University of Southern California Dissertations and Theses

Cartographic approaches to the visual exploration of violent crime patterns in space and time: a user performance based comparison of methods

(USC Thesis Other)

Cartographic approaches to the visual exploration of violent crime patterns in space and time: a user performance based comparison of methods

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content CARTOGRAPHIC APPROACHES TO THE VISUAL EXPLORATION OF VIOLENT
CRIME PATTERNS IN SPACE AND TIME:

A USER PERFORMANCE BASED COMPARISON OF METHODS

by

Benjamin Green Anderson

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

May 2015

Copyright 2015 Benjamin Green Anderson
ii

ACKNOWLEDGMENTS
I want to thank Dr. Robert Vos and Dr. Katsuhiko Oda for their support and advice as I worked
on this project. I also want to thank the rest of the Spatial Sciences Institute faculty for their
support as I completed the program. Finally, thank you to my family and friends, without whom
I could not have made it this far.

TABLE OF CONTENTS
ACKNOWLEDGMENTS ii
LIST OF TABLES vi
LIST OF FIGURES vii
LIST OF ABBREVIATIONS viii
ABSTRACT ix
CHAPTER 1: INTRODUCTION 1
1.1 Motivation 2
1.2 Experiment Design Overview 4
1.3 Research Questions 5
CHAPTER 2: RELATED WORK 6
2.1 Visualization of Spatiotemporal Data with Static Maps 6
2.2 Visualization of Spatiotemporal Data with Animated Maps 7
2.3 Interactivity of Spatiotemporal Maps 9
2.4 Visualization of Crime Patterns 10
2.5 Comparison between Static and Animated Maps 12
2.5.1 Previous Comparative Studies 14
2.5.1.1 Socia's Comparative Study 14
2.5.1.2 Adjustments to Socia's Methods 15
CHAPTER 3: METHODOLOGY 17
3.1 Mapped Data 18

3.1.1 Data Aggregation Methods 19
3.2 Cartographic Design 20
3.3 User Performance Experiment Design 22
3.3.1 Map Test Format and Performance Measurements 24
3.3.2 User-interface Design 25
3.3.2.1 Static Time-series Map Display Format 25
3.3.2.2 Animated Time-series Map Display Format 26
3.3.3 Recruitment of Participants 27
3.3.4 Testing Procedures 28
3.4 Methodology for Analyzing the Results of the Experiment 29
3.4.1 Statistical Analysis Methodology 30
CHAPTER 4: RESULTS AND ANALYSIS 32
4.1 Experimental Population 32
4.2 Overall Performance Metrics 34
4.3 Task Accuracy 36
4.4 Confidence 39
4.5 Completion Time 41
4.6 User-preferences 44
4.7 Correlations 46

CHAPTER 5: DISCUSSION AND CONCLUSIONS 48
5.1 Task Accuracy 49
5.2 Confidence 50
5.3 Completion Time 50
5.4 User-preferences 51
5.4.1 Performance and User-preference Correlations 52
5.5 Conclusions 52
5.5.1 Parallels with Socia’s Study 53
5.5.2 Study Strengths and Weaknesses and Suggestions for Future Research 54
5.5.3 Implications for the Cartography Community 56
REFERENCES 58
APPENDIX A: Chicago Homicide Hot Spot Maps, 2009-2013 63
APPENDIX B: Map Test Survey Interface Images 70

vi

LIST OF TABLES
Table 4.1 Age Distribution of Study Participants 33
Table 4.2 Education Level of Study Participants 33
Table 4.3: Accuracy, Confidence, Completion Time, and User-preferences 35
Table 4.4 Test Score Descriptive Statistics 36
Table 4.5: Confidence Score Descriptive Statistics 39
Table 4.6 Static and Animated Completion Time 42
Table 4.7 User-preference Descriptive Statistics 44
Table 4.8 Static Performance and Preference Correlations 46
Table 4.9 Animated Performance and Preference Correlations 47

vii

LIST OF FIGURES
Figure 3.1 Hot Spot Analysis Tool Settings 21
Figure 3.2 Chicago Homicide Hot Spots, 2009-2013, 12:00 AM to 3:59 AM 22
Figure 3.3 Animated Time-series Map Display Format 27
Figure 4.1 Static and Animated Accuracy Score Histograms 37
Figure 4.2 Static and Animated Confidence Score Histograms 40
Figure 4.3 Static and Animated Completion Time Box-plots 42
Figure 4.4 Static and Animated Completion Time Histograms 43
Figure 4.5 Static and Animated Preference Score Histograms 45

viii

LIST OF ABBREVIATIONS
GIS Geographic Information System(s)
NIBRS National Incident-based Reporting System
SMMD Small-multiple Map-display

ix

ABSTRACT
This study provided an empirical comparison of static and animated cartographic representations
of spatiotemporal phenomena in their application to basic choropleth map-based knowledge-
extraction tasks to answer the following research questions: 1) Do animated maps provide
heightened potential for accuracy in completing basic knowledge-extraction tasks over static
time-series maps, or vice versa? 2) Do animated maps provide heightened potential for efficiency
in completing basic knowledge-extraction tasks over static time-series maps, or vice versa? and
3) How do user preferences align or not align with measurements of accuracy and efficiency?
To this end, this study examined map readers’ accuracy and efficiency in completing
knowledge-extraction tasks through static and animated time-series maps about homicide
patterns in the Chicago metropolitan area. Through an online user performance experiment,
participants answered a series of questions about homicide hot spots and cold spots using both
static and animated versions of the maps as the basis for their answers. They were also asked to
indicate their level of confidence in the accuracy of their responses and to indicate which map
type they preferred for completing the tasks. Task completion times were recorded for efficiency
measurements. The results of independent samples t-tests indicate statistically significant
differences between the static and animated maps in terms of task accuracy and completion time.
Generally, users were able to complete the assigned tasks more accurately and much more
efficiently using the static maps, as compared with their animated counterparts. Additionally,
user-preferences were checked for correlations with task accuracy and completion time via
Pearson’s product-moment correlation coefficient calculations. The results indicate no significant
correlations between performance measurements and user-preferences.
1

CHAPTER 1: INTRODUCTION
Technological advances continue to stimulate new methods for the visualization of spatial and
spatiotemporal information (Blok et al. 2001). While these methods can provide new and
exciting ways to view and interpret spatial data, care must be taken to understand and prevent
potential usability issues. With animated maps, for example, the images can sometimes be too
fleeting to be perceived correctly (Betrancourt and Tversky 2002). Unfortunately, the excitement
that often surrounds new technological developments in cartography, particularly temporally
animated maps, interactive web-maps, and 3D spatiotemporal data visualizations, can sometimes
distract cartographers and their audiences from any potential limitations that might accompany
these new visualization tools. Furthermore, Andrienko et al. (2008) suggest that as animated
maps become more and more common, it is increasingly important to understand the utility of
temporal map animation for supporting basic map-based knowledge-extraction tasks.
This study endeavors to provide an empirical comparison of static and animated
cartographic representations of spatiotemporal phenomena in their application to basic
choropleth map-based knowledge-extraction tasks. These tasks simply prompt study participants
to interpret basic information from the maps and use this information to answer a series of
questions. To this end, this study examines map readers’ performance and efficiency in
completing choropleth map-based knowledge-extraction tasks, using static time-series maps and
animated maps that depict homicide patterns in the Chicago metropolitan area as the basis for
doing so. Through an in-depth user performance experiment, this study helps to provide insight
as to the strengths and weaknesses of static time-series maps and animated maps as the basis for
choropleth map-based knowledge-extraction tasks, to determine which of these visualization
2

methods map users prefer for carrying out these knowledge-extraction tasks, and to better
understand which tool inspires more confidence in response accuracy.

1.1 Motivation
Historically, maps have been essential to mankind’s understanding of geographic features and
phenomena. The very nature of studying geographic phenomena requires a visual element, as
without one there is no way for researchers to understand the spatial context of their data. As
such, few would argue against the notion that cartographic visualization facilitates learning about
geography. After all, a single map image can quickly and effectively communicate complex
spatial processes, whereas it might be virtually impossible to effectively communicate the same
volume of information without these visual tools. Furthermore, according to the picture
superiority effect principle (Shepard 1967) complex concepts that are learned by viewing
pictures are much more likely to be remembered than those learned by reading written words.
While there is little debate over the utility of maps for conveying spatial processes, there
is still some contention over the practicality of modern visualization tools like animated maps
and 3D data visualizations for supporting basic knowledge-extraction tasks. While static maps
have been used for thousands of years, animated maps became popular much more recently, in
the late 1990s (Harrower 2009). As such, cartographic design principles relating to the
communicative effectiveness of static maps have been the subject of countless years of research
while similar research regarding animated maps began very recently, by comparison (Fabrikant
and Harrower 2007). Unfortunately, as suggested by Andrienko et al. (2008), cartographers still
know rather little about the effectiveness of interactive graphical data depictions and
visualization methods for knowledge-extraction, learning, and understanding dynamic processes.
3

A common theme underlying modern visualization research challenges is the lack of verified
methods for identifying any positive or negative influences on people’s map-based knowledge-
extraction or decision-making through interactive visualization tools such as animated maps,
interactive web-maps, 3D maps, and static time-series maps (MacEachren and Kraak 2001;
Fabrikant 2005; Harrower 2007; Andrienko et al. 2008).
While maps are produced and used for many different purposes, they are most often used
as tools for communication of complex spatial phenomena to an audience. The communication
model, as defined by Board (2011), describes the map as a conduit for the transmission of a
message from the mapmaker to the map user. As Board describes, “Cartographic communication
emphasizes not only the medium but both the initiator and receiver of the information being
communicated. It emphasizes a process rather than a product” (Board 2011, p. 37). As such,
cartographers should take account of users’ perceptual and cognitive limits, as well as their
preferences, when designing maps.
Crime maps are most often produced and used by crime analysts at police departments
and various other government agencies and non-governmental organizations. As such, many
crime maps are too specialized or too advanced for the general public to easily understand.
Efforts need to be made to provide the general public with user-friendly and informative
visualization tools so that they too can take part in the conversation about crime in their cities.
Another significant issue in the realm of crime mapping is that the vast majority of crime
maps are produced without any attention paid to temporal variances in the distribution of crime
incidents, as if time played no part in these events. Particularly with violent crime, time plays an
important role in the occurrence and spatial distribution of these events. Assaults, for example,
are much more likely to be clustered near bars late at night, or near sports venues during game
4

time (Ratcliffe 2010). Clearly, space and time interact to create criminal opportunities. As such,
efforts need to be made to understand these temporal variances in criminal events, and to
communicate any patterns that might exist in the distribution of these events to the general
public. Armed with this knowledge, communities may be better prepared to assist in combatting
the proliferation of violent crime in their neighborhoods. Access to this information might also
help concerned citizens to protect themselves and their families from becoming victimized.

