useful.  Poorly designed dialog boxes might cause children to perceive themselves as incompetent with little hope of gaining mastery. This perceived incompetence could affect learning in children (Miserandino, 1996). This is especially important for pre-literates as they are actively trying to test their own limits and the limits of the objects in the world around them (Baumgarten, 2003). Poor designs that cause them to doubt their own competence might leave lasting impressions about their ability to master the digital domain. Hence, we believe that this is an important problem to solve.  3.5 Summary In our exploratory study we found that although children needed help while interacting with software, they rarely read help text or any other forms of text. Though there is potential 48  to help children, the way in which the help is delivered has many poorly understood nuances which could be addressed by future research. Communicating information effectively to children is a challenge in general with the prevalent impatience and their preference to interact rather than consume information passively. We identified several reasons that impeded children‟s ability to interact with dialog boxes, such as the inability to recognize the causal link, hindrance by modal dialogs, poor affordances of dialoguing, inability to understand the information on the dialog and make informed choices with an understanding of the consequences. We have discussed several potential solutions to make dialog boxes more usable for children of age 3 to 12, such as using split-dialogs to improve affordances, controlling the interaction on choices to limit disruptions caused by arbitrary clicks, supporting skimming by layering information for addressing impatience and using non-textual cues to communicate the information on the dialog effectively. In the next chapter we describe our prototypes which implement designs of dialog boxes incorporating some of the discussed ideas. Then, in Chapter 5 we describe a follow-up exploratory study evaluating how children from different age groups interacted with those prototypes.  49  Chapter 4: Designing effective dialogs In Chapter 3 we presented an exploratory study in which we identified that dialogs boxes in the form presented in Tux Paint, which are typical of dialog boxes used in most applications, pose significant challenges for children. We identified several problems (3.2.1.2) and proposed corresponding design solutions (3.4). We described how the problems identified could be addressed by modifying elements of dialogs such as their physical structure and visual representation (3.4, 3.2.1.2). In this chapter we present the various design solutions based on the findings from the previous chapter. These design solutions form the basis for the second study presented in Chapter 5 where we evaluate the designs. We used a series of rapid prototyping techniques during the four days of the second study in order to evolve the prototypes used by the children from one day to the next. Detailed debriefing amongst the researchers took place at the end of each day. Design factors that clearly showed no effect on the children‟s interaction with the dialog boxes were removed and new design factors that were expected to impact their interactions were implemented for the next day of the study. The iterative evolution of the design factors is summarized in Table 11. The following table summarizes the design factors, the corresponding modifications made to the interface elements and the rationale behind them.  50  Design factor Split structure  Support for skimming  Interface modifications Modifying physical structure Splitting content  Adding title text Adding title icon Adding Button icon Changing contrast levels of the body text  Delayed-click “I don‟t know” button  Displaying choices sequentially Adding a third button  Details of the modifications  Rationale  To resemble a comic book style Call-out (Figure 7). Split into two: placing question and answer in their own separate dialogs instead of one (Figure 7). Add a title summarizing the purpose of the dialog.  Improving affordances  Add an icon accompanying the title and complementing it (Figure 8). Add an icon inside each of the buttons to complement the button text (Figure 21).  Improving communication, Addressing impatience Improving communication, Addressing impatience  Displaying text using multiple levels of contrast emphasizing the important words and deemphasizing the unimportant words (Figure 21). Enabling safer choices earlier in time (Figure 9). Adding an “I don‟t know” button (Figure 8).  Addressing impatience  Improving affordances  Improving communication, Addressing impatience  Consequence: Safer arbitrary choices Consequence: Safer arbitrary choices  Color coding  Color coding of Buttons  Coloring the safer option using green and the potentially disruptive option using orange or red (Figure 15)  Illustrating consequence  Highlighting  Highlighting the safest button  Highlighting the safer option by drawing a circle around it in yellow. (Figure 18)  Illustrating consequence  Position  Manipulating position of buttons  Switching the button positions  To understand the role of position in designing safer arbitrary choices  Visibility of title, body text  Manipulating visibility of title and body text  Hiding the title and/or body text  To understand the role of title and body text in communicating the content of the dialog  Table 4 Design factors manipulated in the study, their respective rationale and problems addressed.  We note that the interface modifications shown in Table 4 are not exhaustive. There were additional design factors but we selected those given in the table based on several reasons  51  which can be found in Appendix B.1. Some of the intermediate mockups created during brainstorming these designs are presented in Appendix A. We selected the three dialogs Save, Quit and Erase – to be used in the study based on factors indicated in Appendix B.2.  4.1.1 Improving affordances 4.1.1.1  Split structure  We designed the split dialog to help especially the pre-literates in understanding and responding to the dialog. One dialog originating from an icon representing the computer would contain the question and the other originating from an icon representing the user would contain the choices. We designed the software such that the user icon would be either of a boy or a girl depending upon the gender of the child participating in the session.  Figure 7 A split dialog without the factors to support skimming.  4.1.2 Facilitating safer arbitrary choices The following solutions, of including the “I don‟t know button” and using delayed-click, were aimed at making it easier overall to interact with dialogs while also being safer to those who are either impatient or incapable of processing the information on the dialog. These could lower the decision time and minimize the negative consequences in situations where the child makes a random choice in haste. 52  4.1.2.1  “I don’t know” button  We added a third button to the dialog – the “I don‟t know” – that might be functionally redundant with either of the other choices in that it dismissed the dialog without performing any operation, but had the advantages of visual constancy – displayed at the same position in the dialog with the same visual elements in all the dialogs (Figure 8). It would offer an easier and safer way out of the dialog for children who cannot or do not want to process the information on the dialog.  Figure 8 Dialog with the third “I don‟t know” button.  4.1.2.2  Delayed-click  We designed delayed-click in order to minimize the potentially disruptive effects when a child is clicking arbitrarily without understanding. In delayed-click the buttons are revealed one after another in the clockwise direction with a very short delay between each. This results in the relatively safest “I don‟t know” choice being revealed first, followed by the other choices (Figure 9). If the child‟s intention was to get rid of the dialog without understanding the information on it (either for lack of ability or patience), the “I don‟t know” button could be clicked before the other buttons were revealed completely. The movement caused by the appearance of the buttons, “I don‟t know” in particular, could attract the attention towards it and hence increase the probability of clicking on it. 53  Figure 9 A split dialog revealing its options in sequence. The buttons are revealed in the clockwise direction.  4.1.2.3  Switching button positions  We wanted to understand the role of the position of the buttons in dialog interaction especially when a child wants to arbitrarily click on a choice. Is there a preference towards a side? Understanding this might provide us a better idea on where to position the relatively safer choices on the dialog. To learn more we manipulated the position, i.e. switched the  54  button positions randomly during sessions (Figure 10, Figure 11). The default button positions are as shown in Figure 15.  Figure 10 The Quit dialog with the button positions switched (shown both with and without button color coding).  Figure 11 The Quit dialog with the button positions switched and without title or body text (shown both with and without button color coding).  4.1.3 Improving communication We wanted to better understand the role of title and body text in communicating the content of the dialog. 4.1.3.1  Visibility of title and body text  By the end of second day, we were not sure if children were relying on body text and title to understand what the dialog was communicating to them. Hence we made the visibility of title and body text as a factor on day three of the study (Figure 12, Figure 13).  55  Figure 12 The Erase, Quit and Save dialog with title and body text.  Figure 13 The Quit dialog being displayed without title or body text.  4.1.3.2  Visibility of title  At the end of third day, we suspected that the visibility of the body text was having little or no impact. In order to assess its impact we removed body text completely from all dialogs. We retained the visibility of the title as a factor to help us better differentiate the effect of the hiding the body text (Figure 14).  Figure 14 The Save dialog without the body text but with just the title.  4.1.4 Indicating consequences We wanted to examine whether color coding buttons and highlighting the safe options could nudge children towards the safer options and deter them away from the more potentially disruptive options.  56  4.1.4.1  Color coding of buttons  The dialogs were color coded such that the buttons representing safer options were in green and the buttons representing more potentially disruptive options were in orange (Figure 12). Based on observations from the first two days of the study, some of the children, especially the younger ones with lower reading levels, seemed to be relying on non-textual cues such as color coding of the buttons (e.g., going for the green buttons) or position (e.g., going for the right-hand side buttons). We wanted to better understand this and hence introduced them as factors on day three. In the conditions where the color coding was not present, all the options were colored using a neutral blue (Figure 15).  Figure 15 The Quit dialog without button color coding (shown both with and without title and body text).  Changes to color coding on the final day. On the final day of the study the potentially disruptive option in the Erase dialog “Erase and start over” was colored red instead of orange (Figure 12, Figure 16). We thought that red might better match the relatively higher potentially disruptive nature of the Erase dialog and wanted to see if red could deter children from clicking on the option more than orange.  57  Figure 16 The Erase dialog with the safer option highlighted in yellow, disruptive option coded in red, without body text, with the button positions switched.  4.1.4.2  Highlighting the safest button  Highlighting attracts attention and improves learning of the highlighted material (Chi, Gumbrecht, & Hong, 2007). If the children were not sure about what to click, we thought that highlighting might steer them towards clicking on the safer choice. We started by using green as the color for highlighting (Figure 17). But the green used for highlighting seemed to interfere with the green button coloring. We thought that it would be better to have a color that was not in any of the buttons (i.e., green, orange, and blue) and could also be considered a neutral color, in addition to having an affinity for attracting attention. So we changed the color of highlighting eventually to yellow (Figure 18, Figure 19).  Figure 17 The Quit and Erase dialogs with the safer options highlighted in green. Quit dialog is also color coded while Erase is not.  Figure 18 The Quit dialog with the safer option highlighted in yellow (shown both with and without color coding).  58  Figure 19 The Erase and Quit dialogs with the safer option highlighted in yellow.  4.1.5 Addressing impatience Impatience can be supported by including elements that support skimming of the dialog content (Figure 20, Figure 21). Typographical and graphical design elements have been used to enhance text, supporting skimmability by lowering the barrier of entry and thereby increasing the chances of the user reading further into the content (Kay & Terry, 2010). We supported skimming with the following factors: Title text, that is visually salient and unique per dialog, that acts like a heading for the dialog A title icon, that acts as a unique symbol for the dialog and complements the title text Icons in buttons, intended to be graphical equivalents of the button text Body text, rendered in varying levels of contrast, aimed at emphasizing the key words required for interaction and de-emphasizing the noise words.  These factors were intended to make it easier and faster to interact with the dialogs by reducing the effort to consume information from them and by improving learning and recognition of the dialog‟s content during subsequent interactions. Creating a clear visual hierarchy helps users skim the information by reducing the noise (Krug, 2000). Creation of a visually salient title is aimed at improving the dialog‟s visual hierarchy and rendering the body text using multiple levels of contrast is aimed at reducing the noise on the dialog by 59  deemphasizing unimportant noise words that are not required to communicate the content of the dialog (NASA, 2011). Adding the button icons is also relevant for Indicating Consequences (4.1.4) and Improving Communication (4.1.3), while using title icon as a symbol for the dialog is also relevant for Improving Communication (4.1.3).  Figure 20 Dialog without any of the factors to support skimming.  Title with icon and text. Body text with varying levels of contrast. Buttons with text and icons inside them. Figure 21 Dialogs containing factors to support skimming - emphasizing key words using varying levels of contrast in the body text, with icons inside buttons, and a clear title with an accompanying icon.  Thus, in this chapter we have presented several design solutions aimed at mitigating the 60  different challenges that children faced with dialog boxes. In the next chapter we describe our evaluation of these design factors. We present our observational study and the corresponding findings.  61  Chapter 5: Study 2 - Investigating the impact of different design factors on children’s interaction with dialog boxes In the previous chapter we presented various design solutions aimed at solving the problems that children in our first study (Chapter 3) faced with dialogs. These designs were evaluated through the observational study presented in this chapter. The designs themselves (as described in Chapter 4) were iteratively refined over the course of the four days of the study. In this chapter we describe the methods employed to evaluate the designs followed by the findings. We report on the success and failures of the design factors we explored with children, along with some unexpected, yet interesting, findings. Next, we discuss the underlying interaction and information-consumption patterns of the children we observed. Finally, we conclude with a set of design implications. The goal of our research – how to present information effectively according to children‟s age and gender – has many unknown variables owing to its nascent nature, making a controlled study premature. Hence the follow-up study reported in this chapter is an observational study designed to narrow down the scope and, as stipulated by Glaser, generating hypotheses for future research (Glaser & Strauss, 1967).  5.1 Methodology Compared to the broad exploratory study presented in Chapter 3, the study presented in this chapter was driven by a relatively more defined goal to investigate how children interact with dialog boxes in response to different design factors. Both studies were conducted in the 62  same public venue, Science World, using the same recruitment method through a poster display and the same two-computer setup as shown in Figure 2. This study spanned four days during spring break 2011. At least one researcher was in close proximity to observe the children‟s interactions with the computer during the study sessions. Notes were taken in pen and paper during observations.  Figure 22 Children participating in the study, using our instrumented version of Tux Paint software. In the picture on the right, a research assistant can be seen at the far end assisting a child.  5.1.1 Software The Tux Paint software used in the study was modified as a result of the following process. First, dialog boxes used in Tux Paint (Figure 4) were analyzed to identify interface elements that could be augmented by manipulating different design factors. Second, the design solutions derived from the first study (Chapter 3) were used to improve the constituent elements of a dialog box. Third, a set of low and medium fidelity prototypes (Appendix A) was created correspondingly in paper and PowerPoint and evaluated by peer HCI researchers in an informal pilot study. The dialog designs were revised based on the feedback gathered from the pilot study. Finally, the code of the open source Tux Paint software was modified to incorporate the design factors for use in the study. The final set of design factors (as presented in Chapter 4) used in the study are the following eight: split dialogs, “I don‟t 63  know” button, delayed-click, switching button positions, manipulating visibility of body text and title, color coding buttons, highlighting buttons, and augmenting text with typographical and graphical information. By default, dialog boxes in Tux Paint pop up in response to either a user or system generated event. But we modified the source code such that the dialog boxes popped up at pre-defined intervals regularly and repeatedly. This was done in order to collect more data points as otherwise children would face just a few dialogs or none at all in their typically short sessions. Children‟s interactions with dialog boxes, such as what button they clicked, and the design factors employed were logged for analysis.  