. : ...."I‘
N ,.. ,.

  

 

up» —— - «‘

J __i. L 5 - .a a
Méoi‘iiga State
University

 

This is to certify that the
thesis entitled

CHANGE DETECTION IN ANIMATED CHOROPLETH MAPS

presented by

CAROLYN FISH

has been accepted towards fulfillment
of the requirements for the

MS. degree in GEOGRAPHY

 

 

 

‘3
/ ‘UMajfipProfessor’s Signature
/

/ ,

l

9”” 5/14/10

 

Date

MSU is an Affirmative Action/Equal Opportunin Employer

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KzlProleoc8Prele|RCIDateDueindd

CHANGE DETECTION IN ANIMATED CHOROPLETH MAPS
By

Carolyn Fish

A THESIS

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

MASTER OF SCIENCE
Geography

2010

ABSTRACT

CHANGE DETECTION 1N ANIMATED C HOROPLETH MAPS
By
Carolyn Fish

Computer animation enables cartographers to visualize time-series data as never
before; we can build dynamic map sequences that congruently depict change over time.
However, map readers have difficulty comprehending these animations, and they often
fail to detect important changes between adjacent scenes, this is called change blindness.
These potentially overwhelming perceptual burdens, such as change blindness, threaten
the effectiveness of animated maps in which several important changes can occur
simultaneously throughout the display. Animated maps also require viewers not only to
notice changes but also understand the transitions within these dynamic displays. Graphic
interpolation between display frames, also known as “in-betweening” or “tweening”,
smoothes transitions and lengthens the duration of the transition between scenes in an
animated map series. Previous cartographic literature suggests tweening as one potential
solution for change blindness in animated cartography. This thesis tested the influence of
change blindness on animated choropleth map reading and evaluated the influence of
tweening to increase change detection abilities of map readers. Empirical results from
this research indicate that map readers, 1) have difficulty detecting changes in these types
of maps, 2) often fail to comprehend these maps fully, and 3) are influenced by the use of

tweening between the scenes of the animated map.

 

 

Copyright by
CAROLYN FISH
2010

ACKNOWLEDGEMENTS

First, and foremost I would like to thank my advisor, Kirk Goldsberry, who was
always excited about my work, starting with our very first conversation. Kirk always
encouraged and pushed me to be the best. Without his guidance, I would never have gone
to the International Cartographic Conference in Santiago, Chile, a defining moment in my
graduate career. I would also like to thank my committee of Mark Becker and Judy
Olson; their advice and guidance were vital to the development, analysis, and writing of
this thesis.

My sincere thanks the Department of Geography at Michigan State, my fellow i.
graduate students for helping me both in and outside of school, the faculty for always
offering great advice, the students of Michigan State who participated in this study, and
especially to my undergraduate students in Geography 423 and 426, for inspiring part of
this thesis; I have learned more from you than you could ever imagine.

I would like to thank my colleagues from across the country. Anthony Robinson
and Cindy Brewer, you both introduced me to the world of cartography and have
continued to inspire me. Robert Roth and Andy Woodruff, your help and advice were
vital during my many standstills and setbacks while working on this project.

Funding for this research was provided by the MSU Department of Geography
Graduate Office Fellowship, the Master Thesis Grant from the Cartography Specialty
Group of the Association of American Geographers, and Kirk Goldsberry.

Finally, I would like to thank my parents. They have been here for me through
everything I have done, through my successes, set backs, and failures. They have been

my educational guides since childhood and without their aid I would never be here.

iv

 

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... VIII
LIST OF FIGURES .......................................................................................................... IX
Chapter 1
INTRODUCTION .............................................................................................................. 1
1.1 Introduction ............................................................................................................... 1
1.2 Research Goals ......................................................................................................... 3
1.3 Research Questions ................................................................................................... 4
1.4 Hypotheses ................................................................................................................ 4
1.5 Research Relevance .................................................................................................. 5
Chapter 2
LITURATURE REVIEW ................................................................................................... 7
2.1 Experimental Cartography ........................................................................................ 7
2.2 Representation of Spatial-Temporal Change ............................................................ 8
2.3 Characteristics of Animation .................................................................................. 11
2. 3.1 The Congruence Principle ............................................................................... 11
2.3.2 Informational Equivalence ............................................................................... 12
2.3.3 The Apprehension Principle ............................................................................ 13
2. 3. 4 Computational Equivalence ............................................................................. 13
2.4 Perceptual Obstacles ............................................................................................... 14
2.4.1 Change Blindness ............................................................................................. 14
2.4.2 Change Blindness Blindness ............................................................................ 18
2.4.3 Attention and Its Impact on Change Detection ................................................ 1 9
2.5 Animated maps ....................................................................................................... 20
2. 5.1 Perceptual Issues of Animated Maps ............................................................... 20
2.5.2 The Dynamic Variables .................................................................................... 21
2.5.3 T weening .......................................................................................................... 23
2.5.4 Influence of Smoothed Transitions ................................................................... 24
2.5.5 Animated choropleth maps .............................................................................. 26
2.6 Summary ................................................................................................................. 30
Chapter 3
METHODS ....................................................................................................................... 3 1
3.1 Overview and Goals ................................................................................................ 31
3.2 Overall Change Detection in Animated Choropleth Maps ..................................... 31
3.3 Change Detection Levels and Question types ........................................................ 32
3.4 Design Conditions ................................................................................................... 34
3.5 Change Characterization Accuracy ........................................................................ 37
3.6 Participant Confidence of Personal Performance ................................................... 38
3.7 Expected Results ..................................................................................................... 38
3. 7.1 Change Detection Level Expectations ............................................................. 39

I

 

3. 7.2 Transition Type Expectations .......................................................................... 39

3. 7.3 Change Behavior Expectations ........................................................................ 40
3. 7.4 Participant Confidence Expectations .............................................................. 40
3.8 Participants .............................................................................................................. 41
3.9 Materials ................................................................................................................. 41
3. 9. 1 Hardware ......................................................................................................... 41
3. 9.2 Map Data ......................................................................................................... 42
3.9.3 Legend Design ................................................................................................. 44
3.9.4 Classification Scheme ...................................................................................... 44
3. 9.5 Color Schemes ................................................................................................. 46
3. 9. 6 Interactivity ...................................................................................................... 46
3. 9. 7 Magnitude of Change ....................................................................................... 4 7
3.10 Procedure .............................................................................................................. 48
Chapter 4
RESULTS AND ANALYSIS ........................................................................................... 50
4. 1 Change Detection Performance ............................................................................. 50
4.1.1 Overall Change Detection Performance ......................................................... 51
4.1.2 Change Detection Level Performance ............................................................. 54
4.1.3 Change Detection Performance across Design Conditions ............................ 58
4.1.4 Change Characterization Accuracy- Cross Classification Arrays .................. 69
4.2 Confidence .............................................................................................................. 73
4.2.1 Overall Confidence Ratings for the Entire Test ............................................... 74
4.2.2 C onfidence for Different Levels of Change Detection ..................................... 75
4.3.3 Confidence Ratings for Change Detection Level 1 Across Conditions ........... 76
4. 2. 4 Confidence Ratingsfor Change Detection Level 3 Across Conditions ........... 77
Chapter 5
DISCUSSION ................................................................................................................... 78
5.1 Change Detection Abilities ..................................................................................... 78
5.2 Different Levels of Change Detection .................................................................... 79
5.3 Transitional Design ................................................................................................. 82
5.4 Transitional Salience ............................................................................................... 86
5.5 Tendency toward Change Blindness Blindness ...................................................... 88
5.6 Implicit Representations of Change ........................................................................ 90
5.7 Static Cartographic Conventions ............................................................................ 91
5.8 Attention and its Impact on Change Detection ....................................................... 92
5.9 Data Driven Change Detection Effects ................................................................... 95
5.10 Duration Effects .................................................................................................... 98
5.11 Limitations ............................................................................................................ 99
5.11.1 The Splat Contingent of the Highlight Rectangle .......................................... 99
5.11.2 Spatial Location of the Highlighted Unit ..................................................... 100
5. 1 1. 3 Transitional Behaviors ................................................................................ 1 00
5.11.4 Static Scene Duration .................................................................................. 100
5.11.5 Hits versus Correct Rejections ..................................................................... 102

vi

5.11.6 Magnitude of Change and Map Complexity ................................................ 103

5.11.7 Congruent Data Representations ................................................................. 104
5.11.8 Animated Simultaneous Contrast ................................................................. 104
5.11.9 Map Tasks .................................................................................................... 105
5.11.10 Signal Detection Theory, for Three Choice Answers .................................. 105
Chapter 6
CONCLUSION ............................................................................................................... 107
APPENDICES
APPENDIX A
CONSENT FORM ..................................................................................................... 112
APPENDIX B
PRETEST ................................................................................................................... 115
APPENDIX C
THE TRAINING ....................................................................................................... 118
APPENDIX D
POST- TEST .............................................................................................................. 132
LIST OF REFERENCES ......................................................................... 135

vii

LIST OF TABLES

Table 1. The SDT outcome array. Categorization of map readers potential decisions
based on what they were shown ................................................................................ 51

Table 2. Descriptive statistics of total change detection performance accuracy for all 108
questions for all 78 participants (hits plus correct rejections). ................................. 52

Table 3. Descriptive statistics of hits and correct rejections for the entire experiment for
the entire population (all 108 questions) ................................................................... 53

Table 4. Population level change detection performance for both levels 1 and 3 of change
detection (hits plus correct rejections) ...................................................................... 55

Table 5. Hits, correct rejections and false alarms for all Change Detection Level 1
questions ................................................................................................................... 57

Table 6. Hits, correct rejections and false alarms for all Change Detection Level 1
questions ................................................................................................................... 57

Table 7. Participant accuracy for all three conditions for Change Detection Level 1. To
answer these questions correctly, the participant only needed to notice whether a
change occurred in the highlighted unit. ................................................................... 59

Table 8. Hits, correct rejections and false alarms for the Abrupt Condition for Change
Detection Level 1. ..................................................................................................... 62

Table 9. Hits, correct rejections and false alarms for the Delayed Smooth Condition for
Change Detection Level 1 ......................................................................................... 62

Table 10. Hits, correct rejections and false alarms for the Continuous Smooth Condition
for Change Detection Level I. .................................................................................. 63

Table 11. Participant accuracy for all three conditions for Change Detection Level 3. 64

Table 12. Hits, correct rejections and false alarms for the Abrupt Condition for Change
Detection Level 3. ..................................................................................................... 67

Table 13. Hits, correct rejections and false alarms for the Delayed Smooth Condition for
Change Detection Level 3 ......................................................................................... 67

Table 14. Hits, correct rejections and false alarms for the Continuous Smooth Condition
for Change Detection Level 3. .................................................................................. 68

viii

LIST OF FIGURES
Images in this thesis are presented in color

Figure 1. Demonstration of possible data changes by enumeration unit for classed and

unclassed maps .......................................................................................................... 29
Figure 2. Example of a transition between two scenes ..................................................... 32
Figure 3. Example of Question Type #1 ........................................................................... 33
Figure 4. Example of Question Type #2 ........................................................................... 34
Figure 5. The Abrupt Condition schematic ....................................................................... 35
Figure 6. The Delayed Smooth Condition schematic ....................................................... 36
Figure 7. The Continuous Smooth Condition schematic .................................................. 37

Figure 8. A change characterization array for a pair of three (3) class animated choropleth

maps .......................................................................................................................... 38
Figure 9.Example of confidence question ........................................................................ 38
Figure 10. The sixty-four counties used for the stimuli .................................................... 42
Figure 11. Illustration of the rotation of the choropleth map stimuli ................................ 43
Figure 12. The legend. ...................................................................................................... 44

Figure 13. Graph illustrating the magnitudes of change for all of the 18 transitions used in
the experiment ........................................................................................................... 47

Figure 14. Box and whisker plot of total performance accuracy on all 108 questions for
all participants (hits plus correct rejections) ............................................................. 52

Figure 15. Box and whisker plot illustrating change detection performance by Change
Detection Level ......................................................................................................... 56

Figure 16. Box and whisker plot illustrating change detection performance for all three
conditions for Change Detection Level 1 ................................................................. 60

Figure 17. Mean performance and standard error bars for Change Detection Level 1.. .. 61

Figure 18. Box and whisker plot illustrating change detection performance for all three
conditions for Change Detection Level 3 ................................................................. 65

Figure 19. Mean performance and standard error bars for Change Detection Level 3.. .. 66

Figure 20. Cross-classification array of accuracy rates for all 108 questions in the
experiment ................................................................................................................. 69

ix

Figure 21. Cross-classification arrays of accuracy rates for all questions for the two
question types ............................................................................................................ 70

Figure 22. Cross-classification arrays of accuracy rates for all questions regarding
Change Detection Type 1 for each of the conditions ................................................ 71

Figure 23. Cross-classification arrays of accuracy rates for all questions regarding
Change Detection Type 3 for each of the conditions ................................................ 72

Figure 24. Stacked bar graph of correct and incorrect answers for the entire test (108
questions for 78 participants) categorized by participants’ confidence ratings. ....... 74

Figure 25. Column graph of correct and incorrect answers for each of levels of change
detection categorized by participants’ confidence ratings ........................................ 75

Figure 26. Column graph of correct and incorrect answers for each of the three conditions
for Change Detection Level 1 categorized by participants’ confidence ratings. ...... 76

Figure 27. Column graph of correct and incorrect answers for each of the three conditions
for Change Detection Level 3 categorized by participants’ confidence ratings. ...... 77

Figure 28. Illustration of participants responses to Change Detection Level 3 questions.
Often the correct answer is not always the most popular choice .............................. 81

Figure 29. Illustration of the differences between the ColorBrewer color schemes and the
in-between states of a tweened color scheme with the same origin and destination

states .......................................................................................................................... 85
Figure 30. Illustration of how the current legend design differs from what the map reader

sees during a transition state ..................................................................................... 86
Figure 31. Robinson’s highlighting methods ............. . ...................................................... 94
Figure 32. Original conditions ........................................................................................ 101
Figure 33. Proposed conditions ....................................................................................... 102

Chapter I
INTRODUCTION

1.1 Introduction

Computer animation enables cartographers to depict change over time
congruently by visualizing time-series data more realistically. We can use dynamic
temporal displays to mimic the passage of time. However, while animation is
advantageous for the depiction of time, viewers have difficulty apprehending the changes
that occur in dynamic displays (Tversky et a1. , 2002), and often fail to detect important
changes between adjacent scenes. This phenomenon is known as change blindness
(Grimes, 1996; Rensink, 2002a; Harrower, 2003; Simons and Rensink, 2005; Fabrikant et
al., 2008; Goldsberry and Battersby, 2009). One inherent problem with animated
thematic map displays is that they require users to notice, attend to, and decode many
simultaneous change signals during rapid scene transitions. Since several important
changes can occur simultaneously throughout the display during a single scene transition,
cartographic animation demands potentially overwhelming perceptual requirements of
map readers that threaten the effectiveness of animated maps (Harrower, 2007; Fabrikant,
2008; Goldsberry and Battersby, 2009).

In a more artistic context, Lasseter (1987) suggests that animators control changes
on the display so only one important change occurs at a time. Unfortunately, cartographic
animations are representations of underlying geographic information, and unlike more
artistic animations, their appearance cannot be completely controlled by the cartographer.

Cartographic animations also offer a second additional perceptual burden to

readers. To comprehend map animations fully, it not only is important to notice a change

between scenes of a map animation, readers must also understand the meaning of the
change. To comprehend a change in an animated map, a reader must first notice the
change, take notice of the geographic pattern before and after the change, and decode the
meaning of the transformation. Goldsberry and Battersby (2009) identified these tasks as
a part of the three levels of change detection applicable to thematic map reading:

° Change Detection Level 1: the reader notices the presence or absence of a
change, the persisting or shifting behavior of the map,

° Change Detection Level 2: the reader understands whether the change was an
increase or decrease in the mapped phenomena (only in quantitative thematic
maPS),

° Change Detection Level 3: the reader apprehends the origin state of the map and
makes the connection between the origin and destination states to understand the
meaning of the visual transformation.

To achieve the highest level of change detection and to understand the meaning of the
cartographic animation fully (Level 3), map designs need to afford readers time not only
to notice the origin and destination states. but also perceive the transition between the
two. This represents an important emerging cartographic challenge.

Finally, animated maps are vulnerable to the phenomenon of “change blindness
blindness”, which involves the overestimation of one’s own change detection ability
(Levin et al. 2000, 2002). Change blindness blindness literature has shown that often
people believe they will notice a change when they see one, yet often they still miss
important changes (Levin et al. 2000, 2002). Serious implications could occur if map

readers overestimate their change detection abilities. For example, dynamic cartographic

 

literature has suggested that interactivity can help map readers to read animated maps
more accurately (Tverksy et al.2002; Harrower, 2007; F abrikant et al., 2008); however,
change blindness blindness may mean that map readers are less likely to use the
interactivity available because they believe they have completely comprehended the
animated map.

Recent research has suggested one option to help mitigate change blindness in
animated thematic maps: gradual transitions (Goldsberry, 2004; Fabrikant et al., 2008). 4
By graphically interpolating between scenes, cartographers can smooth and lengthen the
duration of the transition in an animation this process is known as tweening. Tweening
can extend the amount of display time dedicated to the transition, consequently affording
readers with more time to View the change between display scenes. However, it remains
unclear to what extent gradual transitions will facilitate or impede the map reader’s
understanding of changes in these dynamic displays. This thesis investigates how well

map readers detect changes in dynamic thematic maps.

1.2 Research Goals
The goals of this thesis are to 1) determine to what extent change blindness affects
animated map reading and 2) examine how different transition designs affect change

detection.

1.3 Research Questions
To investigate the impact of change blindness on animated map reading, this
research attempts to answer three main questions:
1. How well do map readers detectchanges between scenes in animated maps?
2. Do map readers have greater difficulty detecting changes at higher levels of
change detection (Goldsberry and Battersby, 2009)?
3. Does the design of map transitions influence the change detection abilities of
animated map readers?
4. How will map readers’ confidence in their own change detection abilities

differ from their accuracy in detecting changes?

1.4 Hypotheses

Hypothesis 1. Rates of change blindness in animated choropleth maps will be high
Studies in psychology have shown that humans are innately blind to change (for review,
Simons and Rensink, 2005). Animated maps present additional perceptual burdens to
their viewers because change cannot be completely controlled by the cartographer and

thus many changes occur across the display at once.

Hypothesis 2. Map readers will have increased difficulty detecting changes at higher
levels. Higher levels of change detection require that map readers not only simply notice
the change, but also perceive its cartographic meaning by apprehending both the origin
and destination states of the map before and after a cartographic transition. It currently

remains unknown how well humans detect changes at different change detection levels.

Hypothesis 3. Utilizing tweening while still allowing participants to view the static
map scene before the change will allow participants to best comprehend changes
between map scenes. Previous research has suggested that tweening and smooth
transitions may mitigate change blindness by elongating durations of the transitions
between scenes, and afford map readers time to scan the display during a transition
(Goldsberry, 2004; Fabrikant et al., 2008). A combination of these longer transitions with
static scene display will allow map readers time both to read the map, and perceive the

transition.

Hypothesis 4. Despite missing many of the changes in animated thematic maps, map
readers would be highly confident in their change detection abilities. Previous
research in psychology has shown that while viewers often believe they would notice a
change, more often than not they failed to detect the change. The same results are

expected in this experiment.

1.5 Research Relevance

As a means to inform cartographic design, it is necessary to investigate how perceptual
limitations affect contemporary map reading tasks. Cartographic literature, specifically
Robinson’s The Look of Maps (1952), has called for empirical studies to investigate the
influence of cartographic design on map cognition. With the rapid emergence of dynamic

displays as medium to communicate spatial and temporal information, it is necessary to

 

investigate how the design of animated maps may influence the effectiveness of animated
maps.

It remains unknown how well map readers comprehend animated maps, and more
specifically how well they detect changes in these maps. Without this knowledge, it is
impossible for cartographers to improve the design of animated maps. Results from this
research will inform future design strategies that may potentially increase change
detection and indicate whether these types of animated maps are useful to their readers.
This study aims to uncover how well map readers detect changes, and how different

design strategies affect map readers’ abilities to detect change.

 

Chapter 2
LITURATURE REVIEW

In 1952, Arthur Robinson, published The Look of Maps, the seminal work of
academic cartography in the United States. The goal of the book was to establish that
maps are communication devices and research is needed to understand how best to
communicate cartographic ideas. Robinson specifically called for empirical research as a
means to reveal how cartographic design influences cognition of cartographic displays.
Specifically, Robinson argued that map quality should be judged according to user
performance opposed to aesthetic subjective opinions. This review aims to outline the
animation, psychology, and cartographic literature as it applies to how map readers
comprehend animated maps and how geographic information is communicated in these

displays.