1.2 Experiment Design Overview
The primary aim of the experiment conducted for this study is to measure user performance in
carrying out various map-based knowledge-extraction tasks. This experiment is modeled on the
work of Kristie Socia, specifically on her 2011 thesis entitled Small-multiples and Animation:
Measuring User Performance with Wildfire Visualization. Socia measured user performance via
task accuracy and response time using static time-series maps (small-multiple map displays) and
animated maps that depicted the progression of a wildfire outside of San Diego, California. She
also conducted a survey of user preferences between static and animated maps, and of users’
confidence in the accuracy of their responses. By comparing users’ performance measurements
including response accuracy and response time to their reported preferences and their confidence
in the accuracy of their responses, Socia found that user preferences in her study did not coincide
with the practical application of visualization tools for basic knowledge-extraction tasks.
Like Socia’s study, performance measurements for this study are based on panel
participants’ response accuracy for each knowledge-extraction task and on the average amount of
time it takes them to complete each task. By comparing participants’ test scores and response
times between the animated and static maps, this study helps to provide insight as to the
5

strengths and weaknesses of each visualization tool as the basis for various knowledge-extraction
tasks.

1.3 Research Questions
To reiterate, this study concentrated primarily on measuring user performance (accuracy and
efficiency) in carrying out various map-based knowledge-extraction tasks using both static and
animated time-series maps of Chicago crime incidents as the basis for doing so. This study also
prompted participants to provide a subjective review of each map type based on their personal
preferences and their confidence in the accuracy of their responses. Finally, user performance
measurements were analyzed for correlations with user-preferences to find potential statistical
associations that might suggest patterns and relationships among the different variables. This
work was conducted to answer the following research questions:

1. Do animated maps provide heightened potential for accuracy in completing basic
knowledge-extraction tasks over static time-series maps, or vice versa?

2. Do animated maps provide heightened potential for efficiency in completing basic
knowledge-extraction tasks over static time-series maps, or vice versa?

3. How do user preferences align or not align with measurements of accuracy and
efficiency?

6

CHAPTER 2: RELATED WORK
As described in the previous chapter, this study endeavors to measure user performance, user
preferences, and user’s confidence in the accuracy of their responses in the contexts of static and
animated crime map interpretation, as well as to reveal any relationships that might exist
between these different variables. To provide the necessary background for this study, this
chapter covers previous work that directly relates to this study in the fields of spatiotemporal
data visualization, interactive maps, cartographic experimentation, and crime incident data
visualization. This chapter is divided into five sections. Section 2.1 discusses the visualization of
spatiotemporal data with static maps. Section 2.2 discusses the visualization of spatiotemporal
data with animated maps. Section 2.3 discusses the interactivity of spatiotemporal maps. Section
2.4 discusses the visualization of crime patterns. Finally, Section 2.5 discusses the comparison of
static and animated maps,

2.1 Visualization of Spatiotemporal Data with Static Maps
Static maps can depict change over time with temporal snap-shots (Thrower 1959). The two
most common types of static spatiotemporal maps are small-multiple map displays (SMMDs)
and planimetric overlay map series (Baldwin 2014).
The term SMMD describes a series of small maps arranged next to each other that are
used to portray change over time or to convey multiple thematic attributes for comparison to one
another (Tufte 1995). As Tufte describes them, small multiple map displays are "illustrations of
postage-stamp size [that] are indexed by category or a label, sequenced over time like the frames
of a movie, or ordered by a quantitative variable not used in the single image itself" (Tufte 1995,
p. 67). It is important to note that small-multiples do not necessarily have to be the size of a
7

postage stamp. In fact, the term has been used to describe map series of widely variable sizes.
The key element of SMMDs is that they are series of static maps that can be used to depict
change in the element of interest from one frame to the next.
Planimetric temporal overlay maps are similar to SMMDs in that several map layers
representing different time periods or points in time are used together to communicate change
over time. The key difference between planimetric overlay maps and SMMDs is that in
planimetric overlay, the images are stacked one on top of another, like a layer cake, rather than
side by side. Paper planimetric overlay maps are generally designed to be viewed from an
oblique angle, which allows the user to view each layer individually, and thereby to differentiate
between the different time periods or thematic elements (Baldwin 2014).
Boscoe et al. (1999) demonstrated the utility of static geographic visualizations as
platforms for the exploration, analysis, synthesis, and presentation of georeferenced
spatiotemporal information. Boscoe and his colleagues also explored the utility of static time-
series maps for examining time sequences and displaying changes over time within the confines
of a Geographic Information System. Through an empirical comparison of methods, similar to
the experiment conducted for this study, Boscoe and his colleagues confirmed the utility of time-
series maps in the presentation of spatiotemporal phenomena in multiple fields of research.

2.2 Visualization of Spatiotemporal Data with Animated Maps
The animated map, as defined by Peterson (2014) “is a cartographic statement that occurs in
time. Its interpretation is based on the human sensitivity to detect movement or change in a
display” (Peterson 2014, p. 1). Change in this context, as defined by DiBiase et al. (1992), is
divided into three distinct categories, each emphasizing a distinct type of change: change in
8

either position or an attribute, change in the location of some phenomenon, or change in the
spatial distribution of an attribute. DiBiase et al. also note that animations can be subdivided into
three distinct categories: time series (which depict chronological change), re-expressions (which
depict attribute changes), and flybys (which depict spatial change). For the sake of brevity, re-
expressions and flybys are not discussed in detail here as they are not relevant to this study. Time
series, as Slocum et al. (2009) suggest, are by far the most common form of animated map.
Time-series animations operate in much the same way as a movie clip. Map images (each frame
depicting a specific moment in time or a specific timeframe) are sequenced chronologically and
compiled into a video sequence.
As suggested by Slocum et al. (2009), the first animated time-series maps were
developed in the 1930s, and by the late 1950s cartographers had acknowledged the potential
utility of animated maps for conveying dynamic processes. Thrower (1959), one of the earliest
proponents of animated maps, describes animation “by the use of animated cartography we are
able to create the impression of continuous change and thereby approach the ideal in historical
geography, where phenomena appear as dynamic rather than static entities” (Thrower 1959, p.
10). Despite their acceptance as useful tools for conveying dynamic processes, cartographic
animations remained very rare until the early 1990s (Slocum 2009), due to the extremely high
monetary costs associated with their production. Technological advancements in the early 1990s
allowed the development of much more affordable hardware (Socia 2011). As production costs
have come down over time, animated maps have grown increasingly common.

9

2.3 Interactivity of Spatiotemporal Maps
Cartographic interaction, as defined by Roth (2013), is the dialog between a human and a map.
This interaction is the basis for our ability to read maps and interpret their contents. The mode of
interaction varies widely between different cartographic tools, particularly between static and
animated maps. As Roth points out, the cartographic interaction dialog is often mediated through
a computing device, though it applies to analog cartographic visualization as well, as the simple
act of interpreting the information presented in a static map is one form of cartographic
interaction. While he acknowledges that all maps are interactive to some extent, Roth suggests
that digital map mediums typically provide a much wider array of interaction forms for
manipulating cartographic representations, thereby allowing more flexible interaction. He also
notes that maps with high interactivity are quickly growing in popularity. It can be expected
then, as Roth suggests, that making design decisions that account for the different modes of
cartographic interaction that are made possible by digital map media will only grow more
fundamental to cartographic design as the dominant map prototype shifts from analog to digital.
It is important to note that interactivity has varying levels of intensity (Andrienko et al.
2008). Static maps have the lowest level of interactivity, but not zero. As Andrienko et al. (2008)
suggest, static time-series maps, particularly SMMDs, afford mental interactivity in that people
can control the viewing order of the static sequence, they can choose to go back to the beginning,
and they can study the sequence in any order they choose at their own pace. While animated
maps are a bit more externally interactive, in that they feature start, stop, and rewind buttons,
they are actually less internally interactive, as the animation must be passively viewed in a pre-
defined sequence (Fabrikant 2005; Andrienko et al. 2008).
10

Animations are transient by nature, often requiring viewers to keep track of multiple
symbols and map elements that are changing simultaneously (Socia 2011). As Socia points out,
when animations become too complex, it can become very difficult, if not impossible, to keep
track of all of the different dynamic elements. Betrancourt and Tversky (2002) suggest that
animations may be less effective than static representations because animations are often too
complex or too fleeting to be perceived accurately. Harrower (2007) coined the term split
attention to describe this effect. Split attention, according to Harrower, is a significant weakness
unique to animated maps, particularly when temporal legends or other dynamic elements are
employed.
While all animated maps are interactive in the sense that they provide play and stop
buttons, not all animations provide the same level of interactivity. Without additional interactive
elements such as time-sliders or other interface tools that allow the user to easily navigate the
temporal extent of the animation, the user must attempt to remember and integrate changes
between scenes, which may overload users’ working (short-term) memory (Andrienko et al.
2008). Though static maps are not traditionally viewed as being interactive, SMMDs are
interactive in the sense that the viewer can toggle between images at will, viewing the images in
any order they choose or spending as much time on each image as they deem necessary
(Fabrikant 2005; Andrienko et al. 2008).