5.1.2 Participants There were 84 children who took part in this study. Each child faced 5 dialogs on average in a session lasting 14 minutes on average. Due to an oversight, the log files of 11 children from the first day of the study were incomplete and are hence not included in the quantitative data analysis. Our participants were roughly equally distributed across age groups and genders as shown in Table 5. The details of the distribution of our participants over each day and each iteration through the design factors can be found in Table 11 in Appendix B. Age group Number of children Boys Girls Total [3, 5] (5, 7] (7, 12]  14 12 11  9 15 12  23 27 23  Table 5 Distribution of the participants of the study across genders and age groups.  5.1.3 Methods We analyzed the data in terms of the age groups identified during our first study (3.2.1.1) using similar methods as the first study (3.1.3.1). Additionally, we used regression tests to 64  confirm significant interactions among design factors and children‟s age, gender.  5.1.4 Reading test We employed a reading test based on the San Diego quick assessment (LaPray & Ross, 1969), which required a child to read a series of words from a list that is segmented according to grade levels. The reading grade level of a child was determined when a child made three errors at a particular level. Due to the delayed approval of the ethics protocol amendment, we could only administer the reading test from day three, thus limiting the number of participants.  5.2 Findings In this section we report on the impact of the design factors presented in Chapter 4. We describe how factors such as skimming and split structure did not help the children as expected, while factors such as delayed-click, color coding, highlighting and switching positions had mixed results, and finally how factors such as visibility of body text/title gave us unexpected but interesting results. Data, graphs and more detailed reporting corresponding to the findings described here can be found in Appendix B. Judging by reaction time – when all you have is a hammer: everything looks like a nail The following quantitative results are based mostly on reaction time (RT) as it was the primary quantitative unit that was measured in the study. A feature is regarded as successful or beneficial if it speeds up RT and as unsuccessful if it slows down RT. Though this approach has pitfalls due to various sources of noises (5.3.1), we believe that it provides us with some useful albeit limited insights on children‟s interactions. Owing to the various sources of noise in the RT data and the uncontrolled nature of our 65  study which resulted in data that was not balanced across factors, the validity of the results and the conclusions we draw upon are limited.  5.2.1 Improving affordances 5.2.1.1  Split structures did not improve affordances  The split dialog structure did not improve the affordances of dialoguing. Indifference to split structure. The children did not seem to benefit from the split design. They did not, however, seem to be confused by it either. From our observation we could see no benefit of the split dialogs over the normal ones. The children seemed to interact with them indifferently. They did not even seem to notice the icon representing the user. Older children and boys were faster with split dialogs. The differences in RT between split and non-split dialogs increases with age; older children have faster RTs with split dialogs than younger children. Girls respond slower (by 40%) with split dialogs while boys respond faster (by 31%). But older girls are less affected by split dialogs than younger girls. Please refer to Appendix B.9 for more details.  5.2.2 Improving communication 5.2.2.1  Visibility of body text and title superfluous to communication  Body text and title do not seem to be necessary to communicate the content of the dialog, probably since the dialogs we designed had descriptive button text. Body text and title did not have much impact. We had expected children to be slowed down by hiding the body text and title. But none of the children seemed to be affected by the missing body text or title. Their interactions seemed no different regardless of whether or not the dialogs had body text and title. They did not ask the researchers for clarification when 66  they were missing. Text sped up literates but slowed down others. Pre- and semi-literates were slowed down when the title or body text was present, while literates got faster. Text consumption influenced by color coding. With color coding, the RTs of pre-literates remained similar irrespective of the amount of text on the dialog, while semi-literates and literates interacted slowly with increasing amounts of text. Pre-literates were the only group who seemed to process the text in a linear fashion with their RTs being directly proportional to the amount of text on the dialog, when buttons are not color coded. For other groups, RTs are typically not proportional to the amount of text on the dialog. Less was faster. Hiding body text and title seems to lead to faster interaction. Making the title visible increases RT by about 47% and additionally making the body text visible increases RT by another 23%. Hiding body text and title largely removes the gender difference in RTs. Please refer to Appendix B.15 for more details. 5.2.2.2  Level of difficulty of dialogs – Not all dialogs were created equal  We ran several computerized reading level analysis tests (described in Appendix B.17) on the text of the three dialogs - including the title, body and the response texts - to see if we could co-relate the reading level score of the dialogs to children‟s RTs. The tests gave scores indicative of the reading level required to comprehend the text on the dialogs. Dialog type Erase Save Quit  Flesch–Kincaid Grade Level 1.2 -0.8 -1.6  Gunning fog index 1.9 1.5 1.2  Coleman–Liau Index 8.1 6.6 4.3  Table 6 Readability scores from the three reading tests performed on the three dialogs.  RTs not proportional to the reading score of dialogs. According to the reading tests (B.17), 67  the most difficult among the dialogs is Erase, followed by Save and then Quit. But the RTs do not reflect their respective levels of reading difficulty. There seems to be other factors influencing the level of difficulty of the dialog. Please refer to Appendix B.17 for more details.  5.2.3 Indicating consequences to improve decision making 5.2.3.1  Color coding was partly successful at indicating consequences  Color coding seems successful at indicating the consequences as it slowed down clicking on the potentially disruptive option, possibly because children were made to think twice before clicking. But it interacts negatively with age and works better for girls. Disruptive options clicked more slowly but more frequently. Regardless of color coding, the safer option was always the one clicked most often. With color coding, the potentially disruptive options (coded in orange) required more time to be clicked upon on average but unexpectedly the frequency of clicking them also increases. In other words, color coding seemed to increase the probability of clicking on a potentially disruptive choice while delaying the interaction on it. The attention grabbing nature of the orange color (Ware, 2000) could be attributed to this increase in frequency. Pre-literates benefitted most while literates slowed down the most. Color coding benefited pre-literates the most with a 30% speedup. This was expected as pre-literates would be the most inclined to look for non-textual cues when interacting with dialogs. But with color coding, RTs of semi-literates and literates were slowed down by 8% and 98% respectively. As age and literacy levels increase, color coding slows down performance proportionately thus bringing the RTs across age groups to the same level. Color coding of buttons also seems to slow down children with high reading ability. RT 68  and reading level are inversely related, as expected, when the buttons are not color coded; but with color coding, no such relationship exists. Girls seemed to imbibe color coding better. Overall girls‟ RTs remain relatively unaffected by color coding while boys are slowed down a little (9%). But when color coded, girls are faster than boys at hitting the safer alternatives (coded in green) while slower at hitting the potentially disruptive options (coded in orange), implying that they could be reacting to the color coding better than boys. Color coding benefitted difficult dialogs more. The benefit of color coding is not uniform across dialogs. Higher the reading level score of a dialog (B.17) more is the speed-up by color coding, especially for pre-literates. Please refer to Appendix B.11 for more details. 5.2.3.2  Highlighting – the runner up to color coding  Highlighting sped up the interaction when the safer options ended up being clicked, which makes sense as the safer options were the ones highlighted. But it does not seem to affect the interaction when the non-highlighted option ended up being clicked upon. It is possible that highlighting drew attention towards the safer option and hence helped speedup clicking on it. Not as powerful as color coding. We noticed that though highlighting attracted attention, it did not seem to act as a strong cue. For example it was not as strong as color coding signal, convincing children to act on its cue. Children seemed to pause a moment before selecting the highlighted option, unlike in color coding. It could be because the highlighting as a cue was not strong enough or could be that it was not clear enough as how the cue should be interpreted. In fact the quantitative analysis shows that color coding made highlighting 69  redundant. Highlighting also did not seem to slow down children with higher reading ability as much as color coding did. Literates performed poorly, girls unaffected while boys benefitted. Pre- and semi-literates benefitted similarly from highlighting (23% and 28% speed-up respectively), while literates were hindered (slowed down by 36%). Girls were relatively unaffected by highlighting (slowed down by 6%) while boys benefitted by it greatly (56% speed-up). Yellow better than green to highlight. The color used for highlighting seemed to make a difference. Highlighting with green seemed to slow down the interaction especially when the buttons were color coded, while yellow seemed to speed up the interaction. The slowing down resulting from the green color before we switched to yellow could be attributed to the interaction between green used in color coding the safe button and the green used in highlighting. This may be similar to the Stroop effect (Stroop, 1935). Please refer to Appendix B.12 for more details.  5.2.4 Making arbitrary clicks safer 5.2.4.1  Delayed-click made it safer at a cost  Though delayed-click increased the chances of clicking on the safer “I don‟t know” button, children did not like being made to wait before clicking. Disliked waiting. Children did not like this feature, as it made them wait before they could click. We kept tweaking and reducing the time interval. But we were not successful in finding the right timing interval that satisfied the children and could also solve the problem. Boys seemed to express more dislike for this feature. Delayed-click increased RTs by about 2.5 seconds (30%). But this is roughly the time taken for the delayed-click to come to effect, if we consider a 1 second delay between 70  buttons. If we remove the time delay introduced by the effect itself, we could say that the children‟s RTs are not affected by delayed-click. “I don’t know” gets clicked more with delayed-click. Without delayed-click majority of the children did not click the “I don‟t know” button. With delayed-click, the frequency and speed of clicking the “I don‟t know” button, which was revealed first as part of the effect, goes up at the expense of the last revealed button. In this respect, delayed-click accomplishes its design goal. Please refer to Appendix B.10 for more details. 5.2.4.2  Switching position provides mixed results on children’s interaction  Switching button positions provides a better understanding of the role of button position as a cue when interacting with the dialog. But the results are complex, requiring nuanced interpretation and further research. Poorly understood feature. It seemed like the position of the button (left vs. right side) was one of the cues that children used to interact with the dialog, as we initially observed some children clicking more frequently on buttons of a particular side. But we were not able to observe strong recurring patterns among children showing preference for a particular side. When the position was switched, we did not observe any visible reaction from the children, such as surprise or annoyance. Girls unaffected, boys benefitted. Boys seemed to benefit by switching (35% speed-up) while girls seemed to be relatively unaffected by switching (12% slow-down). Older children perform poorly when compared to younger children when the button positions were switched. Interaction with other factors. Color coding the buttons or having more text on the dialog 71  (title, body text) seemed to reduce the negative impact of switching button positions in terms of RT. Highlighting did not seem to help similarly. Unexpectedly, the potentially disruptive options seemed to enjoy a better chance of getting clicked (and also clicked faster) after their positions were switched while the safer options seemed to be less popular after their positions were switched. Please refer to Appendix B.13 for more details.  5.2.5 Addressing impatience 5.2.5.1  Impatience could not be addressed by skimming  The children did not react any faster for the most part with skimming factors. The factors intended to supporting skimming thus failed to achieve their goal. Preference for icons across ages. Contradicting our expectations, children neither seemed to notice the contrast differences in the body text nor glean the meaning of the icons. The button icons that we used in our study were designed in haste and hence had come out poorly8. This might have made it difficult for pre-literates to make use of the icons. But from study interviews we gathered that pre-literates (to a large degree) and semi-literates (to a lesser extent) preferred having icons in the buttons even though most of them had no clue what they meant. Girls seemed to show more preference for icons than boys. On the other hand, most literates expressed negative or no preference for the icons. They were of the opinion that icons were meant for younger children. This echoes observations from past research (Budiu & Nielsen, 2010; Large & Beheshti, 2005). We realized that designing effective icons for  8  The quality of the icons that we used in the study is considered poor based on the opinions of peers and HCI researchers at UBC. In retrospect we could clearly see how poorly the icons have been designed. But since the opinions had been sought very close to the time of the study, we did not have sufficient time to redesign them.  72  children across ages and gender is in itself an important research topic deserving more attention that is beyond the scope of our research. Pre-literates and girls slowed down while literates were indifferent. Pre-literates were slowed down (by 60%) by the presence of the skimming factors while semi-literates showed a speed-up (of 30%). Literates were relatively unaffected (2% slow-down). Girls were slowed down (by 44%). Boys showed mixed results by age but were more influenced by these factors than girls - slowed down in the case of pre-literates or sped up in case of other age groups. Overall buttons without icons sped up the interaction (by 15%). Please refer to Appendix B.8 for more details.  5.2.6 Other results 5.2.6.1  Children exhibited symptoms of learning as the session progressed  We were interested to see if there were any learning effects or other such similar effects of familiarization as the session progressed and children interacted with increasing number of dialogs. Apart from that the children usually stopped seeking help after the first 2 to 4 dialogs, we did not observe any interaction patterns indicative of learning as the sessions progressed. But if we defined learning as reduction of RTs as sequence increases, we did observe some patterns. Semi-literates and girls showed learning. Semi-literates seemed to exhibit a stronger learning effect than other groups. Girls seemed to experience a learning effect while boys did not. Factors influencing learning. Color coding did not seem to affect learning, unexpectedly even for pre-literates. Visibility of title did not seem to influence learning while visibility of body text seemed to affect it negatively. This might be because without the body text there 73  were far fewer words on the dialog to learn and it might hence become faster to interact with in subsequent passes. Please refer to Appendix B.18 for more details. 5.2.6.2  Differences in reaction times  Semi-literates and girls faster. Overall, pre-literates, not surprisingly, had the slowest RTs. Semi-literates had the fastest RTs; unexpectedly even faster than literates (by 20%). Boys had longer sessions than girls (similar to the first study). But overall girls had faster RTs than boys (by 21%) even across the same reading level. Age was not a good indicator of reading level. There seems to be no simple linear relationship between reading level and RT as we had expected. RTs decreased steadily as reading level increased from 1 to 5, but started to increase after 5. This might be due to the small number of children with a reading level higher than 5 in our study. It is also worth noting the huge variability of reading level within age groups. Pre-literates‟ reading levels were spread from 1 to 4, semi-literates from 1 to 6 and literates from 4 to 11. This suggests that age might not serve as a good indicator of reading level. In fact, Riley also observed a 5 year range in terms of variation in literacy skills among children in same classrooms (Riley, 1996). Please refer to Appendix B.16 for more details. 5.2.6.3  Interaction with split structure  There was a negative interaction between split dialog and the skimming factors; this negative interaction reduced with age. This interaction seemed to be further worsened when delayed-click was also employed. Hence support for skimming generally did well with nonsplit (normal) dialogs without delayed-click. 74  Split structure and delayed-click. With delayed-click, the difference between the default and the split dialog was lost. With delayed-click, boys seemed to lose a bit of their advantage with split dialogs and the girls got much slower.  5.2.7 Summary of findings Table 7 summarizes the quantitative results showing how children of different age groups and gender interacted with each of the design factors. The relative RT speed-up and slowdown between age groups and gender as caused by a design factor is denoted using two shades of green (speed-up) and two shades of red (slow-down). The darker shades represent a relatively larger speed-up or slow-down and are double coded with two arrows. Gray denotes RTs that were relatively unaffected, while a white cell denotes that there was insufficient data to represent the particular condition. The relative RTs as denoted by the different shades of green and red are based on the comparison of data within a row (e.g., pre-literates vs. semi-literates on split-structure) and do not necessarily hold as a comparison between rows (e.g., split-structure vs. highlighting for boys). Design factor  Age groups Gender Pre-literates Semi-literates Literates Boys Girls  Split-structure Making body text, title visible Color coding buttons Highlighting “I don‟t know” button with Delayed-click Switching button positions Supporting skimming Table 7 Summary of how different age groups and gender interacted with the design factors.  