2.1 Experimental Cartography

Robinson’s suggestions that maps should be developed based on map readers’
performance could be argued as the beginning of experimental cartographic research in
the United States. Robinson’s graduate student, J .J . Flannery, following Robinson’s
suggestion tested human subjects and their estimation of graduated symbols on maps.
F lannery’s (1971) findings indicate that generally, map readers underestimated the size of
these symbols. Over the next half a century, academic cartographic research became
synonymous with experimental cognitive psychological research on maps and map

readers. Robinson mentored and advised many academic cartographers many of who

 

continued his tradition, such as Flannery, to evaluate the effectiveness of maps based on
the user and not the map maker.

Montello (2002) indicated that the “heyday” of academic cartography was during
the 19705 based on the number of PhDs awarded during that time; as technology has
improved, it has allowed cartographers to investigate many of the cognitive map design
issues more easily by assessing speed, accuracy, and eye movements, which have all
become part of this academic cartographic research. Finally, many cartographers argue
that the rise of Geographic Information Science has decreased the interest in cognitive
map design research, but others argue that GIS and other new technological
developments may become new fodder for cognitive cartographic research in the future.
For example, little research has empirically investigated interactive and animated maps

and the perceptual abilities of the map readers who use them.

2.2 Representation of Spatial-Temporal Change

Cartographers have a multitude of ways to represent spatial-temporal change. In
the static domain, cartographers can depict change over time with a single map or a series
of maps, called small multiples (Campbell and Egbert, 1990; Tufte, 1983). With
animation, cartographers can display time series geographic data dynamically by
simulating changes of time, space, and attributes.

With a single map, a cartographer can represent temporal changes in attributes
using a “change map” (Monmonier, 1990) or with a map that depicts change using the
visual variables (Bertin, 1983). A change map illustrates attribute differences between

two time periods on a single map. For example, in a choropleth map, a map made up of

 

polygonal units (e. g. counties) colored or shaded based on a value or range of values in
the data, each enumeration unit is colored based on the attribute difference between the
two time periods. Using a sequential color scheme, ranging from light to dark, one could
imagine areas with high rates of change represented with dark hues, while areas with

little change represented with light hues. The visual variables also offer cartographers
additional ways to represent change on a single map. One example is the US. Geological
Survey’s 24,000 scale map series (Campbell and Egbert, 1990). In these topo quads,
urban areas are depicted in pink, and in the revised versions of these maps, areas of newer
urban growth are depicted in purple. Map readers can easily interpret how an urban area .
has changed between two or more time periods using this type of map (Campbell and
Egbert, 1990).

Cartographers can also depict change over time as a series of static graphics
called “small multiples”. Tufte (1983) is often credited with popularizing this type of data
representation. Cartographically, small multiples present a chronologically ordered map
sequence of the same geographic area at different time periods. These types of graphics
allow users to identify how one place has changed over time. Users can internally interact
with small multiples, by viewing the graphic at their own pace in their own preferred
order (Fabrikant et al., 2008).

While static maps can effectively illustrate change over time, dynamic
illustrations of spatiotemporal phenomena, specifically animated maps, offer one
significant advantage by allowing the cartographer to congruently represent change over
time. Using computer animation we can now depict temporal processes using temporally

dynamic representations. Several researchers have recognized the advantages of animated

 

maps as ways to provide new insights into dynamically changing data (Thrower, 1959;
Tobler, 1970). For example, perhaps the first computer animated map was designed by
Waldo Tobler in the early 19705 to visualize and understand urban growth in the Detroit,
Michigan region. However, researchers have also suggested that more research is needed
to provide insight into the power of these types of graphics (Comwell and Robinson,
1966). Until the 19905, however, little empirical research examined the utility and
practicality of animated maps. Previously, the costs and limitations of computing, along
with labor intensity, prevented cartographers from creating animated maps, but as
technology and computing have improved, creating animated maps has become “more a}
practical than ever” (MacEachren and DiBiase, 1991, p.221; Harrower, 2007).

Despite these remarkable technological developments, an important set of new
challenges has emerged. Since cost and access are less of an obstacle, contemporary
cartographic research questions have less to do with computation and more to do with the
design, perception, and cognition of animated maps. “When it comes to animated maps,
the bottleneck is no longer the hardware, the software, or the data - it is the limited visual
and cognitive processing capabilities of the map reader” (Harrower, 2007, p. 269). In this
way, animated maps present human beings with amplified map reading burdens. Recent
research has revealed, for example, that readers have difficulty understanding animated
graphics (e.g. Goldsberry and Battersby, 2009; F abrikant et al., 2008; Harrower, 2003,
2005; Tversky et al., 2002). One problem is that, human beings have difficulty
comprehending changes within these animations, and regularly fail to detect important

changes between adjacent scenes. This is particularly troubling because advocates of

10

 

cartographic animations argue that the power of animated maps is their ability to depict
change over time.

Animated displays are fleeting by design; as a consequence, one inherent problem
with animated displays is that they require users to notice, attend to, and decode multiple
simultaneous change signals during rapid scene transitions. These potentially
overwhelming perceptual requirements threaten the effectiveness of animated maps
because several important changes can occur simultaneously throughout the display
during a single scene transition. The goal of this chapter is to summarize previous

relevant research as a means to give context to the content of this thesis.

.-

2.3 Characteristics of Animation

2.3.1 The Congruence Principle

Computer animation enables cartographers to visualize time-series data as never
before; we can build dynamic sequences that congruently depict change over time. A
graphic is said to adhere to the congruence principle when the format and content of the
graphic corresponds to the format and content of the concepts to be conveyed (Tversky et
al., 2002). Animations are congruent in their ability to mimic the passage of world time
with the congruent passing of display time. Map animations allow us to illustrate change
dynamically in the same way as changes occur in the natural world. Fabrikant et al.
(2008) suggest that well-designed animations that adhere to the congruence principle will
allow their viewers to more easily understand complex spatiotemporal geographic

phenomena.

ll

 

2.3.2 Informational Equivalence

Several studies in psychology and educational psychology have compared
animated and static graphics to identify the benefits of using the two different types of
representations (e.g. Rieber, 1990; Thompson and Riding, 1990; Rieber, I991a; Rieber,
1991b; Large et al., 1996). These studies identified that animation fostered learning and
insight better than static graphics versions of the same ideas; however, when comparing
animated and static graphics these studies did not compare informationally equivalent
graphics (Tversky et al. 2002). Two representations are said to be informationally
equivalent when the same information can be inferred from both representations (Larkin “r
and Simon, 1987). When two graphical representations are informationally inequivalent,
one graphic is believed to present the viewer with more information than the other. In
other words, one graphic is superior to the other because it allows the viewer greater
inference leading to better apprehension when viewing the superior graphic. For example,
Thompson and Riding (1990) designed a program to teach the Pythagorean theorem to
junior high school students. The participants were divided into three groups, one group
viewed a static diagram, the second group viewed a discrete animation, and the third
group viewed a continuous animation. The third group, viewing the continuous animation
performed the best in learning the theorem. In the experiment, the authors state that the
paper graphic and the discrete animation illustrated the same information, while the
continuous animation was not informationally equivalent and included additional
information that lead to successful comprehension of the Pythagorean theorem. In the

cartographic domain, one might argue that small multiples and animated maps are not

informationally equivalent graphics because the two graphics present different

12

 

information to readers. Tversky et al. (2002) argue that informationally inequivalent
graphics cannot be compared because different information is presented to the users.

2.3.3 The Apprehension Principle

The goal of scientific visualization is to communicate complex ideas. In some
cases, these representations facilitate readers understanding; however, poorly designed
maps and graphics can be misunderstood and difficult to read. The apprehension
principle states that external representations must be “readily and accurately perceived
and comprehended” (Tversky et al. , 2002, p 256). Humans have difficulty perceiving
motion, preventing accurate apprehension of animated displays that are informationally
equivalent to their static counterparts (Tversky et al. 2002). As a means to achieve
apprehension, often animations present viewers with different and perhaps more
information than static graphics; however, by presenting more information in the
animated domain the two graphical forms become informationally inequivalent. For
example, interactivity is one known way to increase apprehension; however, interactivity
renders animated and static graphics informationally inequivalent by allowing the user to
interact with the animated graphic differently than they would with a static graphic.

2.3.4 Computational Equivalence

Fabrikant et al. (2008) argue that maintaining informational equivalence for
“good experimental control for comparisons may actually mean degrading [animation’s]
potential power for certain tasks” (Fabrikant et al., p. 202). Computational equivalence,
instead of informational equivalence, provides a better way to compare two different
representations of the same concept. Two representations are said to be computationally

equivalent when both representations allow the viewer to infer the same concepts and

13

 

ideas “quickly and easily” from explicitly designed information (Larkin and Simon,
1987). For example, in a map the cartographer can use a scale bar or a map scale ratio to
express the scale in a particular map, however, these two different representations of the
same idea are computationally equivalent because they both lead to the same inference
without difficulty for the viewer.

While Tversky et al. believe that interactivity in animation creates informational
inequivalence, Fabrikant et al., (2008) suggest there is internal interactivity in static
graphics. Users are able to go back and forth with their eyes to examine the various
scenes of a set of small multiples for any duration of time. The external interactivity of
animated displays, however, is more obvious and requires users to view the animation in
a specific order with developer specified buttons. Sweller (1994) suggests that this
external interactivity may not be as helpful to viewers as the internal interactivity of static

displays and may add cognitive load onto a viewer’s working memory.

2.4 Perceptual Obstacles

2.4.] Change Blindness

Change blindness, the “surprising difficulty observers have in noticing large
changes to visual scenes” may be one perceptual difficulty map viewers face when
reading animated maps and graphics (Simons and Rensink, 2005, p. 16). Change
detection, the opposite of change blindness, is the “apprehension of change in the world
around us” (Rensink, 2002, p. 246). Recent research in psychology indicates that
observers often fail to notice or perceive seemingly important changes, in both the real

world and on a display screen (for overview: Rensink, 2002; Simons and Rensink, 2005).

14

While change blindness/detection studies have changed over the years, the overall
experimental design has remained similar. In nearly all change blindness studies, an
observer is shown a scene, a change is made to the stimuli, and the accuracy of detection
or the response time of the observer is recorded (Rensink, 2002).

In the early change blindness studies, the changes to the visual stimuli were small
and thus changes understandably went unnoticed. These early studies identified that
people were surprisingly poor at detecting changes when the change occurred after a
visual occlusion or during an eye movement (saccade) (Rensink, 2002). For example, in
1953, Robert French tested how well participants in his study were able to detect when
small dots were displaced within a display of two to seven dots. He found that
participants were most successful at detecting the change when there were only a few
(two to three) dots in the display and were less successful at change detection as the
number of dots increased (French, 1953).

By the early 19905, change blindness and detection research no longer focused on
small changes within the visual field, but instead used large realistic stimuli changes.
Even when observers were presented with large magnitudes of change, the alteration
remained unnoticed (Rensink, 2002). For example, one study presented participants with
a photograph that changed to a similar, but slightly altered photo during a visual
disruption (Grimes, 1996). Participants had a surprising difficulty noticing the change
between the two pictures. Once the change was pointed out, participants had no difficulty
recognizing the difference between the two photos. In another study, observers were
shown a movie clip where an actor was swapped with another actor (Levin and Simons,

1997). Once again participants had difficulty noticing the change. Finally, even when

15

 

participants were asked to engage in conversation with another person, they often failed
to notice when their conversation partner was replaced by a different person during a
visual occlusion (Simons and Levin, 1998; Levin et al., 2002)!

Rensink (2002) identified several stimuli contingencies of change that lead to
change blindness.

1. Gap contingent- change occurs during a visual gap between the first and second
(altered) scene, also called the flicker paradigm (for reviews: Simons, 2000 and
Simons and Levin, 1997),

2. Saccade contingent- change occurs during an eye movement (e. g. Henderson and
Hollingworth, 1999),

3. Shift contingent- change occurs during a movement (shift) of the entire stimuli
(Blackmore et al., 1995),

4. Blink contingent— change occurs during an eye blink (e.g. O’Regan et al. 2000),

5. Splat contingent- change occurs when a distracter change, often resembling a
“mud splat” appears on the display at the same time as the change occurs.
Observers are distracted by the visual onset of the “splats” and they miss the
change (e.g. O’Regan et‘ al. 1999),

6. Occlusion contingent- change occurs during a brief occlusion (e.g. Simons and
Levin, 1998),

7. Cut contingent- change occurs during a change in viewing position (e.g. Levin
and Simons 1997). For example, movie directors rely on this type of change

blindness during camera cuts and movements,

16

8. Gradual change- change occurs slowly and gradually between scenes resulting in

change blindness (e.g. Simons et al., 2000).

Each of these contingencies have been tested and it has been determined that each leads
to some level of change blindness.

Animated maps present map readers with similar obstacles that may inhibit
change detection. Five of the eight change contingencies that have been proven to lead to
change blindness in psychology may also apply to animated map reading. Animated
maps are subject to change blindness for the saccade and blink contingencies because at
any point while viewing an animated map people may blink or move their eyes and miss
the change. The splat contingent also relates to animated map reading because often
many changes occur at one point in time on an animated map and have the possibility of
producing the same change blindness effect as the splat contingent. Observers may be
distracted by the onset of change in another place on the map and miss the change. The
cut contingent relates to 3-D fly though maps, or any map where viewing position may
change during the animation. During these types of cartographic animations, cuts may
occur between scenes and the viewer may miss the change. Finally, it has been suggested
that smooth transitions may mitigate change blindness (F abrikant et al., 2008), however,
Simons et al. (2000) has shown that a gradual change may cause change blindness
instead of reducing it.

While participants in all of the change blindness studies mentioned were
surprisingly blind to change even when the salience of change was large (conversation
partner changes), studies have shown that salience of change does have an effect on the

accuracy of change detection (Williams and Simons, 2000). The results indicate that

17

participants were best at correctly identifying persistence, followed by highly salient
changes, and performed the worst at detecting changes when the change was not salient.
The results of their experiments suggest that change detection is dependent on the
saliency of change that occurs between scenes. In animated maps, map readers are often
required to detect changes that may not always be the most salient on the map, and may
be blind to these changes.

2.4.2 Change Blindness Blindness

Despite the large rates of change blindness in both controlled and real-world
environments two studies identified that most people believed they would see these
changes within their visual field. This is called change blindness blindness (Levin et al.
2000, 2002). Levin et al. (2000 and 2002) tested this phenomenon using stimuli from
previous change blindness experiments (Levin and Simons, 1997 and Simons and Levin,
1998). In all of the experiments, participants overestimated their change detection
abilities. In one experiment, 76% of participants rated that they believed they would see a
change, yet 0% of the population actually noticed the change (Levin et al., 2000).
Because of this problem, people may believe they have noticed everything in their visual
field, but in reality they may have missed a significant amount of changes in the world
around them.

This metacognitive error of overestimating change detection abilities has large
implications not only for cognitive science, but also for the real world (Simons and
Rensink, 2005). Change blindness blindness also has serious implications for scientific
visualization including map reading, where missing display changes can result in

misinterpretation of the data presented by the map. For example, if a viewer were to miss

l8

the extreme increase in unemployment during the 19305 in an animated map of
unemployment in the United States during the 20th century, one would assume that the
viewer would go back and review the animation again. However, because of the change
blindness blindness phenomenon, and the overestimation of change detection abilities,
viewers might miss an important change such as this, yet would be less inclined to review
the animation. In short, people believe they are better map readers than they actually are.

2.4.3 Attention and Its Impact on Change Detection

Results from empirical studies on change detection (Rensink, 2002) suggest that
observers should be alerted of upcoming changes because focused attention is needed to
perceive changes in the visual field. Attention allows observers to keep a “memory” of
the scene before the change during a scene transition (Wolfe, 1999). Inattentional
blindness is then the inability to see objects within the area of fixation because attention
is focused somewhere else (Mack and Rock, 1998). Studies testing inattentional
blindness aim to answer the question, why do we sometimes look, but fail to see? Simons
and Chabris (1999) tested inattentional blindness in an experiment where participants
watched a video of several people passing a ball to each other. The experimental
participants were instructed to count the number of times the ball was tossed. During the
ball passing activity, a person in a gorilla costume walked directly through the center of
the display. Despite this activity in the area of fixation, 46% of participants, failed to
notice the unexpected person walking across the display.

Inattentional and change blindness studies beg the question, what captures
attention? Studies have shown that visual salience within the visual field captures

attention more than any other factor (Yantis, 2005). However, other research also

19

indicates that both abrupt onsets (Yantis and Jonides, 1984) and abrupt luminance
(Theeuwes, 1995) help to capture attention, allowing viewers to attend to changes more
easily.

Finally, while humans have the ability to attend to multiple things, four to five
different items at the same time, only one change can be perceived at a time (Rensink,
2002b). Animated maps present a special case of animation, where changes can occur at
multiple places throughout the display at the same time and thus potentially overwhelm

our perceptual system.

2.5 Animated maps

2. 5.1 Perceptual Issues of Animated Maps

While the primary utility of animated maps is to depict change over time
(Harrower, 2007), perceptual limitations like change blindness and inattentional
blindness obstruct map reading tasks and threaten the effectiveness of these maps. As
cartographers embrace the ease of creating animated graphics, the perceptual difficulties
associated with apprehending their message become more noticeable. In animated maps,
time-series geographic information changes constantly across the map display; thus,
several important changes can occur simultaneously throughout the display during a
single scene transition. For example, in an animated dot map, where dots represent the
occurrence of an event in time, (e. g. an earthquake) dots appear and disappear on the map
display as the data changes over time. In some instances, many dots may appear or
disappear at once. Because of the complexity of change that may occur at the same time,

viewers may have difficulty noticing the appearance or disappearance of a particular dot

20

(Harrower, 2003). F rench’s (1953) stimuli was not unlike an animated dot map and his
study revealed the difficulty people have in detecting these types of changes. Lasseter
(1987) indicates “it is important... that only one idea be seen by the audience at a time. If
a lot of action is happening at once, the eye does not know where to look and the main
idea of the action will be [missed]” (Lasseter, 1987, p. 38). Because in animated maps,
multiple important changes occur at many places at one time, change blindness,
inattentional blindness, and change blindness blindness each represent potentially serious
concerns for animated map reading.

2.5.2 The Dynamic Variables

To create effective animated maps, DiBiase et al. (1992) identified the dynamic
visual variables, which attempted to mirror, function as, and extend Bertin’s (1983)
visual variables for static displays. In other words, the cartographer can use the dynamic
variables to create an animated map in the same way as they would use Bertin’s visual
variables to create static graphics. The dynamic variables are: 1) duration 2) rate of
change, and 3) order (DiBiase et al., 1992). The dynamic variables are intended to help
cartographers create effective displays allowing the cartographer to “treat display time as
a controllable dimension of the map” (Goldsberry and Battersby, 2009, p. 206).

The dynamic variable, rate of change is defined as m/d, where m is the magnitude
of change and d is the duration. The cartographer can adjust the rate of change by
adjusting either magnitude of change or duration. The rate of change determines the pace
of the animation. “As m is increased (while holding d constant), the apparent rate of
change in the animation increases and the character of the illusory motion becomes

less smooth and more abrupt” (DiBiase et. al., 1992, p.206). Magnitude of change (m),

21

however, is directly related to the dynamic nature of the underlying data, and can only be
controlled by the cartographer to a certain extent. On the other hand, it is easier to adjust

duration. If duration (d) is increased and m is held constant, rate of change will decrease

and the animation will become smoother and less abrupt.

There are two key components to an animated map, a scene and an event A scene
is a representation of an instant in time called a situation. A scene can be represented as a
static map or as a frame in an animation. An event is composed of a series of situations
(Szego, 1987; DiBiase et al., 1992). Multiple scenes can be displayed as a set of small
multiples to portray an event in static form (DiBiase et al., 1992). In the dynamic domain,
an event is represented as a series of scenes or frames played one after another.

The dynamic variable duration or scene duration is defined as the length of a
scene in time (DiBiase et al., 1992). For example, a scene displayed for ten seconds
would have a longer duration than a scene displayed for one second. The amount of time
a scene is displayed affects the pace and temporal scale of the animation. Fast-paced
animations have short scene durations, while slow-paced animations have long scene
durations. When transitions are elongated in an animation, duration is defined as the
amount of time devoted to the transition (Goldsberry, 2004). Thus, an animation with
long transitions has a longer duration than an animation with short transitions. In the
context of this research, Goldsberry’s definition is a more appropriate indicator of
duration in a map with elongated transitions.

. The quantification of the amount of change between frames in an animation is
known as the magnitude of change (Goldsberry and Battersby, 2009). In effect,

magnitude of change describes the amount of change that occurs between scenes in a

22

dynamic map display (DiBiase et al., 1992, Goldsberry and Battersby, 2009). The design
of the static map frames, the dynamics of the mapped phenomena, and the sampling
interval of the phenomena (DiBiase et. al., 1992) typically determine the magnitude of
change. For example, the United States Census is conducted every ten years, while mean
daily temperatures are collected everyday, thus the magnitude of change is different
between the two because the sampling method and the dynamics of the phenomena are
different. Magnitude of change, however, can also be quantified by the amount of display
change from scene to scene. In thematic map animations, the classification scheme and
the number of classes, influence the magnitude of change between display frames
(Goldsberry and Battersby, 2009). For example, a five class animated choropleth map
will typically have a higher magnitude of change than a three class animated choropleth
map because more enumeration units are likely to change between scenes when there are
a greater number of classes. As a result, readers would have to notice and attend to more
changes throughout the animation.