2.4 Visualization of Crime Patterns
GIS-based visualization and spatial analysis of crime are commonly used to reveal patterns in the
distribution of crime incidents (Nakaya and Yano 2010; Chainey and Ratcliffe 2005). Compstat,
for example, is a system that was developed by the NYPD. “Compstat is a goal-oriented,
11

strategic-management process that uses information technology [including Geographic
Information Systems], operational strategy, and managerial accountability to guide police
operations” (Vito and Walsh 2004, p. 51). According to Vito and Walsh, Compstat combines
accurate and timely intelligence in the form of geocoded criminal incident data, rapid
deployment, and effective tactics which allow police departments to react to crime outbreaks
very quickly. This rapid response to crime outbreaks is made possible, in large part, by the
efficient data collection, visualization, and spatial analysis tools that GIS provides. Upon
implementing Compstat, New York City experienced a dramatic reduction in crime rates across
the board (Vito and Walsh 2004). As Vito and Walsh describe, this success story lead to the
implementation of similar programs around the nation. Today, GIS-based crime analysis
software packages are used, in some fashion, in virtually every police precinct in the United
States.
Significant effort has been devoted to detecting geographic areas with particularly high
crime density, commonly referred to as crime hotspots. Several methods have been utilized for
visualizing these hotspots, including: pin maps, choropleth maps, shaded grid maps, risk terrain
maps, kernel-density estimation maps, Getis-Ord hotspot maps, and inverse-distance weighted
interpolation maps. However, the majority of this work has been done from an entirely spatial
perspective. This is unfortunate because spatial analysis alone ignores the necessary interaction
of space and time to create criminal opportunities (Grubesic and Mack 2008).
Previous crime studies suggest that the spatial distribution of crime incidents varies from
one year to the next, between seasons of the year, between weekdays and weekends, and within
the span of a single day (Bowers and Johnson 2004). Unfortunately, because the vast majority of
crime visualization and analysis has been done from a wholly spatial perspective (Grubesic and
12

Mack 2008), many of the tools for crime data visualization are of limited use for comparing
crime patterns between different time periods.

2.5 Comparison between Static and Animated Maps
Socia (2011), drawing on previous work done by Larkin and Simon (1987) and Andrienko et al.
(2008), indicates that in order to conduct a fair comparison between static time-series and
animations for a specific purpose, the two visualizations must be informationally equivalent
(Andrienko et al. 2008). As Andrienko et al. (2008) describe, informational equivalence (a term
coined by Larkin and Simon 1987) describes the notion that any information inferable from one
representation must also be inferable from the other, and vice versa, for any fair comparison to
be made between them. Like Socia’s study, this study was designed with the necessity of
informational equivalence in mind. The static and animated maps that provided the basis for this
experiment are identical, aside from the fact that one series is static and the other animated.
As suggested by Andrienko et al. (2008), several previous comparative cartographic
experiments have been deemed inconclusive because these experiments attempted to determine
which cartographic approaches were universally superior for representing dynamic processes. In
opposition to these previous works, Andrienko et al. (2008) argue that the question of whether
one cartographic method is comprehensively superior to another is not only an ill-conceived
question, but an unanswerable one. They go on to suggest that visualization designers should,
instead, be interested in determining how interactive visual displays work, determining when
they are successful, and why. As such, this study was conducted only to understand the strengths
and weaknesses of static time-series maps and animated maps in the specific context of this
study, not to determine which tool is generally superior to the other.
13

Usability engineering, a method for evaluating a product or system’s ease of use
(Coltekin et al. 2009), can be used to measure the effectiveness of cartographic representations as
tools for spatial knowledge-extraction. As Coltekin et al. (2009) explain, “Users are provided
with a specific set of tasks based on a particular usage scenario, and in a specific context.
Usability performance metrics such as satisfaction, efficiency, and effectiveness (SEE) are
employed to assess how easy the product or system is to use. Satisfaction refers to the user’s
attitude or preferences about the system, efficiency refers to how quickly the tasks are
completed, and effectiveness refers to whether or not a task is successfully completed” (Coltekin
et al. 2009, p. 6). This study uses the usability engineering principles described above to evaluate
the different visualization tools that were the focus of this comparison.
By focusing on the merits of different visualizations methods and cartographic design
elements for supporting specific knowledge-extraction tasks in a specific context, it may be
possible to gauge the strengths and weaknesses of each cartographic approach in its application
to certain tasks. While one visualization technique might be better for a given task or type of
tasks in the context of this study, that does not mean it is a universally superior visualization
method. There are far too many potential applications for these tools to make a blanket claim of
superiority. Acknowledging the importance of the intended application of these tools, this study
sought only to investigate the suitability of each tool for supporting specific tasks pertaining to a
specific set of maps. While this process did reveal patterns in tool usability and user
performance, these patterns were not and should not be assumed to apply universally outside the
specific context of this study.

14

2.5.1 Previous Comparative Studies
Previous comparative studies of visualization methods for spatiotemporal information (Brunsdon
et al. 2006; Grubesic and Mack 2008) have compared fairly user-friendly visualization tools and
discussed the importance of user-accessibility, but neglected to carry out the necessary panel
review experiments to measure user performance on knowledge-extraction tasks with these tools.
This study develops user-friendly static time-series maps and animations and emphasized the
usability issue via the user performance experiment described in the next chapter.
Other previous studies (Bekele et al. 2009; Midtbo and Larsen 2005) compared static
maps and animated maps via user performance experiments, seemingly to determine which
methods were superior for demonstrating dynamic spatiotemporal processes. This study seeks
only to understand the strengths and limitations of each tool in the specific context of this
experiment. While this study hints at the strengths and weaknesses of each tool for
demonstrating dynamic spatiotemporal processes, the patterns in user performance that are
revealed by this study are not assumed to apply universally.

2.5.1.1 Socia’s Comparative Study
The experiment design for this study is based on a 2011 University of Michigan geography thesis
by Kristie Marie Socia entitled Small Multiples and Animation: Measuring User Performance
with Wildfire Animation. Socia measured user performance via task accuracy and response-time
using static time-series SMMDs and animated maps that depicted the progression of a wildfire
outside of San Diego, California, over time. Socia also conducted a survey of user preferences
and of users’ confidence in the accuracy of their responses.
15

In reviewing the results of her experiment, Socia found that small-multiples afforded
study participants statistically significantly higher response accuracy scores (85.4% for small-
multiples and 80.4% for animation). She also found that small-multiples provided users with a
statistically significant advantage in terms of efficiency. Socia’s study participants were able to
complete the assigned tasks in an average time of 21.8 seconds using the small-multiple series,
while it took them 26.1 seconds, on average, to complete each task using the animated maps.
Socia also found that her study participants tended to be more confident in their responses when
using the small-multiple series than they were when using the animated maps as the basis for
their responses.
By comparing users’ response accuracy and response time to the subjective feedback they
provided on each of the visualization tools, Socia also found evidence to suggest that, at least in
the context of her experiment, user preferences generally did not coincide with the practical
application of the visualization tools for the basic knowledge-extraction tasks. Despite scoring
better in both accuracy and efficiency using the static small-multiple series, the vast majority
(72%) of Socia’s study participants preferred the animated version.

2.5.1.2 Adjustments to Socia’s Methods
In her concluding discussion, Socia (2011) notes several problems with the design of her study.
Chief among these was her decision to include skip-to-time-stamp buttons as the primary
navigation tool for her animated maps. Several of her study participants reported having
difficulty navigating the animations as necessary to complete the tasks she assigned. Study
participants attributed this difficulty to the skip-to-time-stamp navigation interface. This study
endeavors to further Socia’s work by using a time-slider (a scroll bar with which users are able to
16

seamlessly navigate the temporal extent of the animation) rather than skip-to-time-stamp buttons.
This change both helps to simplify the user-interface and allows users to easily navigate the
entire temporal extent of the animation, whereas Socia’s participants were only permitted to skip
back and forth between certain timestamps. This was particularly problematic in Socia’s study
because some of the test questions asked for information that was located in between two skip-to
time-stamp-markers. Several of Socia’s study participants indicated having difficulty with the
skip-to-time-stamp interface, particularly for the test questions that required them to mentally
interpolate between scenes.
Further improvements are being made by ensuring that study participants are properly
briefed on cartographic design and task format before beginning the test, and by ensuring that all
tasks and/or questions are posed in very clear language. Like the decision to use a time-slider
rather than skip-to-time-stamp buttons, these improvements are being made based on feedback
from participants in Socia’s study.
This study also furthers Socia’s research by applying her general methodology to entirely
different subject matter. By replicating and improving upon her methods and applying them to
choropleth homicide incident visualization, rather than to raster-based wildfire visualization, this
study helps to determine whether the results of Socia’s study are applicable outside the specific
context of her experiment.

17

CHAPTER 3: METHODOLOGY
This study endeavored to provide an objective comparison of static and animated cartographic
representations of spatiotemporal phenomena in their application to basic choropleth map-based
knowledge-extraction tasks. To this end, this study compared static time-series maps with
animated maps in their application to the visualization of homicide patterns in the Chicago
metropolitan area. Through an in-depth user performance experiment, this study helped to
provide insight as to the strengths and weaknesses of static time-series maps and animated maps
as the basis for choropleth map-based knowledge-extraction tasks. It also helped to determine
which of these cartographic tools map users prefer for carrying out these tasks, and to better
understand which tools inspired the most confidence in response accuracy.
This chapter discusses the methodology for the user performance experiment itself and
the maps developed for the experiment. The general experiment design that was used in this
study was based loosely on the experiment design that Socia developed for her study, even
though the maps that provided the basis for the two experiments were very different. As
described in the previous chapter, this study aims to further Socia’s work by attempting to adjust
for some of the issues she encountered while conducting her experiment.
Chapter 3 is composed of four sections. Section 3.1 provides details on the data that
provided the basis for the maps that were used for the experiment. Section 3.2 discusses the
cartographic design for the static and animated map series. Section 3.3 discusses the user
performance experiment design. Finally, Section 3.4 discusses the methodology for analyzing
the results of the experiment.