Pre-literates were sped up by non-textual cues such as color coding, highlighting and switching of button positions. They were slowed down by having more text or factors supporting skimming on the dialog and by delayed-click. Semi-literates were slowed down by having more text on the dialog. They were neutral  75  to delayed click. They were sped up by all other factors such as split structure, color coding, highlighting, switching and skimming. Literates were sped up by split structure and having more text on the dialog. They were neutral to skimming factors. They were slowed down by all other factors such as color coding, highlighting, delayed-click and switching. Overall hiding body text and title seemed to lead to faster interaction. They seemed to be superfluous to communicating the content of the dialog especially in the presence of descriptive button text. Color coding delayed the clicking of the potentially disruptive choice. Color coding interacts negatively with age. Delayed-click increased the chances of clicking on the “I don‟t know button” so much so that we could not analyze them separately on their own. Girls interacted differently than boys. Girls interacted faster than boys - even among those who were at the same reading level. Girls seemed to experience a learning effect while boys did not. Girls‟ RTs were unaffected by factors such as color coding, highlighting and switching of button positions while boys were benefitted by each of these factors. Highlighting vs. color coding. Highlighting seemed to be a weaker signal when compared to button color coding in terms of the effect on interaction. For example, highlighting did not interact negatively with reading ability in the way that color coding did. When both were present, color coding made highlighting redundant.  5.3 Discussion In this section we list the challenges and limitations in conducting filed studies with  76  children in the wild, followed by a discussion of some general differences in the interaction patterns of children across age groups. We then present our theory of how children process information from dialogs in two phases, followed by a model of information consumption based on how children from different age groups seemed to be consuming information from the different information channels (textual vs. non-textual). A discussion on how the different information channels interact with each other and the effects of their interaction on children‟s behavior follows. We conclude this section with a discussion of how text on dialogs constitutes contextual cues for interaction and on formulating a level of difficulty of interacting with a dialog.  5.3.1 Challenges with conducting field studies with children in the wild The analysis that follows is based primarily on the Reaction time (RT) data collected from the log files. RT is the time elapsed between the dialog appearing on the screen and a choice on the dialog being clicked upon. But the collected RT data has the following sources of noise. 5.3.1.1  Intention: an elusive measurement  With the current study design we were not able to differentiate between dialogs that were dismissed randomly with the child possibly having no clue on what the dialog was about, and those where the child playfully interacted with the dialog choosing options out of curiosity to see the outcome, or those where the child had some idea on what was going on but then incorrectly made a choice that did not match his/her intention, or the perfect case where the child chose what he/she wanted with full knowledge. This implies that a faster RT might mean any of the following: The child is actually faster at reading, processing information and clicking on the  77  responses. The child processed only a subset of information on the dialog. The child clicked on a choice randomly without processing any information on the dialog.  It should be noted that the amount of information that is considered sufficient or good enough to make the decision at hand, i.e. satisficing (Klein G. , 1998) could vary individually. The behavior of a child could have also changed with time. A child dismissing a dialog in a certain manner, for example randomly without processing much information does not mean that he/she dismisses all other dialogs similarly. This could be the reason why we did not have any clear and simple patterns with respect to the RTs of an individual child over the course of time nor did we find similar RTs for instances of the same dialog or dialogs with similar design factors. 5.3.1.2  Parents’ involvement  Parents of pre-literates (and semi-literates to a lesser degree) typically sat with them during the sessions. They also took it upon themselves to provide all the help needed by their children, including - especially in the case of pre-literates - handling the mouse and interacting with the software at times when the child faced difficulty, for example when interacting with a dialog box. Though this behavior was quite understandable, it also meant that it acted as a source of noise in our data. 5.3.1.3  Delivering help  Most of the pre- and semi-literates required some kind of help from an adult (their parents or us, the researchers) to interact with the first few dialogs. The help offered typically involved verbally instructing the child on the purpose of the dialog and explaining the 78  consequences of the choices9. Delivering help and having the child understand it consumed some amount of time. This could have contributed to the observed pattern that the first couple of RTs in a session were higher than the average RT of the session. 5.3.1.4  Fitts’s law  Children interacted with a dialog much faster, if the mouse cursor happened to be near the dialog when it appeared on screen. We found evidence in our analysis where screen captures corresponding to very low RTs showed that the cursor happened to be at or near the part of the screen where the dialog appeared. Hence due to the above factors, faster RTs do not necessarily mean that the child was faster in processing the information on the dialog and similarly slower RTs do not necessarily mean the child was slower in processing10.  5.3.2 Differences in interaction across age and gender Brain development puts semi-literates at the top. Semi-literates faced the most number of dialogs and had the fastest RTs. The differences in number of dialogs could be stemming from the differences in the average session lengths. Brain development of children reaches a peak at around age 5 (Herman & Epstein, 1986). This could explain why semi-literates had the longest session lengths, stemming probably from sustained interest and also the fastest RTs. Girls interact faster. The development of fine motor skills is faster for girls than boys (Sax, 2005). This might have given an advantage to girls in handling the mouse and might have  9  We avoided interacting with the application on behalf of the children even in cases where they seemed to have no clue about the dialog. 10 But we are inclined to believe that the mean RTs of a child over the course of the session could be a better indicator than individual RTs.  79  translated to faster RTs. Girls also demonstrated higher reading ability than boys of the same age. This advantage in reading could have also contributed to faster RTs. Spatial processing. Older children and boys interacted faster with split dialogs. This could be attributed to how children become better at dealing with spatially distributed information structures such as split dialogs as they get older (Hanlon, Thatcher, & Cline, 1999). For semiliterates and literates, this increased ability to integrate spatially spread-out information could make them faster. This could be explained on the same grounds for girls as they exhibit lower spatial processing skills (Hanlon, Thatcher, & Cline, 1999). Attitude and Skills affect Learning. Both attitude and skills are required for effective learning (Miserandino, 1996). Semi-literates exhibited a stronger learning effect than other groups. This could be because that they have the right mix of both the skills and the attitude for learning to happen. They might have taken learning in this particular context (of a painting application in an environment such as Science World) seriously enough and they also have the supporting skills for effective learning to take place. Pre-literates and literates might have one but not the other. This difference in attitude along with differences in reading skills might explain why girls seemed to experience a learning effect while boys did not.  5.3.3 Two phases of information processing Based on observations and data analysis, we theorize that the processing of information on a dialog by a child is characterized by two fundamental phases. Similar phases in information processing have been identified by other researchers (Kuhlthau, 1991; Spencer, 2006) but our contribution is unique in that it is based on GUI elements such as dialogs. Phase1 - Learning phase. In this phase, which typically lasted between 2 to 4 dialogs, the children seemed to scan the information on the dialog and register key information points 80  such as what the dialog was about and what the choices and implications were. This could have been accompanied, either consciously or autonomously, by registering the corresponding visual cues (Chun & Jiang, 1998) that could help in later recognition. Phase2 - Recognition phase. In this phase, children typically would not seek to gather new information from the dialog. It would primarily be a task of matching with memory. The children would match the cues of the dialog from memory and recall what it was about (Chun & Jiang, 1998), then match the intended purpose of the dialog with the current context and then decide on a choice depending upon the context and desired outcome. If recalling from memory failed at any point, the learning phase could be re-initiated. The learning and recognition phases could also be thought of as occurring in stepped cycles. During a recognition phase when a child feels that the knowledge held about the dialog is insufficient, for example when a choice made without fully understanding its consequences leads to undesirable outcomes, the learning phase might be initiated. These stepped phases might explain why the sequence of RTs does not decrease linearly over time as one might expect if learning were to take place in a linear fashion. It has been observed that users tend to read instructions only after their trial and error approach has failed to give results (Krug, 2000). While the learning phase is expected to occur before the recognition phase, it might not always be the case. At first, the child might have dismissed the dialog randomly or without sufficiently processing the information on the dialog to make an informed choice (Klein G. , 1998), probably thinking of it as an infrequent anomaly that must be dealt with in order to continue working on the application. But when the dialogs started popping up one after another at frequent intervals, the child might have felt a little uncomfortable in dismissing it randomly and hence might have initiated or switched to the 81  learning phase. Hawthorne effect. These phases might have been affected by the Hawthorne effect. The fact that there were adults, researchers and parents, watching over their shoulders might have made the children, especially the literates (Baumgarten, 2003), less inclined to dismiss dialogs randomly. Though we did not reveal to the children or the parents that dialogs were the object of our study, it is possible that the children got a sense of it after interacting with a couple of dialogs and facing questions from us regarding dialogs.  5.3.4 Towards a model for information consumption by age We present an information consumption model based on the differences in how children of different age groups rely on different channels of information (textual vs. non-textual). 5.3.4.1  Role of position and content  By the length of the text contained, the body text can be considered to have more information on it than the title. But the processing times hinted by the difference in the RTs are not proportional. This could be because not all the information on the dialog is processed. For example, more of the information from the title could have been processed than the body text. Similar patterns have been noticed in web users, who read the top portions of pages more thoroughly but resort to scanning as they move down the page (Nielsen, 2006). The percentage of information that is processed from a source has been observed to be indirectly proportional to the amount of information on it (Nielsen, 2003). Children with growing impatience fuelled by the abundance of information around them (Rideout, Foehr, & Roberts, 2010) could be trying to complete the task at hand with as little information processing as possible (Krug, 2000). These might explain the disproportional difference in RTs.  82  5.3.4.2  Mechanisms for information consumption across age groups  As we move across the age groups from pre-literates to literates, adding the body text makes less of a difference. The differences in how children from various age groups process information, more than the speed with which they process the information could explain this behavior. Moreover, pre-literates were slowed down by skimming factors (5.2.5.1) while semi-literates showed a speed-up and literates were indifferent. We theorize that this is owing to the difference in how the different age groups rely on different channels of information (textual vs. non-textual). We imagine that the textual channel with its higher cost of consumption along with higher accuracy11 and reliability12 lies at one end of the spectrum, while the non-textual channel with its lower cost of consumption accompanied with lower accuracy and reliability lies at the other end (Figure 23). Each channel could be associated with different sets of optimizations for consuming information from them – for example interacting by cues such as color, highlighting for the non-textual channel, while interacting by skimming, keywords for the textual channel. Children from different age groups could be seen consuming information from these channels differently using various mechanisms optimized to varying degrees. For example, pre-literates could learn to interact with the dialog by using non-textual cues such as color coding, position of buttons and button icons. Though such cues might not be very accurate or reliable, they provide the best ROI13 for their skill set.  11  Accuracy is the success associated with the communication of information on the dialog. Reliability is the ability to successfully reuse the knowledge/cues associated with the channel across dialogs or with repeated instances of the same dialog. 13 ROI – Return on investment is used here in the context of the return on time/effort by investing in a set 12  83  Age / Reading level Pre-literate  Graphical objects  Semi-literate  Literate  Both graphical and text objects  Text objects  Figure 23 Reliance on different sources of information to varying degrees by different age groups.  Semi-literates could be using a mixture of cues from both channels - going for more objects objects reliable and sophisticated mechanisms such as cueing in on a set of keywords in the body text along with less sophisticated and less reliable cues such as color and position. They could also be using the graphical information as cues to enrich their visual snapshots in memory (Chun & Jiang, 1998) thus aiding speedier interaction. Though it is likely that they did not understand the intended meaning of icons – they might have used them primarily as graphical cues without decoding their meaning (DeLoache, Uttal, & Pierroutsakos, 1998). This could explain why they interact faster with factors intended to support skimming. Literates might be relying primarily on textual information ignoring non-textual cues due to their relatively higher ability to process textual information in a reliable manner. This could be why the skimming factors did not make a difference to them. They could also be using more sophisticated mechanisms on the textual channel, such as narrowing down on a  of mechanisms that optimize information retrieval  84  specific section of the interface to glean the information required for the interaction (Kunar, Flusberg, & Wolfe, 2008), for example focusing on the title, along with few other cue words (Chun & Jiang, 1998) from body text and button text, to identify and interact with the dialog. Thus at each stage along their development curve, the children could be using mechanisms that best suit their skill set and hence provide an attractive ROI.  5.3.5 Interaction of information channels We believe that the two channels of information (non-textual vs. textual) interact with each other under specific circumstances and that this interaction could explain some observed interaction patterns. Pre-literates‟ RTs were proportional to the amount of text on the dialog without color coding. This could mean that pre-literates are trying to process text information in a simple linear fashion instead of using sophisticated mechanisms such as skimming for keywords. But with color coding, the RTs remained the same irrespective of the amount of text. This could mean that the pre-literates ignored the text information when color coding is available, suggesting lack of interference between channels. Semi-literates on the other hand seemed to be slowed down proportionally to the amount of text on the dialog, with color coding. This suggests that semi-literates could be using a mixture of information channels containing both text and color and also the presence of interaction. Literates had similar RTs irrespective of the amount of text on the dialog, without color coding. This could mean that this group is relying primarily on textual cues. But they performed poorly with color coding irrespective of the amount of text on the dialog. This suggests the presence of a stronger interaction between the channels. Thus the interaction 85  grows stronger as we move from pre-literates to literates. Color coding of buttons seemed to slow down children with high reading ability similarly. Coded with strong primary colors such as green, orange and red, the color information channel seems to interfere with their textual channel and slow them down when they are consuming information from the dialog. Whereas the children from other groups, especially pre-literates are prone to rely primarily on non-textual cues such as color, and are hence in most cases indifferent to information from textual channel. The strength of this negative interaction between color coding and textual channels seems to increase with age.  5.3.6 Text as contextual cue Having more text on the dialog seems to offset the slowing down of RTs when switching button positions. If we imagined the words on the dialog as providing more contextual cues (Chun & Jiang, 1998) in identifying the dialog, then having more words could reduce the effect of some missing or misplaced cues (e.g., switched words). If we can think of words as being the signal in the textual channel and switching positions as adding noise, then having more words reduces the signal to noise ratio. We can also think of the various information objects on the dialog being stored as sort of a picture map in memory (Chun & Jiang, 1998) in learning phase (5.3.3). During the recognition phase, if the various information objects on the dialog are present closer to the original form learned into memory, it might facilitate better recall to that extent and consequently facilitate faster interaction with the dialog. The children who rely more on textual cues to interact with the dialogs will hence be the most affected by the switching of the button position. Reading ability and hence the reliance on words to interact with the dialog increases with age and gender, with girls possessing better reading ability than boys of the same age (Sax, 2005). This could explain why older 86  children and girls perform poorly when the button positions are switched.  