2. 5.3 T weening

Animation software offers the designer several other dynamic capabilities beyond
the dynamic variables, including tweening. Tweening or “in-betweening” is a process
first used for hand-drawn animations, and was one of the first uses of the computer for
early computer animations. In hand-drawn animations, the “master” animator would draw
the important “keyframes” and a less experienced animator would fill in the “in-between”
frames (Kochanek and Bartels, 1984). In contemporary animation computers are used not

only to create the “tweens”, but also to adjust and even draw the keyframes.

23

In dynamic thematic maps, the keyframes of animation are usually a set of static
maps that could be displayed as a set of small multiples in the static domain. The
keyframes in these maps are often associated with the temporal sampling interval. While
many geographic phenomena are temporally continuous in nature, this data can only be
collected at discrete times and places. Tweening allows the cartographer to interpolate
graphically between temporal samples (keyframes) and smooth the transitions between
the static maps to produce a continuous effect. For example, in a map based on United
States Census data, the keyframes would correspond to the first year of each decade,
when the Census was conducted. Tweening can be used to smooth the transition between
decades. This technique is similar to the spatial interpolation techniques used by
geographers to develop Digital Elevation Models (DEMs). A DEM is created by
interpolating from a few survey points to create a continuous 2.5 dimensional model of
elevation on the Earth’s surface. Tweening is the graphic and temporal interpolation
between known keyframes.

2.5.4 Influence of Smoother! Transitions

It has been suggested that tweening and “applying smooth transitions” (p.202)
may even help to reduce change blindness (Fabrikant et al., 2008) when the transition
stages are just long enough but not too long (Heer and Robertson, 2007). However, two
studies using the gradual change contingent of change blindness (Simons et al., 2000;
Seeber, 2003) concluded that smoothed transitions did not result in better change
detection. Simons et a1. (2000) tested change detection by altering a photograph over a 12
second gradual transition The authors concluded that gradual transitions resulted in the .

same rate of change blindness as the flicker paradigm used on the same photograph.

24

Goldsberry and Battersby (2009), however, attribute this to the length of the transition
used in the experiment; however, Simons et al. note that while the transition was slow, it
was not unnoticeable.

In 2003, Joanna Seeber was the first to investigate the benefits of smoothing in
the cartographic domain empirically with her study on animated dot maps for her Masters
thesis at the University of Wisconsin-Madison. The goals of her research were to
“ascertain if pattern recognition [was] more accurate with smooth or abrupt animations”
and to determine the type of transition users preferred (Seeber, 2003, p. 33). Because
DiBiase et al. (1992) suggested that people find smooth animations more visually
appealing, she hypothesized that users would perform better with abrupt animations on
more difficult map reading tasks, but users would prefer smooth animations (Seeber,
2003)

Seeber (2003) tested her hypothesis on two groups of 30 subjects, mostly graduate
students in geography. The groups viewed two dot map animations of the same dataset,
one smooth and one abrupt, in different orders. The abrupt animation played for 28
frames at a rate of 1 frame per second over 28 seconds. The smooth animation was
created with tweening and the dots faded in and out with transparency during the
animation. This “animation played at a frame rate of 5 frames per second, was 145
frames long, and thus played for a total of 29 seconds” (p. 28). Twenty-eight of the
frames were data collection points and the remainder of the frames were created with
tweening.

The experiment consisted of quantitative and qualitative multiple-choice and

short-answer questions of three difficulty levels: easy, medium, and hard. The questions

25

were aimed to ascertain performance on several different map reading tasks from specific
to general. Seeber’s results for the quantitative questions indicated that there was little
difference between the smooth and abrupt animations for task performance. On the
qualitative questions, participants indicated the smooth animations were “too fast”, but
found the abrupt animations to be “just right”. However, she stated that map viewing
order had an impact on her results. Participants preferred the abrupt maps to answer
questions about the map, but preferred the smooth map for overall aesthetics (Seeber,
2003)

Finally, Seeber’s animated maps included interactive buttons. Each map (abrupt
and smooth) had the same buttons, play, stop, step forward, and step backward. In her
conclusion, Seeber (2003) noted that participants relied heavily on the step buttons and
suggested that future similar experiments exclude interactive elements such as buttons.

2. 5.5 Animated choropleth maps

A choropleth map is a map made up of polygons called enumeration units colored
or shaded based on the data values or qualitative phenomena associated with each unit
(Slocum et al., 2009). The famous “Red-State, Blue-State” map, illustrating majority
political affiliation by state, is perhaps the most popular and recognized choropleth map
in the United States. An animated choropleth map is often made up of a time-series of
choropleth maps depicting different time periods, shown one after the other. For example,
one could imagine an animated “Red-State, Blue-State” map to illustrate changes in
political affiliation over time.

The primary utility of animated choropleth maps is to depict geographic change,

aggregated to enumeration units, over time and space (Harrower, 2007). Readers of

26

animated choropleth maps may use the maps to locate three different geographic extents:
l) a single enumeration unit, 2) a region of the map, or 3) the entire extent of the map.
Users of these maps must apprehend if a particular place is changing and the magnitude
of change in comparison to the surrounding enumeration units (Goldsberry and Battersby,
2009).

Goldsberry and Battersby (2009) describe three levels of change detection in
animated choropleth maps. All three levels are needed for the map reader to comprehend
the dynamics of these maps fully. The three levels of change detection are:

l. The reader simply noticed there was a change between scenes in the
animated choropleth map but cannot identify how exactly the map
changed.

2. The reader noticed the change and was able to detect that the change
resulted in an increase or decrease from the original state (only
applicable for quantitative maps),

3. The reader noticed the change and is able to recall the origin state of the
map after the transition, the highest level ofchange detection.

To apprehend the geographic information encoded within animated choropleth maps, the
reader must achieve the highest level of change detection.

Finally, one reason animated choropleth maps are difficult to read is due to
classification (Harrower, 2007; Goldsberry and Battersby, 2009). Data values for
choropleth maps depicting quantitative data by enumeration unit are typically divided
into groups called classes. The cartographer determines the grouping of these data values.

In a choropleth map, enumeration units within the same class are shaded the same color.

27

However, technological advances have allowed cartographers to create unclassed maps
more easily, where each individual data value is represented by a unique color value
(Slocum et al. , 2009). In classed animated choropleth maps, some units may shift in color
many times throughout an animation when the data values associated with those units fall
near class breaks. Thus the changes in these maps give a flicker illusion (Figure 1). On
the other hand, other units remain the same color or persist through time when data

values only change within a class and never cross class breaks (Harrower, 2007). Some of

this is due to classification, which may hide changes in the data.

28

 

 

   

.5, 9w sawdust-w“ . a '5‘

       

allow iii! tinting“
~r, .‘ . -.,_. ’

H " Enumeration Unit 1
Enumeration Unit 2

    
 

data values

 

 

Unclauod Map

' . . v“.‘ _ .- . , , _ ,A.

. 32:55"

     

Enumeration Unit 1
Enumeration Unit 2

data values

 

—r.me—>

Figure l. Demonstration of possible data changes by enumeration unit for classed and unclassed
maps. Enumeration Unit 1 changes subtly over time, however according to Harrower, this unit “flickers”
by switching back and forth between classes in the classed map, while Enumeration Unit 2 changes
drastically over time, but remains in one class in the classed map. Harrower argues that in the unclassed
map, the changes in the map more congruently represent the changes in the data (graphic modeled afier
Harrower 2007 and Goldsberry and Battersby, 2009).

Harrower suggests using unclassed animated choropleth maps because they do not
‘flicker’ as much and slight changes are more easily apprehended in unclassed maps. The
prior notion (Monmonier, 1985; Dykes and Unwin, 1998) that classed maps are more
desirable than unclassed maps does not hold true for animated choropleth maps because
of the availability of interactivity which gives the user the ability to click or mouse-over

an enumeration unit for its data value (Harrower, 2007). However, Goldsberry and

29

Battersby (2009) state that there are still instances when classed animated choropleth
maps may be necessary, and cartographers must do their best to create animated

choropleth maps that allow for the greatest apprehension by map readers.

2.6 Summary

Due to recent technological advances, and reduced computing costs, animated
scientific graphics are increasing in popularity. However, whether or not these new kinds
of visual representations offer advantages over their static counterparts remains unknown.
For instance, many cognitive and perceptual limitations that have only recently been
uncovered threaten the effectiveness of animated maps. Several researchers (Fabrikant et
al.; Goldsberry and Battersby, 2009) have hypothesized that smoothing transitions in
animated maps may improve apprehension of these graphics. However, Simons et al.
(2000) and Seeber’s (2003) studies tested the effect of smooth transitions on change and
pattern detection and refute this suggestion. Their results indicate that smooth transitions
did not facilitate change and pattern detection. This thesis investigates the incidence of
change blindness in animated choropleth maps and how different design strategies

influence change detection in these types of maps.

30

 

Chapter 3
METHODS

3.1 Overview and Goals

An experiment was designed to investigate the influence of change blindness on
animated choropleth map reading. In the experiment, map readers were asked to detect
changes in animated choropleth maps. Accuracy, bias criterion, and discriminability of

the changes were each assessed. There were several goals of the study:

1. To assess human change detection abilities for animated choropleth map reading,

2. To evaluate human change detection performance for different levels of change
detection (Goldsberry and Battersby 2009),

3. To explore how different transition designs influence map readers abilities to detect
changes,

4. To investigate change detection performance for different “change characterizations”,

5. To investigate the relationship between participants’ confidence and their abilities to

detect changes in animated choropleth map reading tasks.

3.2 Overall Change Detection in Animated Choropleth Maps

Participants in the study saw 108 total questions. In each question, a choropleth
map was shown. Following a transition, one unit, of the total 64-unit map, was
highlighted, and the participant was asked to indicate whether the highlighted unit had
changed between the two scenes (Figure 2). Participants were assessed on whether they

correctly or incorrectly detected changes.

31

 

rHANSl‘UO“

Figure 2. Example of a transition between two scenes. The red rectangle was intended to highlight an
enumeration unit of the map. The participant was required to determine whether the highlighted unit
changed in lightness between the two scenes.

3.3 Change Detection Levels and Question types

Two question types were asked to correspond with the different levels identified
by Goldsberry and Battersby (2009). In Question Type #1, the participant was asked a
simple yes/no question about whether he or she was able to detect a change in the unit
highlighted with the red rectangle (Figure 3). These questions were designed to
understand accuracy for Change Detection Level. The first level of change detection is
the ability simply to notice that a change has occurred. Goldsberry and Battersby (2009)
classify it as the lowest level of detection. Half of the questions in the test were of Type

#1, for a total of 54 questions.

32

 

Did the highlighted unit
change between scenes?

 

Figure 3. Example of Question Type #1. In this question, participants simply had to indicate whether they
had noticed a change or not. This question was designed to understand how well participants were able to
detect changes at Level 1.

For Question Type #2, the participant was asked to recall the origin state of the
highlighted unit prior to the scene change by selecting “high”, “medium”, or “low” from
the answer choices (Figure 4). If the participant wished to indicate that the unit had
persisted, he or she could select the current color of the unit from the options. This type
of change detection is considered an advanced level of detection; the reader is able not
only to notice the change, but is also able to recall the origin state of the map prior to the
change (Goldsberry and Battersby, 2009). This level of change detection is necessary to
understand the fully meaning encoded in the map. F ifty-four questions of the total 108

question test were of Question Type #2.

33

 

Please indicate the color
of the highlighted unit
before the scene change:

DUI

 

Figure 4. Example of Question Type #2. In this question, participants were instructed to indicate the
shade of highlighted unit before the scene change. This question was designed to understand how well
participants were able to detect changes at Level 3. Participants not only had to grasp whether the unit had
changed, but also the origin state of the unit before the change.

3.4 Design Conditions
Previous research suggests that transitional designs variables might influence

change detection accuracy in animated maps(Gold5berry, 2004; F abrikant et al. , 2008).
F abrikant et al. (2008) suggested that gradual transitions created with tweening may
allow map readers more time to view the transitions and thus more easily apprehend the
changes that occur during a transition state. In this experiment, performance was assessed
for three different transition design conditions. The three conditions were:

1. Abrupt (Figure 5),

2. Delayed Smooth (Figure 6),

3. Continuous Smooth (Figure 7).

34

In all three conditions participants saw a two second map animation; however,
duration of the transition and duration of the static scene before the transition differed
across conditions. In the “Abrupt Condition”, the static map (origin state) was shown for
two seconds then abruptly transitioned to the second scene (destination state) of the map
(Figure 5). The second scene remained visible for 250 milliseconds before the red
highlight rectangle appeared to highlight one enumeration unit within the second scene.
The 250-millisecond delay was added to ensure that subjects focused on the change
within the map and not the appearance of the question and the red rectangle. In other
words, the experiment was designed to avoid problems like the splat contingent of change
blindness (O’Regan et al., 1999). The participant could view the second scene for as long
as he or she wished until answering the question. Once the question was answered, the
next map would appear. Thirty-six questions were asked about maps within the Abrupt

Condition.

 

 

 

 

 

Two (2) Seconds .9910

Figure 5. The Abrupt Condition schematic. In this condition. the static first scene (purple) was shown for
two (2) seconds. followed by the second scene (orange) which was shown for 250 milliseconds before the
highlight rectangle and question appeared.

In the “Delayed Smooth Condition”, a static map (origin state) was shown for one

second and the map gradually transitioned for one second to the second scene

35

(destination state) through the use of tweening (Figure 6). The red highlight rectangle and
the question appeared immediately after the transition ended and the second scene was
shown. Once again. participants had an unlimited amount of time to View the second
scene and answer the question. Thirty-six questions were asked about maps within the

Delayed Smooth Condition.

 

 

 

 

 

V V
One (1) Second One (1) Second

Figure 6. The Delayed Smooth Condition schematic. In this condition, the static first scene (purple) was
shown for one (I) second. followed by a one (I) second transition between the first and second scene. Once
the transition ended. the highlight rectangle and question appeared.

In the “Continuous Smooth Condition”, no static map was shown. This transition
design was the most congruent to the passage of world time. The map was in a perpetual
state of transition from the onset of the first scene (origin state) to the transition to the
second scene (destination state) of the map upon appearance on the display (Figure 7).
The map transitioned for a total of two seconds. This condition adheres to the congruence
principle by mimicking the passage of time with a continuous transition (Tversky et al.
2002). Immediately following the transition, the red highlight rectangle and the question
appeared. Thirty-six questions were asked about maps within the Continuous Smooth

Condition.

36

 

 

 

 

 

W

Two (2) Seconds

Figure 7. The Continuous Smooth Condition schematic. In this condition. no static scene was shown,
the map immediately began to transition upon appearance on the screen and continued to transition for two
(2) seconds. Once the transition ended. the highlight rectangle and question appeared.

Each participant saw all three conditions. The conditions were ordered in six
different ways for a total of six different test forms to ensure that learning effects did not
play a role in the experiment. An equal number of participants took each of the six test

forms.

3.5 Change Characterization Accuracy

Changes within an animated choropleth map can be effectively categorized using
cross-change characterization arrays (Monmonier, 1975; Goldsberry 2004). In the array,
the diagonal elements from upper left to lower right indicate the units that did not change
while the off-diagonals indicate the units that did change and how they changed. The
maps used in this experiment were three class maps with “high”, “medium”, and “low”
categories. A three-class map has nine possible transition behaviors shown in a 3 x 3
matrix (Figure 8). In the experiment, 12 questions were asked about each of the possible
change detection behaviors, with the exception of the “low” to “medium” transition and
the “high” to “medium” transition. Fifteen questions were asked about units transitioning
from “low” to “medium”, while only nine questions were asked about units transitioning

from “high” to “medium”

37

 

destination state

 

L noA L-rM L—>H

M M—>L noA M->H

origin state

H-9L H-tM

   

Figure 8. A change characterization array for a pair of three (3) class animated choropleth maps. In a
three-class map, there are nine (9) potential change behaviors. Three of these behaviors are examples of
persistence, while the other six change in various ways. The persistent behaviors are represented by the
diagonals from top left to bottom right on the array.

3.6 Participant Confidence of Personal Performance
Following each of the 108 questions, participants were asked to indicate their

confidence on the previous question by indicating “confident , neutral”, or “unsure”

(Figure 9). There were a total of 108 confidence questions.

How confident are you about your answer?

Confident Neutral Unsure

Figure 9.Example of confidence question. Participants were able to indicate that they were “confident”,
“neutral”, or “unsure” of their answer on the previous question. There were 108 confidence questions, one
following each of the change detection questions.

3.7 Expected Results
Previous change detection research in psychology has shown that humans have
great difficulty detecting changes in dynamic displays (for review: Rensink, 2002;

Simons and Rensink, 2005). Since animated maps present a uniquely complex stimulus,

38

it was predicted that participants would have difficulty detecting and apprehending
messages encoded within the animations.

3. 7.1 Change Detection Level Expectations

Due to the assumption that it would be easier for participants simply to detect
whether an enumeration unit within the map shifted or persisted during a transition, than
to comprehend both the origin and destination state of a particular unit, it was predicted
that participants would perform better on Question Type #1. In this question type,
participants were asked simply to indicate whether the unit changed, or remained the
same. There are two bases for this prediction: 1) participants would have an easier time
answering questions about Change Detection Level 1 because they had a 50% chance of
guessing correctly and 2) because it was assumed that it would be easier simply to notice
the presence of a change, than both to notice it and recall the unit’s initial origin state
(Question Type #2).

3. 7.2 Transition Type Expectations

By elongating transitions between scenes, tweening can reduce the rate of change,
and lengthen the amount of time devoted to cartographic transitions. This increase in the
duration of the transition results in a lower rate of change. Rate of change, defined by
DiBiase et al. (1992), is the magnitude of change (m) divided by the duration of the static
scene ((1). However, duration of the static scene does not account for transitions between
scenes, thus a second definition can be used. Duration of change (D) is the duration of
the transition between two scenes. In a non-tweened animation, the duration of change is
the amount of time between the two frames, and in a tweened animation it is the “length

of time devoted to the tween” (Goldsberry, 2004, p. 52).

39

 

It was predicted that the Delayed Smooth condition would elicit the highest
change detection accuracy because it affords the reader enough time to view the static
scene before the transition as well as the gradual change between the two scenes.
Participants would have time to 1) view the static map for one-second and 2) View the
transition for another second. This design would more likely allow participants to notice,
attend to, and perceive a greater number of changes within the map, making it possible to
perceive the specific change of the highlighted enumeration unit.

3. 7.3 Change Behavior Expectations

Accuracy for each type of change behavior, as shown in the change
characterization arrays, was expected to vary. Because changes where the unit
transitioned between the “highest” and “lowest” class are more salient, it was expected
that these types of changes would elicit higher change detection performance scores.

3. 7.4 Participant C onfidence Expectations

Previously perceptual research has indicated that while many times people miss
changes within their visual field, they firmly believe that they would see the changes
within their visual field (Levin et al., 2000, 2002); this phenomenon is known as change
blindness blindness. In this study, it was expected that participants would be more
confident than accurate. While participants would not always be highly accurate, it was
predicted that questions with higher accuracy scores would also have higher confidence
ratings. Because of the expectation that participants would perform better when asked
about lower levels of change detection, it was predicted that participants would be more

confident on these types of questions. Finally because it was predicted that participants

40

would perform better on the Delayed Smooth Condition, it was also expected that

participants would rate these questions with higher confidence.

3.8 Participants

A total of eighty—one students at Michigan State University participated in the
study. The participants signed up for the study voluntarily in several Geography and
Integrative Social Science classes; however, only three of the total 81 participants were
Geography majors. The other 78 participants were made up of 46 different majors from
across campus. Each participant was compensated $10 for about 30 minutes of his or her
time. An equal number of males and females participated in the study, 39 of each gender.
The age of the participants ranged from 18 to 29 (mode: 20). All of the participants were
students; they indicated that they were freshmen, sophomores, juniors, seniors, other, or

grad students in school (mode: sophomore).

3.9 Materials

3.9.1 Hardware

The study was conducted on Dell Precision 690 Computers with Dell 1907 FPt
computer displays with a screen resolution of 1280 x 1024, and a refresh rate of 60 Hertz.
The brightness was set to 100 and the contrast was set to 50 on all of the displays. The
questions in the study were designed in Adobe Flash CS3 (Adobe Systems, Inc., 2007)
using the default frame rate of 12 frames per second. The experiment was administered

using Mozilla F irefox (Mozilla Corporation, 2009) in “full-screen” mode.

41

3.9.2 Map Data
Sixty-four of the total 154 counties of the state of Georgia were used to create the

animated stimuli maps (Figure 10).