18

3.1 Mapped Data
Crime hot spot maps provided the basis for the aforementioned comparison. A six-step time-
series of static hot spot maps and an animated version of the same maps were created to visualize
how the spatial distribution of homicide incidents varied from one time period to the next. While
the visualization methods that provided the basis for this study could be applied to any point-
incident-based datasets that provide specific point locations and specific times for each data
point, this study focused on homicide incidents in Chicago, Illinois. Chicago was selected as the
basis for this study for three reasons. First, Chicago has one of the highest homicide rates in the
entire United States (Huffington Post 2013). Second, the city of Chicago provides a very
detailed, accurate, and up-to-date crime dataset that is accessible via the City of Chicago Data
Portal (https://data.cityofchicago.org/). Finally, among the several datasets for different cities
studied (Atlanta, Chicago, Denver, San Diego, and Seattle), Chicago stood out because it
contains records on the time of day at which each incident was reported, which have been
thoroughly checked by the Chicago GIS team for consistency and accuracy (City of Chicago
Department of Innovation and Technology 2014). Many of the other datasets that were
considered contained incomplete records or limited metadata with which to verify the suitability
of the data for this study.
All data used for this study are available through the City of Chicago Data Portal
(https://data.cityofchicago.org/). This includes a very detailed National Incident-based Reporting
System (NIBRS) crime dataset going back to 2001, containing point incident data for individual
crime events in the Chicago metropolitan area, as well as police precinct and police beat
shapefiles which helped to provide context for the crime data. The dataset chosen for this study
is one of very few NIBRS datasets that has accurate time of day data (in addition to accurate
19

spatial data). Most local datasets have date information only. Importantly, time values in the
dataset are based as closely as possible on the times at which the incidents actually occurred,
rather than the time at which law enforcement officers arrived on scene (City of Chicago
Department of Innovation and Technology 2014). The dataset also has detailed metadata that
was used to verify its suitability for this study, something many other NIBRS datasets are
lacking. This metadata file provides details on the method of data acquisition, expanded
definitions of attribute data and data types, and estimations of spatial and temporal accuracy for
the incident data.

3.1.1 Data Aggregation Methods
Since one day’s worth of incident data does not provide a sufficient number of points for the
effective visualization of spatiotemporal trends in these events, all homicide events that occurred
over the course of several years (2009-2013) were visualized together as if they occurred in the
same 24-hour period. The homicide incident points were divided into six separate datasets based
on the time of day at which they took place. The six time periods were 12:00 AM to 3:59 AM,
4:00 AM to 7:59 AM, 8:00 AM to 11:59 AM, 12:00 PM to 3:59 PM, 4:00 PM to 7:59 PM, and
8:00 PM to 11:59 PM. Once the data points were divided by time period, the incident data for
each time period were spatially joined with a polygon shapefile containing all of the police beats
in the city of Chicago. Area measurements for the police beats were then utilized to produce a
homicide rate (number of incidents per square mile) for each police beat and each time period.
These homicide rate values for the police beat polygons provided the basis for the Getis-Ord Gi*
hotspot analysis calculations described in the next section.

20

3.2 Cartographic Design
Getis-Ord Gi* hot spot maps provided the basis for the comparison between static and animated
visualizations in this study. To construct the hot spot maps, the Hot Spot Analysis (Getis-Ord
Gi*) tool available in ArcGIS 10.2 (Figure 3.1) was used. Given a set of weighted features (the
police beat polygons, weighted according to the homicide rate for each beat, in this case) the
Getis-Ord Gi* statistic identifies statistically significant clusters of hot spots and cold spots
(clusters of police beats with particularly high or low homicide rates, in this case). In addition,
the tool required the researcher to select an option for conceptualizing the spatial relationships
between the features they wish to analyze. Since the incident data for this study were aggregated
to police beat polygons and converted into homicide rates (number incidents per square mile
calculated individually for each police beat), spatial relationships for this analysis were
conceptualized using the Contiguity_Edges_Corners option. This setting allows researchers to
include all polygons that share an edge or corner with the target polygon in the computations for
that polygon. This setting was chosen because the police beats varied considerably in size. If one
of the distance-based conceptualization options had been selected, the output might have been
distorted due to the highly variable size of the police beat polygons.
21

Figure 3.1: Hot Spot Analysis Tool Settings

Once the Getis-Ord Gi* statistic has identified the hot spots and cold spots (as well as
areas that are neither hot or cold spots, referred to as not significant by the Getis-Ord Gi*
output), each police beat polygon is assigned a color, based on its status as a hot spot, cold spot,
or neither. Hot spots and cold spots each have three confidence levels: 99%, 95%, and 90%.
Each of these levels is assigned a different shade of either red or blue. The 99% confidence hot
spots are marked in dark red, the 95% confidence hot spots in dark orange, and the 90%
confidence hot spots in light orange. The 99% confidence cold spots are marked in dark blue;
the 95% confidence cold spots are marked in light blue, and the 90% confidence cold spots are
marked in light blue/green. Those areas determined to be insignificant, neither hot nor cold
spots, are marked in pale yellow. Figure 3.2 is an example of one frame of a six map time-series.
22

Figure 3.2: Chicago Homicide Hot Spots, 2009-2013, 12:00 AM to 3:59 AM

3.3 User Performance Experiment Design
In order to measure the effectiveness of static time-series and animated maps as tools for the
communication of complex spatiotemporal information to an audience, a panel of study
participants was asked to complete a series of choropleth map-based knowledge-extraction tasks,
using each of the aforementioned visualization tools in turn as the basis for doing so. Based on
task accuracy and response time, it was possible to gain valuable insights as to which tools better
facilitated successful and timely completion of the assigned tasks.
23

The user performance experiment outlined above provided the basis for the evaluation of
each visualization technique as a tool for supporting users’ spatial knowledge-extraction. The
results of this evaluation were not intended to prove any cartographic methods to be universally
superior or inferior in their application to the proposed tasks. This study was intended only to
gain some general insights into the strengths and weaknesses of each technique for certain types
of map-based knowledge-extraction tasks in the specific context of time-series crime maps.
The user performance experiment consisted of three distinct segments: a pre-test, map
test, and post-test. The pre-test consisted of a brief tutorial on interpreting time-series hot spot
maps, a description of what study participants could expect to encounter as they completed the
exercise, and a short questionnaire that collected basic background information on study
participants (age, sex, and level of education). The map interpretation tutorial gave participants
an opportunity to learn how to interpret and interact with the maps before accuracy and
efficiency measurements began. The map test measured study participants’ accuracy, response
time, and confidence in responses using each of the cartographic tools as the basis for their
answers. The post-test consisted of a series of questions on user-preferences between the static
and animated maps as well as a series of qualitative open-ended questions aimed at gaining
insights on users’ map-reading strategies and whatever other feedback they offered.
The user-preference assessment in the post-test was conducted to determine which of the
proposed cartographic tools were best liked by review participants, which tools inspired the most
confidence in responses, and to understand which tools participants felt to be most effective for
the assigned tasks. These user-preference questions were structured using a forced-choice Likert-
scale. For each of a series of statements, which were designed to gauge user-preferences between
the static and animated maps, participants were asked to indicate their level of agreement
24

(strongly disagree, disagree, agree, or strongly agree). The neutral (neither agree nor disagree)
option was omitted from the possible responses to prevent central tendency bias. Acquiescence
bias was avoided by including an equal number of positive and negative statements in the
qualitative review questions. Two statements favored the static maps and two statements favored
animated maps. Participants were asked to indicate their level of agreement with each
statement.

3.3.1 Map Test Format and Performance Measurements
The panel review process for this study focused primarily on user performance in carrying out
map-based knowledge-extraction tasks through a series of static and animated choropleth maps.
More specifically, this study was designed to test user’s ability to visually discern temporal
change in choropleth maps of homicide rates in Chicago analyzed at the scale of police precincts
(while the maps are designed to show hotspots, the hotspots are displayed similarly to a
choropleth map, so the interpretability of these maps is similar to choropleth map interpretation).
The test questions asked participants to both discern the overall distribution of homicide hot
spots and cold spots and to identify particular time periods where certain areas experienced
particularly high or low homicide rates.
Performance measurements were based on each panel participant’s accuracy scores for
each set of map-based knowledge-extraction tasks and on the average amount of time it took
them to complete each task. Each participant answered three questions using the static maps as
the basis for their answers and three additional questions using the animated maps as the basis
for their answers. Tight experimental control was maintained by ensuring that the questions for
the static version of each cartographic technique corresponded very closely with the questions
25

for the animated version, without permitting the review participants to answer questions from
memory based on previous exposure. This helped to ensure that neither map type was
inadvertently put at a disadvantage. The map test consisted of six multiple-choice questions,
three for the static map series and three for the animation. The testing sequence was randomized
as much as possible while still ensuring that the testing sequence alternated between static and
animated questions in turn. See Appendix B for a detailed account of the survey interface,
including the test questions.

3.3.2 User-Interface Design
The user performance experiment was administered using Qualtrics, a web-based survey design
and distribution platform. This platform was chosen because it provided an array of advanced
tools for formatting questions, embedding images and videos at specific display sizes, and
tracking response times. The user-interface displayed the static maps and animations along with
test questions and radio buttons for the multiple choice responses that participants were asked to
provide. Time measurements were recorded from the time at which participants submitted the
previous page to the time they submitted the current page. Since each test question was on a
separate page of the survey, this resulted in completion time values for each participant and each
map test question.