5.3.7 Formulating a level of difficulty for a dialog Users are confused if the distinctions between the various choices are not clear enough (Krug, 2000). But what other factors could determine the difficulty in making a choice? We propose that the level of difficulty of interacting with a dialog to be based on The degree of separation among the various choices and their implications on the dialog, i.e. ambiguity among the options on the dialog The degree of complexity of the information and underlying abstractions required to be understood to interact with the dialog  As reading level of a child increases, the speed with which the Erase dialog is dismissed increased. This could be attributed to the grade-level readability score (B.17) of the Erase dialog which is the highest among the 3 dialogs and also to the clear dichotomy between the choices in the Erase dialog relative to other dialogs. Save seemed to require the most time to respond compared to other dialogs irrespective of reading level as it could be considered to be most difficult in both aspects of abstraction and dichotomy. The dichotomy between the implications of the options - Saving now and not saving now – could be considered as the lowest among the 3 dialogs. Quit with the implications of being done and wanting to paint and, Erase with the implications of erasing everything to start over and continuing to paint can be thought of as offering relatively larger degrees of separation among the choices than the Save dialog. Further, to understand the Save dialog, one needs to understand the file system abstractions and the concept of saving, which might be difficult especially for pre- and semi-literates. The level of difficulty of interacting with a dialog box thus seems to depend on both the 87  grade level index of the text, the characteristics of the choices on the dialog and the abstractions behind it.  5.4 Design implications We present the design implications based on our observations and analysis according to the age and gender of the target audience. This could help designers to construct interfaces that adapt to the age and gender of the children.  5.4.1 Designing for gender Girls exhibit better reading ability, patience and thoroughness in processing the information on interfaces than boys. They are not benefited by non-textual cues such as color coding or highlighting. This implies that we could present textual information to girls and expect them to interact reliably without non-textual cues. For boys, information should be presented in a format that assists easy and fast consumption – text augmented with nontextual cues such as icons, support for skimming, color coding etc. Girls do not perform well with spatially distributed structures such as split dialogs. Hence information should be presented to them in spatially cohesive structures. Girls also tend to take fewer risks, concentrate on the task at hand and do not explore the software much. Hence designers could put more effort in trying to expose the functionality of the system, for example by revealing undiscovered functions related to their current task.  5.4.2 Designing for age Literates perform well with textual information and poorly with non-textual cues. Hence powerful cues such as color-coding should be avoided for literates. Subtler cues such as highlighting do not interfere as much. Hence if cues need to be used on the interface, subtler 88  cues such as highlighting14could be used. Literates seem to have evolved highly efficient information processing mechanisms such as focusing on the relevant sub-section of the interface. They perform poorly when interface elements such as buttons are switched positions. Hence they could take advantage of information presented hierarchically in perceptual layers (e.g., prominent title first, then body text) and in consistent positions. Since semi-literates seem to be using a mixture of information channels, information should be presented to them using both text and non-textual cues such as color-coding, icons, and highlighting. They also do better with minimal text information such as just title text, possibly owing to their limited ability in processing text. Hence the text information presented should contain just a few essential key words and be supplemented with nontextual cues. Pre-literates are the biggest benefactors of non-textual cues such as color coding and highlighting. They perform the worst when there is just textual information without any nontextual cues. They also have a preference for graphical objects such as icons. Hence interfaces designed for pre-literates should communicate the information primarily using non-textual cues such as icons, color coding and highlighting. A minimal amount of text can be added to the interface for the benefit of those pre-literates who can read, to encourage and educate others who cannot read yet and for the benefit of the adults who want to help the children. When designing for pre-literates and semi-literates, care should be taken to avoid  14  The manner in which the cue is implemented could also determine its power. For example, highlighting using bright red color instead of yellow could make highlighting as powerful and hence as interfering as color coding.  89  representing abstractions unique to the digital domain such as file systems and actions associated with them such as save and overwrite as children will likely have difficulty understanding them. Reading level varies widely within a particular age and appears to influence interaction more strongly than age. Hence designers should consider building quick reading tests into their software in order to profile children and customize the presentation of information accordingly.  5.4.3 Designing for children in general In general, when presenting multiple alternative choices, designers should strive to maximize the distinctiveness among the choices while minimizing ambiguity – as it plays an important role in influencing interaction. Moreover, since children dislike waiting, intentional delays of any kind should be avoided. Care should also be exerted in choosing colors to code information on the interface. Color coding a choice with orange increases the probability of clicking on it, while delaying the interaction on it. This could imply that though colors such as orange delays users by acting as a visual warning, they could also attract attention and result in more clicks. When the same color is used to code multiple pieces of information on the interface (e.g., using green to both highlight and color code a button), performance suffered, suggesting a possible interference in processing the visual signals. Similarly, using multiple non-textual cues might be redundant. For example, highlighting was ineffective when buttons were also color coded. Hence when we, as designers, are tempted to use more than one non-textual cue, especially directed towards achieving the same goal, we need to carefully reconsider our decision. The negative effect of switching buttons was offset by increasing the textual information 90  on the dialog. Thus when there are changes on the interface that affects some contextual cues, care should be taken to maintain a healthy signal to noise ratio, by having other stable cues that offset the changes.  5.5 Summary In this observational study we observed how children interacted with the various design factors of the dialogs according to their age and gender. The design goals, motivated by our first exploratory study presented in Chapter 3, were aimed at improving affordances and communication, illustrating consequences and addressing impatience. In this study we employed split structure to improve affordances, tested the role of title and body text in communicating information on dialogs by manipulating their visibility, color coded and highlighted the buttons to indicate the consequences of clicking on them and encourage or guide children towards clicking the safer choice, made arbitrary clicks safer by adding a “I don‟t know” button and by revealing buttons in a sequence with a delay in between, switched position of the buttons to better understand the role of positions in children‟s choices, and addressed impatience by supporting skimming of the dialog‟s content. We found that the split structure did not improve affordances. Visibility of title and body text did not contribute much to communication. Color coding and highlighting were successful at indicating consequences only for pre-literates and semi-literates. Minimizing arbitrary clicks by introducing a delay and “I don‟t know” button is not straightforward. Impatience could not be addressed by the skimming factors. Switching button positions did not help us to clearly understand the role of positions. 91  We found that girls have different interaction patterns than boys and that it also varies greatly by age group. We put forth a theory of how children consumed information from the dialog. We discussed a model of information consumption in which children of different age and gender consumed information differently from different information channels (textual and non-textual) – with pre-literates relying primarily on non-textual cues at one end of the information spectrum, literates relying primarily on textual cues at the other end and semiliterates relying on a mixture of sources in the middle of the spectrum. Children also develop sophisticated mechanisms for consuming information as they age, such as using keywords and focusing on a subset of the information space. We discussed on how these channels interact with each other and how this affects children‟s interaction. We formulated a level of difficulty of interacting with a dialog that depends both on the ambiguity or clear separation among choices as well as the complexity of the abstractions. We presented the design implications that will help designers in constructing adaptive interfaces that improve the interaction by taking the age and gender into account.  92  Chapter 6: Conclusions and future work Children‟s abilities vary greatly by gender (Sax, 2005) and age (Lueder & Rice, 2008). We started this research with the broad goal of designing adaptive interfaces that improve usability by taking the age and gender of a child into account.  6.1 Contributions In order to observe how children interact differently by age and gender, and the general difficulties they face when interacting with software, we conducted an exploratory study (Chapter 3). We were initially interested in designing a help system that could enable the children to interact independently with minimal adult help. We ended up identifying several challenges in designing an effective help system for children. We also identified dialog boxes to be one of the significant challenges faced by children and classified the various challenges by age groups. We came up with several design solutions targeted at different age groups (Chapter 4). We conducted a follow-up observational study (Chapter 5) to explore how children of different age groups interact with these design factors. We found that a number of the design factors did not work as we had expected. Based on the data, we put forth several theories of information processing – on how children consume information from the dialog, on the existence of information channels and how children of different age groups and gender consume information differently from these different channels, on the interaction between these channels, and the effect of their interaction on children‟s behavior. We proposed several design implications based on our observations and analysis. We believe that these can be used as a foundation for building age and gender appropriate adaptive interfaces. 93  In summary, our contributions are identifying and classifying children‟s problems with dialogs by age groups, proposing design solutions targeted at different age groups, observing how children of different age groups interacted with the designs, proposing theories on how children process information differently by age and gender, and putting forth design implications based on our observations on how to design for children taking age and gender into account.  6.2 Limitations Uncontrolled study. The biggest limitation of our research stems from the uncontrolled nature of the study which led to unbalanced factors and also contributed to various sources of noise in the RT data (5.3.1). This limits the strength and validity of our conclusions. The informal study setup, in addition to resulting in limited (and non-uniform) session lengths, also limited our ability to converse with or observe children in depth. Source of age-related differences. We assume that the differences in observed behavior among children of different age groups are a result of differences in capabilities rooted in developmental differences. But the differences could be attributed to other factors such as the differences in the cultural, social and technological environments into which the particular group of children studied were born into. We ignored, the effect of culture and the social standing of the parents on children (Tizard & Hughes, 1984) in our study, although we could imagine them to be powerful factors influencing parenting style, exposure to and outlook on technology, which in turn could influence children‟s interaction. Timed dialogs. Most of the dialogs in Tux Paint (and quite possibly in other applications as well), by default, are triggered in response to user/system events. But the dialogs used in our 94  study were triggered automatically at pre-determined intervals in order to maximize observation and data points. This could have affected many aspects of interaction such as causality. Under-represented literates. Children above the age of 10 in general did not seem enthusiastic to participate in our study. Their low participation made it challenging to compare the effect of many design factors across age groups. Considering the above limitations, this research should be considered as a preliminary step towards constructing adaptive interfaces for children.  6.3 Future work Controlled and counterbalanced studies. As our studies were uncontrolled and since some of the design factors did not work as we had expected, controlled experiments investigating a subset of design factors in more detail, with counter balanced factors across age and gender, should be the next step. Help system. Though the potential for an intelligent help system was clear, we identified several challenges in designing an effective help system for children. The way in which the help is presented has many poorly understood nuances. More research is needed in this area to validate the challenges and find solutions. Design of icons. Designing effective icons for children, especially for different ages, fell out of scope of our research. Research in design and evaluation of icons, similar to that performed with other demographics (Leung, McGrenere, & Graf, 2011), should be replicated for children. Further, icons could be animated to see if they attract attention and improve communicability (Baecker, Small, & Mander, 1991), for example, using a shrugging action 95  for the “I don‟t know” button. Less is more? The dialogs without either title or body text had the fastest RTs. But apart from RT, can we say that interaction with reduced information was better? It is possible that though the children seemed to have reacted faster they might have done so at lesser comfort levels. This might have translated into lesser hope of learning more about the software or mastering it eventually (Miserandino, 1996). More research is needed to verify that lesser amount of text is indeed better, not just faster. The button text of our dialogs had ample information to support interacting with the dialog. For semi-literates and literates this information was probably sufficient. We could test if simple button texts, such as „Yes‟, „No‟ instead of descriptive texts such as „Yes, I am done‟, „No, I want to paint‟, are sufficient. Switching positions. This research fails to find plausible explanations as to why certain groups of children get faster when the button positions are switched. More research is needed to learn about how children use position as a cue to interact with dialogs. Touch screens. As the motor skills of pre-literates are not fine tuned yet (Lueder & Rice, 2008), they had great difficulty in using the mouse for interacting. It made them slower and dependent on adults. This could have also disproportionately affected the performance of boys, as their motor skills lag behind girls (Sax, 2005). Touch-screens have been observed to improve children‟s interaction (Battenberg & Merbler, 1989) and even bring their performance closer to adults (Scaife & Bond, 1991). We wonder how our observations and analysis would have been different if we had used more usable input technologies such as touch interfaces. Culture. Our learning to interact with computers is not culture-agnostic, but rather a complex process of socialization and enculturation. But cross-cultural empirical studies from the 96  developmental perspective are limited (Yan & Fischer, 2004) pointing to the potential for more research in this area. Such research, along with the research on age and gender differences, could improve the intelligence behind the adaptive interfaces we envision. Longitudinal studies. Technological expertise influences interaction (Yan & Fischer, 2004; Budiu & Nielsen, 2010; Baecker, Small, & Mander, 1991). In two studies conducted a decade apart, Budiu & Nielsen found that the technological expertise has shifted down among age groups – behaviors of middle-aged children (6 to 8) seen a decade ago were now the behavior of the youngest children (3 to 5) (Budiu & Nielsen, 2010). As familiarity with technology increases and children‟s exposure to technology happening earlier with each passing year (Rideout, Foehr, & Roberts, 2010), it remains to be seen how the increased technological expertise of younger children matures over time and how it solves existing problems and introduces new ones. A world without dialogs. Several alternative solutions have been suggested to avoid using dialog boxes. A general recommendation is to make the application smart enough so that it does not bother users with confirmations, while providing helpful modeless feedback (Cooper, Reimann, & Cronin, 2007). For example designers could make all actions reversible by having the application support undo operation instead of warning the user about a potentially disruptive situation (Cooper, Reimann, & Cronin, 2007; Raskin, 2007). Future research is required to see if these are indeed viable solutions, especially for children.  