 

Figure 10. The sixty-four counties used for the stimuli are highlighted in purple; the other counties of
the state of Georgia were excluded from the stimuli.

The stimuli maps depicted average yearly unemployment rates in Georgia
counties between 1990 and 2008. This data is freely available from the United States
Bureau of Labor Statistics for all of the counties in the United States. Nineteen keyframes
were used to represent each year between 1990 and 2008, thus there were 18 total
transitions.

The geographic area used in the animations was unfamiliar to the participants of
the study. Due to the demographics of the participants, the geographic area selected for
this study was located outside the Great Lakes region. To avoid biases from participants,
they were not informed of the geographic area or underlying data. In addition, the
counties used for the study were rotated in four different directions throughout the

experiment (Figure 11).

42

 

ROTATEP

Figure 1]. Illustration of the rotation of the choropleth map stimuli. In the experiment, each of the
questions used the same basemap. To avoid having participants “memorize” the basemap, the map was
rotated in one of four directions.

To create the animations for the stimuli, the data was first manipulated in
Microsoft Excel (Microsoft Corporation, 2008), joined to a shapefile in ArcMap 9.2
(Environmental Systems Research Institute, Inc., 2006), rotated in Adobe Illustrator CS3
(Adobe Systems Inc., 2007), and then imported directly into Adobe Flash CS3 (Adobe
Systems Inc., 2007) where the set of small multiples was animated into a sequence.

There were a total of 19 maps and 18 transitions. These maps were repeated six
times throughout the test, once per condition type and once per question type. Within
each set of maps, the years were randomly ordered using a random number generator.
Within each condition, the different question types were inter-mixed. For example, the

participant might see three Question Type #1 ’s followed by two Question Type #2’5.

43

Throughout the 108 questions of the experiment, 36 of the 64 enumeration units
of the map were highlighted, thus the same unit was highlighted 3 times (once per
condition).

3. 9.3 Legend Design

The layouts for the map animations were simple. To prevent biases related to the
geographic location and theme of the map, no map title was included. Previous
cartographic animation research suggests simple legends (Harrower, 2003), thus the
legend only included the legend definitions of: “high”, “medium”, and “low” as opposed
to more detailed numerical definitions (Figure 12).

High

Medium

 

 

 

Low

 

 

 

Figure 12. The legend used in the experiment was simple and only included: “high”, “medium”, and
“low”.

3. 9.4 Classification Scheme

Several different classification schemes can be used in choropleth mapping;
however, no one classification scheme works optimally for all data. Different
classification schemes result in a different number of data points or observations within
each class. The scheme also has the ability to emphasize or de-emphasize different
characteristics within the data.

According to Harrower (2003), additional issues of classification come into play
in animated mapping. The classification scheme can lead to empty classes and skewed
distribution and must be heeded when choosing how to class the data. It is also important
to have a matched legend throughout an animated choropleth map sequence to avoid

44

confusing map readers. In other words, the classification scheme chosen should attempt
to avoid the problems noted by Harrower (2003).

Brewer and Pickle (2002) suggest that quantiles are the best classification scheme
for map comparison. The quantile classification scheme places the same number of
observations into each class. The time series used in this experiment consisted of 1,216
observations (19 years of data for 64 counties). The three-class quantile classification
results in about 405 observations in each class. Often class breaks occurred between two
observations of the same value and the classification was altered slightly to move class
breaks between different observation values.

Because of the additional perceptual requirements necessary to read animated
choropleth maps, cartographers must design these maps differently than traditional static
choropleth maps. In static choropleth maps, five to seven classes are often considered an
appropriate number of classes to use (Slocum et al., 2009). However, five to seven
classes are too many for animated choropleth maps (Harrower, 2003), where users have a
limited amount of time to view each scene of the animation. To reduce complexity,
Harrower (2003) suggests using only two to three classes for animated choropleth maps.
Because animated map displays are fleeting by nature, users often do not have much time
to view the display and must apprehend important geographical distributions in a matter
of moments. Goldsberry and Battersby (2009) also indicate that increases in the number
of classes result in higher complexities and magnitudes of change. The stimuli in this
thesis experiment have three classes to reduce strain on working memory as well as to

reduce magnitude of change.

45

3. 9.5 Color Schemes

Choosing an appropriate color scheme is very important for choropleth maps. In
these maps, a gray color scheme was used to avoid color effects and protect biases
against colorblind participants; a lightness-based sequential color scheme was used to
imply a unipolar data distribution (Brewer, 1994). Specifically, because the maps in this
experiment utilized tweening, the colors ranged from dark to light and when a unit
transitioned between light and dark, the color passed through the middle gray color. The
differences in lightness between the shades of gray used were salient enough that even
between just two frames of the animation, it was a noticeable difference for the map
reader. This way, participants could distinguish even the slightest color shift that occurred
between two single frames.

3.9.6 Interactivity

Although research suggests interactivity can help users better comprehend
animated maps, and that users often get frustrated with maps they cannot control
(Harrower, 2003), these maps did not include any interactivity. The reason for this is two-
fold: 1) interactivity would allow each participant to view the animation differently and
would no longer be controlled for time, and 2) in Seeber’s study (2003), interactivity was
available and her participants’ tended to rely too much on the buttons to move between
scenes. In her study, each condition included the same buttons, play, stop, step forward,
and step backward. After noting the reliance on the buttons, Seeber stated, “it would be

useful to design future experiments so that the animations had no step buttons” (p. 66).

46

3. 9. 7 Magnitude of Change

DiBiase et al. defined magnitude of change, m, as the difference in position and
attributes between scenes. Goldsberry and Battersby (2009) extended this definition, as.
the amount of symbology change between two scenes. In a classed animated choropleth
map, m is the number of enumeration units that shift between classes (Goldsberry, 2004).
Overall this dataset included many maps with relatively high magnitudes of change
(Figure 13). For example, the highest magnitude of change between two scenes was 50 of
the 64 units, while the lowest magnitude of change between was when 11 of the 64 units

changed between scenes.

Magnitude of Change

 

 

8

 

8

 

 

B

Magnitude of Change
8

 

 

Figure 13. Graph illustrating the magnitudes of change for all of the 18 transitions used in the
experiment. The lowest MOC is 1 I, while the highest MOC is 50. In all cases, the MOC is relatively high
because unemployment in Georgia was very dynamic.

47

3.10 Procedure

Each participant completed the experiment in the Windows computer lab in the
Geography Building at Michigan State University, and spent between 25 and 45 minutes
participating in the study. Between one and eight subjects participated at the same time.
Cardboard tri-folds were used to block the participants from disturbing each other.

During the study, subjects were asked to enter the testing room, were assigned
computers (8 of the total 16 computers in the lab were used for testing), and were thanked
for their participation. The computers screens were set 30 centimeters from the edge of
the desk; however, participants were allowed to sit at whatever distance from the screen
they felt was comfortable.

The participants were handed two consent forms following the Michigan State
University Institutional Review Board protocol. They were instructed to read, Sign, and
date one copy of the consent form (see Appendix A) and keep the second copy for his or
her records. Following the completion of the consent form, each participant was given the
“pretest” (see Appendix B). The pretest consisted of questions about age, gender, school
year, and the participants’ familiarity with animated and interactive maps. The goal of the
pretest was to gather basic information about each of the participants’ background and
abilities.

Upon completion of the pretest, before starting the actual test, each participant
took part in a training session on the computer. In the training the participants practiced
the questions they would be asked and were given general information about the study
(see Appendix C). The goal of the training was to familiarize the participants with the

questions they would see during the test. The training did not inform the participants

48

about the different transition types they would see. Each participant had to navigate
through the training before starting the test. Forward and back buttons were used to move
through the training to allow participants to go back if necessary; however these buttons
were not available during the actual experimental test.

Immediately following the training, the experimental test began. Participants had
to answer all 108 questions to complete the test. Upon completion, participants were
handed the posttest (see Appendix D). The posttest consisted mostly of open-ended
opinion questions about their feelings towards the test, the transitions, and the training.
The goals of the posttest were to 1) determine if the participants felt the test was
effective, 2) assess whether participants had different opinions about the different types
of transitions they saw, and 3) ascertain whether the training was adequate at
familiarizing the participants with the types of questions they would see. Once all of the
participants in the room had finished the test, the participants were allowed to collect
their $10 compensation and leave the testing room. The response data was immediately

recorded and stored in a spreadsheet for analysis.

49

Chapter 4
RESULTS AND ANALYSIS

4. 1 Change Detection Performance

Change detection performance was evaluated for all 108 questions of the study.
Accuracy was evaluated for:

l. The entire study (108 questions),

2. By change detection level (54 questions of each),

3. For each transition design (36 questions of each),

4. For each change characterization.

Participants were also evaluated on their abilities to detect changes as well as
persisting behaviors between map scenes. The ability to detect a change is classified as a
hit, while the ability to detect persistence is called a correct rejection. Accuracy in this
study is defined as hits plus correct rejections.

Signal detection theory (SDT) was also used to investigate the decision-making
strategies of the participants. SDT is widely used by psychologists, but in recent years has
been used to assess decision making for geographic information tasks (Griffin and Bell,
2009). Decision responses specifically for “yes/no” type questions can be categorized
into four types of potential responses; hits, correct rejections, false alarms, and misses
(Macmillian and Creelman, 1991). Both hits and correct rejections have been discussed
above and are each important for comprehending animated maps. False alarms in this
experiment occur when a participant indicates seeing a change when there was no
change. Misses occur when there is a change, but the participant indicates no change.

These four possible outcomes are shown in (Table 1).

50

 

 

 

 

 

 

 

 

 

Participant Decision
Reality Change No Change
Change Hit (H) Miss (M)
No Change False Alarm (FA) Correct Rejection (CR)
Table l. The SDT outcome array. Categorization of map readers potential decisions based on what they

were shown.

In this study, SDT was used to measure discriminability of a change and
participant bias to indicate either “yes” or “no”. The measure of discriminability is a' ’
which is defined as:

a"= 2(H) — 2(FA)
As d ' increases the discriminability ofa change increases.

There are several ways to measure participant bias, however in this study the
formula used was:

criterion (c) = ’/.~ (2 (H) + 2 (FA))
A positive 0 indicates a tendency to say “no” while a negative c indicates a tendency to
say “yes”. A zero c indicates no bias toward either “yes” or “no”.

4.1.] Overall Change Detection Performance

Table 2 summarizes the overall performance of all participants. The mean score
out of 108 was 57.26. The median was similar to the mean at 57.5 correct responses. The
minimum was 45 of 108 questions and corresponds to chance for the entire experiment.

The maximum was 72 of 108 question or two-thirds correct (Figure 14).

51

 

 

 

 

 

 

 

 

 

Total Performance Accuracy
Total Possible 108 (100%)
Mean 57.26 (53.02%)
Median 57.5 (53.24%)
Minimum 45 (41.67%)
25th Percentile 54 (50%)
75th Percentile 60 (55.56%)
Maximum 72 (66.67%)
Standard Deviation 5.17 (4.79%)

 

 

 

 

Table 2. Descriptive statistics of total change detection performance accuracy for all 108 questions
for all 78 participants (hits plus correct rejections).

Total Experiment Performance

1n
IW

 

1M
IW

 

 

 

 

 

 

 

80
g "Maximum: 72
i:
<
3rd Quartile: 60
60 l
g I [Modlanz 57.5
1st Ouartlle: 54
‘6 F ----------- “Minimum: 45 — - — _ 7:3,.ng
i 40 '~-"L"’i.1 V1.37
E
:i
2 Mean: 57.26
20
o I

 

 

Figure 14. Box and whisker plot of total performance accuracy on all 108 questions for all
participants (hits plus correct rejections). The chance score is equivalent to the minimum number of
correct answers (45) for all participants for the entire test.

52

Overall participants in the study commonly failed to detect changes between map
scenes. Chance for the entire experiment (all 108 questions) was evaluated as the chance
for Change Detection Level 1 questions (Question Type #1) plus chance for Change
Detection Level 3 questions (Question Type #2). For Change Detection Level 1
questions, with two options for responses, chance was 50% (27 of 54 questions). For
Change Detection Level 3 questions, with three choices, chance would be expected to
result in an accuracy of 33.3% ( l 8 of 54 questions). The chance accuracy score for the
total 108 question test would then be 41.67% (45 of 108 questions). Participants averaged
57.26/108 (53.02%) questions correct. A one-sample t-test at a 0.05 significance level
was used to determine whether accuracy was significantly different from chance. The
mean accuracy score of the entire experiment was significantly different from the chance

score of 45, t(77) = 20.933, p<0.0001.

 

 

 

 

 

 

 

 

 

 

 

 

 

Total Performance
Hits Correct Rejections False Alarms
Total Possible 72 (100%) 36 ( 100%) 36 (100%)
Mean 32.69 (45.41%) 23.62 (65.60%) 12.38 (34.39%)
Median 33 (45.83%) 24.5 (68.06) 11.5 (31.94%)
Minimum 17(23.6l%) 606.67%) 4(11.11%)
Maximum 51 (70.83%) 32 (88.89%) 30 (83.33%)
Standard Deviation 7.02 (9.75%) 4.78 (13.27%) 4.78 (13.27%)
Mean d’ 0.3]
Mean criterion c 0.27

 

 

 

Table 3. Descriptive statistics of hits and correct rejections for the entire experiment for the entire
population (all 108 questions).

The participants in this study maintained higher correct rejection rates (65.6%)

than hit rates (45.41%) (Table 3). A paired t-test at a 0.05 significance level was used to

53

test if the hit and correct rejection rates were significantly different. The paired t-test
indicated that the two rates were significantly different at ((77) = -8.444, p<0.0001. The
d ' was 0.31, indicating that overall it was hard to distinguish between noise and the signal
(the change). The criterion c indicates there was a bias to say there was no change rather
than change.

4.1.2 Change Detection Level Performance

Participants were evaluated on their abilities to detect changes at the different
levels of change detection (Table 4, Figure 15). Level 1 Change Detection refers to
questions where participants were asked simply to notice the change, while Level 3
Change Detection refers to questions where participants were required to recall the origin
state of the highlighted unit. Change detection accuracy was determined as hits plus
correct rejections. A total of 54 questions were asked about each change detection level.
It was expected that participants would perform better when asked whether they had
simply noticed the change than when they were required to recall how the unit changed

over time.

54

 

 

 

 

 

 

 

 

 

Change Detection Change Detection
Level 1 Level 3
Total Possible 54 (100%) A 54 (100%)
Mean 34.49 (63.87%) 22.77 (42.17%)

Median 34.5 (63.89%) 23 (42.59%)
Minimum 22 (40.74%) 13 (24.07%)

25th Percentile 32 (59.26%) 21 (38.89)
75th Percentile 37 (68.52%) 25 (46.30%)
Maximum 45 (83.33%) 30 (55.56%)
Standard Deviation 4.03 (7.45%) 3.24 (6.01%)

 

 

 

 

 

Table 4. Population level change detection performance for both levels 1 and 3 of change detection
(hits plus correct rejections). Change Detection Level 1 corresponds to Question Type #1 where
participants were asked simply to indicate whether the highlighted unit changed or remained the same.
Change Detection Level 3 corresponds to Question Type #2 where participants were asked to indicate the
class of the origin state ofthe highlighted unit.

55

Change Detection Levels Performance

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

54
50
45 Maximum: 45
g 40
3 3rd Quartile: 37
g 35 *Median: 34.5
<
30 tst Quartilegjz Maxjmmo
“r ‘ if, 5,},- ' - ’ ‘i """"""""
25 "”‘m‘ t ““’ ' 3rd Quartile: 25
‘5 .. Median: 23
2 20 Minimum: 22 _ _1_st gee—”99:21 . -----------
E criunce (78)
a
z 15
uh
Minimum: 13
10
5
Mean: 34.49 Mean: 22.77
0 a .
Level 1 Level 3
Change Detectlon Level

Figure 15. Box and whisker plot illustrating change detection performance by Change Detection
Level. Participants performed better on Change Detection Level 1 questions, by simply noticing the
change. Performance was not as high on Change Detection Level 3, when participants needed to recall the

origin state ofthe highlighted unit.

Participants performed better at Change Detection Level 1 (mean: 34.39) than
Change Detection Level 3 (mean: 22.77). The accuracy scores for the two question types
were significantly different from each other, t(77)=20.029, p<0.0001.

When simply noticing the change, the mean accuracy score (34.49 of 54
questions) was within two standard deviations of chance (50%, 27 of 54 questions).
When asked to identify how the unit changed, the mean accuracy score (22.77 of 54
questions) was within 1.5 standard deviations of chance (33.3%, 18 of 54 questions).

Two one-sample t—tests were used to determine if accuracy scores were significantly

56

different than chance. For Change Detection Level 1 questions, t(77)=16.428, p<0.0001
indicates that the scores were significantly different than chance. The accuracy scores
were also significantly different than chance for Change Detection Level 3, t(77)=12.988,

p<0.0001.

 

Change Detection Level 1

 

 

 

 

 

 

 

 

 

 

 

 

 

Hits Correct Rejections False Alarms
Total Possible 36 (100%) 18 (100%) 18 (100%)
Mean 22.45 (62.36%) 11.09 (61.61%) 6.91 (38.39%)
Median 22 (61.1 1%) 11 (61.11%) 7 (38.89%)
Minimum 11 (30.56%) 3 (16.67%) 3 (16.67%)
Maximum 34 (94.44%) 15 (83.33%) 15 (83.33%)
Standard Deviation 4.71 (13.07%) 2.39 (13.29%) 2.39 (13.29%)
Mean d' 0.65
Mean criterion c -0.02

 

Table 5. Hits, correct rejections and false alarms for all Change Detection Level 1 questions

 

Change Detection Level 3

 

 

 

 

 

 

 

 

 

 

 

 

 

Hits Correct Rejections False Alarms
Total Possible 36 (100%) 18 (100%) 18 (100%)
Mean 10.24 (28.45%) 12.53 (69.59%) 5.47 (30.41%)
Median 10 (27.78%) 13 (72.22%) 5 (27.78%)
Minimum 2 (5.56%) l (5.56%) 0 (0%)
Maximum 19 (52.78%) 18 (100%) 17 (94.44%)
Standard Deviation 3.34 (9.27%) 3.25 (18.06%) 3.25 (18.06%)
Mean (1' -0.03
Mean criterion c 0.58

 

Table 6. Hits, correct rejections and false alarms for all Change Detection Level 1 questions

57

 

 

Hits and correct rejection rates differed for the two question types (Table 5, Table
6). Hi rates were highest for Change Detection Level 1 questions. Correct rejections rates
were higher for Change Detection Level 3 questions than Change Detection Level 1.
Thus false alarm rates were higher for Change Detection Level 1 questions than for
Change Detection Level 3 questions. A paired t-test was used to distinguish differences
between hits for both question types at a 0.05 significance level. The hit rate differences
were significant, t(77)=25.910, p<0.0001. The difference between correct rejection rates
for the two detection levels were also significant, t(77)=-4.060, p<0.0001.

The 61’ indicates it was easier to distinguish a change for Change Detection Level
1 than Change Detection Level 2 with the higher a' ' for Change Detection Level 1. There
also was a stronger bias for Change Detection Level 3, and participants tended to say
there was no change rather than change. The criterion c value, which is very close to 0 for
Change Detection Level 1, indicates very little bias for that level.

4.1.3 Change Detection Performance across Design Conditions

To test the influence of different transition types on animated choropleth map
reading tasks, participants’ were evaluated on their abilities to detect changes with three
different transition types: Abrupt, Delayed Smooth, and Continuous Smooth. In the
Abrupt Condition, one scene abruptly transitioned to a second map scene. In the Delayed
Smooth Condition, the map remained static for one second and then smoothly
transitioned to the second map scene for another second. In the Continuous Smooth
Condition, the map immediately began to transition from the first map scene to the

second over two seconds. It was hypothesized that participants would perform best in the

58

Delayed Smooth Condition. This condition allows the participants to view the static

scene and also watch the transition.

4.1.3.1 Change Detection Level I

 

Change Detection Level 1

 

 

 

 

 

Abrupt Delayed Smooth Cgrgi:otlt:us
Total Possible 18 (100%) I8 (100%) 18 (100%)
Mean 11.2] (62.25%) 11. 68 (64.89%) 11.60 (64.46%)
Median 11 (61.11%) 12 (66.67%) 11 (61.11%)
Minimum 6 (33.33%) 7 (38.89%) 6 (33.3%)

 

25th Percentile

10 (55.56%)

11 (61.11%)

10 (55.56%)

 

 

 

 

 

 

 

75th Percentile 12 (66.67%) 12 (72.22%) 13 (72.22%)
Maximum 16 (88.89%) 16 (88.89%) 17 (94.44%)
Standard 0
Deviation 1.93 (10.72%) 1.96 (10.91 /o) 2.24 (12.42%)

 

Table 7. Participant accuracy for all three conditions for Change Detection Level 1. To answer these
questions correctly, the participant only needed to notice whether a change occurred in the highlighted unit.