3.3.2.1 Static Time-series Map Display Format
The static time-series maps were presented full-size, the same size as the animated map video
clip, approximately 6.5 inches tall by 8.5 inches wide on a single page. HTML was utilized to
ensure that the images were displayed the exact same size on all screens, irrespective of potential
26

variances in browser and screen size choices among study participants. A scroll-bar was
employed to allow study participants to easily scroll from one full-size map image/time-period to
the next. It also allowed for the map images to be displayed at a much larger scale, which helped
to avoid issues with label legibility.

3.3.2.2 Animated Time-series Map Display Format
The animated time-series were presented using a video clip embedded into the survey page,
much in the same way as the static version. Like the static maps, the video image was
approximately 6.5 inches tall by 8.5 inches wide. The imbedded video interface provided play,
pause and start over buttons, as well as a time-slider with which users were able to seamlessly
navigate the temporal extent of the series as necessary to complete the assigned tasks.
The time-slider helped to avoid problems with split-attention, one of the cognitive issues
commonly associated with animated maps discussed in Chapter 2, by utilizing the temporal
labeling of the animation itself as the temporal legend for navigating between time periods.
Additionally, Adobe Premiere’s cross-dissolve feature, which gradually transitions from one
frame to the next, was employed to help prevent participants from being surprised by the
transition between animation frames, which can negatively influence comprehension and
memorization. See figure 3.4 on the next page for an image of the animated time-series map
display interface.
27

Figure 3.3 Animated Time-series Map Display Format. See Appendix B for larger image.

3.3.3 Recruitment of Participants
Participants were recruited via Qualtrics Panels. This service allows the researcher to define a set
of criteria for selecting participants, as well as criteria for determining which participants and
their responses should be included in the final analysis, based on completion time and various
other measures of response quality. In total, 1,300 participants were recruited. To be included in
the study, respondents simply had to be eighteen or older and have some level of college
education (either in school or graduated). Due to screen-size requirements, participants were
excluded from the final analysis for accessing the survey on tablets or smartphones. A quota
system was used to ensure that respondents were 50 % male and 50% female. Attention
questions were dispersed throughout the survey to ensure that participants were paying attention
28

and trying to answer the test questions correctly. For example, one question simply requested
that respondents select the strongly disagree option. Any participants who failed to do so were
disqualified. Additionally, any submissions completed in less than 30% of the mean completion
time were excluded from the final analysis. Since participants were compensated for
participation in the study, these measures helped to weed-out any participants who might have
tried to game the system by rushing through the survey without actually trying to answer the
questions correctly. Out of the 1,300 recruited participants, fifty met all of the conditions for
inclusion in the final analysis. Once the fifty valid, in-quota, completes were collected, data
collection ceased.

3.3.4 Testing Procedures
Testing for this study was conducted via internet distribution, as described in the previous
section. Upon clicking the link to begin the survey, study participants were asked to begin the
pre-test by following the on-screen instructions.
The pre-test portion of the web-form provided participants with a brief tutorial on the
cartographic techniques that provided the basis for the map exercise they were asked to
complete. The pre-test also provided a brief description of the test format and collected some
basic demographic information from participants before they began the map test. No informed-
consent documentation was included in the pre-test, as this study was exempted from
Institutional Review Board oversight.
Following the pre-test, participants were informed that the map test was about to begin,
and that all map questions would be timed to measure task performance efficiency. Each page of
the map test consisted of a static time-series or animation followed by a content question and a
follow-up question regarding participants’ confidence in the accuracy of their response. The
29

web-form automatically recorded the time it took for participants to move from one page to the
next, as well as the timing of each click on each page.
Upon completion of the map test, participants were asked to answer a series of multiple-
choice Likert-scale questions that were designed to gauge user-preferences. Participants were
also given the opportunity to provide open-ended feedback or to describe their strategies for
completing the exercises if they desired. This concluded post-test and the exercise.

3.4 Methodology for Analyzing the Results of the Experiment
By comparing participants’ aggregate test scores and response times between the animated and
static map series, it was possible to examine the strengths and weaknesses of static and animated
maps in their application to choropleth map-based knowledge-extraction tasks. Additionally,
statistical analysis was employed to measure statistical significance and to better understand
patterns and relationships in the test results and feedback provided by study participants. The
first two research questions for this study are restated below as null and alternative hypotheses to
help provide context for the subsequent statistical analysis discussion. Following these is
another pair of null and alternative hypotheses for static and animated confidence scores. While
the confidence data were not used directly to answer the research questions for this study, they
were still included in the statistical analysis.

There is no significant difference between static and animated mean accuracy scores:
H0: µstatic accuracy score = µanimated accuracy score
There is a significant difference between static and animated mean accuracy scores:
H1: µstatic accuracy score ≠ µanimated accuracy score
30

There is no significant difference between static and animated mean completion times:
H0: µstatic completion time = µanimated completion time
There is a significant difference between static and animated mean completion times:
H1: µstatic completion time ≠ µanimated completion time
There is no significant difference between static and animated mean confidence scores:
H0: µstatic confidence score = µanimated confidence score
There is a significant difference between static and animated mean confidence scores:
H1: µstatic confidence score ≠ µanimated confidence score

3.4.1 Statistical Analysis Methodology
The statistical analysis conducted for this study consisted of basic independent samples (two-
tailed) t-tests and Pearson’s product-moment correlation coefficient calculations. The
independent samples t-test allows researchers to examine if the means of two different data sets
are significantly different from each other. The t-tests were utilized to compare the mean test
scores, mean confidence levels, and mean completion times between the static and animated map
series, and to measure the statistical significance of these findings. These comparisons provided
the information necessary to determine whether the static or animated condition better facilitated
accurate and efficient retrieval of the information needed to complete the assigned tasks. They
also helped to determine whether one visualization method inspired more confidence in response
accuracy than the other. The Pearson’s product-moment correlation coefficient allows the
31

calculation of correlation by dividing the covariance of the two variables by the product of their
standard deviations. This results in a correlation coefficient between -1 and 1, with -1 indicating
total negative correlation, 0 indicating no correlation, and 1 indicating total positive correlation.
The Pearson’s product-moment correlation coefficient calculations were conducted to determine
whether or not user-preferences are correlated with accuracy or efficiency measurements, and to
measure the statistical significance of these findings.

32

CHAPTER 4: RESULTS AND ANALYSIS

This chapter summarizes the results of the user performance experiment and discusses the results
of the statistical analyses that were conducted on the survey data. The findings of the experiment,
which are the focus of this chapter, are divided into six sections. Section 4.1 discusses the
experimental population. Section 4.2 discusses overall performance metrics. Section 4.3
discusses task accuracy. Section 4.4 discusses participants’ confidence in their responses. Section
4.5 discusses completion-time. Section 4.6 discusses user-preferences. Finally, Section 4.7
discusses the relevant statistical correlations between these different variables.

4.1 Experimental Population
Approximately thirteen-hundred individuals agreed to take part in the study. The vast majority
of these individuals were screened out because they didn’t meet all of the requirements for
inclusion in the final analysis. Many of the excluded responses were screened out for not meeting
the college education requirement. Some were screened out for inadequate screen size or because
their Flash Player version was out of date. Several more were screened out because they failed to
enter the correct answers for the preliminary attention questions as well. Fifty-eight participants
completed the entire survey. Of these, eight failed to pass the second series of attention-filters
(questions designed to test whether participants are paying attention) and were disqualified from
the final analysis, leaving fifty complete responses, twenty-five male and twenty-five female.
All fifty respondents were college-educated and between the ages of eighteen and sixty-eight.
See Tables 4.1 and 4.2, respectively, for more details on the age and level of education of study
participants. Each participant answered three static map questions and three animated map
questions. They also answered six confidence questions. These questions were asked after each
33

of the six static or animated map questions. They also answered a series of questions that was
intended to measure user-preferences between the static and animated map series.

Table 4.1 Age Distribution
Age Response %
18-25 1 2%
26-34 9 18%
35-54 21 42%
55-64 14 28%
65 or over 5 10%
Under 18 0 0%
Total 50 100%

Table 4.2 Education

Level of Education Response %
Less than High School 0 0%
High School / GED 0 0%
Some College 11 22%
2-year College Degree 4 8%
4-year College Degree 21 42%
Master’s Degree 9 18%
Doctoral Degree 1 2%
Professional Degree (JD, MD) 4 8%
Total 50 100%
34

4.2 Overall Performance Metrics
Table 4.3 shows the descriptive statistics for the accuracy scores, confidence scores, completion
times and user-preferences. These scores and completion time values are grouped according to
map type (static and animated). The accuracy scores represent how participants answered the
static and animated map questions correctly. If a participant answered a question correctly, he or
she received 1 point. Then, his or her total score was calculated by map type. The total score was
averaged for further analysis. The confidence measurements were handled in essentially the
same way. Confidence scores, however, were recorded out of three points possible. Each of the
confidence questions has three scales of confidence: not confident, somewhat confident, and very
confident. These scales were assigned the values of 1, 2, and 3 respectively to quantify users’
confidence. Then, each user’s scores were added together by map type and averaged. For the
timing values, the number of seconds it took each participant to answer each test question was
added together by map type. Then, the totaled timing values were averaged. Users’ preferences
between the static and animated maps were measured through the Likert-scale questions. For
each of a series of statements, participants were asked to indicate their level of agreement. The
six scales of agreement were very strongly disagree, strongly disagree, disagree, agree, strongly
agree and very strongly agree. These scales were assigned values of 1 to 6, respectively. The
three preference scores for each map type were then averaged together. This quantification
enabled the author to calculate user-preference scores for each map type.
The static map series afforded users slightly higher accuracy scores, confidence scores,
and preference scores while greatly reducing completion time. No substantial differences were
discovered in test scores, confidence scores, or completion times between male and female
35

respondents. The age distribution of the sample population was not sufficiently diverse to
warrant an analysis of differences in performance metrics between age groups.
The mean accuracy score for the static map series was .826 out of 1 (82.6%), while the
mean accuracy score for the animated map series was .733 out of 1 (73.3%). The mean
confidence score for the static map series was 2.59 out of 3, while the mean confidence score for
the animated series was 2.48 out of 3. The mean completion time for the static map series was
50.64 seconds, while the mean completion time for the animated map series was 67.94 seconds.
Finally, the mean preference score for the static map series was 4.05 out of 6, while the mean
preference score for the animated series was 3.34 out of 6. The results of the descriptive and
inferential statistics are discussed in greater detail in the following sections.