6.4 Closing comments Children are being exposed to increasing doses of technology and are starting to use computers at a much earlier age than ever before. Parents are busy and are often unavailable 97  to assist their children in the home; teachers too are busy and cannot attend fully to each child in the school. This hurts children‟s learning as do interfaces that are designed in ignorance of the enormous differences in age and gender among children. Hence, interfaces that adapt themselves to the age and gender of the child and help the child to interact independently are valuable. A review of literature reveals that most of the interaction design research with children is of applied nature, thus highlighting the need for conducting more basic research aimed at a deeper understanding of children‟s evolving abilities as they age and at the development of guidelines based on how this affects children‟s interaction with technologies (Hourcade, 2008). Our research contributes preliminary guidelines for the design and evaluation of adaptive software that takes age and gender of children into account. We hope that researchers will continue to work on the important problem of making software more usable and developmentally appropriate for children.  98  Bibliography Baecker, R., Small, I., & Mander, R. (1991). Bringing icons to life. Proceedings of the SIGCHI conference on Human factors in computing systems: Reaching through technology (pp. 1--6). New York, NY, US: ACM. Battenberg, J. K., & Merbler, J. B. (1989). Touch Screen versus Keyboard: A Comparison of Task Performance of Young Children. Journal of Special Education Technology , v10 n2 p24-28. Baumgarten, M. (2003). Kids and the internet: a developmental summary. Comput. Entertain. , Volume 1 Issue 1. Berghel, H. (1997, February). Cyberspace 2000: dealing with information overload. Communications, ACM 40, 2 , pp. 19-24. Budiu, R., & Nielsen, J. (2010). Usability of Websites for Children: Design guidelines for Targeting Users Aged 3-12 Years, 2nd edition. USA: Nielson Norman Group. Campbell, F. A., Goldman, B. D., Boccia, M. L., & Skinner, M. (2004). The effect of format modifications and reading comprehension on recall of informed consent information by low-income parents: a comparison of print, video, and computer-based presentations. Patient Education and Counseling , 53, 205 - 216. Carmody, T. (2010). 10 Reading Revolutions Before E-Books. Retrieved August 2011, from The Atlantic: http://www.theatlantic.com/technology/archive/2010/08/10-readingrevolutions-before-e-books/62004/ Chi, E., Gumbrecht, M., & Hong, L. (2007). Visual Foraging of Highlighted Text: An EyeTracking Study. In J. Jacko, Human-Computer Interaction. HCI Intelligent Multimodal Interaction Environments (pp. 589-598). Berlin / Heidelberg: Springer . 99  Chun, M. M., & Jiang, Y. (1998). Contextual Cueing: Implicit Learning and Memory of Visual Context Guides Spatial Attention. Cognitive Psychology , 36, 28 - 71. Cohen, M. A. (2009). Auditory recognition memory is inferior to visual recognition memory. PNAS : Proceedings of the National Academy of Sciences , 6008. Cooper, A., Reimann, R., & Cronin, D. (2007). About Face 3: The Essentials of Interaction Design. New York, NY, USA: John Wiley & Sons, Inc. Criticker. (2011). Personalized movie recommendation service. Retrieved from Criticker: http://criticker.com/ Davenport, T. H., & Beck, J. C. (May 2001). The Attention economy. Ubiquity . DeLoache, J. S., Uttal, D. H., & Pierroutsakos, S. L. (1998). The Development of early Symbolization: Educational implications. Learning and Instructions , Vol. 8, pp 325. Druin, A. (2002). Age matters. SIGCHI Bull.: suppl. interactions , 5--5. Druin, A. (2002). The Role of Children in the Design of New Technology. Behaviour and Information Technology , 21(1), pp. 1--25. Eisner, W. (2011). PS Magazine: The Best of The Preventive Maintenance Monthly . Abrams ComicArts . Floyd, M., Bocarro, J., Smith, W., & Baran, P. (2001). Park-based physical activity among children and adolescents. Am J Prev Med , 41(3):258-65. Gelderblom, H., & Kotzé, P. (2008). Designing technology for young children: what we can learn from theories of cognitive development. Proceedings of the 2008 annual research conference of the South African Institute of Computer Scientists and Information Technologists on IT research in developing countries: riding the wave of technology (pp. 66--75). New York, NY, USA: ACM. 100  Gelderblom, H., & Kotzé, P. (2009). Ten design lessons from the literature on child development and children's use of technology. Proceedings of the 8th International Conference on Interaction Design and Children (pp. 52--60). New York, NY, USA: ACM. Glaser, B., & Strauss, A. (1967). The discovery of Grounded Theory: Strategies for qualitative research. Chicago: Aldine. Grossman, T., & Fitzmaurice, G. (2010). ToolClips: an investigation of contextual video assistance for functionality understanding. In Proceedings of the 28th international conference on Human factors in computing systems (CHI '10). (pp. 1515-1524). New York, NY, USA: ACM. Gulli, A. (2005). The anatomy of a news search engine. In Special interest tracks and posters of the 14th international conference on World Wide Web (WWW '05) (pp. 880-881). New York, NY, USA: ACM. Hanlon, H., Thatcher, R., & Cline, M. (1999). Gender differences in the development of EEG coherence in normal children. Developmental Neuropsychology , 16(3):479–506. Herman, T., & Epstein. (1986). Stages in human brain development. Developmental Brain Research , Volume 30, Issue 1, 114 - 119. Higginbotham, S. (2011). Children and streaming don’t always mix. Retrieved Oct 2011, from Gigaom: http://gigaom.com/broadband/children-and-streaming-dont-always-mix Hill, W., & Terveen, L. (1996). Using frequency-of-mention in public conversations for social filtering. In Proceedings of the 1996 ACM conference on Computer supported cooperative work (CSCW '96) (pp. 106-11). New York, NY, USA: Mark S. Ackerman (Ed.). ACM. 101  Holt, J. (1991). How Children Learn. Penguin Books. Hourcade, J. (2008). Interaction Design and Children. Foundations and Trends in Human– Computer Interaction , 1(4), 277-392. IDC. (2011). Proceedings of the 10th International Conference on Interaction Design and Children. New York, NY, USA: ACM. Iijima, M., Arisaka, O., Minamoto, F., & Arai, Y. (2001). Sex Differences in Children's Free Drawings: A Study on Girls with Congenital Adrenal Hyperplasia. Hormones and Behavior , Volume 40, Issue 2, Pages 99-104. Johnson, E. J. (2003, 11). Do Defaults Save Lives? Science , p. 1338. Kay, M., & Terry, M. (2010). Textured agreements: re-envisioning electronic consent. In Proceedings of the Sixth Symposium on Usable Privacy and Security (SOUPS '10) (pp. Article 13, 13 pages). New York, NY, USA: ACM. Klein, G. (1998). Sources of Power: How People Make Decisions. Cambridge, MA: MIT Press. Klein, P. S., Nir-Gal, O., & Darom, E. (2000). The Use of Computers in Kindergarten, With or Without Adult Mediation: Effects of Children's Cognitive Performance and Behavior. Computers in Human Behavior , 16(6), 591--608. Krug, S. (2000). Don'T Make Me Think!: A Common Sense Approach to Web Usability (1st ed.). Indianapolis, IN, USA: Que Corp. Kuhlthau, C. C. (1991). Inside the Search Process: Information Seeking from the User's Perspective. Journal of the American Society for Information Science , 42(5), 361-371. Kunar, M., Flusberg, S., & Wolfe, J. (2008). The role of memory and restricted context in repeated visual search. Attention, Perception, & Psychophysics , 70, 314-328. 102  LaPray, M., & Ross, R. (1969). The graded word list: Quick gauge of reading ability. Journal of Reading , 12, 305–307. Large, A., & Beheshti, J. (2005). Interface design, web portals, and children. Library Trends. 54, 2 (2005) , 318--342. Lepper, M. R., & Gurtner, J. (1989). Children and computers: Approaching the twenty-first century. American Psychologist , 44(2), 170-178. Leung, R., McGrenere, J., & Graf, P. (2011). Age-related differences in the initial usability of mobile device icons. Behaviour & Information Technology , 30, 5, 629-642. Lueder, R., & Rice, V. (2008). Ergonomics for children: Designing products and places for toddlers to teens. New York: Taylor & Francis. Massironi, M. (2001). The psychology of graphic images: Seeing, drawing, communicating. Mahwah, N.J: Lawrence Erlbaum. Matejka, J., Li, W., Grossman, T., & Fitzmaurice, G. (2009). CommunityCommands: command recommendations for software applications. In Proceedings of the 22nd annual ACM symposium on User interface software and technology (UIST '09) (pp. 193-202). New York, NY, USA: ACM. McCloud, S. (2011). Google Chrome. Retrieved January 2, 2011, from http://www.google.com/googlebooks/chrome/ McCloud, S. (1993). Understanding Comics: The Invisible Art. New York: HarperPerrenial. Mccloud, S. (1994). Understanding Comics: The Invisible Art. Harper Collins. McGrenere, J., & Moore, G. (2000). Are We All In the Same "Bloat"? In Proceedings of Graphics Interface, (pp. 187-196 ). McKnight, L., & Read, J. C. (2009). Designing the 'record' button: using children's 103  understanding of icons to inform the design of a musical interface. Proceedings of the 8th International Conference on Interaction Design and Children (pp. 258--261). New York, NY, USA: ACM. McNamara, A., Bailey, R., & Grimm, C. (2009). Search task performance using subtle gaze direction with the presence of distractions. ACM Trans. Appl. Percept. , 17:1--17:19. Miserandino, M. (1996). Children who do well in school: Individual differences in perceived competence and autonomy in above-average children. Journal of Educational Psychology , 88(2), 203-214. Mullet, K., & Sano, D. (1995). Designing visual interfaces: communication oriented techniques. Upper Saddle River, NJ, USA: Prentice-Hall, Inc. Murphy-Hill, E., & Murphy, G. C. (2011). Peer Interaction Effectively , yet Infrequently , Enables Programmers to Discover New Tools. Human Factors , 405-414. NASA. (2011). ATTENTION MANAGEMENT. Retrieved August 2011, from Color usage research lab: http://colorusage.arc.nasa.gov/clutter.php NASA. (2011). CREATING PERCEPTUAL LAYERS. Retrieved August 2011, from Color usage research lab: http://colorusage.arc.nasa.gov/Percept_layers_1.php Netflix. (2011). Netflix launches "Just for Kids". Retrieved August 1, 2011, from The Netflix blog: http://blog.netflix.com/2011/08/netflix-launches-just-for-kids.html Nielsen, J. (2000). Designing Web usability: The practice of simplicity. Indianapolis, IN: New Riders. Nielsen, J. (2000). Designing Web usability: The practice of simplicity. Indianapolis, IN: New Riders. Nielsen, J. (2006). F-Shaped Pattern For Reading Web Content. Retrieved August 2011, 104  from Useit.com, Alertbox: http://www.useit.com/alertbox/reading_pattern.html Nielsen, J. (1997). How Users Read on the Web. Retrieved August 2011, from Useit.com, Alertbox: http://www.useit.com/alertbox/9710a.html Nielsen, J. (2003). Information Pollution. Retrieved August 2011, from Useit.com, Alertbox: http://www.useit.com/alertbox/20030811.html Nielsen, J. (2005). Lower-Literacy Users: Writing for a Broad Consumer Audience. Retrieved August 2011, from Useit.com, Alertbox: http://www.useit.com/alertbox/20050314.html Nielsen, J. (2005). Lower-Literacy Users: Writing for a Broad Consumer Audience. Retrieved August 2011, from Useit.com, Alertbox: http://www.useit.com/alertbox/20050314.html Nielsen, J. (2005). Usability of Websites for Teenagers. Retrieved August 2011, from Useit.com, Alertbox: http://www.useit.com/alertbox/teenagers.html Ofcom. (2008). Media Literacy Audit: Report on UK children’s media literacy. Piaget, J. (1959). The Language and Thought of the Child. 3d ed. London: Routledge and Kegan Paul. Pirolli, P. L. (2007). Information Foraging Theory: Adaptive Interaction with Information. USA: Oxford University Press. Porter, J. (2010, October). HOW DOES TECHNOLOGY CHANGE READING? Retrieved August 2011, from Armstrong institute for interactive media studies: http://aims.muohio.edu/2010/10/24/how-does-technology-change-reading/ PostRank Labs. (2011). Get the news and posts that matter. Retrieved August 1, 2011, from http://labs.postrank.com/gr 105  Pressley, M. (1977). Imagery and Children‟s Learning: Putting the Picture in Developmental Perspective. Review of Educational Research , 47, 585-622. Queeky. (2011). Queeky - an online drawing community. Retrieved August 1, 2011, from Queeky: http://www.queeky.com/ Raskin, A. (2007, July 13). Never Use a Warning When you Mean Undo. Retrieved August 2011, from A list apart: http://www.alistapart.com/articles/neveruseawarning Redish, J. (2007). Letting Go of the Words: Writing Web Content that Works. Morgan Kaufmann. Rideout, V., Foehr, U., & Roberts, D. (2010). Generation M2: Media in the Lives of 8- to 18Year Olds. Kaiser foundation. Riley, J. (1996). The teaching of reading. London: Paul Chapman. Sax, L. (2005). Why gender matters. New York: Doubleday. Scaife, M., & Bond, R. (1991). Developmental changes in children's use of computer input devices. Early Childhood Development and Care , 69, 19–38. Shardanand, U., & Maes, P. (1995). Social information filtering: algorithms for automating “word of mouth”. In Proceedings of the SIGCHI conference on Human factors in computing systems (CHI '95) (pp. 210-217). New York, NY, USA: Irvin R. Katz, Robert Mack, Linn Marks, Mary Beth Rosson, and Jakob Nielsen (Eds.). ACM Press/AddisonWesley Publishing Co.,. Shaw, E., LaBore, C., Chiu, Y. C., & Johnson, W. L. (2004). Animating 2D digital puppets with limited autonomy. Smart Graphics: Proceedings of the 4th International Symposium (pp. 1-10). Banff, Canada: Heidelberg: Springer. Shields, M. K., & Behrman, R. E. (2000). Children and Computer Technology: Analysis and 106  Recommendations. The Future of Children , Vol. 10, No. 2 , pp. 4-30. Shneiderman, B. (1986). Designing the User Interface: Strategies for Effective HumanComputer Interaction. Boston, MA, USA: Addison-Wesley Longman Publishing Co., Inc. Spencer, D. (2006). Four Modes of Seeking Information and How to Design for Them. Retrieved August 1, 2011, from boxesandarrows. Stone, S. J. (1995). Teaching Strategies: Strategies for Teaching Children in Multiage Classrooms. Childhood Education , v71 n2 p102-05. Stroop, J. R. (1935). Studies of interference in serial verbal reactions. . Journal of Experimental Psychology , 18, 643-662. Tizard, B., & Hughes, M. (1984). Young children learning. Cambridge, Massachusetts: Harvard University Press. Uden, L., & Dix, A. (2000). Iconic Interfaces for Kids on the Internet. Proceedings of IFIP WorldComputer Congress, (pp. 279--286). Beijing. Ware, C. (2000). Information visualization: perception for design. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc. Weinreich, H., Obendorf, H., Herder, E., & Mayer, M. (2008). Not quite the average: An empirical study of Web use. ACM Trans. Web , 2, 1, Article 5. Williams, R. (1994). The non-designer's design book: design and typographic principles for the visual novice. Berkeley, CA, USA: Peachpit Press. Yan, Z., & Fischer, K. W. (2004). How children and adults learn to use computers: A developmental approach. New Directions for Child and Adolescent Development , 105, 41--61. 107  Zammit, K. (2000). Computer Icons: A Picture Says a Thousand Words. Or Does It? Journal of Educational Computing Research , v23 n2 p217-31 . Zite. (2011). A personalized iPad magazine that gets smarter as you use it. Retrieved from Zite: http://www.zite.com/  108  Appendices  109  Appendix A Mockups of potential design solutions for effective dialogs This section contains some of the mockups based on the findings of Study 1 (Chapter 3). These mockups represent intermediate stages of the refined designs presented in Chapter 4.  A.1  Improving affordances  Infusing a comic book style into the dialogs by splitting it into two could improve the affordances of dialoguing.  Figure 24 Call-out style dialog box originating from the relevant icon.  Figure 25 Two way conversational dialog between the computer (represented by Penguin, Tux paint‟s mascot) and the user.  A.2  Illustrating consequences The consequences of the various choices can be made more evident by dynamically  revealing pictograms and symbols as the child hovers over the corresponding choice (Figure 26, Figure 27, Figure 28). 110  A think out cloud illustrating the consequences that appears on hover.  Figure 26 Visual illustration of the consequences of the choices with the mouse pointer hovering over “Yes, Save it”.  Figure 27. Visual illustration of the consequences of the choices with mouse pointer hovering over “No, Don‟t bother”  Consequences implied using typography and images.  Figure 28 Mock up of button icons conveying potential consequences using images and typography  111  A.3  Safer arbitrary choices  Revealing the dialog text gradually in parts, where the order of appearance of a word or phrase depends on its importance (Figure 29), could optimize information consumption.  Figure 29. Gradually revealing information and choices to solve some problems with dialogs. Dialogs also have the persistent „Back‟ button.  A.4  Addressing impatience  Adding information using multiple layers (body text with contrast, a title that stands out) and multiple modalities (text and icons) could lower the cost of consuming information.  Figure 30 Mockup of a dialog with multiple layers to support skimming such as the title with an icon and body text with varying contrast levels.  112  Appendix B Additional details of the second study This section contains additional information in the form of data and corresponding graphs to support the data analysis presented in Chapter 5.  B.1  Narrowing down of design choices  There were several choices when enumerating the design factors of a dialog box. We chose some of them over others based on the following reasons 1. Intuition on how it would work, based on our previous exploratory study with children (Chapter 3) 2. Feedback from fellow researchers on the prototypes that we had developed 3. Simplicity of the solution in terms of practical constraints such as for example the amount of time required for implementing (programming) a particular design choice over another  113  Design factor Call-out style of dialoguing  Button icons  Dialog text  Button text  Cueing view point  Temporal adaptation  Choices Call-out originates from A character The icon on the interface representing the action Icon is Embedded within the button Emerges on-hover on a think-out style dialog Icon is Static Animated Action oriented icons Simple smiley icons Icons of faces conveying simple emotions corresponding to „Yes‟, „No‟ Changing attributes of the dialog text such as its Contrast levels Hue Typography in order to influence the perception of its meaning for example urgency of the text Consisting of Simple words for example „Yes‟, „No‟ Descriptive actions for example „Save the picture‟ Attributes of the dialog could be modified so as to be cued from the viewpoint of either the Computer, or the User for example „Yes‟ button in „Quit?‟ dialog could be either red or green depending on the viewpoint Gradual appearance of different interface components such as Words of the dialog text Buttons Other elements of the dialog such as title, text, buttons  The chosen one Originating from the character (Figure 24, Figure 25).  