59

 

Change Detection across Design Conditions (CD Level 1)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

18
Maximum: 17
w-
16 Maximum: 16 Maximlim: 16
14 T
g I 3rd Quartile: 13 3rd Quartile: 13
E 12 3rd Quartile: 12 Median: 12 __fi
‘ Med'a“: 1‘ 1st Quartile: 11 ”em“: 11
g 10 1st Quartile: 10 1st Quartile: 10
8 “’"1 """""""""""""""""" ”"Enér'i'céigi‘i
oo- 8
S Minimlim: 7
§ 6 Minimum: 6 Minimum: 6
z
4
2
Mean: 11.21 Mean: 11.68 Mean: 11.60
0 T Y I
Abrupt Delayed Smooth Continuous Smooth
Conditlon Type

Figure 16. Box and whisker plot illustrating change detection performance for all three conditions
for Change Detection Level 1. Participant performance was similar for all three conditions for the lowest

level of Change Detection.

60

 

Abrupt Delayed Continuous
Smooth Smooth

Condltlona

Figure 17. Mean performance and standard error bars for Change Detection Level 1. The accuracy
results indicate there were no large differences between the three conditions.

Based on the mean scores for Change Detection Level 1 (Abrupt: 11.21; Delayed
Smooth: 11.68; Continuous Smooth: 11.60), participants performed best on the Delayed
Smooth Condition, as expected (Figure 17, Table 7). However, performance in all of the
conditions was similar and median scores only varied by 1 question (Figure 16). The
Continuous Smooth Condition had the highest maximum score of all three conditions (17
of 18 questions answered correctly). A one-way repeated measures ANOVA indicated
that differences in accuracy for all three conditions were not significantly different from

each other. F (2, 154)=1 .409, p=0.248

61

Mean and median scores were higher than chance (9 of 18 questions), but
minimum scores were all below chance. Three one-sample t—tests indicated that all of the
accuracy scores for each of the conditions were significantly different from chance
(Abrupt: t(77)=10.093, p<0.0001; Delayed Smooth: t(77)=12.050, p<0.0001; Continuous

Smooth: t(77)=10.284, p<0.0001).

 

 

 

 

 

 

 

 

 

 

 

 

 

Abrupt Condition:
Change Detection Level 1
Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 7.53 (62.71%) 2.73 (45.51%) 3.27 (54.5%)
Median 8 (66.67%) 3 (50%) 3 (50%)
Minimum 3 (25%) 0 (0%) l (16.67%)
Maximum 12 (100%) 5 (83.33%) 6 (100%)
3:223: 2.02 (16.87%) 1.05 (17.55%) 1.05 (17.5%)
Mean d' 0.27
Mean c -0.23

 

 

 

 

Table 8. Hits, correct rejections and false alarms for the Abrupt Condition for Change Detection
Level 1.

 

Delayed Smooth Condition:
Change Detection Level 1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 7.77 (64.74%) 3.91 (65.17%) 2.09 (34.83%)
Median 8 (66.67%) 4 (66.67%) 2 (33.3%)
Minimum 3 (25%) 0 (0%) 0 (0%)
Maximum 12 (100%) 6 (100%) 6 (100%)
3:12:33: 2.20 (18.36%) 1.21 (20.13%) 1.21 (20.17%)
Mean 11' 0.86
Mean c -0.01

 

Table 9. Hits, correct rejections and false alarms for the Delayed Smooth Condition for Change
Detection Level 1.

62

 

 

 

 

 

 

 

 

 

 

 

 

 

Continuous Smooth Condition:
CHEESE Detection Level 1
Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 7.15 (59.62%) 4.45 (74.15%) 1.55 (25.83%)
Median 7 (58.33%) 5 (83.33%) 1 (16.67%)
Minimum 1 (8.33%) 2 (33.33%) 0 (0%)
Maximum 12 (100%) 6 (100%) 4 (66.67%)
3:13;: 2.40 (19.97%) 1.17 (19.49%) 1.17 (19.5%)
Mean (1' 0.97
Mean c 0.19

 

 

 

 

Table 10. Hits, correct rejections and false alarms for the Continuous Smooth Condition for Change
Detection Level 1.

Hit rates were similar across conditions, while correct rejection rates varied
(Table 8, Table 9, Table 10). The highest hit rate was in the Delayed Smooth Condition,
while the highest correct rejection rate was for the Continuous Smooth Condition.
Dividing accuracy into hits and correct rejection can help to understand how the different
condition types affect change detection. A one-way repeated measures ANOVA was used
to distinguish differences between hits for all three conditions at a 0.05 significance level.
The hit rate differences were not significant, F (2, 154)=2.049 p=0.132. The differences
between correct rejection rates for the three conditions were significant, F (2,
154)=59.366 p<0.0001. To understand how the conditions were different, a comparison
of the main effect using a Bonferroni corrected paired comparison indicated there was a
significant difference between all of the conditions (Abrupt and Delayed Smooth,
p<0.0001; Abrupt and Continuous Smooth, p<0.0001; Delayed Smooth and Continuous
Smooth, p=0.003).

The d’ measure was highest for the Continuous Smooth condition because of the

large amount of false alarms in the Abrupt Condition. This indicates that it was easier to

63

discriminate a change for the Continuous Smooth Condition than for the Abrupt
Condition. The criterion c indicates that participants were the least biased in the Delayed
Smooth Condition. Thus SDT indicates that both the smoothly changing conditions were
better than the Abrupt Condition.

4.1.3.2 Change Detection Level 3

 

 

 

 

 

 

 

 

 

 

 

Change Detection Level 3

Abrupt Delayed Smooth Cglgigioutzus

Total Possible 18 (100%) 18 (100%) 18 (100%)
Mean 8.21 (45.66%) 7.68 (42.66%) 6.87 (38.18%)
Median 8 (44.44%) 8 (44.44%) 7 (38.89%)
Minimum 3 (16.67%) 4 (22.22%) 3 (16.67%)
25th Percentile 7 (38.89%) 7 (38.89%) 6 (33.33%)
75th Percentile 10 (55.56%) 8.75 (48.61%) 8 (44.44%)
Maximum 13 (72.22%) 13 (72.22%) 12 (66.67%)
3:12:23: 1.94 (10.81%) 1.78 (9.87%) 1.61 (8.92%)

 

 

 

 

 

Table 1]. Participant accuracy for all three conditions for Change Detection Level 3, the highest level
of change detection. In these questions, participants were asked to recall the origin state of the unit
following the transition.

64

Change Detection across Design Conditions (CD Level 3)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

18
16
14
Maximum: 13 Maxim_um: 13
: 1
E 12 Maximum 2
<
E 10 3rd Quartile: 10
a 3rd Quartile: 8.75
o .
'6 8 IMedlan: 8 1|Median: 8 3'” mam" 8
. 7 , (Median: 7
i 6 1st Quartile: 7 tst Quartile: 7
§ * """"""""""" {3151,5118 10 chance :81
2
4 _
Minimum: 4 .1
b I
Minimum: 3 Minimum: 3
2
Mean: 8.22 Mean: 7.68 Mean: 6.87
O Y Y I
Abrupt Delayed Smooth Continuous Smooth
Conditlon Type

Figure 18. Box and whisker plot illustrating change detection performance for all three conditions
for Change Detection Level 3. Participant performance varied depending on the transition condition for
this higher level of change detection.

65

 

Abrupt Delayed Continuous
Smooth Smooth

Conditlon.

Figure 19. Mean performance and standard error bars for Change Detection Level 3. The accuracy
results indicate that participants were more accurate in the Abrupt Condition than the other two conditions.

Mean performance scores for Change Detection Level 3 were highest in the
Abrupt Condition (mean: 8.22) (Figure 19, Table 11). However, performance in all of the
conditions was similar and median scores only varied by 1 question (Figure 18). There
was a significant difference between the three conditions, F (2, 154)=35.808, p<0.0001.
To understand fully how the conditions were different for Change Detection Level 3
questions, a comparison of the main effect using a Bonferroni corrected paired
comparison indicated no significance between the Abrupt and Delayed Smooth
conditions (p=0.187), but did indicate significant differences between the Delayed
Smooth and the Continuous Smooth conditions (p= 0.007) and the Abrupt and the

Continuous Smooth conditions (p<0.0001).

66

Mean and median scores were higher than chance (6 of 18 questions), but only
accuracy scores for the Abrupt Condition were more than one standard deviation above
chance. Minimum scores were all below chance, and the 1St quartile for the Continuous
Smooth Condition coincided with chance. Three one-sample t-tests, however, indicated
that the scores were significantly different from chance (Abrupt: t(77)=10.093, p<0.0001;

Delayed Smooth: t(77)=12.050, p<0.0001; Continuous Smooth: t(77)=10.284, p<0.0001).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Abrupt Condition:
Change Detection Level 3
Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 4.44 (36.97%) 3.78 (63.03%) 2.218 (36.97%)
Median 4 (33.33%) 4 (66.67%) 2 (33.33%)
Minimum 0 (0%) 0 (0%) 0 (0%)
Maximum 8 (66.67%) 6 (100%) 6 (100%)
3:11:33: 1.83 (15.23%) 1.58 (26.41%) 1.58 (26.41%)
Mean d’ -0.01
Mean c 0.37

 

 

 

Table 12. Hits, correct rejections and false alarms for the Abrupt Condition for Change

Level 3.

 

Delayed Smooth Condition:

 

Change Detection Level 3

 

 

 

 

 

 

 

 

 

 

 

 

 

Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 3.40 (28.31%) 4.28 (71.37%) 1.72 (28.63%)
Median 3 (25%) 4 (66.67%) 2 (33.33%)
Minimum 0 (0%) 0 (0%) 0 (0%)
Maximum 9 (75%) 6 (100%) 6 (100%)
3:11:33: 1.78 (14.79%) 1.25 (20.79%) 1.25 (20.79%)
Mean (1’ -0.03
Mean c 0.61

 

 

Detection

Table 13. Hits, correct rejections and false alarms for the Delayed Smooth Condition for Change
Detection Level 3.

67

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Continuous Smooth Condition:
Change Detection Level 3
Hits Correct Rejects False Alarms
Total Possible 12 (100%) 6 (100%) 6 (100%)
Mean 2.41 (20.09%) 4.46 (74.36%) 1.54 (25.64%)
Median 2 (16.67%) 5 (83.33%) 1 (16.67%)
Minimum 0 (0%) 0 (0%) 0 (0%)
Maximum 6 (50%) 6 (100%) 6 (100%)
3:11:32: 1.38 (1 1.51%) 1.44 (23.98%) 1.44 (23.98%)
Mean 11’ -0.29
Mean 0 0.79

 

 

 

 

Table 14. Hits, correct rejections and false alarms for the Continuous Smooth Condition for Change
Detection Level 3.

Participants produced the most hits in the Abrupt Condition and the most correct
rejections in the Continuous Smooth Condition (Table 12, Table 13, Table 14). The
hypothesized Delayed Smooth Condition was the second best in both cases. Two one-
way repeated measures ANOVA’s were used to detect differences between hits and
correct rejections across conditions. Hits were significantly different across conditions,
F (2,1 54)=34.177, p<0.0001. A comparison of the main effect with Bonferroni corrected
paired comparison indicated that all conditions were significantly different from each
other (p<0.0001 for all conditions). Differences were also significant for correct
rejections across conditions, F (2,154)=7.398, p<0.0001. A comparison of the main effect
using a Bonferroni corrected paired comparison indicated no difference between the two
gradual transition conditions (p=0.728). However, differences were significant between
the Abrupt and the Delayed Smooth (p=0.035) and the Abrupt and the Continuous
Smooth Conditions (p=0.004).

Signal detection theory is designed to understand differences between noise and

signals and is typically used for two-choice answers. For Change Detection Level 3, all

68

wrong answers were lumped into misses and false alarms, while all correct answers were
denoted as hits or correct rejections. Using these measures, d ’ was greatest for the Abrupt
Condition and smallest for the Continuous Smooth Condition. Although d’ was small for
all three conditions. There was also the least bias for the Abrupt Condition with a c of
0.37 and there was the greatest bias to indicate no change for the Continuous Smooth
condition with a c of 0.79.

4.1.4 Change Characterization Accuracy- Cross Classification Arrays

Overall Performance
destination state

 

0.72 0.47 0.46
n:936 11:1170 n=936
0.43 0.56

   

n:936 n=936

0.4 0.41

112936 11:682 r165?”
Figure 20. Cross-classification array of accuracy rates for all 108 questions in the experiment. These
rates indicate how well participants performed on questions for each of the nine possible transitions. The
results indicate that the highest rates of detection occurred when the unit remained the same, the diagonals
from top left to bottom right. For example, participants were 73% accurate on detecting the absence of a
change when the unit remained “high” between the two scenes.

origin state

      

The overall performance accuracy was highest for correct rejections, when
participants identified a persisting behavior between map scenes (the diagonals from top
left to bottom right on the cross-classification array) (Figure 20). Participants were least
accurate when the unit shifted from “high” to “low” with only a 40% accuracy rate. Hit
rates, where participants correctly identified there was a change, however, were similar

for all types of change, ranging from 40% to 56%.

69

ChengeDetection Levelt ChangeDetectionLeveB

11. 1 . n . destination state

 

0.68 0.71 0.67 0.76 0.27 0.24

11' $1»? 11: Ir ’1 11:468 n=782 11:468

      
 

M , 0.49 062 M 1 0.37 0.49
E 1 E n:468 n=468
0.12 0.09

 

n=468 n=214
Figure 21. Cross-classification arrays of accuracy rates for all questions for the two question types.
The rates for Change Detection Level 1 indicate how well participants performed on simple change
detection for each of the nine possible transition behaviors, while Change Detection Level 3 indicates how
well participants identified the origin state of the highlighted unit.

Accuracy rates for Change Detection Level 1 were highest when the mapped unit
persisted within the “high” class between two scenes. Participants were also highly
accurate (71%) when the unit transitioned from the “low” class to the “middle” class
between scenes. Overall, rates of change detection were higher for Change Detection
Level 1 than for Change Detection Level 3; however, chance was also higher for Change
Detection Level 1 questions than Level 3 questions (Figure 21). For Change Detection
Level 3, participants remained highly accurate at detecting persistence behavior,
however, when changes were more substantial (a larger class shifi) participants often
failed to detect the change correctly. The differences between hit rates of the two types of
change detection were different, especially when the unit changed fiom the “highest”
class. Hit rates for the change of “high” to “low” and “high” to “medium” for Change
Detection Level 1 were 69% and 56%, respectively, while for Change Detection Level 3
hit rates were only 12% and 9%, respectively. A paired t-test indicated that the two

matrices were significantly different from each other, ((8) = 2.780, p= 0.024.

70

Abrupt Condition Delayed Smooth Condition Continuous Smooth Condition

('11),t1jy)\f

L .M3 3- L f M - L --
L 0.69 0.76 0.51 L 0.58 0.69 0.72 L 0.77 0.68 0.78

'—1.'r'r 1" 3' .3' ' 3. 1121513 11:15611:156

5,4054 039 0.62 M058056 0.59 ;5.035 068 0.66
g: ' 1 . § n 168r.—156n: 156
‘non 0600.76 “073 0.58081 'l'0.62 0.490.78

1 1 ::1‘i‘5n_156n:156

Figure 22. Cross-classification arrays of accuracy rates for all questions regarding Change Detection
Type 1 for each of the conditions. The rates indicate how well participants performed on simple change
identification. 1n the Abrupt Condition, the map transitioned abruptly from one scene to the next. In the
Delayed Smooth Condition, the map remained static for one second followed by a smooth transition for one
second. The Continuous Smooth Transition maps began to transition immediately after the map was
presented.

Correct rejection rates, the rate of correctly identifying persisting behaviors, were
typically higher than hit rates, the rate of correctly identifying there was a change, for
Change Detection Level 1 across conditions (Figure 22). The highest correct rejection
rate was in the Delayed Smooth Condition when the highlighted unit remained in the
“highest” class (81%). The lowest correct rejection rate was in the Abrupt Condition,
when the unit remained in the “middle” class (39%). The highest hit rate was in the
Continuous Smooth condition when the highlighted unit changed from “low” to “high”
(78%). The lowest hit rate was when the unit transitioned from the “middle” class to the
“lowest” class (35%).

There were some notable differences across conditions depending on the nature of
the change. First, the correct rejection rate was 39% for the Abrupt Condition when the
highlighted unit remained in the “middle” class. While for the two smoothly changing
conditions, correct rejection rates were 56% and 68% for the Delayed Smooth and
Continuous Smooth Conditions, respectively. For the transition from the “middle” class
to the “lowest” class, hit rates were over 50% for the Abrupt (54%) and Delayed Smooth
(58%) conditions while participants seemed to have difficulty identifying this same

71

change in the Continuous Smooth condition and only had a hit rate of 35%. Finally, there
were also differences in hit rates when the highlighted unit transitioned from the “lowest”
to the “highest” class. Hit rates were over 70% for the two gradually changing conditions
(78% for Continuous Smooth and 72% for Delayed Smooth) while hit rates were only
51% for the Abrupt Condition. Despite these differences, the Friedman non-parametric
repeated measures ANOVA, indicated no significant differences between the means of

the matrices for the three conditions, p < 0.905.

Abrupt Conditlon Delayed Smooth Conditlon Continuous Smooth Condltlon
., E- L ‘ M- L
L 069 044 0.38 L 076 026 021 L 0.83 0.21 0.14

7'3'33'11" 1.3331313131'. 3: "1‘{;3'1:r"11ir".'
M047 060 0.45 M036 065 0.58 0.27 0.70 0.44
)171 .11 ) 1 1::‘1111:1 _ n:156n=156n=156
I019 012 060 .012 0090.74 .1. 0.03 0.040.70

n:156 [1:234 [1:155

Oflglll state

(nu-1n All .1.)

1

1:-1i’11’; 1‘.‘

v. . E1 1 1 n:156 n=78 n=156
Figure 23. Cross-classification arrays of accuracy rates for all questions regarding Change Detection
Type 3 for each of the conditions. The rates indicate how well participants performed on the highest level
of change identification, where participants were required to recall the origin state of the highlighted unit.
In the Abrupt Condition, the map transitioned abruptly from one scene to the next. In the Delayed Smooth
Condition, the map remained static for one second followed by a smooth transition for one second. The
Continuous Smooth Transition maps began to transition immediately after the map was presented.

There was a greater gradation between accuracy rates for each of the nine types of
change by condition type for Question Type #2 (Figure 23). Participants were
consistently more accurate at correct rejections than hits. Correct rejection rates for
Change Detection Level 3 were similar to those in Change Detection Level 1, despite the
participant having three choices to choose from instead of two. Hits, however, were lower
than for Change Detection Level 1. The highest correct rejection rate was 83% in the
Continuous Smooth condition when the unit remained in the highest class. The lowest
correct rejection rate was 60% in the Abrupt Condition, when the unit remained in the
“middle” and “highest” classes.

72

There are several notable differences between the three conditions made apparent
by the cross-classification array. The greatest gradation between the three conditions was
for the change from the “lowest” class to the “highest” class. The rate of accuracy for this
particular type of change was highest for the Abrupt Condition (39%) and lowest for the
Continuous Smooth Condition (14%). Differences were also evident for the transition
from the “lowest” class to the “middle” class and from the “middle” class to the “lowest”
class. In these two cases, with large differences between the three conditions, participants
performed best in the Abrupt Condition and worst in the Continuous Smooth Condition.
Despite these differences, the Friedman non-parametric repeated measures ANOVA,
indicated no significant differences between the means of the matrices for the three
conditions, p < 0.19. Finally, the transition from the “high” to the “middle” class for all
three conditions had an overall low rate (12% for the Abrupt Condition, 9% for the
Delayed Smooth Condition, and 4% for the Continuous Smooth Condition).

4.2 Confidence

Following each change detection question, participants were asked to rate their

confidence on the presiding question as either “confident”, “neutral”, or “unsure”, for a

total of 108 confidence questions.

73

4.2.1 Overall Confidence Ratings for the Entire Test

Participant Confidence for Entire Test .

 

3,958 questions

" ”Confident“
"Neutral"
" "Unsure"

Number of Questions

 

Correct Incorrect

Figure 24. Stacked bar graph of correct and incorrect answers for the entire test (108 questions for
78 participants) categorized by participants’ confidence ratings.

When participants correctly identified a change or persisting behavior, more than
half of the time they indicated that they were “confident” about their answer (Figure 24).
Of the total 4,466 correct answers, the “confident” choice was chosen 2,397 times
(53.67%). Participants chose “neutral” or “unsure” only 47.33% of the time they
answered correctly. For the 3,958 incorrect responses, participants indicated they were
“neutral” (40.98%) more than “confident” and “unsure”. However, despite answering the
questions incorrectly, participants indicated they were more “confident” (37.85%) than
“unsure” (21.17%). The percentage of correct answers did significantly differ by

confidence rating, x2(2, N = 8424) = 215.78, p <0.0001.

74

4.2.2 Confidence for Different Levels of Change Detection

Participant Confidence by
Change Detection Level

 

2.690 correct

2,436 incorrect

  
 

 

 

  

 

1,522inoorrect

‘5 150° "Confident“
g 1041. - “Neutral“

1000 ,. ‘
a 799 626 ( l' i ‘1 ="Unsure"
z (30%) 53‘ . <4

500 J— —— (38%) —— (35%) :-

0 4
Correct Incorrect Correct Incorrect
Level 1 Level 3

Figure 25. Column graph of correct and incorrect answers for each of levels of change detection
categorized by participants’ confidence ratings. Level 1 is the lowest level of change detection while
Level 3 is the highest level of change detection identified by Goldsberry and Battersby (2009). In the
experiment, for Change Detection Level I, the participants were asked simply if they noticed the change,
while for Change Detection Level 3, the participants were asked to identify the origin state of the
highlighted unit.

Participants’ confidence ratings varied between question types (Figure 25).
Participants were more accurate on Level 1 Change Detection, and were also more
confident when they answered these questions correctly (57.92% for 2,690 correct
answers). When asked to detect changes at the highest level of change detection,
participants were less confident and less accurate (47.24% for 1,776 correct answers)
than they were for lower levels of change detection. In all cases, except incorrect
responses for Change Detection Level 3, “confident” was the most popular answer. For
incorrect responses for Change Detection Level 3, participants selected “neutral” more
than any other possible confidence rating (42.73%). In all cases, “unsure” was the least

popular answer. Correct answers significantly differed by confidence rating for Change

75

Detection Level 1, x2(2, N = 4212) = 111.57, p <0.0001 as well as for Change Detection
Level 3, x2(2, N = 4212) = 58.33,p <0.0001.

4.3.3 Confidence Ratings for Change Detection Level I Across Conditions

Participant Confidence for
Change Detection Level 1

91 1 905

-— 74
1 000 8 correct correct

.. correct

   

 

“ "Confident"
”Neutral“
1 92

266 —— —- 290 .

243 191 198
.. . _ _ 29°/.)_ _ 32% _, -_*" "
2 200 (28%) (36%) (‘A H . _ VA ( I ~ ( . r). (m. Unsure
Correct Incorrect Correct Incorrect Correct Incorrect
Abrupt Delayed Smooth Continuous Smooth

  
  

 
  

 

v—

 

 

Figure 26. Column graph of correct and incorrect answers for each of the three conditions for
Change Detection Level I categorized by participants’ confidence ratings.

Between conditions, confidence ratings were similar for Change Detection Level
1 (Figure 26). Participants were more confident when answering questions in the Abrupt
Condition even when they answered incorrectly (47%), while confidence ratings for the
other two conditions were 38% (Delayed Smooth) and 39% (Continuous Smooth). In all
cases correct answers significantly differed by confidence rating (Abrupt: x2(2, N = 1404)
= 25.84,p <0.0001; Delayed Smooth: x2(2, N= 1404) = 48.34, p <0.0001; Continuous

Smooth: x2(2, N= 1404) = 42.15, p <0.0001).

76

4.2.4 Confidence Ratings for Change Detection Level 3 Across Conditions

Participant Confidence for
Change Detection Level 3

 

  
  
  
     

     

      

 

__ _ 805 —. 358 _
900 i 763 incorrect incorrect
800 r 641
700
600 ‘
'5 m 344 364: "Confident“
300 4- 224 _ (44%) _. 224 _ (43%) 173 443%}; "Neutral"
5 20° ‘" (35%) — — (37%) — ”" (33%) “ ‘ L ”"3””
100 ‘ - _ 7 .
Correct Incorrect Correct Incorrect Correct Incorrect
Abrupt Delayed Smooth Continuous Smooth

Figure 27. Column graph of correct and incorrect answers for each of the three conditions for
Change Detection Level 3 categorized by participants’ confidence ratings.

Confidence ratings were similar between the three conditions for Change
Detection Level 3 (Figure 27). The percentage of correct answers where participants
chose the “confident” choice ranged from 46.41% to 48.21%. Participants chose
“confident” between 34.21% and 36.27% of the time when they answered the question
incorrectly. The two widest ranges between conditional confidence ratings were when
participants answered questions correctly and selected either “neutral” or “unsure”.
Neutral ratings ranged between 34% and 37% and unsure ratings ranged between 16%
and 20%. In all cases correct answers significantly differed by confidence rating (Abrupt:
x2(2, N= 1404) = 28.42, p <0.0001; Delayed Smooth: x2(2, N= 1404) = 20.61,p

<0.0001; Continuous Smooth: x2(2, N = 1404) = 49.79, p <0.0001)

77

Chapter 5
DISCUSSION

5.1 Change Detection Abilities

The results from this experiment suggest that map readers have difficulty
detecting changes between scenes in animated choropleth maps. Participants averaged
53% accuracy at detecting a change or a persisting behavior, meaning nearly half of the
responses were incorrect. Finally, the results of SDT indicate that, in general, participants
had a bias to believe there was less change in the maps than in reality. This finding
supports the first hypothesis of this thesis, that rates of change blindness would be high
for basic map reading tasks.

Maps graphically depict geographic information, and cartographers, unlike other
animators, cannot fully control the look and behavior of dynamic maps. Consequently,
many changes may occur across the display at a single moment in time. This is
problematic because humans are unable to perceive more than one change at a time
(Rensink, 2002) and thus many changes are left unnoticed in an animated map.
Furthermore, the primary utility of animated maps is their ability to portray changes over
time and space (Harrower, 2007), meaning the potential implications of cartographic
change blindness could be substantial. When map readers miss changes they fail to
comprehend the meaning encoded within the map. The inherent complexity of animated
maps requires readers to see more than just a gorilla in a basketball game (Chabris and
Simons, 1999). For example, in an animated choropleth map of unemployment changes
over time, failing to detect the change from the lowest class to the highest class indicates
a significant failure to understand the dynamics of unemployment over space and time.

Because of the meaningful data these maps represent, map readers must complete more

78

burdensome perceptual tasks in limited periods of time. Unlike movies or other stimuli
used in psychology change blindness literature, simply noticing the changes is only part

of the map reading task at hand.

5.2 Different Levels of Change Detection

The results of this study indicate that map readers are better at detecting changes
at Change Detection Level 1 than Change Detection Level 3. Participants were more
accurate at detecting lower levels of change when they simply needed to identify whether
the highlighted unit had changed. These results support the hypothesis that people would
perform better on Question Type #1, Change Detection Level 1 than Question Type #2,
Change Detection Level 3. Using the calculations for d ’ and c as noted in the Results
section, the SDT results indicated that it was easier to distinguish a change for Change
Detection Level 1 than for Level 3 and there was a greater bias to indicate no change for
Level 3, however there are limitations to the d ’ and 6 measures for Change Detection
Level 3 questions where there were three possible answers.

To comprehend the meaning of an animated choropleth map completely, the
reader must not only notice a change in a particular unit, but also must understand and
recall how that unit changed over time. Map readers are able to deduce that if a map
symbol has changed, it must have changed from a different symbol, however, the full
meaning of the transition is lost if the reader is not able to identify the origin state of that
symbol. While it is important to detect changes at Level 3 to apprehend an animated
choropleth map fully, this research indicates that this task is very difficult in the way it

has been currently presented to our readers.

79

The disparity between the accuracy rates of the two levels of change detection
may be linked to the two different questions types. Participants had a 50% chance on
Change Detection Level 1 questions compared to a 33.3% chance on Change Detection
Level 3 questions. However, it is clear that Level 3 Change Detection is more difficult
because it requires the participant to notice, attend to, and perceive more information
about map transitions.

While complete comprehension of animated choropleth maps requires the map
reader to achieve Change Detection Level 3, this is not to demean the necessity of
detecting changes at lower levels. Lower levels of change detection are essential to
understanding these types of maps because detection of a simple change can lead to
further exploration that, upon reviewing, leads to Change Detection Levels 2 and 3.
Finally, this research did not specifically explore Change Detection Level 2, the detection
of whether change in a particular area increased or decreased, however, Figure 28 may
help visually to understand Level 2 Change Detection by exploring participants responses

for Change Detection Level 3 questions in more detail.

80

origin state

destination state

 

high

 

 

 

 

low medium high
0.76
5
0.19

L M
E
3 o 57
‘8 .
E 0.37 0.41

L M M H

 

 

 

H L M

0.09

H

 

 

 

 

Correct Answer

Figure 28. Illustration of participants responses to Change Detection Level 3 questions. Ofien the
correct answer is not always the most popular choice. This graphic may help to understand Change

Detection Level 2, the detection of an increase, decrease, or persistence between origin and destination
states.

The Signal Detection analysis in this thesis does not take into account the

apparent bias to believe that if there was a change, it was from the adjacent class and not

the more distant class. For example, when the unit transitioned from “high” to “low”,

there was a large bias to select “medium” instead of “low”. Once again, when the unit

transitioned from “low” to “high” more participants selected “medium” as the possible

answer than either of the other two possible answers.

81

5.3 Transitional Design

The design of the transitions between scenes had different effects on change
detection accuracy. For lower levels of change detection, participants performed best
with the smoothly changing conditions based on the results of SDT. While accuracy was
similar across design conditions, d’ and c indicated that it was easier for participants to
distinguish a change and were less biased to assume there was a change in these two
conditions. Basically, while the Abrupt Condition had the highest accuracy in terms of
hits, false alarms were also very high and indicate that participants had a bias to select the
“yes” option. For higher levels of change detection, the Abrupt Condition appeared to be
superior based on simple accuracy analysis, however, more analysis is needed to
understand SDT for this type of question with three potential answers.

In the posttest (see Appendix D) participants were asked to indicate whether they
felt that some of the condition types helped them to detect the changes between the
scenes more easily. In general the participants in this study were divided over which
condition they felt helped them the most. Some participants indicated that the Abrupt
Condition was the best because the time devoted to the static scene before the transition
was the longest. Those who preferred the Abrupt Condition also indicated that because
the change was so sudden it drew their attention to places where the map had changed.
However, participants who felt they performed better with the smoothly changing
transitions typically believed that the time devoted to the transition allowed them to
apprehend the transitional behavior of the map more effectively.

It was expected that the Delayed Smooth Condition would elicit the highest

change detection accuracy because it allowed for an equal duration of time devoted to the

82

transition and the static scene before the change. Participants in the study were not told
explicitly about the different condition types they would see although they did see
questions during the training of each of the different transition designs. Thus, in general,
responses on the posttest did not directly refer to the Delayed Smooth Condition, perhaps
because they did not notice the difference between this condition and the Continuous
Smooth Condition. However, one participant did refer directly to this condition and felt
that it would result in their highest score by condition type.

One reason participants may have produced a large number of false alarms in the
Abrupt condition may be related to the salience of overall change in the Abrupt
Condition. Because all of the change occurs in a fraction of a second, it may affect the
participants’ bias to indicate “yes”; the abruptness of the change seems to make it appear
to the reader that there is change even when there was no change in the highlighted unit.
The abrupt condition seems to induce Harrower’s idea of animated simultaneous contrast.

One problem with the smoothly changing transition designs may be related to the
Rensink’s (2002) two types of change that a person may perceive: dynamic and
completed change. Dynamic change is a change that the viewer sees during a transition
event. For example, in this experiment, in the smoothly changing conditions, it was
possible for the participant to catch the change while it was occurring, called dynamic
change detection. However, the Abrupt Condition represented an example of completed
change detection; it is possible for map readers to become confused when watching the
smoothly changing transition because they may only notice the change during a dynamic
change and misread the transition. For example, when a unit transitioned from the “low”

class to the “high” class, it was possible that a map reader may have noticed the change

83

during the midpoint of the transition (while the unit was in the “middle” class) and
thought that the unit changed from the “middle” class to the “high” class instead from the
“low” class. In effect, this explicit representation of change may lead to confusion of map
readers especially when they are not watching the entire transition event.

One issue with tweened maps is that tweened transitions produce colors on the
map that differ from those in the legend (Battersby and Goldsberry, 2010). This
experiment was designed with this in mind thus the shade of gray during the tween from

the “low” class to the “high” class matched the “middle” class shade of gray. Battersby

and Goldsberry (2010) suggest using color schemes that are appropriate for tweening, ,1
such as the one used in this experiment. Map readers may misread maps if the transition

does not progress through the colors of the middle classes. For instance, sequential color

schemes using a combination of both hue and lightness do not transition through the

middle classes of the scheme (Figure 29). For example, when a transition occurs between

the highest and lowest classes using a yellow, green, blue color scheme, the color that

occurs during the transition does not “travel” through the middle class color (green).

Basically, a transition between the yellow color and the dark blue color in this particular

color scheme is not the green color that is shown on the legend as the middle class, but as

the graphical interpolation of color between yellow and blue.

84

ColorBrewer Yellow Green Blue Sequential Color Scheme

Same
Tweened Sequential Color Scheme

Figure 29. Illustration of the differences between the ColorBrewer color schemes and the in-between
states of a tweened color scheme with the same origin and destination states.

 

   

 

 

 

 

 

 

 

A
Same
7

 

 

   

 

 

 

 

 

 

 

Because additional colors are presented on the map, map readers may be able to
identify the differences and may be more likely to detect changes at higher levels. It may
be easier for map readers to identify how the unit is changing during dynamic change
detection if tweened colors do not visually “pass through” the colors of the middle
classes. More colors are shown on the map then on the legend during a transition and thus
each unique color during a transition could correspond to a unique transition event.
Finally, non-tweened or abruptly changing maps do not suffer from this same effect
because no interpolated colors are shown on the map. Several participants indicated that
it was difficult to identify whether the unit had changed from the neighboring class or not
because in a tweened map a unit that changed from black to white or vice versa must
visually progress through gray.

Tweened maps also suffer from a lack of congruence between the legend and the
transition (Figure 30). In this study, the maps had three classes: “high”, “medium”, and
“low”, however during a transition event in the smoothly changing conditions, the
intermediary colors on the map did not correspond to any of the three colors shown on
the legend and is technically meaningless. For example, if a unit transitioned between

middle gray and white, the light gray color that occurs during the transition was not

85

shown on the legend. This could lead to confusion for the map reader when viewing these

tweened maps. A gradient legend may be less confusing for the map reader.

Low Medium High

 

Gradual Legend Current Legend

Low |:] Low

 

High

Figure 30. Illustration of how the current legend design differs from what the map reader sees during
a transition state. Using a gradual legend may make this less confusing to the map reader.

Finally, based on the results of SDT, it appears that the smoothly changing
conditions were superior for Change Detection Level 1. Specifically, the continuous
tween outperformed the other two conditions because participants had an easier time
distinguishing a choroplethic shifting behavior from persisting behavior and participants
were less biased to select a particular answer. The Continuous Smooth Condition was
also the most congruent to the passage of time. This supports previous research by
Fabrikant et al. (2009) that says that more congruent representations would be more

easily understood and comprehended.

5.4 Transitional Salience
The cross-classification change arrays enable the exploration of the different types
of shifting that elicits the highest change detection rates. The results suggest that correct

86

rejections (detection of a persistence behavior) were easier to detect. Hit rates (detection
of a class shifi) varied depending on the question type and the condition. However,
throughout entire experiment, hit rates were similar across all six shifting behaviors. This
finding refutes the hypothesis that more drastic shifting behaviors would be more easily
apprehended.

There was no clear pattern as to which shifting behaviors were the most
perceptually salient. Previous literature suggests that persistent and more salient changes
are more easily noticed than less salient changes (Williams and Simons, 2000), however,
it is currently unknown what changes are most salient in these types of maps. The
previous research tested how easily people noticed changes when only one change
happened after a transition and it was believed that more drastic changes were more
perceptually salient. Specifically in the Williams and Simons study (2000), the
researchers tested participants" abilities to detect whether a change had occurred on a
small fictional character developed for the study. The participants were instructed to try
to detect a change to this character in several ways. For example, arms and legs were
removed or added to the character and participants were asked whether they were able to
detect the change of the character. In this thesis study, changes were less obvious than
those presented in the Williams and Simons experiment and this may be why there is no
clear pattern as to which types of change were the most easily detected.

Figure 28 illustrates the responses given to Question Type #2. In cases where
enumeration units changed drastically, from the “high” class to the “low” class or vice
versa, participants incorrectly believed the unit was changing from the neighboring class

(the “middle” class). This is related to dynamic vs. completed change detection,

87

discussed in the previous section. The results may indicate that map readers in this study
assumed that changes would not be drastic between scenes. Map readers may have a bias
to think the unit persisted or only changed from the neighboring class. This prejudice
may stem from a prior notion that more drastic change may seem unlikely. If map readers
already assume that drastic changes are unlikely, what does this mean for this type of
map reading? In other words, if people are assuming a particular change will happen,
what is the point of this type of visualization? More research is needed to attempt to

break this bias habit especially in highly dynamic geographic data.

5.5 Tendency toward Change Blindness Blindness

Finally, this research attempted to understand to what extent animated map
readers experience change blindness blindness. Participants in this study were generally
over confident, but not very accurate. One participant even noted, “it was very easy”
(although many participants indicated it was very difficult and appreciated not being
graded on the task). More importantly, participants rarely choose the “unsure” choice
when rating their confidence, despite the overall failure to detect changes in many of the
maps. This result supports the hypothesis that people would be highly confident but not
highly accurate when detecting changes in animated choropleth maps.

This result is consistent with other change blindness blindness research. Levin and
his colleagues (2000 and 2002) found that viewers believed they would see a change,
even when 100% of viewers failed to notice the change. In Levin’s experiments,
participants were asked to indicate whether they believed they would see a change before

they saw it. In this study, participants were asked to rate their confidence in change

88

detection after each change detection question. However, even when asked after failing to
detect the change, participants were confident they saw it correctly.

The bias towards selecting “confident” may indicate that participants had a prior
notion to believe they saw a change. This bias may undermine the effectiveness of these
types of maps and from a cartographic standpoint is difficult to combat. In general,
participants in this study underestimated the dynamics of the geographic information
presented in the maps they saw, and were confident that they knew how the data changed
over time despite missing many of the changes. More research is needed to explore how
cartographers can indicate to their readers that while they may believe they understand
the underlying data presented in the map, in fact, they probably did not fully understand
the meaning of the geographic information presented.

One known method that helps map readers avoid change blindness is to
incorporate interactivity and allow users to replay the scenes of an animation. However
change blindness blindness may cause map readers to believe they have correctly
identified changes between scenes, despite missing them, and thus map readers may be
reluctant to use the available interactivity to its full potential. Basically, map readers who
are sure they know what happened are not going to use a replay button, because they
think they understand the underlying data better than they actually do.

The map task in this experiment was fairly difficult. It has been noted that this
particular task of noticing whether a specific unit has changed between two moments in
time, is probably more difficult then other types of animated choropleth map reading
tasks such as detecting the overall change trend over time, but this remains unknown.

Despite the difficulty of the task at hand, participants still felt “confident” in their change

89

detection abilities. However, difficult tasks, such as the one tested in this thesis, may help
map readers to realize that they are missing important changes. Many of the participants
noted that they were glad they were not graded and apologized for being particularly bad

at this type of task.

5.6 Implicit Representations of Change

One problem with animated maps as compared to their static counterparts is that
often the differences between map scenes are implicit and only implied. Therefore much
of the problem with these types of displays stems from the map readers’ failure to
apprehend the implied transitions buried within the animation. Basically, the map reader
is forced to “read between the lines” to apprehend the change in the map. In contrast,
static maps such as small multiples allow map readers to internally interact and scan the
maps in a different way (Fabrikant et al., 2008), and may make it easier to comprehend
the changes between the scenes.

Cartographers have many possible ways to increase performance on different
change detection levels in animated maps. Two choices to help improve detection are:
interactivity, and explicit representations of change. Interactivity is commonly cited as a
way to assist map readers with complex dynamic map tasks (Tversky ei al., 2002;
Harrower, 2007; Fabrikant et al., 2008). Maps that enable users to click or mouse over a
particular place to identify how the unit changed, or have buttons that allow the user to go
back and review the previous state of the map may help to improve performance on both
levels of change detection. Similarly, explicit representations of spatial-temporal change

such as change maps or cross-classification arrays may also assist map readers to detect

90

changes at higher levels. The addition of change maps and cross-classification arrays
within the map display during an animation may also help make changes more salient by
depicting them explicitly. Currently, if map readers do not achieve the first level of
change detection, there will be no reason for the map reader to go back to investigate the
higher levels of change detection. By explicitly representing change in more than one
way (animation, change maps, and cross-classification arrays) map readers may be more

likely to understand the changes presented.

5.7 Static Cartographic Conventions

As cartographers continue to use the dynamic domain for portraying
spatiotemporal data, the concepts and conventions that are “tried and true” in the static
domain may not be acceptable for use in the dynamic domain. One example of this is the
number of classes used in an animated choropleth map. In the static domain, it has been
convention to use five to seven classes to design maps appropriately for the map reader’s
perceptual system (Slocum et al., 2009). However, Harrower (2003) and Goldsberry and
Battersby (2009) note that when using this number of classes, 1) the reader only has a
limited amount of time to take note of the geographic pattern in the animated map before
a change (Harrower, 2003) and 2) the number of possible transition behaviors increases
exponentially with the number of classes in the animated choropleth map (Goldsberry
and Battersby, 2009).

Similarly, ColorBrewer (Brewer and Harrower, 2002) color schemes have been
proven to allow cartographers to create visually appealing maps, as well as account for

colorblindness, the color limitations of different visual media, black and white

91

conversion, and allow map readers to distinguish between the different colors easily.
However, these colors were also designed to allow for map labels and thus the color
schemes do not span the entire lightness ramp (from the darkest version of a hue to the
lightest version of a hue). Differences between the colors may not be as noticeable as
they need to be in the dynamic domain. For example, in a tweened map there must be
enough difference between the colors for the map reader to notice, attend to, and perceive
the change, also called a just noticeable difference (JND). More research is needed to

explore JND’s to create more effective tweened animated displays.

5.8 Attention and its Impact on Change Detection

Previous change blindness research in psychology suggests that attention is

needed for change detection (e.g. Rensink, 2002). In this thesis research, participants

were not given any cues as to where a change would occur. One possible solution would
be to alert map readers of an upcoming change. In fact, one participant indicated that she
would appreciate being told where to look in the map to help identify and decode the
changes. Similar to the highlight rectangle used in this thesis experiment, one could
imagine several of these rectangles appearing prior to the change to direct the map
reader’s attention to places in the map where change will occur.

Robinson (2009) has suggested seven methods to highlight specific display
elements within interactive displays (Figure 31). Although, these ideas were designed to
direct attention across displays with multiple views, they could also be used to represent

change in dynamic cartographic displays explicitly. Of the seven highlighting techniques

92

that can be used to direct attention in multi-view displays, six of these methods could

potentially be used to signify change in these animations explicitly:

1.

Color- can be used to symbolize change with the adjustment of the fill or the
stroke color,

Depth of Field- used in photography to adjust areas in and out of focus. Gives a
pop-out effect of the in-focus/highlighted unit,

Transparency- adjustment of the alpha value for non-highlighted units. Gives the
effect of pop-out of the highlighted unit,

Color Desaturation- desaturation of non-highlighted units. This works in color
maps by “graying out” all non-highlighted units,

Contouring- applies contour lines to the outside of the highlighted unit. This
produces a similar effect as the color highlight but potentially decreases the
impact of simultaneous contrast,

Style Reduction- can be used when all units have a particular style applied to
them by removing that style from non-changing units when a change is about to
occur. Robinson uses the example with labeling where all the units of a
choropleth map are labeled and to highlight a particular unit, all of the labels are

removed except from the “highlighted” unit.

93

 

Robinson’s Highlighting Methods

 

2. Depth of Field

   

3. Transparency 4. Color Desaturation

   

5. Contouring 6. Style Reduction

Figure 31. Robinson’s highlighting methods. These methods could potentially be used to direct attention
to specific changes to help mitigate cartographic change blindness.

94

One additional method can be used to denote a change not included in Robinson’s (2009)
set of highlighting methods:

7. Point Symbolization- can also explicitly denote whether a unit will change or
remain the same. This technique works by placing a point symbol on each of the
changing units before or during a change.

By explicitly symbolizing change during transitions, cartographers may be able to direct

map readers’ attention to the changes and increase change detection.

5.9 Data Driven Change Detection Effects
The results of this research indicate that several data driven elements that affect
change detection in animated choropleth maps:
1. Saliency of Change- based on the cross-classification arrays presented in this
thesis, the nature of a given change affects detection in animated choropleth maps.
It was expected that more drastic changes would elicit higher change detection
rates because of the indications from previous research (Williams and Simons,
2000). However, it seems that map readers often are confused as to whether the
origin unit transitioned from the adjacent class or from a more distant class.
Participants also tended to look at the dark areas more often than the light areas as
indicated in the posttests and tended to notice an onset or offset of the darkest
color in the scheme during a transition event. While cartographers cannot fully
control this data driven effect, they should beware of the implications of using
lightness based color schemes and realize that darker colors appear more salient to

map readers against a white background.

95

 

2. Spatial Location- some participants in the study noted that during the experiment
they spent their time focusing on the center of the map or at least a specific region
within the map. Cartographers must realize that only one change can be detected
at a time (Rensink, 2002), and only four to five places can be attended to at any
given time making change detection outside of the attended area difficult. If
changes occur outside of this area of fixation, cartographers must determine a way
to direct attention to those changes.

3. Relative Size of Changing Area- the size of the unit compared to its surrounding
units could potentially affect change detection in animated choropleth maps. For
example, it may be easier to notice a change in a large unit surrounded by several
small units, than to see change in a small unit surrounded by several larger units,
or when all the units are the same size. Attention must be directed to units that are
smaller than surrounding units or else changes may go unnoticed.

4. Overall Direction of Change- this is the effect of the overall change in the map
on change detection in a particular unit of the map. For instance, if most units in
the animated choropleth map increase, it may imply to the reader that in general
the data increased across the map. In this experiment, there were several years
where the magnitude of change was high and many of the units changed in the
same way. One would assume then that all of the units changed in the same
direction (all increased or all decreased), however, in some cases, this was not
always true. Map readers may have a bias to answer a question in a certain way

based on the overall direction of the change. Participants also seemed to find it

96

easier to detect changes when there was one prominent color change on the map
based on posttest responses.

. Change in Neighboring Areas- Harrower’s (2007) animated simultaneous
contrast indicates that changes in neighboring units of an animated choropleth
map affect the apparent change in a particular unit. This problem affects overall
change detection as well as the different levels of change detection noted by
Goldsberry and Battersby (2009). If apparent motion or changes in surrounding
units affect the perception of change in a neighboring area, it might be safe to
assume that this also affects the perception of how a particular place is changing.
. Magnitude of Change- in animated choropleth maps, many changes can occur
during a single scene transition. The number of changes that occur during one
transition event is defined as the magnitude of change. Despite the multiple
changes that occur during a transition event in an animated choropleth map, map
readers can only perceive one change at any given moment (Rensink, 2002). One
participant in this study noted that it was very difficult to identify the particular
change of the highlighted unit because so many changes happened at once. In
animated thematic maps cartographers can limit the magnitude of change by
choosing a limited number of classes. Harrower (2003) suggests using between
two and three classes because of the limited amount of time the map reader has to

encode the information.

97

5.10 Duration Effects

Along with testing change detection in animated choropleth maps, this study also
evaluated the components of animated maps the cartographer can control. Cartographers
have full control of the pace and speed of the animated through the use of duration. In
these types of maps there are two types of duration available to the cartographer: duration
of static scene and duration of the transition. DiBiase et al. (1992) defined duration as the
amount of time devoted to static scene display. However, as tweening and smooth
transitions become easier to implement in animated maps, it has become the
cartographer’sjob to take into account both duration of the static scene as well as
duration of the transition. The results of this thesis indicated that duration of the transition
may be important for change detection, limited duration of the static scene may not give
our map readers enough time to view the static scene.

1. Duration of Transition - the duration of the transition affects animated
choropleth map reading and specifically change detection. In this thesis, when the
duration of the transition was increased, SDT indicated that participants had an
easier time distinguishing a change. Smooth transitions may not always allow for
the greatest accuracy of detection; however, these congruent representations of
some spatiotemporal data. do not elicit the same bias to indicate there was a
change, thus false alarms were lowest for these types of transitions. It still remains
unknown how these smooth transitions affect accuracy of detecting a change at
higher levels.

2. Duration of Static Scene- the duration of transition also has an effect on change

detection in animated maps. The results indicate that transitional design affected

98

apprehension of change in animated choropleth maps. All of the conditions in this
experiment had an equal amount of time devoted to the entire animation from
initial onset to the appearance of the question; however, duration of the static
scene varied for each of the condition designs. Research in psychology has shown
that the time given to encode the information before the change affects the
amount of information that is encoded (Hollingworth et al., 2001). It is important
that map readers are given adequate time to view the static scene before the
change due to the effects of this type of duration on change detection abilities,
although the SDT analysis indicates that this may not affect map reading as much

as the psychological literature suggests.

5.11 Limitations

This study included some important limitations, however the results still provide
valuable incite into the change detection abilities of animated choropleth map readers.

5. I 1.1 The Splat Contingent of the Highlight Rectangle

The visual onset of the red rectangle at the end of the transition in this study may
cause change blindness. In the Abrupt Condition, a delay of 250 milliseconds was added
between the onset of the second map scene and the onset of the rectangle. However, this
delay may not be enough time to avoid the “splat contingent” of change blindness
(O’Regan, et al, 1999). Finally, no delay was added for either of the smooth transition

conditions.

99

5.11.2 Spatial Location of the Highlighted Unit

In this experiment, 36 of the 64 units in the map were highlighted three times after
a transition. The experiment was designed so these 36 units were not clustered in one
particular area of the map. Instead, the goal was to have the highlighted unit appear in
many different places across the map. However, the cross-classification arrays indicate
that some change characterizations had relatively high accuracy rates while some had
very low rates. Low rates may be related to the spatial location of the highlighted unit.
For example, the transition from the “highest” class to the “lowest” class resulted in a

12% accuracy rate for Question Type #2 (Change Detection Level 3). The highlighted

 

unit for these particular questions was located on the periphery of the map and none of
the questions about this type of change were located in the center of the map.

5. 11.3 Transitional Behaviors

During the experiment, only one question was asked about the transition from the
“high” class to the “middle” class for Change Detection Level 3 questions. The change
detection rate was only 9% for all participants for Change Detection Level 3 and may be
explained by the lower n for this type of change (“high” to “middle”) than for all of the
other types of change. The particular unit highlighted for this type of change was also on
the periphery of the map and thus could be easily missed. It would be ideal in future
experiments, to have more questions about each of the nine possible transition behaviors
to avoid problems such as this in similar experiments.

5.11.4 Static Scene Duration

This experiment was controlled for the total duration of the animation from first

onset to the appearance of the highlight rectangle and question (Figure 32). However, the

100

time given to view the static scene before the transition varied by condition. The Abrupt
Condition had the longest static scene duration (two seconds), followed by the Delayed
Smooth Condition (one second), and finally the Continuous Smooth Condition (nearly
zero). According to previous research in psychology, the amount of time given to encode
information is directly related to the amount of information the map reader is able to
retain after the change took place (Hollingworth et al., 2001). It would be useful to design

a future study where duration of the static scene was equivalent across conditions (Figure

 

 

 

 

 

33).
ORIGINAL CONDITIONS
ABRUPT CONDITION
VJ
06"
Two (2) Seconds , (58°C
we
(90
’L

DELAYED SMOOTH CONDITION

 

 

 

 

 

WW

One (I) Second One (1) Second

CONTINUOUS SMOOTH CONDITION

 

 

 

 

Two (2) Seconds

Figure 32. Original conditions. Total duration for all conditions was equal, but duration ofthe static scene
and the transition were unequal.

101

PROPOSED CONDITIONS
DELAY WITH SHORT TWEEN

 

 

 

 

 

W4

Two (2) Seconds One (1) Second

DELAY WITH LONG TWEEN

 

 

Two (2) Seconds Two (2) Seconds

Figure 33. Proposed conditions. Total duration for all conditions was unequal, but duration ofthe static
scene was unequal.

5.11.5 Hits versus Correct Rejections

In this study, the results indicated that there was a difference in accuracy when
there was a change to detect (hits) versus when the participant was required to detect a
persisting behavior (correct rejections). In similar psychology studies, correct rejections
are often ignored and only hits are counted. However, in this type of change detection, it
is not only important to notice a change, but also to notice a persistence behavior between
map scenes. For example, in an animation of the Red-State, Blue-State map, between the
years 2000 and 2004, only three states changed from red to blue or vice versa. However,
while these three states changed, it is also important to note that 47 states did not change.
In these types of maps, persistence is meaningful. In this particular example, it means that
47 states did not have a significant change in political affiliation to change the electoral
votes for those states. This study suggests that persistence between map scenes (correct

rejection) was easier to distinguish than when there is a change (hit). However, in this

102

 

study there were more questions were the correct answer was a hit rather than a correct
rejection. This could have an effect on the results because a blindly guessing participant
would be more likely to get a higher correct rejection score than hit score. This also may
account for the high false alarm rate across the entire experiment.

5.11.6 Magnitude of Change and Map Complexity

This experiment tested change detection when viewing an animated choropleth
map with 64 enumeration units each exhibiting one of nine unique transition behaviors.
Several participants found it difficult to detect changes with the high number of units in
the map, as well as the large magnitude of change that occurred between many of the
map scenes. In this experiment, the lowest magnitude of change was 11 units, and the
highest magnitude of change was when 41 of the 64 units changed. These high
magnitudes of change are a result of the underlying data as well as the classification
scheme and the number of classes used. The data used to create these maps was yearly
average unemployment data between 1990 and 2008. Unemployment can change
dramatically in short periods of time, and in these particular years for this geographic
place, unemployment was very dynamic. The data was classed using a quantile
classification scheme where the same number of data points are placed into each the
classes. This scheme was appropriate to avoid empty classes and a skewed distribution,
but may have been a major factor influencing the high magnitude of change in these
animations. Finally, it would have been more appropriate to limit the overall size and
complexity of the map by limiting the number of units in the map to a number less than

64.

103

 

“‘7'

5. 11. 7 C ongruent Data Representations

Battersby and Goldsberry (2010) argue that classed maps should not be tweened
because it implies there is a smooth transition between the two classes, when in fact there
no gap between two classes of a classed choropleth map, thus interpolation is
unnecessary. This study used tweening for a classed map and may have caused confusion
for the study participants.

5.11.8 Animated Simultaneous Contrast

Harrower (2007) extended the idea of simultaneous contrast a well-known
problem in static maps to the animated domain. This idea of simultaneous contrast occurs
when a particular unit in a choropleth map perceptually shifts in color due to the different
colors of the surrounding units. For example, dark units tend to appear darker when
surrounded by light units, and light units appear lighter when surrounded by dark units.
Harrower suggests the idea of “animated simultaneous contrast” is when a particular unit
may look as though it is changing more or changing less because of the apparent change
in the units surrounding it. For example, one unit may persist throughout an animation,
but if many drastically changing units surround that unit, it may appear as though the
persisting unit is changing too. In Harrower’s study, participants evaluated change in a
small area of a choropleth map. His participants noted how the change in the surrounding
areas affected their abilities to answer the questions in his study. Animated simultaneous
contrast can lead to misinterpretation of changes in the map.

In the case of this experiment, the maps were created with real unemployment
data, thus there were instances where some units persisted throughout the 18 years used

in the animation. In other cases these units were surrounded by drastically changing units

104

that may have affected the change detection abilities of the participants in this study.
Basically, map readers may not understand how unemployment shifted or persisted
because the enumeration unit in question may be surrounded by units that change
differently from it.

5.11.9 Map Tasks

It should be noted that this study only tested one particular map task; participants
were required to notice change in one enumeration unit during a singular complex
transition. This map task may be more difficult than some other map tasks for animated
choropleth maps and could in part explain the low detection rates presented here. Other
map tasks include evaluating one unit over many transitions times, many units between
two times, or many units over many times. Detection rates may differ when readers are
asked to perform other map tasks. In general, however, as a form of geovisualization, the
goal of these maps should be to allow users to search for unknown patterns that would
not be apparent in any other way. Change blindness, however, may seriously limit to the
goal of these types of geovisualizations.

5.11.10 Signal Detection Theory for Three Choice Answers

Finally, signal detection theory as it was used in this thesis is designed for only
two- choice answers. In the yes/no questions, d ’ and c were conclusive that the
Continuous Smooth Condition allowed for the easiest discriminability of the change from
the noise, while the Abrupt Condition produced too much noise for the participants to
correctly select the answer. Instead in the Abrupt Condition, participants were biased to
select “yes” because of the amount of noise leading to the large amount of correct

rejections for this condition. In the three-choice questions, d’ and 0 did not give a clear

105

 

answer because SDT failed to take into account the differences in wrong answers.
Basically, as it was calculated, all wrong answers were lumped together, however, Figure
28 illustrates that it seems that participants had a bias to indicate that the enumeration
unit had transitioned from the neighboring class. Further SDT analysis is needed to fully

understand how participants were biased in these questions.

 

106

Chapter 7
CONCLUSION

Change blindness is a serious problem that limits the effectiveness of dynamic
geovisual displays. As cartographers we are unable to control the actions and transitions
in animated maps completely because of the inherent tie to underlying geographic data.
Thus many changes may occur during a single scene transition. However, the realistic
depiction of real world data changes may require more cognitive power of map readers 5.;
than humanly possible. As Harrower (2007) states, the limits of computing are no longer

a problem for the development of these types of displays, it is the limited cognitive

 

capacity of our map readers that potential undermines these new ways to represent data. I
Finally, it is the connection to underlying spatiotemporal data that requires map readers to
perceive more than just simple changes between scenes of a map animation. In other
words, map readers must fully comprehend the meaning of the change to understand the
change in the worldly phenomena it is depicting.

The results of this empirical study have revealed that people are surprisingly poor
at detecting changes in animated choropleth maps. This thesis has also shown that map
readers are not only poor at detecting basic changes, but also are unable to detect changes
at higher levels. In recent years, several researchers have suggested gradual transitions as
a potential strategy to mitigate change blindness (Goldsberry, 2004; Fabrikant et al.,
2008); however, more analysis is needed to understand how gradual transitions affect
change detection especially at higher levels.

More research is needed to understand how to design these maps to allow for
better apprehension of cartographic change. Previous research suggests that interactivity

may help to alleviate problems such as change blindness. However, these findings

107

suggest that interactivity may not solve these problems because map readers are not only
missing changes in these displays, they are confident they did not miss them. In effect,
interactivity may not be enough to make these types of displays understandable. Based on

the findings in this thesis research, these questions deserve further attention:

° Can change detection be improved by cueing map readers to a particular place
within a dynamic display? Previous research suggests that attention is necessary for
accurate change detection (e.g. Rensink, 2002). By cuing our readers where to look

we are directing their attention to where a change may or may not occur. This way

 

our readers are not fruitlessly searching for a potential change in a relatively large
visual area. Robinson’s (2009) highlighting methods may provide potential ways to

cue our readers’ attention to a specific place within these types of displays.

0 What is the effect of changing the duration of the static scene? In this study, each
condition controlled for duration of the entire animation from the onset of the first
scene of the animation to the onset of the question and the red highlight rectangle.
However, research suggests that the amount of time given to encode information
affects how much of the scene before the change can be retained after the change

(Hollingworth et al., 2001).

108

 

Could stabilizing the rate of change (Goldsberry, 2004) affect change detection
in animated choropleth maps, by adjusting the duration of the transition based
on the magnitude of change? Many participants felt they would more easily
apprehend changes in the display if less changes happened at once. In other words,
they felt the magnitude of change was too high. Often the cartographer cannot adjust
magnitude of change, however, by increasing the duration of the transition when there

is a higher magnitude of change may help to improve change detection in these types

 

of maps.

How well do map readers apprehend changes in other types of dynamic maps?
While animated choropleth maps present a fairly obscure mapping technique, this
change blindness phenomena could also extend to maps where missing a change has
more serious consequences. The use of static navigation maps has decreased
significantly as use of electronic in-vehicle navigation units have increased. These
devices require the user to be aware of display changes that occur to make key way
finding decisions. In-vehicle navigation units also cannot always be the focus of the
user’s attention. If a driver is confused because of change blindness, wrong turns
could be made, and potentially life-threatening consequences could ensue. Testing
change detection abilities for these types of maps and others may provide insight into

how changes in different types of maps are comprehended.

109

0 Can we create maps that express the difficulty of perceiving the changes for our
map readers? This study indicates that despite missing a large percentage of changes
in these types of maps, confidence in change detection was still high. This potentially
undermines the use of interactivity, allowing map readers to go back and replay the
animation. There may be better ways to design these maps so our map readers
understand how unlikely it is that they have correctly comprehended the full meaning

of these types of maps.

Perception of change in animated map displays is fundamental to map cognition
and understanding. As cartographers and map readers continue to use these types of maps
to portray spatiotemporal information, it is essential that map designers attempt to
alleviate some of the stress on the cognitive limits of map readers by creating more easily
understood and perceived animated maps. Cartographers must generate more effective

animations and make an effort to reduce change blindness in animated choropleth maps.

110

 

 

APPENDICES

111

Appendix A
CONSENT FORM

Research Participant Information and Consent Form
You are being asked to participate in a research project on animated map reading. This
study is being conducted by Carolyn Fish under the direction of Drs. Kirk Goldsberry and
Judy Olson. Participation in this study is completely voluntary. This document provides
information about the study so you can make an informed decision about whether you
would like to participate. You should feel free to ask the researchers any questions you
may have.

Evaluating Human Change Detection in Animated Maps

Carolyn Fish, Graduate Student
Kirk Goldsberry, Assistant Professor
Judy Olson, Professor Emeritus

Department of Geography
Michigan State University

116 Geography Building
Michigan State University
East Lansing, MI 48824

1. PURPOSE OF RESEARCH:

This study evaluates how well humans detect changes in animated maps. Often humans

' miss the changes between two scenes in a map animation. From this study, the
researchers hope to learn how different ways of animating may allow for you to better see
those changes. Your participation in this study will take about 30 minutes of your time.

2. WHAT YOU WILL DO:

You must be 18 years of age or older to participate. If you agree to participate in this
study, you will be asked to complete: a pretest, training session, test, and a posttest. In the
pretest you will be asked simple questions about your age and gender. In the training, you
will learn about map animations, the types of questions you will be asked in the test, and
how to answer these questions using the software. In the test you will answer
approximately 100 questions about the changes you saw in the simple map animations. In
the posttest you will indicate to the researcher what you thought of the test and how you
think it could be improved.

3. POTENTIAL BENEFITS:

You will not directly benefit from your participation in this study. However, your
participation in this study may contribute to the understanding of human change detection
in animated maps.

112

 

4. POTENTIAL RISKS:

The potential risk to you is minimal. At the most, participation in this study will result in
fatigue, due to the number of questions and the length of time it will take you to complete
this study.

5. PRIVACY AND CONFIDENTIALITY:

In this study, your name will not be connected with your answers. Only the consent form
will contain your name and it will not be connected with the answers you give in the test.
Your answers in the test may be presented at professional meetings or used in
publications, but your identity will remain anonymous.

6. COSTS AND COMPENSATION FOR BEING IN THE STUDY:

You will not incur any cost for participating in this study. You will be compensated $10
for completing your participation.

7. YOUR RIGHTS TO PARTICIPATE, SAY NO, OR WITHDRAW:

Participation in this research project is completely voluntary. You have the right to say
no. You may change your mind at any time and withdraw. Your answers will then be
destroyed and will not be included in the data for the study. If you chose to withdrawal
from the study, before completion, you will not receive the $10 compensation. You will
not, however, be penalized in any other way.

8. CONTACT INFORMATION FOR QUESTIONS AND CONCERNS:

If you have concerns or questions about this study, such as scientific issues, please
contact the researcher Carolyn Fish (Graduate Student, Department of Geography,
Michigan State University, 116 Geography Building, East Lansing, MI 48824,

cfishl 1@msu.edu, 517-353-9940) or Dr. Kirk Goldsberry (Assistant Professor,
Department of Geography, Michigan State University, 116 Geography Building, East
Lansing, MI 48824, kg@msu.edu, 517-353-0308) or Dr. Judy Olson (Professor Emeritus,
Department of Geography, Michigan State University, 116 Geography Building, East
Lansing, MI 48824, olsonj@msu.edu, 517-353-8757).

If you have questions or concerns about your role and rights as a research participant,
would like to obtain information or offer input about those rights, or would like to
register a complaint about this study, you may contact, anonymously if you wish, the
Michigan State University’s Human Research Protection Program at 517-355-2180, Fax
517-432-4503, or e—mail irb@msu.edu or regular mail at 202 Olds Hall, MSU, East
Lansing, MI 48824.

113

9. DOCUMENTATION OF INFORMED CONSENT

Your signature below means that you voluntarily agree to participate in this research
study.

 

Print Name

 

 

Signature Date

Please keep one copy of this form for your records.

114

Appendix B
PRETEST

Map Study- Pretest questions

1. Have you taken this test before? If yes, please raise your hand and inform the

researcher. If no, please continue on to the next question.

2. Please indicate your gender (circle 1): Male / Female

3. What year are you in school:
a. Freshman
b. Sophomore
c. Junior
d. Senior
e. Grad Student
f. I am not in school
g. Other
4. Do you wear glasses or contacts? (Circle one)

5. Are you colorblind?

a. Yes
b. No
c. Unsure

6. Do you play video games?

a. Often (more than 5 times a week)

: Yes/No

b. Sometimes (between 2 and 5 times a week)

c. Rarely (less than once per week)

d. Never

115

 

7. How often do you use the computer?
a. Often (more than 5 times a week)
b. Sometimes (between 2 and 5 times a week)
c. Rarely (less than once per week)
(I. Never
8. What is your age: __

9. What is your major?

 

An animated map is a map that shows change overtime. The most popular type of
animated maps are TV weather maps like you might see on your local news channel.
These maps show how the weather is changing in a given place over a period of
time. These maps can be interactive, such as those you might see on a weather
website, which allow you to stop, slow or adjust the animation. However, not all
animated maps are interactive. Also interactive maps are not necessarily animated.
For example, online navigation maps (e.g. GoogleMaps or MapQuest) are often
interactive and allow you to adjust them, but do not move on their own. (The maps

you will see in this test will be animated but not interactive).

10. Have you ever seen an animated map?
a. I have never seen an animated map
b. I am familiar with animated maps

c. l have made animated maps

116

11. If yes, where have you seen animated maps? (circle all that apply)
a. TV weather maps
b. news media
c. online (world wide web)
(I. as a part of course material

e. other, explain:

 

12. Have you ever used an interactive map?

a. Yes
b. No
c. Unsure

13. If yes, where did you use one? (circle all that apply)
a. online (world wide web)
b. as a part of course material

c. other, explain:

117

 

Appendix C
THE TRAINING

 

Thank you for your participation!

 

I Confinue I

 

This is a test designed to understand
how people read and see changes
in animated maps.

 

 

r 60 Back I I Continue I

118

 

You will not recognize the geographic
place used for this test.

Do not worry about trying to figure
this out, it is unimportant for this study.

 

 

I Go Back I I Continue I

 

 

 

This training session will introduce you
to the maps you will see and the types
of questions you will encounter during
the test.

 

 

[ Go Back I I Continue ]

 

 

119

 

You will see three different question types
throughout the test.

 

 

r 60 Back J I Continue I

 

For each question, you will see a map and will
be asked a question about the change you
saw while viewing the animation.

The map you will see will look similar to the
county map shown below, however, it will
have more counties.

5

 

 

 

I7 Go Back I I Continue 1

120

 

 

While you watch the county map, some of
the counties will transition in color.

For example, a county that starts out as the lightest
shade of grey may transition to the darkest shade of grey.
‘RANSITION
Notice: the two
highlighted counties
changed during the
transition.

      

I Go Back I I Continue J

 

 

 

 

 

 

 

After each transition, you will be asked a
question about the changes that you see
after the transition.

Make sure to try your best in answering
all of the questions.

 

 

F Go Back I I Continue I

121

 

Question Type #1

   

‘ After the transition, one unit
will be highlighted in red.

Then you must decide
whether the unit changed
color by selecting "Yes" or "No".

L Go Back I I Continue 1

  

 

 

 

Now let's try a practice question.

 

 

r Go Back I I Continue 1

122

 

Practice Question #1

 

 

Practice Question #I

   

Did the highlighted unit
change between scenes?

\ m

L

123

 

Great that's correct!
Now let's look at the
second type of question
you will see on the test.

 

 

 

L Go Back I I Continue I

 

 

 

Question Type #2

 

Please indicate the color
of the highlighted unit
before the scene change:

"- I

This question type is the similar to question
type 1, except that the question is a little
more complex.

After the transition, you must decide
how the unit changed color by
selecting one of the three possible colors

 

 

 

 

 

 

 

 

 

 

 

I Go Back I I Continue I

 

 

124

 

Now let's try another practice question.

 

 

I Go Back 1 I Continue I

 

 

 

Practice Question #2

 

125

 

 

Practice Question #2

Please indicate the color
of the highlighted unit
before the scene change:

 

EDI

 

 

Great that's correct!
Now let's look at the
third type of question
you will see on the test.

 

 

Go Back I I Continue I

 

 

126

 

Question Type #3

The third question will consist of a confidency scale
and will follow every question in the test.

After each of the questions, you will be asked to rate
how confident you were about your answer.

g

Confident Neutral Unsure

 

 

I Go Back I I Continue 1

 

Now, let's try a couple more questions
that are more similar to the type
you will see on the test.

 

 

r Go Back 1 I Continue I

127

 

 

- High
[:I Medium
:I Low

 

 

Did the highlighted unit
change between scenes?

High

[3 Medium
I:I Low

 

128

 

How confident are you about your answer?

 

r--- '- -'-“- "mm:- n
Confident Neutral Unsure

 

Great, you are correct!
Let's tryjust one more practice question.

 

 

I Go Back 1 I Continue I

129

 

 

 

- High
I: Medium
":7? Low

 

 

Please indicate the color
of the highlighted unit
before the scene change:

DUI

 

130

 

How confident are you about your answer?

r-r—-— _-._...._.____-_ ._.._..

Confident Neutral Unsure

 

Great,
now you are ready to begin the test.

 

 

I Go Back I I Continue 1

 

 

131

Appendix D
POST- TEST

Map Study- Posttest questions

Have you ever taken a test like this before?

a. Yes
b. No
c. Unsure

Did you feel that the training session of this test was useful to you in learning how

to take the test and introducing you to the types of questions you answered?

a. Yes
b. No
c. Neutral

Was it easy to navigate through the test program?

a. Yes
b. No
c. Neutral

What was your overall strategy to try to identify the changes?

 

 

 

132

,..l

5. Did you feel that there were some maps that allowed you to detect the change

between scenes easier? Describe.

 

 

 

6. Did you feel that some of the maps transitioned slower than others? If so, do you

think that some of these maps allowed you to more easily detect the changes?

 

 

 

7. Did you feel that some of the maps had smoother transitions than others? If so, do

you think that some of these maps allowed you to more easily detect the changes?

 

 

 

8. Do you have any other comments that might help to improve the test or the design

of the maps?

 

 

 

133

LIST OF REFERENCES

134

Adobe Flash CS3 (Version Creative Suite 3). (2007). San Jose, California: Adobe
Systems, Inc.

Adobe Illustrator CS3 (Version Creative Suite 3). (2007). San Jose, California: Adobe
Systems, Inc.

ArcMap 9.2. (2006). Redlands, California: Environmental Systems Research Institute,
Inc.

Battersby, S., & Goldsberry, K. (2010). Considerations in design of transition behaviors
for dynamic thematic maps. Cartographic Perspectives, 65.

Bertin, J. (1983). Semiology of graphics: University of Wisconsin Press.

Blackmore, S. J ., Brelstaff, G., Nelson, K., & Troscianko, T. (1995). Is the richness of
our visual world an illusion? Transsaccadic memory for complex scenes.
PERCEPTION-LONDON-, 24, 1075-1075.

Brewer. C. A. ( 1994). Color use guidelines for mapping and visualization. Visualization
in modern cartography, 2, 123-148.

Brewer, C. A., & Pickle, L. (2002). Evaluation of Methods for Classifying
Epidemiological Data on Choropleth Maps in Series. Annals of the Association of
American Geographers, 92(4), 662-681.

Campbell, C. S., & Egbert, S. L. (1990). Animated cartography/Thirty years of scratching
the surface. Cartographica: The International Journal for Geographic Information
and Geovisualization, 27(2), 24-46.

Cornwell, B. (1966). Possibilities for computer animated films in cartography.
Cartographic Journal, The, 3(2), 79-82.

DiBiase, D., MacEachren, A. M., Krygier, J. B., & Reeves, C. (1992). Animation and the
Role of Map Design in Scientific Visualization. Cartography and Geographic
Information Science, 19(4), 201-214, 265-266.

Dykes, J., & Unwin, D. Maps of the Census: a rough guide. Case Studies ofVisualization
in the Social Sciences, Technical Report, 43, 29-54.

F abrikant, S. I. (2005, Mar. 21-23, 2005). Towards an Understanding of Geovisualization
with Dynamic Displays: Issues and Prospects. Paper presented at the American
Association of Artificial Intelligence (AAAI) 2005 Spring Symposium Series:
Reasoning with Mental and External Diagrams: Computational Modeling and
Spatial Assistance, Stanford University, Stanford, CA.

135

F abrikant, S. I., Rebich—Hespanha, S., Andrienko, N., Andrienko, G., & Montello, D. R.
(2008). Novel Method to Measure Inference Affordance in Static Small-Multiple

Map Displays Representing Dynamic Processes. The Cartographic Journal, 45(3),
201-215.

F lannery, J. J. (1971). The relative effectiveness of some common graduated point
symbols in the presentation of quantitative data. Cartographica: The International
Journal for Geographic Information and Geovisualization, 8(2), 96-109.

French, R. S. (1953). The discrimination of dot patterns as a function of number and
average separation of dots. Journal of experimental psychology, 46(1), 1-9.

Goldsberry, K. (2004). Stabilizing Rate of Change in Thematic Map Animations.
Unpublished Masters Thesis. University of California Santa Barbara, Santa
Barbara, CA.

Goldsberry, K., & Battersby, S. (2009). Issues of Change Detection in Animated
Choropleth Maps. Cartographica: The International Journal for Geographic
Information and Geovisualization, 44(3), 201-215.

Griffin, A. L., & Bell, S. (2009). Applications of Signal Detection Theory to Geographic
Information Science. C artographica: The International Journal for Geographic
Information and Geovisualization, 44(3), 145-158.

Grimes, J. (1996). Scenes across Saccades. Perception, 5, 89-1 10.

Harrower, M. (2003). Tips for Designing Effective Animated Maps. Cartographic
Perspectives, 44, 63-65 and 82-83.

Harrower, M. (2007). The cognitive limits of animated maps. Cartographica: The

International Journal for Geographic Information and Geovisualization, 42(4),
349-357.

Harrower, M. (2007). Unclassed Animated Choropleth Maps. The Cartographic Journal,
44(4), 313-320.

Harrower, M., & Brewer, C. A. (2003). Colorbrewer. org: an online tool for selecting
colour schemes for maps. Cartographic Journal, The, 40(1), 27-37.

Heer, J ., & Robertson, G. G. (2007). Animated transitions in statistical data graphics.
IEEE transactions on visualization and computer graphics, 13(6), 1240-1247.

Henderson, J. M., & Hollingworth, A. (1999). The role of fixation position in detecting
scene changes across saccades. Psychological Science, 10(5), 438.

136

Hollingworth, A., Williams, C. C., & Henderson, J. M. (2001). To see and remember:
Visually specific information is retained in memory from previously attended
objects in natural scenes. Psychonomic Bulletin & Review, 8(4), 761.

Kochanek, D. H. U., & Bartels, R. H. (1984). Interpolating splines with local tension,
continuity, and bias control. ACM SIGGRAPH Computer Graphics, 18(3), 33-41.

Large, A., Beheshti, J ., Breuleux, A., & Renaud, A. (1996). Effect of animation in
enhancing descriptive and procedural texts in a multimedia learning environment.
Journal of the American Society for Information Science, 47(6).

Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand -.
words. Cognitive science, 11(1), 65-100.

Lasseter, J. (1987). Principles of Traditional Animation Applied to 3D Computer
Animation. Computer Graphics, 21(4), 35-44.

 

Levin, D. T., Drivdahl, S. B., Momen, N., & Beck, M. R. (2002). False predictions about
the detectability of visual changes: The role of beliefs about attention, memory,
and the continuity of attended objects in causing change blindness blindness.
Consciousness and Cognition, 11(4), 507-527.

Levin, D. T., Momen, N., Drivdahl, S. B., & Simons, D. J. (2000). Change blindness
blindness: The metacognitive error of overestimating change-detection ability.
Visual Cognition, 7(1), 397-412.

Levin, D. T., & Simons, D. J. (1997). BRIEF REPORTS Failure to detect changes to
attended objects in motion pictures. Psychonomic Bulletin & Review, 4(4), 501-
506.

Levin, D. T., Simons, D. J., Angelone, B. L., & Chabris, C. F. (2002). Memory for
centrally attended changing objects in an incidental real-world change detection
paradigm. British Journal of Psychology, 93(3), 289-302.

MacEachren, A. M., & DiBiase, D. (1991). Animated maps of aggregate data:
Conceptual and practical problems. Cartography and Geographic Information

Science, 18(4), 221-229.

Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A user’s guide:
Cambridge University Press Cambridge.

Mack, A. (2003). Inattentional blindness. Current Directions in Psychological Science,
12(5), 180.

137

Monmonier, M. (1990). Strategies for the Visualization of Geographic Time-Series Data.
Cartographica, 27(1), 30-45.

Monmonier, M. S. (1975). GLASS INTERVALS TO ENHANCE THE VISUAL
CORRELATION OF CHOROPLETHIC MAPS. Cartographica: The

International Journal for Geographic Information and Geovisualization, 12(2),
161-178.

Monmonier, M. S. (1985). Technological transition in cartography: University of
Wisconsin Pr.

Montello, D. R. (2002). Cognitive map-design research in the twentieth century:

Theoretical and empirical approaches. Cartography and Geographic Information
Science, 29(3), 283-304.

Mozilla Firefox. (2009). Mozilla Corporation.

0 Regan, J. K., Deubel, H., Clark, J. J ., & Rensink, R. A. (2000). Picture changes during
blinks: Looking without seeing and seeing without looking. Visual Cognition,
7(1),191-211.

O'Regan, J. K., Rensink, R. A., & Clark, J. J. (1999). Change-blindness as a result of
émudsplashesi. Nature, 398(6722), 34.

Rensink. R. A. (2002). CHANGE DETECTION. Annual Review of Psychology, 53(1),
245-277.

Rensink, R. A. (2002). Failure to see more than one change at a time. J. Vis, 2(245a).

Rieber, L. P. (1990). Using computer animated graphics in science instruction with
children. Journal of Educational Psychology, 82(1), 135-140.

Rieber, L. P. (1991a). Animation, incidental learning, and continuing motivation. Journal
of Educational Psychology, 83(3), 318-328.

Rieber, L. P. (1991b). Effects of visual grouping strategies of computer-animated

presentations on selective attention in science. Educational Technology Research
and Development, 39(4), 5-15.

Robinson, A. C., & Center, G. (2006). Highlighting techniques to support
geovisualization.

Robinson, A. H. (1952). The Look of Maps: An Examination ofCartographic Design.
Madison, WI: University of Wisconsin Press.

138

Seeber, J. (2003). Do We Benefit from Smooth Animations: An examination of the
impact of temporally smooth dot map animations on map readers. Unpublished
Masters Thesis. University of Wisconsin-Madison, Madison, WI.

Simons, D. J. (2000). Current approaches to change blindness. Visual Cognition, 7(1), 1-
15.

Simons, D. J ., & Chabris, C. F. (1999). Gorillas in our midst: Sustained inattentional
blindness for dynamic events. PERCEPTION-LONDON-, 28, 1059-1074.

Simons. D. J ., F ranconeri, S. L., & Reimer, R. L. (2000). Change blindness in the
absence of a visual disruption. Perception, 29, 1143-1154.

Simons, D. J ., & Levin, D. T. (1997). Change blindness. Trends in Cognitive Sciences,
1(7), 261-267.

Simons, D. J ., & Levin, D. T. (1998). BRIEF REPORTS Failure to detect changes to
people during a real-world interaction. Psychonomic Bulletin & Review, 5(4),
644-649.

Simons, D. J ., & Rensink, R. A. (2005). Change blindness: Past, present, and future.
TRENDS in Cognitive Sciences, 9(1), 16-20.

Slocum, T. A., McMaster, R. B., Kessler, F. C., & Howard, H. H. (2009). Thematic
cartography and geographic visualization: Pearson/Prentice Hall.

Sweller, J. (1994). Cognitive load theory, learning difficulty, and instructional design.
Learning and instruction, 4(4), 295-312.

Szego, J. (1987). Human Cartography: Mapping the World of Man. Stockholm: Swedish
Council for Building Research.

Theeuwes, J. (1995). Abrupt Luminance Change Pops Out; Abrupt Color Change Does
Not. Perception and Psychophysics, 57(5), 637-644.

Thompson, S. V., & Riding, R. J. (1990). The effect of animated diagrams on the
understanding of a mathematical demonstration in 11-to 14-year-old pupils. The

British journal of educational psychology, 60, 93.

Thrower, N. J. W. (1959). Animated cartography. The Professional Geographer, 1 1(6), 9-
12. '

Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit region.
Economic geography, 46, 234-240.

139

Tufte, E. (1983). The Visual Display of Quantitative Information. Cheshire, Connecticut:
Graphics Press.

Tversky, B., Morrison, J. B., & Betrancourt, M. (2002). Animation: Can it Facilitate?
lntemational Journal of Human Computer Studies, 57, 247-262.

Williams, P., & Simons, D. J. (2000). Detecting changes in novel, complex three-
dimensional objects. Change Blindness and Visual Memory, 2000(7), 297-322.

Wolfe, J. M. (1999). Inattentional Amnesia Jeremy M. Wolfe. Fleeting memories:
Cognition of brief visual stimuli, 71.

Yantis, S. (2005). How visual salience wins the battle for awareness. Nature gr“
Neuroscience, 8, 975-977. I
Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective attention: Evidence ”

from visual search. Journal of Experimental Psychology: Human perception and
performance, 10(5), 601-621.

140

  

      
     

SAN T

Willi/31111, IllllllljllllllllllllllEs

0