Table 4.3: Accuracy, Confidence, Completion Time, and User-preferences
N Minimum Maximum Mean Std. Deviation
Static Accuracy Score

50 0.333 1.000 0.827 .226
Animated Accuracy Score

50 0.333 1.000 0.733 .243
Static Confidence Score 50 2.000 3.000 2.593 .352
Animated Confidence Score

50

1.666

3.000

2.480

.331
Static Completion Time 50 21.049 94.559 50.636 16.818
Animated Completion Time

50

16.666

158.018

67.944

30.339
Static Preference Score

50 1.333 6.000 4.053 2.550
Animated Preference Score

50 1.000 5.667 3.347 1.017

36

4.3 Task Accuracy
The mean accuracy score for the static maps was 82.6% while the mean accuracy score for the
animated maps was 73.3%. The static scores are somewhat more skewed than the animated
scores, while the animated scores are somewhat more kurtotic. Table 4.4 summarizes the
relevant descriptive statistics for the static and animated map test scores. Figure 4.1 contains
histograms of the static and animated map test scores.

Table 4.4 Test Score Descriptive Statistics

Static Accuracy
Score
Animated
Accuracy Score
Mean 0.827 0.733
Median 1.000 0.667
Mode 1.000 0.667
Std. Deviation .226 .243
Skewness -.951 -.330
Std. Error of Skewness .337 .337
Kurtosis -.238 -1.023
Std. Error of Kurtosis .662 .662
Minimum 0.333 0.333
Maximum 1.000 1.000
37

Figure 4.1 Static and Animated Accuracy Score Histograms
38

To test the hypothesis that the static and animated maps afforded users statistically
significantly different mean accuracy scores, an independent samples (two-tailed) t-test was
performed. As can be seen in Table 4.4 the animated and static map series test scores are
sufficiently normally distributed for conducting a t-test (i.e., skew < |2.0| and kurtosis < |9.0|;
Bayer, Buhner, Danay, Schmider, and Ziegler 2010). Additionally, the assumption of
homogeneity of variances was tested and satisfied via Levene’s F-test F (98) = .005, p = .945.
The t-test was associated with a statistically significant effect, t (98) = 1.990, p = .049. Thus,
study participants achieved higher test scores with the static map series than with the animated
map series, with a mean difference of approximately 9.3%.

39

4.4 Confidence
The mean confidence score for the static maps was 2.59 out of 3 while the mean confidence
score for the animated maps was 2.48 out of 3. The static confidence scores are somewhat less
skewed and more kurtotic than the animated confidence scores. Table 4.5 summarizes the
relevant descriptive statistics for the static and animated confidence scores. Figure 4.2 contains
histograms of the static and animated confidence scores.

Table 4.5: Confidence Score Descriptive Statistics

Static Confidence Animated
Confidence
Mean 2.593 2.480
Median 2.667 2.667
Mode 3.000 2.667
Std. Deviation .352 .331
Skewness -.297 -.478
Std. Error of Skewness .337 .337
Kurtosis -1.137 .005
Std. Error of Kurtosis .662 .662

40

Figure 4.2 Static and Animated Confidence Score Histograms
41

To test the hypothesis that the static and animated maps afforded users statistically significantly
different mean confidence scores, an independent samples (two-tailed) t-test was performed. As
can be seen in Table 4.5, the animated and static map series confidence scores are sufficiently
normal for conducting a t-test (i.e. skew < |2.0| and kurtosis < |9.0|). Additionally, the
assumption of homogeneity of variances was tested via Levene’s F-test and was not confirmed,
thus equal variances were not assumed, F (98) = .578, p = .049. The t-test was not associated
with a statistically significant effect t (98) = 1.659, p = .100. Thus, the confidence scores for the
static map series were not significantly different from the mean confidence scores for the
animated map series.

4.5 Completion Time
The static maps were associated with a mean completion time of 50.64 seconds. By comparison,
the animated maps were associated with a numerically higher mean completion time of 67.94
seconds. The mean static completion times were slightly less skewed and somewhat less kurtotic
than the mean animated completion times. Table 4.6 summarizes the relevant descriptive
statistics for the static and animated completion times. Figure 4.3 contains box-plots of static and
animated completion times, respectively. Figure 4.4 contains histograms of static and animated
completion times, respectively.

42

Table 4.6 Static and Animated Completion Time

Static Completion Time Animated Completion
Time
Mean 50.636 67.945
Median 50.467 61.212
Mode 21.050 16.667
Std. Deviation 16.818 30.339
Skewness .517 .831
Std. Error of Skewness .337 .337
Kurtosis -.048 .532
Std. Error of Kurtosis .662 .662
Minimum 21.050 16.667
Maximum 94.559 158.018

Figure 4. 3 Static and Animated Completion Time Box-plots

Minimum:1
First
Quartile:
49.01
Median:
61.21
Third
Quartile:
82.88
Maximum:
158.02
0 50 100 150
Animated Completion Time (seconds)
Minimum:
21.05
Median:
50.47
First
Quartile:
39.63
Third
Quartile:
61.09
Maximum:
94.56
0 50 100 150
Static Completion Time (seconds)
43

Figure 4.4 Static and Animated Completion Time Histograms (in seconds)

44

To test the hypothesis that the static and animated maps afforded users statistically
significantly different mean completion times, an independent samples (two-tailed) t-test was
performed. As can be seen in Table 4.6 the animated and static map series completion times are
sufficiently normally distributed for conducting a t-test (i.e. skew < |2.0| and kurtosis < |9.0|).
Additionally, the assumption of homogeneity of variances was tested via Levene’s F-test and
was not confirmed, thus equal variances were not assumed, F (98) = 11.79, p = .001. The t-test
was associated with a statistically significant effect t (98) = -3.528, p = .001. Thus, study
participants were able to complete the static map tasks in a significantly shorter mean time than
the animated map tasks, with a mean difference of 17.31 seconds.

4.6 User-Preferences
The mean static preference score was 4.05 out of 6, while the mean animated preference score
was 3.35 out of 6, resulting in a mean difference of .70 in favor of the static map series. Static
preference scores were slightly more skewed and substantially more kurtotic than animated
preference scores (though both were within acceptable regions for statistical analysis). Table 4.7
provides details on the descriptive statistics of user-preferences. Figure 4.5 contains static and
animated preference score histograms.
Table 4.7 User-preference Descriptive Statistics

Static Preference
Score
Animated
Preference Score
Mean 4.053 3.346
Median 4.000 3.333
Mode 3.667 3.00
Std. Deviation 0.850 1.017
Skewness -.263 .010
Std. Error of Skewness .337 .337
Kurtosis 1.358 .108
Std. Error of Kurtosis .662 .662
45

Figure 4.5 Static and Animated Preference Score Histograms
46

4.7 Correlations
Pearson’s product-moment correlation coefficient calculations were used to determine whether
test scores or completion times were positively or negatively correlated with user-preference
scores. Animated completion time and animated preference score were the most strongly
correlated of the pairs of variables, with a correlation coefficient of -0.165. Static accuracy score
and static preference score were the least strongly correlated of the different pairs of variables,
with a correlation coefficient of -0.001. As can be seen in Tables 4.8 and 4.9, however, no
correlations with significance values below .05 were found between these variables for the static
map test or for the animated map test. Thus, no statistically significant correlations were found.

Table 4.8 Static Performance and Preference Correlations

Static Accuracy
Score
Static Completion
Time
Static Preference
Score
Static Accuracy Score
Pearson Correlation 1 -.001 -.057
Sig. (2-tailed)

.993 .693
N 50 50 50
Static Completion Time
Pearson Correlation -.001 1 .126
Sig. (2-tailed) .993

.383
N 50 50 50
Static Preference Score
Pearson Correlation -.057 .126 1
Sig. (2-tailed) .693 .383

N 50 50 50

47

Table 4.9 Animated Performance and Preference Correlations

Animated
Accuracy Score
Animated
Completion Time
Animated
Preference
Score
Animated Accuracy Score Pearson Correlation 1 .001 .033
Sig. (2-tailed)

.997 .820
N 50 50 50
Animated Completion Time Pearson Correlation .001 1 -.165
Sig. (2-tailed) .997

.252
N 50 50 50
Animated Preference Score
Pearson Correlation .033 -.165 1
Sig. (2-tailed) .820 .252

N 50 50 50

48

CHAPTER 5: DISCUSSION AND CONCLUSIONS
This study endeavored to provide an empirical comparison of static and animated cartographic
representations of spatiotemporal phenomena in their application to basic choropleth map-based
knowledge-extraction tasks. To this end, this study examined map readers’ performance and
efficiency in completing choropleth map-based knowledge-extraction tasks, using static time-
series maps and animated maps that depict homicide patterns in the Chicago metropolitan area as
the basis for doing so. This study helped to provide valuable insight as to the strengths and
weaknesses of static time-series maps and animated maps as the basis for choropleth map-based
knowledge-extraction tasks. It also helped to determine which of these visualization methods
map users prefer for carrying out these knowledge-extraction tasks.
The results of the user performance experiment clearly indicate that the choice between a
static or animated display interface can greatly influence map-readers’ ability to read hot spot
maps accurately and efficiently. Generally, users were able to complete the assigned tasks more
accurately and much more efficiently using the static maps, as compared with their animated
counterparts. Interestingly, while the user-preference metrics, in aggregate, indicate that study
participants preferred the static maps over their animated counterparts, no significant correlations
were found between individual test scores and individual user-preferences. This suggests the
possibility that user-preferences are not based entirely on the practical application of the tools to
the assigned tasks.
Chapter five is devoted to further discussion of these findings. This discussion is divided
into five sections. Section 5.1 discusses task accuracy. Section 5.2 discusses users’ confidence in
the accuracy of their responses. Section 5.3 discusses completion time. Section 5.4 discusses user
preferences and finally, Section 5.5 discusses the conclusions drawn from this study, as well as
49

its limitations, its parallels with Socia’s study, its implications for the cartography community,
and the author’s suggestions for future research.