Reason(s) 1, 3  Embedded within the button (Figure 28, Figure 26).  2, 3  Static icon.  1, 3  Action oriented icons  1  Changing contrast levels of dialog text (Figure 30).  1, 2  Descriptive actions as Button text  1  Viewpoint of the computer  1  No temporal adaptation (Figure 29, Figure 30).  1, 2, 3  Table 8 List of the choices considered during prototyping and reasons as to why we chose some over others.  B.2  Why these three dialogs?  We used three dialogs in the study: Save, Quit and Erase. Initially, we had wanted to have dialogs with different levels of potential disruption. A dialog having potentially 114  disruptive outcome such as causing loss of work might demand more attention and might increase the incentive for the child to consume the information in the dialog. We also wanted a certain number of dialogs that we could rotate through and pop-up at frequent intervals during the user session. We also wanted the dialogs to be useful. For example, from our earlier exploratory study, we had observed that children would go through several iterations of a painting, most often using an eraser tool to completely erase the canvas in between iterations, before settling on a painting that they wanted to print. We thought that the „Erase‟ dialog might help the children in this respect. Dialog  Title  Dialog text  Options  Save  Save?  Do you want to save your painting?  Quit  Quit?  Are you done painting?  Erase  Erase? Do you want to erase your painting and start over?  Save and continue No, don‟t save now Yes, I am done No, I want to paint Yes, erase and start over No, I want to paint  Intended disruption level Low  Medium High  Table 9 Details of the three dialogs used in the study along with their intended disruption level.  B.3  Notes for interpreting the data  o  The unit of RT unless mentioned otherwise is in seconds.  o  The differences in RT for most of the factors in our study are under 10 seconds. When we discuss differences in RT, even a difference of a few seconds (e.g., 2 to 4) is considered to be noteworthy. Differences in RT Significance levels for discussion <1 A trivial difference, not typically worthy of discussion 1–3 A minor difference that is typically considered worthy of discussion >3 A major difference worthy of discussion Table 10 Significance levels of the differences in RT for the purpose of the discussion of results.  o  Almost all of the graphs in the results (in this appendix) are of plots of means. In a typical plot of means graph we have three variables: the Y axis consists of mean RT and  115  the X axis contains a variable of interest. The data points on the graph are usually grouped using another variable of interest resulting in multiple plot lines. The legend helps us to distinguish between the plot lines. 0 in the legend or in the X axis means that the relevant factor was absent while 1 means that it was present. The error bars drawn through the data points in the graphs are measures of standard errors. o  The number of the dialogs, children and or other factors is included in most tables (in this appendix) to show that the numbers are not skewed heavily one way or another. When they are skewed it raises a warning flag that the corresponding data should be interpreted with care.  o  Very low counts in the tables (in this appendix) that call for careful interpretation of data are highlighted using a different color from the rest of the table elements, for example, 1.  o  The children who participated in our study were not checked for any physical, psychological or physiological factors (e.g., color blindness) which might have turned out to be a shortcoming when considered together with the design factors we employed (e.g., color coding).  B.4  Iterative evolution of design factors  Since this was primarily intended as an observational study aimed at reducing the design factors for a subsequent controlled experiment, we started with a set of dialog box designs and kept tweaking the designs during the study based on our observations. Each change in the design factors, or the way in which the study was conducted or data was collected, was accompanied by a corresponding change in the Batch number. We ended up with five batches in our second study. The details of the batches follow.  116  Batch Day  B1  1  B2  2  B3  3  Design factors employed Split structure, Support for skimming, Delayed click Same as B1  Visibility of Title, Body text Color coding of buttons Position of buttons  Differences from previous batch  -  B1 had only one set of design factors per participant. B2 had 2 sets of design factors per participant. We experimented with 2 frequencies of popping up the dialogs – we started with 2 minutes and then changed it to 1 minute. The following design factors were removed: Split structure Delayed click Contrast differences in text Button icons Third “I don‟t know” button Starting words „Yes‟, „No‟ from button text  Number of children Boys Girls Total 3 8 11  17  12  29  15  3  18  1  7  8  4  14  18  We added a reading test to assess the reading ability of children.  B4  4  B5  4  Visibility of Title Color coding of buttons Position of buttons Highlighting the safest button Same as B4  The frequency of popping up the dialogs continued to remain at 1 minute from B3 onwards. The body text was removed completely from all dialogs. B3 had one set of design factors per participant whereas B4 had multiple design factors per child. Erase button is red in color while it was Orange in B3. Color of highlighting is Yellow compared to Green in B4. The title of Quit dialog was changed to „Done?‟ to match the dialog text.  Table 11 Summary of the evolution of design factors through the course of the study  117  B.5  Getting to know the data  An overview  Figure 31 Histogram showing the distribution of the age of the children who participated in the study.  From the histogram we can see that the age distribution seems to be positively skewed. Age Number of children Boys Girls 3 1 0 4 6 0 5 7 9 6 6 6 7 6 9 8 5 5 9 3 3 10 2 3 11 0 1 12 1 0 Table 12 Break up of the number of participants by gender across all ages.  118  Age groups Gender Number of children Boys 37 Girls 36 Table 13 Distribution of gender among participants in the study.  The numbers are roughly equally distributed across age groups and genders. Age group Number of children Boys Girls Total [3, 5] (5, 7] (7, 12]  14 12 11  9 15 12  23 27 23  Table 14 Distribution of the participants of the study across genders and age groups.  B.6  Data pre-processing  Zero RT. Two of the data points in the logged data had a reaction time of 0. Since this was not practically possible, we analyzed the data further. From the screen captures we found that they corresponded to instances where the children had clicked the dialog by accident. The children were engaged in the act of painting and the dialog just happened to appear with its button directly under the mouse pointer when they were clicking in order to paint. Hence these dialogs caught the click accidentally and were dismissed before the children could even realize it. Since this is not representative of behavior wherein the children were intended to interact with the dialogs, the two records were removed from further analysis. It is quite possible that other similar accidental clicks were made but they ended up with non-zero RTs. Many RTs below 3 seconds (78 data points, about 19% of the data) might fall under this category. But since many children, especially literate ones and those who might have been clicking randomly without processing the information, can be expected to have very fast RTs and since we had no way of measuring intention or knowing whether the dialog was clicked on accidentally, randomly or otherwise, we could not identify and eliminate 119  these data points and hence they were retained for further analysis. Removal of outliers. After analyzing the data, there is just one data point which stands out from the rest. It is a RT of 120 seconds. This clearly strikes us as an outlier considering the facts that the average RT for the study is around 9 seconds. The average RT and SD for the same child who had this outlier (a 5 year old boy) is 17 seconds and 31 seconds respectively. Moreover for this child this is the only RT that is over 30 seconds. Hence it seems to us that this data point might not be representative of the overall interaction pattern of the child. We removed the data point identified above as an outlier.  B.7  Reaction times across genders and age group  Interaction patterns across age, gender In terms of interaction, the activity is not equally distributed across age groups, while it is roughly distributed equally across genders with the boys having faced a bit more dialogs. The difference in the number of dialogs across ages or genders likely arises from the differences in the session length as we had dialogs popping up at pre-determined intervals irrespective of any other factor15. Gender Number of Percentage of Average session length dialogs dialogs in seconds Boys 225 54.1 673 Girls 191 45.9 476 Table 15 Activity of participants as measured by the number of dialogs faced and the average length of the session during the study, split across genders.  15  Though we changed the interval a couple of times, it stayed the same for most part of the study.  120  Age group Number of dialogs [3, 5] 108 (5, 7] 205 (7, 12] 103  Percentage of dialogs 26.0 49.3 24.7  Average session length in seconds 540 641 535  Table 16 Activity of participants across age groups. The most active group is highlighted.  It is interesting to note that the groups at both ends of the age spectrum seem to share similar characteristics. Their activity levels measured in terms of the number of dialogs are similar, which could be attributed to their similar average session lengths. Semi-literates come out as the most active in terms of activity. Semi-literates have faced almost half the dialogs in the study. This activity pattern (highlighted) is seen irrespective of the gender. Age group Average session length in seconds Boys Girls [3, 5] 705 285 (5, 7] 654 630 (7, 12] 654 426  Number of dialogs Boys Girls 74 34 92 113 59 44  Percentage of dialogs Boys Girls 17.79 8.17 22.12 27.16 14.18 10.58  Table 17 Activity of participants across age groups and gender. The most active group is highlighted.  We can observe that while the number of dialogs follows the pattern of average session length for girls, it does not do so for boys. We are not sure as to why this might be so. We did tweak the frequency of popping the dialogs a couple of times during the study. It is possible that has contributed to the lack of correspondence between the session length and the number of dialogs among boys. One caveat with the comparisons of the likes of above is that we did not have any 3 or 4 year old girls participating in the study. This makes it difficult to compare pre-literates across genders. Girls in general seem to have faster RT than boys. Gender Average RT Boys Girls  9.59 7.53  Number Number Difference of dialogs of children in RT 225 37 2.06 191 36  Table 18 Average RTs of the children across genders.  121  Age groups Average RT [3,5] (5,7] (7,12]  11.34 7.2 8.67  Number of dialogs 108 205 103  Number Difference of children in RT 23 27 4.14 23 -1.47  Table 19 Average RTs across age groups.  Pre-literates have the slowest RTs. The semi-literates seem to have the fastest RTs unexpectedly even slightly higher than the literates. Age groups Gender Average RT [3,5] (5,7] (7,12]  Boys Girls Boys Girls Boys Girls  12.5 8.82 7.49 6.97 9.2 7.95  Number of dialogs 74 34 92 113 59 44  Number of children 14 9 12 15 11 12  Difference in RT 3.68 0.52 1.25  Table 20 Average RT across age groups and gender.  Figure 32 Average RTs across age groups and gender.  50% of the outliers (RT > 30) are from 5 year old boys (compared to 12.5% from 5 year old girls). The remaining 37.5% is divided equally among boys and girls across ages. The 122  group of 5 year old boys led the group even if we lower the threshold, accounting for 25% of data points with RT > 15. This could be because age 5 could be the age when the boys are beginning to read words. They then could be caught in between two modes of information processing – one based on non-textual cues such as color and the other based on textual information.  B.8  Support for skimming  Quantitative data analysis At first glance it looks like the factors did not make a difference.  Figure 33 Effect of factors to support skimming across gender.  123  Figure 34 Effect of factors to support skimming across age groups.  Age groups Skimming support [3,5] (5,7] (7,12]  Absent Present Absent Present Absent Present  Average RT  Number of dialogs Boys Girls Boys Girls 12.75 14 12 2 21.06 19.71 16 7 13.08 7.26 13 19 5.79 8.1 28 20 11.45 7.23 20 13 9.76 10.67 25 9  Number of children Boys Girls 4 1 4 2 4 4 5 4 6 4 7 6  Difference in RT Boys Girls -8.31 -5.71 7.29  -0.84  1.69  -3.44  Table 21 Effect of factors to support skimming across age groups and gender.  The overall pattern still holds with the girls seeming to be slowed down by these factors across all age groups. Pre-literate boys are slowed down while those of the other groups seem to be sped up, with semi-literates enjoying a clear advantage. Delayed-click It seems that the presence of delayed-click interacts negatively with the factors intended 124  to support skimming resulting in slowing down the children. Delayed-click Skimming support Average RT Number of dialogs Without Absent 10.55 31 Present 8.02 45 With Absent 10.1 48 Present 12.97 60  Number of Difference children in RT 10 2.53 14 13 -2.87 18  Table 22 Interaction of factors to support skimming and delayed-click.  Dialog structure It seems that the skimming support has no effect with the default dialog structure but seems to slow down split dialogs. Dialog structure Skimming support Average RT Number of dialogs Default Absent 11.42 50 Present 11.17 47 Split Absent 8.31 29 Present 10.59 58  Number of Difference children in RT 14 0.25 16 11 -2.28 16  Table 23 Interaction of factors to support skimming and dialog structure.  The split dialog structure seems to interact negatively with the factors intended to support skimming. The effect is more pronounced for girls than for boys. Age groups [3,5]  Dialog structure Default Split  (5,7]  Default Split  (7,12]  Default Split  Skimming support Absent Present Absent Present Absent Present Absent Present Absent Present Absent Present  Average RT 13.12 30.67 12.67 19.15 10.47 7.75 8.38 5.75 11.61 12.35 5.6 6.64  Number of dialogs 8 3 6 20 19 24 13 24 23 20 10 14  Number of children 4 2 2 6 5 6 3 4 5 8 6 6  Table 24 Interaction of skimming factors and dialog structure across age groups.  125  Differenc e in RT -17.55 -6.48 2.72 2.63 -0.74 -1.04  The negative interaction with the dialog structure and the factors supporting skimming seems to reduce with age. Pre-literates seem to be affected the most. For semi-literate and literates there seems to be almost no interaction between the dialog structure and factors intended to support skimming. Semi-literates seem to be helped by the support for skimming and literates seem to be slowed down by it regardless of the dialog structure. Combining delayed-click and dialog structure The support for skimming generally seems to slow down children when the information is displayed in a split dialog structure. But this is even worse when delayed-click is present. Hence we can conclude that the support for skimming generally does well with normal dialogs without delayed-click. Delayed Dialog Skimming click structure support Without Default Absent Present Split Absent Present With Default Absent Present Split Absent Present  Average RT Number of dialogs 11.68 25 9.64 25 5.83 6 6 20 11.16 25 12.91 22 8.96 23 13 38  Number of children 7 7 3 7 7 9 8 9  Difference in RT 2.04 -0.17 -1.75 -4.04  Table 25 Interaction of factors to support skimming, delayed-click and dialog structure.  Icons in buttons Hiding button icon seems to speed up the interaction. In other words the icons inside the buttons seem to take about 2 seconds of processing time. Button icons Average RT Number of Number of Difference dialogs children in RT Visible 11.17 47 16 1.73 Not visible 9.44 18 3 Table 26 Effect of the presence of icons in buttons on RT.  126  B.9  Split structure  Quantitative data analysis Boys seem to benefit from split dialogs, while girls seem to get slower for the most part. Older children seem to benefit more than younger children – with older girls suffering less due to split dialogs than younger girls. Age groups  Gender  [3,5]  Boys Girls  (5,7]  Boys Girls  (7,12]  Boys Girls  Dialog structure Number of dialogs Default 9 Split 19 Default 2 Split 7 Default 28 Split 13 Default 15 Split 24 Default 25 Split 20 Default 18 Split 4  Number of children 3 4 1 2 4 1 2 3 3 5 6 2  Average RT Difference in RT 18.78 16.89 14 19.71 10.82 2.23 5.47 9.08 14 6.15 9.11 6.5  1.89 -5.71 8.59 -3.61 7.85 2.61  Table 27 Effect of dialog structure on RT across age groups and gender.  Delayed-click With delayed-click, the difference between the default and the split dialog seems to have been lost. Delayed-click Dialog structure Number of dialogs Without Default 50 Split 26 With Default 47 Split 61  Number of children 9 8 11 13  Average RT Difference in RT 10.66 5.96 11.98 11.48  4.7 0.5  Table 28 Interaction between dialog structure and delayed-click.  With delayed-click, the boys seem to lose a bit of their advantage to split dialogs and the girls seem to get much slower. 127  Gender Delayed-click Dialog structure Number of dialogs Boys Without Default 27 Split 16 With Default 35 Split 36 Girls Without Default 23 Split 10 With Default 12 Split 25  Number of children 4 5 6 8 5 3 5 5  Average RT 12.41 5.25 13.91 10.81 8.61 7.1 6.33 12.44  Difference in RT 7.16 3.1 1.51 -6.11  Table 29 Interaction between dialog structure and delayed-click across genders.  B.10 Controlling user’s actions – delayed-click Quantitative data analysis Employing delayed-click seems to increase the RT by about 2.5 seconds. Delayed-click Number of Number of Average RT dialogs children Without 76 17 9.05 With 108 22 11.69 Table 30 Effect of delayed-click on RT.  The semi-literates seem to be the least affected group followed by pre-literates and then literates. Age groups Delayed-click Number of dialogs [3,5] Without 12 With 25 (5,7] Without 37 With 43 (7,12] Without 27 With 40  Number of children 4 6 5 7 8 9  Average RT Difference in RT 16.67 18.24 7.95 7.86 7.19 11.72  -1.57 0.09 -4.53  Table 31 Effect of delayed-click on RT across age groups.  Response Clicking the “I don‟t know” (BACK) button becomes faster with delayed-click, whereas the other buttons get clicked slower as they are revealed after the “I don‟t know” button. This 128  is expected, as the “I don‟t know” button is the first revealed and the only button on the dialog for a short time. YES seems to be the slowest because it is the last one to be revealed. We can also see how the percentage of clicks shifts from YES towards BACK with delayedclick. In this respect, delayed-click seems to accomplish what it was designed for. It increases the probability (and decreases the time) of clicking the least disruptive option (BACK) at the expense of making it longer to click the potentially most disruptive option (YES). Response BACK NO YES  Delayed-click Number of dialogs Without 9 With 24 Without 48 With 72 Without 19 With 12  Percentage of responses16 11.84 22.22 63.16 66.67 25.00 11.11  Number of children 6 9 13 17 11 9  Average RT Difference in RT 14.78 8.24 6.54 7.04 -4.65 11.69 11.42 -10.58 22  Table 32 Interaction between delayed-click and type of response clicked.  Caveats to interpretation of the data   The timing interval with which the buttons were revealed in sequence was continuously tweaked but we do not have information about the tweaks made. Hence for the purpose of this analysis the timing interval has been treated though it was uniform. But different values could have elicited different behaviour.  B.11 Color coding of buttons Girls‟ RTs remain the same with or without color coding, but RTs of boys increase slightly with color coding.  16  Percentage is calculated separately for responses that occurred under the delayed-click and those that occurred without the delayed-click. For example, without delayed-click 11.84% of responses were BACK, while with delayed-click it jumped up to 22.22%.  129  Gender Button color coding Boys Without With Girls Without With  Average RT 8.98 9.76 7.53 7.53  Number of dialogs 50 175 57 134  Table 33 Effect of button color coding on RT across genders.  Across groups, the RTs vary more without color coding than with color coding. Overall, the semi-literates show no big impact of color coding. But for boys from this group color coding seems to make them slightly slower. This negative effect goes on to increase with the next group (literates), for both genders. The literates surprise us with their RTs increasing with color coding of buttons (slower by almost 4 seconds). This is similar to the behavior with how the color coding of buttons interacts with high reading levels. The effect is more pronounced for boys, who are slowed down by approximately 6 seconds with color coding, than for girls who are slowed down by only 3 seconds. Age group Button color coding [3, 5] (5, 7] (7, 12]  Average RT Total 14.78 10.41 06.82 07.39 04.76 09.44  Without With Without With Without With  Boys 15.65 11.56 06.09 07.96 04.30 10.20  Girls 12.33 08.07 07.20 06.83 05.43 08.43  Table 34 Effect of button color coding on RT across age groups.  We can observe that with color coding, the RTs of both the responses (YES and NO) are roughly the same. Without color coding it takes a few seconds more to affirm (YES - Erase and start over, I am done, Save and continue). From the percentage of responses we can observe a slight shift from NO to YES with color coding.  130  Response NO  Button color coding Without With Without With  YES  Average RT 7.64 6.09 9.72 6.2  Number of responses 78 79 29 46  Percentage of responses17 72.90 63.20 27.10 36.80  Table 35 Interaction between button color coding and response type.  We can observe that the Erase dialog seems to be the most helped by color coding followed by the Save dialog. Dialog type Button color coding Erase Without With Quit Without With Save Without With  Average RT 8.53 4.39 6.86 6.73 9.76 7.00  Number of children 36 36 42 56 29 33  Table 36 Interaction between button color coding and type of dialog.  For the pre-literates color coding makes a big difference in terms of RT. But for other groups of children, the effect of color coding is not as clear. Some dialogs seem to be faster but some others are slowed down by color coding. It might be because that this group of children is less dependent on the color cue. Dialog type Button color coded  Erase Quit Save  Without With Without With Without With  Average RT [3,5] (5,7] Boys Girls Boys Girls 18 17.33 5.88 7 4.4 2.25 5.62 3.5 11.11 6.56 5.42 7.18 3.75 8.27 5.83 23.5 7.33 5.67 9 5.75 2 9 7.5  (7,12] Boys Girls 3.33 2 5.29 5.2 5.43 12 16.5 3.5 1 8.67  Table 37 Interaction between button color coding and type of dialog across age groups and gender.  Girls seem to be faster at clicking the safer green button, while having similar RTs in  17  Percentage is calculated based on the condition of color coding per response.  131  clicking the potentially disruptive orange button. Button color Blue Green Orange  Average RT Boys Girls 8.98 7.53 6.95 5.13 6.61 7.00  Number of children Boys Girls 50 57 42 46 18 16  Table 38 Effect of button color on RT across genders.  Since we have far fewer counts from the literates and they are not evenly distributed across the button color conditions, we show their data but ignore them from further analysis in this section. The pre-literates perform the worst without color coding (blue color for all buttons). We can also observe that semi-literates become slower with color coding. Button Color  Blue Green Orange  [3,5] Boys Girls 15.65 12.33 6.24 2.62 4.75 4.33  Average RT (5,7] Boys Girls 6.09 7.2 7 5.29 9.62 11.5  (7,12] Boys Girls 4.3 5.43 21 10.8 2 7.88  Number of children [3,5] (5,7] (7,12] Boys Girls Boys Girls Boys Girls 17 6 23 44 10 7 21 13 20 28 1 5 8 6 8 2 2 2  Table 39 Effect of button color on RT across age groups and genders.  Color coding of buttons removes the effect of highlighting. In other words, when the buttons are color coded, highlighting does not seem to make a difference. Whereas when the buttons are not color coded, highlighting seems to make it faster to interact with the dialog (by almost 3.5 seconds). Button color coding Button highlighting Without Without With With Without With  Average RT 9.06 5.54 6.08 6.28  Number of children 81 26 96 29  Table 40 Interaction between button color coding and button highlighting.  132  Figure 35 Interaction between button color coding and button highlighting.  Caveats to interpretation of the data Children, especially the pre-literates, might have interacted with the color of the buttons based on their preference for a certain color rather than relying on any conventions.  B.12 Highlighting the safe option Quantitative data analysis Overall highlighting seems to help speed up interacting with a dialog. Button highlighting Average RT Number of dialogs Without 7.45 177 With 5.93 55 Table 41 Effect of highlighting on RT.  Boys seem to benefit the most from highlighting. Girls‟ performance seems to be relatively unaffected by highlighting.  133  Gender Button highlighting Boys Without With Girls Without With  Average RT 8.36 3.64 6.34 6.71  Number of dialogs 97 14 80 41  Table 42 Effect of highlighting on RT across genders.  Regarding age groups, both the pre- and semi-literates are almost equally helped by highlighting. Unexpectedly the literates seem to be slowed down by highlighting. But since the data points are not well distributed within this group, especially for boys, this might not be representative. Age groups Button highlighting [3,5] Without With (5,7] Without With (7,12] Without With  Average RT 8.17 6.33 7.4 5.36 5.81 7.9  Number of dialogs 65 6 86 39 26 10  Table 43 Effect of highlighting across age groups.  Figure 36 Effect of highlighting across age groups.  134  With respect to responses, highlighting seems to speed up clicking on the affirmative YES options more than clicking on the NO options. It also seems to slightly increase the likelihood of YES being clicked. Response Button highlighting Average RT NO YES  Without With Without With  7.11 6.03 8.18 5.74  Number of dialogs 121 36 56 19  Percentage of dialogs18 68.4 65.5 31.6 34.5  Table 44 Interaction between highlighting and type of response.  We can observe in the table below that the safer options (cells with green background) seem to clearly benefit by highlighting. We cannot say the same about the potentially disruptive options. From the percentage column, we can observe that highlighting did not cause a considerable shift in interaction patterns (e.g., as much as color coding) for the most part (except the Save dialog).  18  Percentage is calculated with respect to highlighting per response.  135  Response text  Button highlighting  Erase and start over Without With I want to continue Without With I am done Without With I want to paint Without With Don‟t save now Without With Save and continue Without With  Number of dialogs  Percentage of dialogs  Average RT  Speedup in RT with highlighting  10 2 49 11 17 6 53 22 19 3 29 11  16.9 15.4 83.1 84.6 24.3 21.4 75.7 78.6 39.6 21.4 60.4 78.6  7.2 2.5 7 4.09 5.82 5.83 7.4 6.32 6.58 11 9.9 6.27  4.7 2.91 -0.01 1.08 -4.42 3.63  Table 45 Interaction between highlighting and response text clicked.  There is a potential negative effect when the button is both color coded and highlighted using green color. In B5, there is no speedup due to highlighting when the button is also color coded. Batch B4  B5  Button color coding Button highlighting Number of dialogs Average RT Difference in RT Without Without 8 4.62 -1.83 With 11 6.45 With Without 12 4.75 -8.25 With 7 13 Without Without 31 9.1 4.23 With 15 4.87 With Without 21 4.38 0.24 With 22 4.14 Table 46 Interaction between the different colors used for highlighting and color coding of buttons.  136  B.13 Switching the button position Quantitative data analysis At first glance, switching button positions seems to yield a small advantage in terms of RT. Button position Number of Number of Average RT dialogs children Default 176 37 7.35 Switched 56 24 6.27 Table 47 Effect of switching the button position on RT.  The older children seem to get slower as the button position is switched. Boys seem to benefit by the switching while girls remain relatively unaffected. Age groups Gender Button position Number of dialogs [3,5] Boys Default 35 Switched 11 Girls Default 23 Switched 2 (5,7] Boys Default 37 Switched 14 Girls Default 60 Switched 14 (7,12] Boys Default 10 Switched 4 Girls Default 11 Switched 11  Number of children 6 3 7 2 5 5 11 8 3 2 5 4  Average RT 10.74 5.36 5.17 7.5 7.62 5.36 6.9 5.29 4.3 6.75 5.36 9.18  Difference in RT 5.38 -2.33 2.26 1.61 -2.45 -3.82  Table 48 Effect of switching the button position on RT across age groups and genders.  137  Figure 37 Effect of switching the button position on RT across age groups.  The YES responses seem to have been clicked faster with switching while the NO responses remain unaffected. We can once again see that girls seem to be relatively unaffected by switching of button positions. For boys it seems like clicking the switched YES option (positioned at the right side after switch) is faster than clicking the switched NO option (positioned at the left side after switch). This could provide us with insights into the order in which the children read the options in the dialog box and/or prioritize them for interaction.  138  Gender  Response  Button position  Boys  NO  Default Switched Default Switched Default Switched Default Switched  YES Girls  NO YES  Number of dialogs 58 14 24 15 68 17 26 10  Number of children 13 9 11 5 18 9 17 6  Average RT 7.88 6.86 10.17 4.33 6.01 6.76 7.04 7.5  Difference in RT 1.02 5.84 -0.75 -0.46  Table 49 Interaction between switching of button positions and the type of response across genders.  We see the same trends as observed earlier – older children seem to get slower when the buttons are switched. The clicking of the switched YES button seems to be in general faster than clicking the switched NO button. Age groups  Response  [3,5]  NO YES  (5,7]  NO YES  (7,12]  NO YES  Button position Default Switched Default Switched Default Switched Default Switched Default Switched Default Switched  Number of dialogs 41 5 17 8 73 16 24 12 12 10 9 5  Number of children 10 3 10 4 16 9 11 5 5 6 7 2  Average RT 8.07 5.2 9.65 6 6.56 6.12 9.04 4.25 4.67 8.7 5.11 8.2  Difference in RT 2.87 3.65 0.44 4.79 -4.03 -3.09  Table 50 Interaction between switching of button positions and the type of response across age groups.  Except for the Quit dialog, clicking on the switched YES responses seems to be faster than clicking the switched NO responses. From the percentage column, we can observe an interesting behavior. Except for the Save dialog, switching of the YES buttons (placed right after the switch) seems to increase their chances of getting clicked while switching of the NO buttons (placed left after the switch) seems to decrease their chances of getting clicked. The 139  Save dialog is different from the other dialogs in that the safe option is placed first on the left of the dialog and the potentially disruptive option is placed on the right. If we took the nature of the options into account, the potentially disruptive options seem to have a better chance of getting clicked after their position is switched while the safer options seem to be less popular after their position is switched. Response  Dialog  Response text  Save  Quit  Erase  Erase and start over YES I want to continue  NO  I am done  YES  I want to paint  NO  Save and continue  YES  Don‟t save now  NO  Nature of the response  Button position  Potentially Default disruptive Switched Safer Default Switched Potentially Default disruptive Switched Safer Default Switched Safer Default Switched Potentially Default disruptive Switched  Number Percentage Number Average Difference of of of RT in RT dialogs dialogs19 children 6 6 53 7 10 13 60 15 34 6 13 9  10.17 46.15 89.83 53.85 14.29 46.43 85.71 53.57 72.34 40 27.66 60  5 4 28 7 10 8 29 10 17 4 9 5  7.67 5.17 6.58 5.57 4.9 6.54 7.1 7 9.76 4 7 7.44  Table 51 Table 52 Interaction between switching of button positions and response text clicked.  Color coding Color coding of buttons seems to remove the effect of switching. But highlighting does not seem to affect switching in a similar way.  19  Percentage is calculated on the number of dialogs for a response, per dialog type per button position.  140  2.5 1.01 -1.64 0.1 5.76 -0.44  Button color coding Without  Button position Default Switched Default Switched  With  Number of dialogs 93 14 83 42  Number of children 26 10 30 16  Average RT 8.65 5.29 5.89 6.6  Difference in RT 3.36 -0.71  Table 53 Interaction between switching of button positions and button color coding.  Button highlighting Without With  Button position Default Switched Default Switched  Number of dialogs 147 30 29 26  Number of children 34 7 17 17  Average RT 7.56 6.9 6.28 5.54  Difference in RT 0.66 0.74  Table 54 Interaction between switching of button positions and highlighting.  Boys seem to be not affected by switching of buttons with color coding, but without color coding switching seems to speed up their RT. But since the numbers for the without color coding condition in boys in unbalanced, the data should be interpreted with caution. For girls, switching seems to interact with color coding in an opposite manner. Without color coding, switching does not seem to make a difference for girls but with color coding they seem to be slowed down. Gender Boys  Girls  Button color coding Button position Number of dialogs Without Default 43 Switched 7 With Default 39 Switched 22 Without Default 50 Switched 7 With Default 44 Switched 20  Number of children 8 5 10 6 18 5 20 10  Average RT Difference in RT 9.86 6.29 3.57 7.1 0.92 6.18 7.6 0.6 7 4.82 -2.23 7.05  Table 55 Interaction between switching of button positions and button color coding across genders.  Pre-literates seem to be slowed down when the buttons are switched with color coding while they seem to react faster when the buttons are switched without color coding. Semi141  literates seem to react faster in general when the buttons are switched, but even faster when there is no color coding. Literates seem to be slowed down in general by the switching irrespective of color coding. Age groups [3,5]  Button color coding Without  Default Switched Default Switched Default Switched Default Switched Default Switched Default Switched  With (5,7]  Without With  (7,12]  Button position  Without With  Number of dialogs 19 4 39 9 61 6 36 22 13 4 8 11  Number of children 5 2 12 3 15 5 14 10 6 3 4 3  Average RT 16.84 5 4.49 6 7.07 4.33 7.36 5.59 4.08 7 6.12 9.09  Difference in RT 11.84 -1.51 2.74 1.77 -2.92 -2.97  Table 56 Interaction between switching of button positions and button color coding across age groups.  Body and title text It seems that as the amount of text information on dialog increases, the speed-up due to switching increases too. Body Text Not visible  Title  Not visible Visible  Visible  Visible  Button position Default Switched Default Switched Default Switched  Number of dialogs 87 30 34 11 55 15  Number of children 22 13 11 7 8 4  Average RT 5.18 5.73 8.21 6.55 10.24 7.13  Difference in RT -0.55 1.66 3.11  Table 57 Interaction between switching of button positions and visibility of body text and title.  Caveats to interpretation of the data We do not know what is causing the speedup of RT for some children when the buttons  142  are switched. That answer could undermine a lot of assumptions in our analysis and hence change the results.  B.14 Preference for a choice in terms of its position and how it is worded Based on our observations, it seemed that position of the button (left vs. right side) was one of the major cues that children used to interact with the dialog. If we took a look at the data from B3 to B5 (which had just 2 buttons), it seems to support this observation. Button side Number of responses Left 81 Right 151 Table 58 Count of the buttons clicked per side of the dialog box from B3, B4, and B5.  It might seem that the children have a preference for the right side. But a closer look at the data reveals alternate views. The data from B2, which had 3 buttons instead of just 2, with the third button being “I don‟t know” is shown in Table 59. Button side Left Middle Right  Number of responses 31 121 33  Table 59 Count of the buttons clicked per side of the dialog box from B2.  Based on this data, right side does not seem to be the preferred side any more. But maybe it is the 2nd button that is preferred. Though the data from B2 seems to support this view, data from B3 to B5 needs further analysis before we can conclude it to support this view. One of the important differences between B2 and the rest of the batches is that the buttons were always displayed in fixed positions in B2 while switching of the button positions was incorporated as a design factor from B3 onwards. We take the switching factor into consideration and look at the data again. After switching of the button positions, the right side loses its dominance. In fact, going by the counts, it seems that after switching positions the left side is the preferred side. 143  Position switched  No  Number Percentage of responses of responses Left Right Left Right 50 126 28.4 71.6  Yes  31  25  55.4  44.6  Table 60 Effect of switching button positions on the count of buttons clicked per side.  We can observe that except for the Save dialog the other two dialogs have the same pattern: the right side shows clear dominance before switching positions. Dialog type Button position Number of responses Left Right Erase Default 6 53 Switched 7 6 Quit Default 10 60 Switched 15 13 Save Default 34 13 Switched 9 6 Table 61 Effect of the interaction between dialog type and switching of the button positions on the count of the responses per side of the dialog.  The Save dialog is different from the others in that the safest option is placed on the left while the disruptive option is placed on the right (Figure 14). Dialog Erase Quit Save  Response text Left Right Erase and start over I want to continue I am done I want to paint Save and continue Don‟t save now  Table 62 The different dialogs and their corresponding response texts divided according to the side in which the responses were placed on the dialog.  