5.1 Task Accuracy
As discussed in Chapter 4, the static maps were associated with a mean test score of 83%, while
the animated maps were associated with a mean test score of 73%. While the difference in mean
test scores between the static and animated maps was fairly small, this difference was found to
be statistically significant. These findings indicate that, at least in the context of low temporal
resolution choropleth time series visualization, static maps are likely to better facilitate the
retrieval of accurate information than their animated counterparts.
These findings are bolstered by some of the feedback participants provided in the open-
ended questions at the end of the map test survey. Two participants indicated that they had to
rewind the animated version and replay it at least once in order to glean the information
necessary to complete the assigned tasks, whereas they were able to quickly and easily jump
back and forth between time-frames using the static map series. Based on the disparity between
accuracy scores for the static and animated maps, in conjunction with this user feedback, it
seems likely that some study participants answered the animated test questions incorrectly
because they failed to recall the information from previous frames in the video sequence
correctly. While this evidence is far from conclusive, it does lend credence to previous claims
(Betrancourt and Tversky 2002; Andrienko et al. 2008) that animations may be too fleeting or
have too many moving parts to be perceived accurately.

50

5.2 Confidence
The static maps were associated with a mean confidence score of 2.59 out of 3 while the
animated maps were associated with a mean confidence score of 2.48 out of 3. Interestingly,
these confidence scores correspond fairly closely with accuracy and efficiency measurements.
Accuracy scores and confidence scores were both higher for the static maps than for their
animated counterparts. Likewise completion times were shorter for the static maps than for their
animated counterparts. These findings, while not conclusive, suggest two possibilities; that
study participants were more likely to be confident in their responses when they had answered
the question correctly, or that the static maps inspired more confidence in response accuracy than
the animated maps, independently from correctness or completion time. The findings of the
correlation calculations support the latter conclusion. However, given that the correlation
calculations produced no statistically significant results, it is difficult to say for certain.

5.3 Completion Time
The static maps were associated with a mean completion time of 50.64 seconds. By comparison,
the animated maps were associated with a mean completion time of 67.94 seconds. It took study
participants 17.3 seconds longer, on average, to complete the animated map test questions than it
took them to complete the static map test questions. As detailed in Chapter 4, this disparity in
completion times was found to be statistically significant. In fact, the difference between static
and animated mean completion times is quite substantial. It took study participants
approximately 35% longer to answer the test questions using the animated maps, as compared
with their static counterparts. Based on these findings, it seems prudent to conclude that, at least
in the context of low temporal resolution homicide hot spot time-series maps, static maps are
51

likely to be more efficient than their animated counterparts. However, this may vary substantially
depending on the temporal resolution of the mapped data, as animation may begin to outperform
static time-series with increased temporal granularity.
There could be many potential explanations for the significant disparity in completion
times between the static and animated map series. Perhaps, as previous researchers have
suggested (Betrancourt and Tversky 2002; Andrienko et al. 2008), animations can be too
complex or too fleeting to be perceived accurately. This would explain why some participants
seemingly had to stop the sequence and replay it, or use the time-slider to navigate back to a
previous time period to retrieve the information necessary to complete the assigned task, thereby
increasing the completion time for that task. It could also be due to the complexity of the user-
interface for the animated maps. The buttons to play, pause and rewind the video sequence, as
well as the time-slider, added a degree of complexity to the user-interface. This might have
interfered with users’ efficient interaction with the animated maps.

5.4 User-Preferences
As detailed in Chapter 4, the mean static user-preference score was 4.05 out of 6, while the mean
animated preference score was 3.35 out of 6, resulting in a mean difference of .70 favoring the
static maps. It is not entirely clear why users tended to prefer the static maps over their animated
counterparts. One study participant indicated that they experienced difficulty navigating back
and forth between time periods as necessary to complete the assigned tasks. Most study
participants, however, reported no such difficulty. Two participants indicated that they had to
stop and rewind the video sequence to complete the assigned animated map interpretation task,
whereas they were able to complete the corresponding static map interpretation task more easily.
52

In these cases, it seems that preferences were closely associated with the practical application of
the static and animated map products to the tasks at hand. Based on the findings of the
correlation coefficient calculations between users’ map task performance metrics and quantified
preferences (which are the focus of the next section), however, it seems inappropriate to
generalize these findings to the entire sample population.

5.4.1 Performance and User-preference Correlations
As discussed in Chapter 4, no statistically significant correlations were found between accuracy
and user-preferences or between efficiency and user-preferences. In this instance, however, the
lack of any strong correlations between performance metrics and user-preferences is an
interesting finding in itself. One might expect to find a positive correlation between static test
scores and static preference scores, for example, based on the assumption that users are likely to
prefer the method that allows them to complete the task at hand most accurately and efficiently.
Surprisingly, however, no such correlations were found for the static maps or for their animated
counterparts. Two study participants indicated that they were displeased with the efficiency of
the animated map interface, but given that two participants only constitutes 4% of the sample
population, this sentiment was far from typical. As such, it seems that user-preferences, at least
in the context of this experiment, were largely independent from user performance metrics.

5.5 Conclusions
Overall, this study produced some very interesting results. Substantial differences were found
between static and animated performance metrics, largely favoring the static map series. This
study also found evidence to suggest that user-preferences are not strongly correlated with the
53

practical application of cartographic tools for specific knowledge-extraction tasks. This is
arguably the most interesting result of this analysis. Since maps are generally used as tools,
rather than for entertainment purposes, one might expect that map-readers would prefer the
cartographic tool that enables them to most effectively and efficiently complete the task at hand.
While this study does not contradict this notion directly, the fact that no statistically significant
correlations were found whatsoever between performance metrics and user preferences certainly
lends credence to previous claims that user-preferences, particularly in relation to animated
maps, might not be based entirely on the practical application of these tools for learning about
geographic features or spatial processes. Perhaps the growing popularity of animated maps really
is due to the ‘fun-factor’ that is commonly associated with innovative new cartographic products.

5.5.1 Parallels with Socia’s Study
Overall, the results of this study tend to reinforce Socia’s findings. In both instances, study
participants were able to complete the assigned tasks more accurately and more efficiently using
the static maps than they were able to using the animated maps as the basis for their answers.
Interestingly, neither study found a statistically significant difference between confidence levels
for the static maps and for the animated maps, despite the significant differences in performance
metrics.
The time-slider was introduced for this study on the basis that it might alleviate users’
operational difficulty and split-attention. The decision to include a time-slider, rather than skip-
to-time-stamp buttons, doesn’t seem to have had a substantial effect on task accuracy or
completion time. Accuracy scores were comparable between the two studies despite the
54

differences in animated map display interface. As such, it seems that the time-slider did not have
a substantial effect on user performance
One area where the two studies differed is in their findings on user-preferences. Socia’s
participants, despite lower test scores and slower response times, expressed a strong preference
for the animated maps over the static maps that provided the basis for her study. Participants in
this study, on the other hand, tended to prefer the static map series. User-preferences in this study
aligned better with user performance metrics than they did in Socia’s study. Participants in this
study tended to prefer the cartographic product that allowed them to complete the assigned tasks
most accurately and efficiently, whereas the same cannot be said for Socia’s study participants.
However, given that no significant correlations were found between performance metrics and
user preferences at the level of the individual participant, one would be ill-advised to conclude
that there is a causal relationship between accuracy or efficiency and user-preferences.

5.5.2 Study Strengths and Weaknesses and Suggestions for Future Research
Overall, this study was quite successful in addressing the research questions it set out to answer.
Statistically significant differences were revealed between static and animated accuracy scores
and between static and animated completion times. Overall, these findings suggest that static
maps are more effective and efficient for communicating complex spatiotemporal phenomena
like homicide patterns to an audience. While this study did not find any statistically significant
correlations between accuracy or completion time and user-preferences, the lack of correlations
are interesting findings in themselves. As such, this study unearthed evidence to support both of
its proposed hypotheses.
55

There were, however, a couple of areas that could be improved upon for future research.
One way to improve upon the methods for this study would be to include a comparison between
different levels of temporal granularity. For example, one map sequence could cover the 24
hours of the day in six four-hour segments, while another could cover the same 24-hour period in
twenty-four 1-hour segments. In this context, perhaps the animated maps would begin to
outperform the static maps as granularity is increased. To facilitate this comparison, two separate
sample populations would need to be tested. One for the low temporal granularity series and one
for the high temporal granularity series. Otherwise participants may become fatigued by the
length of the test. Another area where improvements could be made to this study is in the open-
ended qualitative user-preference questions. Responding to these questions was optional for this
study. As such, many study participants chose not to answer them. If the open-ended questions
had been required, it may have been possible to gain additional insights into the reasons behind
users preferences between the static and animated maps. At the outset of this study, the
researcher assumed that the correlation coefficient calculations between performance metrics and
user-preference metrics would be sufficient to reveal any potential relationships between these
different variables. In practice, however, these data did not reveal any significant correlations
between performance and user-preferences. Future research in this area should emphasize
detailed, open-ended personal accounts of the reasoning behind each participant’s preferences
between static and animated maps. It may also be interesting to inform participants of their
scores and completion times before asking them to indicate their preferences between the two
types of maps.