From the point of view of the response text in descending order of popularity, it seems like the children preferred certain responses over others. The top three most preferred choices also happen to be the safer, less disruptive choices.  144  Response text  Number of responses I want to paint 75 I want to continue 60 Save and continue 40 I am done 23 Don‟t save now 22 Erase and start over 12 Table 63 Count of the responses sorted in descending order. The safer responses are highlighted in green.  Considering the response text data together with age groups, we can observe that both pre- and semi-literates have similar preferences for the response text with the safer (sounding) options clearly preferred over the rest. But how does this hold up when we consider factors such as switching of button positions, color coding of buttons and gender? Response text  Number of children [3, 5] (5, 7] (7, 12] I want to paint 21 44 10 I want to continue 18 36 6 Save and continue 11 26 3 I am done 11 6 6 Don‟t save now 7 9 6 Erase and start over 3 4 5  Percentage of children20 [3, 5] (5, 7] (7, 12] 29.58 35.20 27.78 25.35 28.80 16.67 15.49 20.80 8.33 15.49 4.80 16.67 9.86 7.20 16.67 4.23 3.20 13.89  Table 64 Count and percentage of the responses clicked across age group.  We can observe from the table below that switching button position seems to heavily influence the choices made. The pre-literates are the most affected. The direction of the differences before and after switching is comparable for the most part between the semiliterate and literate children. For example, the potentially disruptive options such as “Don‟t save now”, “Erase and start over” and “I am done” are much more likely to be chosen by preliterates and literates when the button positions are switched. This is a bit confusing in the context of the Save dialog given that after switching positions the “Don‟t save now” option  20  Calculated per age group  145  ends up on the left side of the dialog while the other two options end up on the right side. Hence the idea that the pre-literates might be cueing onto the position of the buttons is not very tenable with this data. On the other hand, the interaction patterns on the relatively safer options such as “I want to continue”, “I want to paint” and “Save and continue” seems to support this idea. For the safer options that are by default on the right side of the dialog, the chances that a pre-literate child clicks on them after they are switched positions (ending up on the left) decreases by over 63%, while such effect is not seen for the safer option that is by default on the left side. This suggests that children especially the pre-literates could be using position as a cue to interact with the choices. In our case they seem to prefer the right side. Dialog Erase  Quit  Save  Response21  Button position  Erase and start over Default Switched I want to continue Default Switched I am done Default Switched I want to paint Default Switched Save and continue Default Switched Don‟t save now Default Switched  Number of children [3, 5] (5, 7] (7, 12] 1 2 3 2 2 2 17 31 5 1 5 1 5 2 3 6 4 3 20 35 5 1 9 5 11 20 3 0 6 0 4 7 2 3 2 4  Percentage of children22 [3, 5] (5, 7] (7, 12] 05.56 06.06 37.50 66.67 28.57 66.67 94.44 93.94 62.50 33.33 71.43 33.33 20.00 05.41 37.50 85.71 30.77 37.50 80.00 94.59 62.50 14.29 69.23 62.50 73.33 74.07 60.00 00.00 75.00 00.00 26.67 25.93 40.00 100.0 25.00 100.0  Table 65 Interaction between the switching of button position and response text across age groups.  Unexpectedly no strong and consistent patterns are seen in terms of Button color coding in this case. But there seems to be an effect of gender. Glancing down the column of percentages, boys in general seem to be affected more by switching of button positions than girls.  21  Responses are listed by their default order in the dialog boxes. E.g., in Quit dialog ‘I am done’ is at the left while ‘I want to paint’ is at the right. 22 Percentages are calculated per dialog, per button position.  146  Response  Button position  Save and continue  Default Switched Don‟t save now Default Switched Erase and start over Default Switched I want to continue Default Switched I am done Default Switched I want to paint Default Switched  Number of children Boys Girls 20 14 2 4 2 11 6 3 2 4 4 2 25 28 3 4 2 8 9 4 31 29 5 10  Percentage of children Boys Girls 90.91 56.00 25.00 57.14 09.09 44.00 75.00 42.86 07.41 12.50 57.14 33.33 92.59 87.50 42.86 66.67 06.06 21.62 64.29 28.57 93.94 78.38 35.71 71.43  Table 66 Interaction between the switching of button position and response text across genders.  Caveats to interpretation of the data Since the number of children who participated and hence the interaction counts for when the button positions were switched are relatively lower in number than the default position, these comparisons might be weak and not representative. We have talked about button position and the nature of the response text (safer vs. disruptive) as separate factors. But in reality it is quite possible that these interact and the cues used by the children are a mixture of these and many more that we have not yet considered.  B.15 Visibility of body text and title Body text refers to the main body of text on the dialog, usually the question on the dialog.  147  Dialog Body text Title Quit Are you done painting? Quit?23 Save Do you want to save your painting? Save? Erase Do you want to erase your painting and start over? Erase? Table 67 Information on the text contained in the three dialogs along with their title  Quantitative data analysis It is interesting to see the fastest RT was when both the title and body text were hidden. Making the title visible seems to add about 2.5 seconds to RT and making the body text visible along with that seems to add about 2 seconds to RT. Body text Title Not visible Not visible Visible Visible Visible  Number of dialogs 117 45 70  Average RT 5.32 7.8 9.57  Table 68 Effect of the visibility of body text and title on RT.  When both the body text and title are not visible, both boys and girls seem to take similar amount of time to interact with the dialog. Since we have two rows with relatively low counts we are not able to compare all the remaining conditions across genders. Gender Boys  Girls  Body text  Title  Number of dialogs Not visible Not visible 44 Visible 6 Visible Visible 61 Not visible Not visible 73 Visible 39 Visible Visible 9  Number of children 9 2 11 17 11 1  Average RT 5.09 1.33 10.33 5.47 8.79 4.44  Table 69 Effect of the visibility of body text and title on RT across genders.  As we move across the age groups from pre-literates to literates, adding the body text seems to make less and less of a difference. Adding the title seems to show a similar but less linear pattern.  23  The title was changed from ‘Quit?’ to ‘Done?’ in B5. But for the analysis it is assumed that this change had no or negligible effect on the interaction.  148  Age groups [3,5]  (5,7]  (7,12]  Body text  Title  Number of dialogs Not visible Not visible 40 Visible 8 Visible Visible 23 Not visible Not visible 61 Visible 29 Visible Visible 35 Not visible Not visible 16 Visible 8 Visible Visible 12  Number of children 8 3 5 12 6 4 6 4 3  Average RT Speedup of RT 5.2 9.12 -3.92 12.52 -3.4 4.85 8.03 -3.18 9.03 -1 7.44 5.62 1.82 5.5 0.12  Table 70 Effect of the visibility of body text and title on RT across age groups.  149  Figure 38 Effect of the visibility of body text on RT across genders and age groups.  Effects of color coding With color coding we could see a small speed-up (of about 1.3 seconds) across all 3 conditions. Body text  Not visible  Title  Not visible Visible  Visible  Visible  Button color coding Without With Without With Without With  Number of dialogs 41 76 25 20 41 29  Number of children 15 23 11 10 7 5  Average RT 6.22 4.84 8.4 7.05 10.07 8.86  Differenc e in RT 1.38 1.35 1.21  Table 71 Interaction between visibility of body text, title and button color coding.  Since the number of children is spread thin across the cells, the data should be interpreted with care. Pre-literates are the biggest benefactor of color coding. The speed-up of RT for 150  this group without color coding seems to be proportional to the amount of text information on the dialog. But with color coding, their RTs are all similar. Semi-literates seem to be slowed down proportional to the amount of text information added to the dialog with color coding. Without color coding, there is no such pattern. Literates seem to perform poorly with color coding in all 3 conditions irrespective of the amount of text on the dialog. Without color coding, they seem to do uniformly well, with similar RTs irrespective of the amount of text on the dialog. Age groups [3,5]  (5,7]  (7,12]  Body text  Title  Button color coding Number of dialogs Not visible Not visible Without 5 With 35 Visible Without 4 With 4 Visible Visible Without 14 With 9 Not visible Not visible Without 32 With 29 Visible Without 17 With 12 Visible Visible Without 18 With 17 Not visible Not visible Without 4 With 12 Visible Without 4 With 4 Visible Visible Without 9 With 3  Number of children 2 8 2 3 3 2 10 10 6 5 2 2 3 5 3 2 2 1  Average RT Difference in RT 10.8 6.4 4.4 12.75 7.25 5.5 16.79 10.9 5.89 5.62 1.62 4 8.24 0.49 7.75 7.61 -2.92 10.53 5.25 -2.92 8.17 4.75 -1.75 6.5 4.56 -3.77 8.33  Table 72 Interaction between visibility of body text, title and button color coding across age groups.  B.16 Reading level We collected reading assessment data from 23 children. The reading assessment was offered to children on the 3rd and 4th day of the study. About 52% of those children (23 out of 44) took the reading assessment. The rest of the children did not take a reading assessment 151  primarily due to their inability to read. There seems to be no clear linear relationship between reading level and RT. RT seems to decrease from reading level 1 to level 5, but after level 6 there seems to be no clear relationship to RT. Reading level 1 2 3 4 5 6 7 11  Average RT 11.89 8.03 8.96 5.32 3.56 6.68 10 7.5  Table 73 Reading level scores of the children against their corresponding average RTs.  Color coding of buttons and highlighting From the figure below we can observe that when the buttons are not color coded, the RT and the reading level seem to be inversely related. But when the buttons are color coded, no such relationship exists. Rather, after level 4 RT increases as the reading level increases.  Figure 39 Interaction between reading level and button color coding (left), highlighting (right).  But highlighting does not seem to interact with reading level as color coding seems to be. 152  But highlighting does not seem to interact with reading level as color coding seems to be.  B.17 Reading level analysis on dialog text Quantitative data analysis The result of three different readability tests on the dialog text for the three dialogs follows. The tests that were run on the text were Flesch–Kincaid Grade Level, Gunning fog index and Coleman–Liau Index24. Dialog type Erase Save Quit25  Flesch–Kincaid Grade Level 1.2 -0.8 -1.6  Gunning fog index 1.9 1.5 1.2  Coleman–Liau Index 8.1 6.6 4.3  Table 74 Readability scores of three tests on the text of the three dialogs.  Though the values of scores of the individual tests vary from each other, they have similar magnitude across dialogs, i.e. they point towards the same direction in terms of the difficulty of the dialogs. According to the scores, the Erase dialog seems to the most difficult, followed by the Save dialog, closely followed by the Quit dialog. Considering the response times of the dialogs, it seems that the Erase and Quit dialogs have similar RTs while the Save dialog seems to have been the slowest to respond to. This does not match with the index of difficulty as revealed by the 3 readability tests. Dialog type Average RT Number of dialogs Erase 6.46 72 Quit 6.79 98 Save 8.29 62  Number of children 36 43 33  Table 75 Average RTs across dialogs.  We translate this into an order on how fast the dialogs were dismissed, with the slowest  24  The three tests were chosen based on their popularity, access to the tools to run the tests and the closeness of the scores that these tests produced to the grade levels. 25 Even though the title was changed to ‘Done?’ in B5, this analysis was performed with the title as ‘Quit?’  153  ones coming first. For the pre-literates it is Save, Erase, Quit. For the semi-literate it is Save, Quit, Erase. For literates it is Quit, Save, Erase. For none of the age groups the order matches that of the order indicated by grade level difficulty. Dialog type  Erase Quit Save  Average RT  Number of dialogs [3,5] (5,7] (7,12] [3,5] (5,7] (7,12] 8.43 5.97 4.45 21 40 11 7 6.4 7.56 32 50 16 9.33 8.17 6.67 18 35 9  Number of children [3,5] (5,7] (7,12] 12 17 7 16 17 10 11 16 6  Table 76 Average RTs of dialogs across age groups.  We take a look at how the different types of dialog (and the information on it) interact with other factors. Reading level Based on the graph below, the Erase dialog which scored the highest in terms of grade level readability index seems to be dismissed faster by children with higher reading levels. On the other hand, the Save dialog which could be thought of being the hardest to interact with owing to the poor dichotomy between the options and the understanding of the abstractions required, has the most fluctuations. It does not seem to get faster to interact with as the reading level increases. We restrict our discussion of the graph to reading level of 6, since the data is sparse above that.  154  Figure 40 Interaction between reading level of children and the different dialogs.  Caveats to interpretation of the data Basing any analysis on readability score assumes that the children are reading all or most part of the text, which might not be true. The reading scores might not account for the information consuming patterns developed by children such as skimming and interaction based on keywords or in the case of most pre-literates – relying on nontextual cues such as color. The reading scores might be indicative of when the children are reading the dialog text for the first time. For subsequent interactions, a completely different aspect of the dialog text, such as for example ability to form a unique visual map that is easy to recall, unique keywords placed at easy to access positions in the dialog etc, might come to play and the validity of the reading score, hence, might be weighed down.  155  The length of the text on the dialog box might not be sufficient to generate a reliable readability score upon it. Even if the generated readability scores are valid for the dialog text the scoring system might not have been validated with children, especially with very young children (pre- and semi-literate).  B.18 Learning effect Quantitative data analysis Since the dialogs were triggered automatically by the system based on a regular time interval, the maximum number of dialogs corresponds to the session length analysis which we saw earlier in Section B.7. Boys tended to have longer sessions and hence faced more dialogs than girls, on average. The same was true for semi-literates. Each child faced 4 dialogs on average. Gender Max. number Average number of dialogs of dialogs Boys 17 4.95 Girls 13 4.37 Table 77 Maximum and average number of dialogs across genders.  Age groups Max. number of dialogs [3,5] 13 (5,7] 16 (7,12] 17  Average number of dialogs 4.12 5.29 4.07  Table 78 Maximum and average number of dialogs across age groups.  Girls seem to have lower RTs for the most part as the number of dialogs increases, while boys do not seem to show such a pattern.  156  Boys Sequence Number of (nth dialog) children / dialogs 1 37 2 32 3 28 4 24 5 22 6 19 7 18 8 10 9 9 10 6 11 5 12 4 13 4 14 2 15 2 16 2 17 1  Girls Average RT Sequence Number of children / dialogs 13.81 1 36 10.47 2 30 9.89 3 26 6.29 4 23 5.36 5 17 8.42 6 15 7.78 7 13 11.3 8 10 13 9 7 6.5 10 4 7 11 4 9.75 12 3 5.75 13 3 11.5 19 10 18  Average RT 8.81 11.37 9.5 5.52 6.71 5.87 3.08 5.3 7.14 2.75 2.75 2.67 10.33  Table 79 Effect of the sequence with which the dialogs appeared on RT across genders  157  Figure 41 Effect of the sequence with which the dialogs appeared on RT across genders.  All age groups seem to show a reduction in RT as the sequence increases up to the 7th dialog. This makes sense as up to the 7th dialog we have representation of data from a number of users. Semi-literates seem to show a relatively strong learning effect when compared to other groups. It might be because they have relatively stronger numbers for most of the sequence (e.g., unlike literates who do not have much representation after the 7th dialog).  158  Sequenc e  [3,5] Number of children  1 2 3 4 5 6 7 8 9 10 11 12 13  23 16 16 14 10 7 7 4 4 3 2 1 1  Average RT  14.57 14.44 10.81 7.14 7.8 11 8.71 15 16.25 4.67 3 20 5  Sequenc e  (5,7] Number of children  Averag e RT  Sequen ce  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16  27 26 24 21 20 18 18 15 11 6 6 5 5 1 1 1  11.07 10.85 8.92 5.29 5.65 6.17 4.22 4.73 6.91 3 4.33 2.6 7 4 22 6  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17  Table 80 Effect of sequence on RT across age groups.  159  (7,12] Numbe r of childre n 23 20 14 12 9 9 6 1 1 1 1 1 1 1 1 1 1  Averag e RT  8.43 8.15 9.79 5.58 4.56 6.67 7.17 35 26 18 14 14 14 19 16 14 18  Figure 42 Effect of sequence on RT across age groups.  Color coding It is interesting that color coding seems to have no effect in reducing RT as the sequence increases.  160  Figure 43 Interaction between the sequences with which the dialog appeared and button color coding.  Body text, title visibility The visibility of title does not seem to affect the learning curve.  Figure 44 Interaction between the sequence with which the dialog appeared and visibility of title.  But the absence of body text seems to accelerate learning. The presence of body text, if 161  anything, seems to worsen the learning curve.  Figure 45 Interaction between the sequence with which the dialog appeared and visibility of body text.  Caveats to interpretation of the data We should also take into account that each child saw several designs for example, during some sequences the button positions could have been switched, during others the buttons would have been without color coding and during some the dialogs might have lacked body text or title. It is possible that the learning effect might have been stronger if the children were exposed to the same designs repeatedly. The effect of sequence: For those children who saw the buttons in the switched position to start with (as their first dialog), then the subsequent default view might seem as the switched one. Hence from the child‟s point of view, what represents a switched condition depends on what he/she started with. We are not sure what effect this might have had on interaction.  162  Appendix C Call for participation poster displayed at Science World The following poster, containing a collage of paintings created by children using Tux Paint and uploaded to tuxpaint.org, was displayed at Science World as a call for participation.  163  Appendix D Consent form for parents and legal guardians  164  165  166  Appendix E Assent form for children aged 3 – 6 years  167  Appendix F Assent form for children aged 7 – 12 years  168  169  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0052139/manifest

Comment

Related Items