56

5.5.3 Implications for the Cartography Community
The primary implication of this research for the cartographic community is quite simple.
Animated choropleth maps just might not be practical for specific knowledge-extraction tasks, as
compared with their static counterparts. While animated maps can be very visually appealing,
this study demonstrated how animation can hinder the effective and efficient retrieval of specific
geographic information. In light of these findings, cartographers should consider the potential
costs and benefits associated with the choice between static and animated display interfaces very
carefully. It seems that animation may be best-suited to providing a very general overview a
geographic phenomenon, whereas static map series are much better suited to more specific
knowledge-extraction tasks. Cartographers should keep this in mind when deciding between
static and animated display interfaces.
Another interesting implication of this research is that user-preferences between static
and animated choropleth maps do not seem to be directly correlated with the practical application
of these tools for routine spatial knowledge-extraction. While this could be attributed to the
likelihood that some users just find interactive maps to be more entertaining, it also suggests the
possibility that many map-users do not fully understand the impact that animation has on their
ability to accurately perceive the information contained in a cartographic animation. The
disparity between static and animated confidence metrics in this study lends credence to this
notion. Participants indicated similar confidence levels for the static and animated tasks, despite
scoring substantially higher on the static map test. Perhaps cartographers should work against the
tide of popular opinion and consider the likely disconnect between performance and user-
preferences when designing maps. In some situations, users may prefer a certain type of
57

visualization, even when the cartographic design choices for that visualization have a markedly
negative impact on their ability to accurately and efficiently interpret its contents.

58

REFERENCES

Andrienko, G., N. Andrienko, S. Fabrikant, D. Montello, and S. Rebich-Hespanha. 2008.
“Novel Method to Measure Inference Affordance in Static Small-multiple Map Displays
Representing Dynamic Processes”. The Cartographic Journal. 45 (3): 201-215

Baldwin, S. 2014. Institutional Inscription in the Minorcan Quarter of Saint Augustine, Florida.
University of Southern California.

Bayer, L., M. Buhner, E. Danay, E. Schmider, and M. Ziegler. 2010. “Is it Really Robust?
Reinvestigating the Robustness of ANOVA against Violations of the Normal Distribution
Assumption”. European Journal of Research Methods for the Behavioral and Social
Sciences. 147-151.

Bekele, T., J. Cinnamon, M. Chipman, M. Cusimano, R. Glazier, T. Hernandez, M. Marshall, C.
Rinner. 2009. “Evaluating Web-based Static, Animated and Interactive maps for Injury
Prevention”. Geospatial Health. 4 (1): 24-46.

Betrancourt, M., B. Tversky. 2002. “Animation: Can it Facilitate?” International Journal of
Human-Computer Studies. 57: 247-262.

59

Blok, C., S. Fuhrman, N. Hedley, B. Jiang, A. Koussoulakou, D. Montello, T. Slocum. 2001.
“Cognitive and Usability Issues in Geo-visualization”. Cartography and Geographic
Information Science. 28 (1):, 61-75.

Board, C. 2011. “Cartographic Communication”. The Map Reader: Theories of Mapping
Practice and Cartographic Representation. 1: 37-47.

Boscoe, F. P., D. Haug, A. MacEachren, L. Pickle. 1999. Geographic Visualization: Designing
Manipulable Maps for Exploring Temporally Varying Georeferenced Statistics.
Pennsylvania State University.

Bowers, K. J., S. Johnson. 2004. “The stability of space-time clusters of burglary”. British
Journal of Criminology.

Chainey, S. J. Ratcliffe. 2005. GIS and Crime Mapping. Wiley Publications.

City of Chicago. 2014. Crime Incidents 2001 to Present. Taken from:
www.data.cityofchicago.org accessed on 7-1-2014.

Coltekin, A., S. Fabrikant, S. Garlandini, B. Heil. 2009. “Evaluating the Effectiveness of
Interactive Map Interface Designs: A Case Study with Eye-Movement Analysis”.
Cartography and Geographic Information Science. 36 (1): 5-17.

60

DiBiase, D., J. Krygier, A. MachEachren, C. Reeves. 1992. “Animation and the Role of Map
Design in Scientific Visualization”. Cartography and Geographic Information Systems.
19 (4): 201-214, 265-266.

Fabrikant, S. 2005. “Towards an Understanding of Geovisualization with Dynamic Displays:
Issues and Prospects”. Proceedings, American Association for Artificial Intelligence
2005 Spring Symposium Series: Reasoning with Mental and External Diagrams.
Stanford University, Mar. 21-23, 2005.

Fabrikant, S., M. Harrower. 2008. The Role of Map Animation for Geographic Visualization.
Geographic Visualization: Concepts, Tools, and Applications. John Wiley and Sons, Ltd.

Grubesic, T. H., E. Mack. 2008. “Spatiotemporal Interaction of Urban Crime”. Quant
Criminol. 24: 285-306.

Harrower, M. 2007. “The Cognitive Limits of Animated Maps”. Cartographica. 42 (4): 349-
357.

Huffington Post. 2013. Fourth Of July Violence in Chicago: Homicide Rate Passes 200 after
Long Holiday Weekend of Shootings. Taken from:
http://www.huffingtonpost.com/2013/07/07/fourth-of-july-violence-c_n_3558463.html
Accessed on 6/1/2014.
61

Kessler, F. C., H. Howard, B. McMaster, T. Slocum. 2009. Thematic Cartography and
Visualization. Pearson, Prentice Hall.

Kraak, M., A. MacEachren. 2001. “Research Challenges in Visualization”. Cartography and
Geographic Information Science. 28 (1): 3-12.

Larkin, J., H. Simon. 1987. “Why a Diagram is (Sometimes) Worth a Thousand Words”.
Cognitive Science. 11: 65-69.

Nakaya, T., K. Yano. 2010. “Visualizing Crime Clusters in a Space-Time Cube: An
Exploratory Data-Analysis Approach Using space-time Kernel Density Estimation and
Scan Statistics”. Transactions in GIS. 14 (3): 223-239.

Ratcliffe, J. 2010. “Crime Mapping: Spatial and Temporal Challenges”. Handbook of
Quantitative Criminology. 5-23.

Shepard R N, 1967. “Recognition memory for words, sentences and pictures.” Journal of Verbal
Learning and Verbal Behavior. 6: 156-163.

Socia, K. M. 2011. Small-Multiples and Animation: Measuring User Performance with Wildfire
Visualization. Michigan State University.

62

Thrower, N. 1959. “Animated Cartography.” The Professional Geographer. 11 (6)

Tufte, E. 1995. Envisioning Information. Graphics Press.

Vito, G., W. Walsh. 2004. “The meaning of Compstat Analysis and Response”. Journal of
Contemporary Criminal Justice, 20 (1): 51-69. Sage Publications.

63

APPENDIX A: CHICAGO HOMICIDE HOT SPOT MAPS, 2009-2013

64

65

66

67

68

.
70

APPENDIX B: IMAGES OF MAP TEST SURVEY INTERFACE

71

72

73

74

75

76

77

*This image was shrunken down to fit on this page. The display size was adjusted to account for
the empty space around the margins so that it was displayed the same size as the static maps in
the survey interface.

78

79

80

81

82

83

84

*The static map series was displayed again here in the survey interface but has been omitted
here.

85

*The animated map series was displayed again here in the survey interface but has been omitted
here.

86

*The static map series was displayed again here in the survey interface but has been omitted
here.

87

*The animated map series was displayed again here in the survey interface but has been omitted
here.

88

89

Abstract (if available)

Abstract This study provided an empirical comparison of static and animated cartographic representations of spatiotemporal phenomena in their application to basic choropleth map‐based knowledge‐extraction tasks to answer the following research questions: 1) Do animated maps provide heightened potential for accuracy in completing basic knowledge‐extraction tasks over static time‐series maps, or vice versa? 2) Do animated maps provide heightened potential for efficiency in completing basic knowledge‐extraction tasks over static time‐series maps, or vice versa? and 3) How do user preferences align or not align with measurements of accuracy and efficiency? ❧ To this end, this study examined map readers’ accuracy and efficiency in completing knowledge‐extraction tasks through static and animated time‐series maps about homicide patterns in the Chicago metropolitan area. Through an online user performance experiment, participants answered a series of questions about homicide hot spots and cold spots using both static and animated versions of the maps as the basis for their answers. They were also asked to indicate their level of confidence in the accuracy of their responses and to indicate which map type they preferred for completing the tasks. Task completion times were recorded for efficiency measurements. The results of independent samples t‐tests indicate statistically significant differences between the static and animated maps in terms of task accuracy and completion time. Generally, users were able to complete the assigned tasks more accurately and much more efficiently using the static maps, as compared with their animated counterparts. Additionally, user‐preferences were checked for correlations with task accuracy and completion time via Pearson’s product‐moment correlation coefficient calculations. The results indicate no significant correlations between performance measurements and user‐preferences.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Visualizing historic space through the integration of geographic information science in secondary school curriculums: a comparison of static versus dynamic methods

PDF

A user study of GIS infused genealogy with dynamic thematic representation and spatiotemporal control

PDF

GIS-aided production of equipment locator maps for Metro rail maintenance and support

Asset Metadata

Creator Anderson, Benjamin Green (author)

Core Title Cartographic approaches to the visual exploration of violent crime patterns in space and time: a user performance based comparison of methods

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 04/14/2015

Defense Date 03/18/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag animated maps,homicide incident visualization,map interpretation,OAI-PMH Harvest,static time‐series maps,user performance

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Oda, Katsuhiko (Kirk) (committee chair), Chiang, Yao-Yi (committee member), Vos, Robert O. (committee member)

Creator Email bganders@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-546770

Unique identifier UC11297849

Identifier etd-AndersonBe-3289.pdf (filename),usctheses-c3-546770 (legacy record id)

Legacy Identifier etd-AndersonBe-3289.pdf

Dmrecord 546770

Document Type Thesis

Format application/pdf (imt)

Rights Anderson, Benjamin Green

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA