DOT/FAA/AM-07/26

Office of Aerospace Medicine

Washington, DC 20591

Information Complexity in
Air Traffic Control Displays

Jing Xing
Civil Aerospace Medical Institute
Federal Aviation Administration
Oklahoma City, OK 73125

September 2007

Final Report
 NOTICE

This document is disseminated under the sponsorship

of the U.S. Department of Transportation in the interest
of information exchange. The United States Government
assumes no liability for the contents thereof.
___________

This publication and all Office of Aerospace Medicine

technical reports are available in full-text from the Civil
Aerospace Medical Institute’s publications Web site:
www.faa.gov/library/reports/medical/oamtechreports/index.cfm
 Technical Report Documentation Page
1. Report No. 2. Government Accession No. 3. Recipient's Catalog No.
DOT/FAA/AM-07/26
4. Title and Subtitle 5. Report Date
Information Complexity in Air Traffic Control Displays September 2007
6. Performing Organization Code

7. Author(s) 8. Performing Organization Report No.

Xing J
9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)

FAA Civil Aerospace Medical Institute

P.O. Box 25082 11. Contract or Grant No.

Oklahoma City, OK 73125

12. Sponsoring Agency name and Address 13. Type of Report and Period Covered

Office of Aerospace Medicine

Federal Aviation Administration
800 Independence Ave., S.W.
Washington, DC 20591 14. Sponsoring Agency Code

15. Supplemental Notes

Work was accomplished under approved task HRRD522

16. Abstract

Air traffic controllers typically use visual displays to interact with various automation systems. Automation tools
are intended to reduce controller task load, but they may also create new tasks associated with acquiring,
integrating, and utilizing information from displays. Consequently, the complexity of information displayed may
reduce the efficiency and effectiveness of an automation system. Moreover, complexity could cause controllers to
miss or misinterpret visual data, thereby reducing safety. Thus, information complexity in air traffic control
(ATC) displays represents a potential bottleneck in ATC systems. To evaluate the cost and benefit of an
automation system, it is important to understand whether the information it provides is too complex for
controllers to process. The purpose of this study was to answer three basic questions: 1) What constitutes
information complexity in automation displays? 2) What level of display complexity is “too complex” for
controllers? 3) Can we objectively measure information complexity in ATC displays? In this study, we first
developed a general framework for measuring information complexity. The framework reduces the concept of
complexity into three underlying factors: quantity, variety, and the relations between basic information elements;
each factor is evaluated at three generic stages of human information processing: perception, cognition, and action.
By this definition, we decompose complexity into a 3x3 matrix, measuring the effects of a complexity factor on
information processing at a given stage. We then take the following steps to develop complexity metrics for ATC
displays: 1) Identify task requirements of using the displays in ATC; 2) Determine corresponding brain functions
pertinent to the task requirements; and 3) Choose the metric that can measure the effects of the complexity
factor on the brain functions. Using this approach, we developed nine metrics of ATC display complexity. These
metrics provide an objective method to evaluate automation displays for acquisition evaluation and design
prototypes.
17. Key Words 18. Distribution Statement
Information Complexity, Display, Interface Design, Document is available to the public through the
Evaluation, Air Traffic Control Defense Technical Information Center, Ft. Belvior, VA
22060; and the National Technical Information
Service, Springfield, VA 22161
19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price
Unclassified Unclassified 19
Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

i
 ACKNOWLEDGMENTS

The author thanks Dr. Carol Manning for her consultations throughout this study. Thanks, also,
to Drs. Lawrence Bailey, Ben Willems, Pam Della Roccco, and Earl Stein for their suggestions and
advice. Additional thanks go to Dino Piccione, Steve Cooley, and Paul Krois for their support of
the Federal Aviation Administration’s requirements for information complexity research. Finally,
the author expresses appreciation for the Federal Aviation Administration internal reviewers, whose
comments greatly improved the quality of this paper.

iii
 CONTENTS

INTRODUCTION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
IDENTIFYING COMPLEXITY METRICS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Understanding Information Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
A Framework for Decomposing Factors of Information Complexity . . . . . . . . . . . . . . . . . . . . . 3
Metrics of Information Complexity for ATC Displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Perceptual Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Cognitive Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Action Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
A PRELIMINARY CASE STUDY: ASSESSING THE COGNITIVE COMPLEXITY OF
THE MICROSOFT POWERPOINT INTERFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Representational Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Processing Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Perceptual Complexity in the Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Comparisons With Other Measures of Cognitive Complexity in the Literature . . . . . . . . . . . 12
Relevant Measures of Action Complexity in the Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

v
 INFORMATION COMPLEXITY IN AIR TRAFFIC CONTROL DISPLAYS

INTRODUCTION amounts of information they cannot use. Thus, to have

a new technology operate safely, efficiently, and at its full
“This thing is too complex to use!” In an age of infor- performance potential, its complexity should be mini-
mation technology, most new technology users have mized within its operational scope. Before the industry will
either said or heard this phrase at one time or another. be able to reliably control the complexity of new systems,
Over the past several decades, computer technologies it will need an objective way to measure the complexity
have achieved enormous progress in the speed at which associated with new information technologies.
they can process and present information. With these The purpose of this study was to develop metrics
processing limitations overcome, the new challenge for of information complexity that can be applied to the
developers of new technologies is to provide systems evaluation of ATC displays. For it to be applicable, the
with more types of information to process. Such efforts complexity measures need to meet the following require-
result in information being presented in complex ways. ments: 1) quantitatively correlated with performance
Conversely, human information processing capabilities measures; 2) independent of specific ATC displays; and
have not experienced the same level of improvement. 3) as much as possible, independent of users’ experience
Therefore, the complexity of information provided by new with a display. As an initial attempt, Xing and Manning
technologies can easily exceed human capacity limitations. (2005) performed an extensive review of previous com-
As a result, complexity has become a big concern in the plexity studies and analyzed the findings of these studies
information technology industry. A survey (Economist, for their potential application to ATC displays. It was
Oct. 28, 2004) indicated that information complexity hoped that previously developed measures could be found
was holding the industry back as many technologies that would meet the purpose and requirements stated
were not implemented because of their complexity. above. Instead, they found that most of the measurements
One survey respondent stated, “66% of all IT projects of complexity focused either on the display system or
either fail outright or take much longer to install than on the information processing of the human operator,
expected because of their complexity.” As the complexity while seldom addressing both. A complete description of
of technology increases, its usefulness decreases, because it complexity for ATC technologies needs to consider the
becomes “too complex to use.” Also, the efficiency of using factors involved with both the human operators and the
these technologies decreases, because managing the use display systems. The literature analysis concluded that, for
of complex systems often increases rather than mitigates ATC display applications, a measurement that integrated
workload. Furthermore, high levels of complexity can both human information processing and display factors
result in an operator missing or misunderstanding even would need to be developed.
critical pieces of information. These types of performance The purpose of this study raised three basic questions:
errors are of particular concern to those work domains What is complexity? How complex is too complex for
where safety is mission-critical. users? Finally, is it possible to quantitatively measure the
Unfortunately, as information technology has become complexity of ATC displays? We organized this paper
an inescapable part of the aviation system, the aviation into two parts to address the above questions. In the first
industry cannot avoid the problem of information com- part, “Identifying Complexity Metrics,” we summarize
plexity. In particular, air traffic control (ATC) work is our understanding of information complexity from a
performed in a dynamic environment where controllers literature analysis. Then we describe a framework that
continuously receive large volumes of information from decomposes complexity into a set of factors associated
multiple sources; they must monitor this information for with human information processing. After that, we
changes in the environment, and they must make deci- present a set of complexity metrics developed for ATC
sions and perform effective actions in a timely manner. As displays. Parts of the preliminary version of these results
technological capabilities have developed over time, there have been published in an earlier report (Xing, 2004).
has been a tendency to add information to ATC displays In the second part, “A Preliminary Case Study: Assessing
and to automate controller functions. Although these new the Cognitive Complexity of the Microsoft PowerPoint
technologies are meant to improve ATC performance, they Interface,” we present a preliminary case study to dem-
can also increase the information complexity with which onstrate how the metrics can be used to evaluate the
the controller has to contend. The last thing controllers complexity of a display.
need from new technologies is to be overloaded by large
1
 IDENTIFYING COMPLEXITY and other researchers have pointed out, complexity lies
METRICS somewhere between order and disorder (Drozdz, Kwapien,
Speth, & Wojcik , 2002). Burleson and Caplan (1998)
Methods summed up the use of variety for defining complexity
We first derived a definition of information complex- when they stated, “The concept of complexity refers to
ity from an analysis of the literature. We next developed diversity of forms, to emergence of coherent patterns
a framework for measuring complexity. The framework out of randomness and also to some ability of frequent
is generally applicable to interactive visual displays. The switching among such patterns.”
framework specifies that complexity is constrained by Relations. Relations among the basic elements (rules
task requirements. We then applied the framework to of structures, interconnections, etc.) of a system also
ATC displays and developed a set of complexity metrics contribute to its complexity. Individual parts of a system
by combining the mechanisms of human information are held together by the relations of its internal structure.
processing with ATC task requirements. An example would be a chess pattern. A chess pattern
can be of great complexity to a player because the player
RESULTS values the relations between the elements, not just the
number and the variety of them.
Understanding Information Complexity Xing and Manning also identified two principles that
Although there are many definitions of complexity in contribute to the diversity of complexity measures. The
the literature, the term has proven to be very difficult to first principle is observer dependency. As Edmonds (1999)
accurately define. For instance, a simple Internet search described, “Complexity only makes sense when consid-
on complexity will yield literally hundreds of definitions ered relative to a given observer.” One example would be
and measures. Xing and Manning (2005) reviewed and the experiment performed by Grassberger (1991) where
synthesized the major contributions to complexity associ- subjects were asked to assess the complexity of a set of
ated with visual displays. In their report, they reviewed images. Figure 1 shows three images that Grassberger used
the literature from several lines of study: general concepts, in the experiment. The variety of the images, measured
information complexity, cognitive complexity, and display as the disorder of image pixels, increased from left to
complexity. While each of these areas is focused on differ- right. Thus, the image on the right is the most complex
ent aspects of human or machine systems, the definitions from the perspective of computer image processing. Yet,
have a great deal in common. Essentially, all the reviewed human subjects perceived the middle one as the most
definitions and measures converged on three factors as-
sociated with complexity: quantity of basic information
elements in a system, variety of elements, and the relations
between elements. Xing and Manning’s analysis revealed
that the concept of complexity is multi–dimensional and
cannot be sufficiently described with a single measure,
and they proposed that complexity is the combination
of the three basic factors. Figure 1: Variety increases from left to right, yet
Quantity. Intuitively, the quantity of basic elements is humans perceives the middle one the most complex.
related to the complexity of a system. Whether referring
to minimum description size of a system (Bennett, 1990; complex. The reason is that the human visual system
Crutchfield & Young, 1989), number of states (McCabe, processes visual features within small local areas rather
1976), or number of “chunks” in cognition (Klemola, than individual pixels, per se. Those features include line
2000), all studies of quantity seem to agree that a larger orientation, spatial frequency, luminance contrast, color,
quantity corresponds to a higher degree of complexity. etc. The image in the middle of Figure 1 contains many
Nevertheless, quantity alone is not sufficient to define distinctive visual features, while the left and right images
complexity in its entirety. are composed of homogeneous texture with little visual
Variety. Indeed, the variety of the elements in a system features. This experiment indicates that the effects of
is also a key component of many definitions of complex- variety on complexity depend on how observers process
ity. The concept of variety has been widely used in the information.
literature. For instance, many researchers have used the The second principle is task dependency. That is, the
degree of disorder or entropy in information theories as the complexity of things depends on which aspect you are
measure for variety or complexity, even though disorder concerned with. For example, if the task is to count peas
alone cannot sufficiently describe complexity. As Drozdz in a basket, then the complexity of the peas does not vary
2
with the number of the peas and variations in the shape with neuronal activities, and the activities end with the
or color of the peas. Therefore, it is essential to determine offset of the stimulus. However, after the stimulus ends,
the task requirements of a visual display before assessing the neuronal activity for cognitive processing continues
its complexity. until an action or a plan for actions is made and executed.
To summarize our understanding of information com- In this sense, our model is similar to Wickens’ model
plexity, we generalized the following definition from the for interface evaluation, where information is processed
literature: Complexity is the combination of three basic at four sequential stages: perception, working memory,
factors: quantity, variety, and relation of basic elements; decision, and action (1991).
all three of which are evaluated by the mechanisms of While our three-stage classification of brain functions
observers’ information processing and constrained by is coarse, those stages have intrinsically different mecha-
task requirements. nisms of information processing. One example would be
the neuronal response patterns over time, as illustrated
A Framework for Decomposing Factors of Informa- in Figure 2. Perceptual neuronal responses (left panel)
tion Complexity start following the onset of a stimulus and ends after the
Since complexity depends on how observers process stimulus is no longer present, while the cognitive responses
information, we looked into the mechanisms of informa- (middle panel) remain for an extended period without
tion processing in the human brain. Based on the extensive the stimulus, such sustained activity is the substrate of
literature, we outlined a conceptual model of human working memory. On the other hand, neuronal responses
information processing associated with the use of visual in the cortical pre-motor areas become activated before an
displays. In this simple model, information presented action and end after the action plan is executed. Another
via visual display devices is processed in three stages: example is the way in which information is encoded.
perception, cognition, and action. Through perception, Perceptual information processing is initially performed
a user acquires visual features of displayed information. in a parallel manner. Thousands of visual neurons are
The perceived information then feeds into the cognition activated by a visual image and they simultaneously
stage, where one’s perception is integrated with informa- encode many pieces of the image. Thus, the perceptual
tion from long-term memory, and an internal (mental) system offers a relatively broad information bandwidth.
representation of the display is generated. Based on this On the other hand, working memory, as the basis of
representation, users can then make action plans to use cognitive activities, has a highly limited capacity. That is,
the information or interact with the display. only a few pieces of information can be simultaneously
Given that several hundreds of brain areas have been encoded (Cowan, 2001). Consequently, the information
identified for special aspects of information processing, bandwidth of the cognition stage is much less than that
the above model is over-simplified and we only use it for perception. Finally, the cortical areas that encode
for application purposes. The definitions of perception action plans are characterized with “one plan at a time,”
and cognition have been controversial; however, opin- yielding an even narrower bandwidth (Georgopoulos,
ions about the basic distinction between perception Schwartz, & Kettner, 1986; Pouget, Zemel, & Dayan,
and cognition are more consistent among researchers: 2000; Xing & Andersen, 2000).
Perception does not necessarily involve working memory Given the inherent differences in the three stages, the
while cognitive activities are based on working memory. three complexity factors should be separately evaluated at
At the perception stage, the brain responds to a stimulus each stage. This results in a 3x3 matrix, as shown in Table

Perception Cognition Action

Action execution
Brain activity

Stimulus
Working memory

Time

Figure 2: Activity patterns over time for information processing in the three stages.
3
 Table 1: Metrics of information complexity

Perception Cognition Action

Quantity

Variety

Relation

1, with the rows being the three complexity factors and Perceptual Complexity
columns being the three stages. Therefore, we decompose We derived the following basic perceptual tasks for
complexity into nine metrics. Each metric describes the using generic ATC displays. The generic tasks include:
effect of the complexity factor on information processing 1) Detect critical messages; 2) Search for data of a given
at a given stage, so it is associated with the operator’s task category; and 3) Scan / rapidly read graphic patterns and
performance. For example, one of the metrics that we text. The perceptual functions involved in these tasks
proposed is the degree of clutter (described in detail later). include target pop-out (Pop-out means that a visual target
Clutter occurs when the perception of a central target is can be effortlessly detected irrespective of the amount
masked by the presence of overlaying and immediately of surrounding visual materials), detection, search, seg-
surrounding stimuli. Clutter directly affects the speed and mentation, text reading, etc. To drive measurements of
accuracy of text reading and target detection. With the the effects of the complexity factors on the performance
known capacity limits of brain information processing, of these tasks, we need to understand the underlying
these metrics can elucidate why a visual display can be mechanisms of these functions.
too complex for human operators. Given the principle The mechanisms of the above visual tasks can be
that complexity is constrained by task requirements, the described with the well-known two-step model of visual
format of each metric may vary with different applica- information processing. Figure 3 illustrates the model
tions. Next, we used this framework to develop metrics in which the visual system processes information in two
specifically for ATC displays. steps. In the first step, a visual image is segmented into
distinctive visual objects, and salient targets pop out of
Metrics of Information Complexity for ATC Displays the image. This step involves parallel processing. Based
Since complexity is constrained by task requirements, on the results of the parallel processing, the second step
developing metrics for ATC displays needs to consider the involves the visual system serially focusing on the salient
generic tasks associated with using the displays. Therefore, targets or selected objects so that information can be ana-
we used the following steps to develop the metrics: lyzed in detail. Next we evaluated the three complexity
• Identify task requirements; factors in this perceptual model to derive the metrics of
• Determine corresponding brain functions pertinent perceptual complexity.
to the task requirements; and Quantity evaluated by perception. According to the
• Choose the metric that can measure the effects of the perceptual model, the quantity factor does not affect im-
complexity factor on the brain functions. age segmentation and target pop-out due to their parallel
processing. However, it does affect the serial processing of
This procedure requires understanding the nature of visual details. Processing time increases with the number
the tasks associated with using displays. Since the purpose of visual elements in a display. Since serial processing is
of this study is to develop complexity measures for generic limited to the information available within retinal fovea
ATC displays, we extracted some typical characteristics where the eyes are fixated, the basic visual element in serial
of those displays: processing is the fixation. Therefore, we propose that the
• The displays contain mainly text, icons, and other metric of Quantity evaluated by perception is the number
binary graphical patterns (symbol, charts, etc.); of fixation groups. A fixation group is defined as a set of
• Many categories or types of data are presented in in- visual stimuli that can be perceived within a foveal fixation
termingled and sometimes overlapped patterns; for detailed analysis. Typically, a foveal fixation spans a
• Displays are often dynamic; information is continu- viewing angle of about 2-4 degrees. The average time to
ously updated; and search for a particular target on a display increases with
• Displays are interactive; users actively access to and the number of fixation groups. While there is no physi-
update information. ological limit on how many fixations one can make on
4
 Object Detailed
segmentation Serial
processing
searching
Salient area (contrast,
pop-out orientation,
speed, …)

Figure 3: Diagram of the two-step perceptual model.

UAL 135 N123 DAL62

UAL 135
350 1210 350 1210 OTP /165 299A
265H-22 265H-22 672HZ23 742 HOLD
DAL62
299A TWA45 TWA45 UAL 260
UAL 135
742 HOLD 310N 310N 350 1210 350 1210
AAL16 485CST 485CST 265H-22 265H-22
350A TWA45
445 460 310N TWA45
485CST UAL 135 AAL116 AAL16
310N 350A
485CST 350 1210 250A
265H-22 470 445 460

Figure 4: Graphic illustration of the effects of the quantity factor on perception.

a display, visual experiments have demonstrated that it Variety evaluated by perception. Variety is related
takes 600-700ms for an observer to perceive the informa- to image segmentation and pop-out, both based on the
tion in one fixation. Therefore, the capacity limit of this uniformity and distinction of the visual features. There-
metric is determined by the time available for users to fore, we proposed that this metric involved the number
monitor a display. For example, if an air traffic controller of different visual features, including distinctive colors,
has ten seconds maximally to scan all the information on luminance contrast, spatial frequency or size, texture,
a display, then the number of fixation groups included and motion signals in a display. Increasing the variety of
in the display should be less than 14. visual features leads to difficulties in visual segmentation
Figure 4 illustrates the concept of the metric. Both and target pop-out; as a result, a complex display cannot
pictures contain aircraft symbols and datablocks, mim- be efficiently organized, and salient targets cannot be
icking those on controllers’ radar displays. Controllers instantly detected without being searched. In addition,
typically scan the datablocks and fixate on important visual studies have demonstrated that switching between
ones for detailed processing. Thus, each datablock can visual features such as color and luminance contrast
be counted as one fixation group. The picture on the left increases search time. This effect is called “cost of switch-
panel has four datablocks and to scan them all would take ing.” Switching also reduces the reliability of text reading
at least 4 × 600ms = 2. 4 sec. In contrast, the picture on and target detection. Consider, for example, that the two
the right panel has 11 datablocks so it would take at least pictures in Figure 5 contain the same materials. The text
11× 600ms = 6.6 sec to scan all of them. in the left panel has the same font and uses three colors.
In many applications, displays are crowded and it The red letters “LA” indicate an alert that controllers
takes many fixations to view all the information. One need to instantly detect and pay attention to. The blue
strategy to reduce perceptual complexity is to use color and black text represents two categories of datablocks.
or other cues to aid visual segmentation, so information This picture uses three visual features (colors) to seg-
can be segregated into several objects. Consequently, ment the displayed information, and the segmentation
the number of fixations required to complete the visual is very effective. However, the figure in the right panel
search is reduced. uses many visual features (colors, font, sizes, etc) and
5
segmentation becomes impossible. As a result, the right patch was presented alone. However, when a blank gap
figure appears to be more complex than the one on the was introduced between the central and surrounding
left due to its variety. patch, the suppression effect became much weaker. These
Relation evaluated by perception. The relationship experimental results implicitly suggested two methods to
of visual elements affects the processing of detailed visual reduce the clutter effect: 1) reducing the amount of text
information. In particular, the perceived contrast of a in a display and, 2) keeping blank surrounds for targets
visual stimulus depends on the physical contrast of the to be quickly read or detected.
stimulus and other visual stimuli in its surrounding area. Figure 6 is an illustration of the clutter effect. The
Contrast is important as the difficulty of text reading picture on the left shows the datablocks in a baseline
and graphic target detection are primarily determined by conflict alert display to alert pilots of potential conflicts
the perceived contrast. Therefore, we proposed that the with other aircraft. The picture on the right shows the
metric for relation evaluated by perception was the degree datablocks on a prototype of the display improvement.
of clutter, defined as the effect of the visual perception of Since pilots primarily need only the aircraft heading and
a stimulus being masked by the presence of other stimuli. altitude information, the prototype displays altitude in
Clutter can increase search time and reduce target detec- the datablock and uses a simple triangle to indicate the
tion as well as text readability. The effect is apparent when current heading direction of the aircraft. The additional
background visual stimuli are spatially superimposed on information is hidden and displayed only upon the user’s
the target. Moreover, the perceived luminance contrast request. The clutter is thus greatly reduced, and the infor-
of a visual target can also be largely suppressed by the mation can be more easily read. Notice that this declutter
presence of surrounding stimuli. Reduction in perceived strategy works for pilot but not controllers who need to
contrast results in significant deterioration of text read- see the hidden information most of the time. Controllers
ability. Xing and Heeger (2001) examined this effect and typically declutter their radar displays by showing part of
found that the perceived contrast of a sine-grating patch the datablock (called limited-datablock) when too many
embedded in a large patch of the same kind of gratings aircraft make the displays crowded.
was about half the contrast perceived when the central

N123 N123
UAL 135
OTP /165 UAL 135 OTP /165
350 1210
DAL62 350 1210 DAL62 672HZ23 DAL62
DAL62 672HZ23
265H-22 299A 265H-22 299A 299A
299A
TWA45 UAL 260 UAL 742 HOLD
135 TWA45 742 HOLDUAL 260 742 HOLD
UAL 135
742 HOLD
310N 350 1210 350 1210 310N 350 1210 350 1210
DAL62 DAL62
LA 485CST 265H-22 265H-22 LA 485CST 265H-22 265H-22
TWA45 299A TWA45 299A
PAA 185 PAA 185
310N 742 HOLD 310N 742 HOLD
350C
TWA45 350C
TWA45
TWA45 AAL16 485CST TWA45
AAL116 AAL16
135 AAL116
310N UAL485CST 310N UAL 135
250A 310N 350A 250A 310N 350A
485CST 350 1210 485CST 350 1210
485CST 445 460 485CST 445 460
265H-22 470 265H-22 470

Figure 5: Illustration of the effects of the variety factor on perception.

UAL 135
N123
1210 165
OTP /165
350 1210
265H-22 DAL62 672HZ23 DAL62 1144 102
299A 299A
340
TWA45 UAL 260 UAL
742 HOLD 742 HOLD
135 1100 1210
310N 350 1210 350 1210
PAA485CST
185 DAL62 350
265H-22 265H-22
350C TWA45 299A 485
310N 742 HOLD
TWA45 485
TWA45
135 AAL116
310N UAL485CST
AAL16 265 4701321 350
310N 350A
485CST 350 1210 250A485CST 445 460
265H-22 470

Figure 6: Illustration of the clutter effect on text reading and target detection.
6
Cognitive Complexity traditional term “manipulation” to be consistent with
We theorized that the following cognitive tasks are the recent literature and to emphasize that this type of
performed when operators interact with ATC displays. memory is for processing information, not for maintain-
The generic cognitive tasks include 1) constructing, main- ing it. The circles in each buffer represent the elements
taining, and updating the mental representation of the of information. Open circles represent items of available
information contained in the display; 2) comprehending information; filled circles represents information selected
text and graphic information; and 3) binding (or asso- for an action. The arrows represent information flow
ciating) items of information to plan an action or make between the buffers.
a decision. All of these cognitive tasks require working The model performs tasks through interactions be-
memory. Therefore, measures of cognitive complexity tween processing memory and maintenance memory.
should be based on quantifying the demand that using Below are the basic ways that the model processes a
a display imposes on working memory. complex task:
Traditionally, working memory has been considered • Maintenance memory keeps items of information
as the limited-capacity storage system involved in the “on-line” without being attended to; such items form a
maintenance and manipulation of information over short “to-do” list for completing a task; they can be quickly
periods of time. However, recent findings suggest that retrieved and are subject to decay if not attended to
working memory for maintenance is different from that for over a period of time.
manipulation. Memory for maintenance is about chunks, • Processing memory binds pieces of information that
or elements that are in our conscious awareness in the are simultaneously needed for planning an action or
absence of sensory inputs, while memory for manipulation making a decision.
is about the independent elements or variables that must • Processing memory retrieves information from sensory
be simultaneously considered to plan an action or make systems, maintenance memory, and / or long-term
a decision (Cowan, 2001; Halford, Wilson, & Phillips, memory. Information needed for an action or deci-
1998). Based on recent findings from psychophysical sion is selected from those sources and associated in
and neurophysiological studies, we generalized a work- processing memory.
ing memory model that incorporates maintenance and • Depending on the task, processing memory can discard
manipulation memory, and we used the model to evaluate information that is no longer needed or register new
the complexity factors. information into maintenance memory for later use.
Figure 7 shows a diagram of the working memory
model. It consists of input mechanisms (long-term The limit for representational complexity. According
memory and inputs from sensory systems) and two work- to our model, “too complex to use” is when the memory
ing memory buffers: processing memory and maintenance demand for processing the displayed information exceeds
memory. We used the term “processing” rather than the the capacity of working memory. We generalized the

Maintenance
memory

Processing
memory Decision-making

Long-term memory Sensory systems

Figure 7: The diagram of a working-memory model for cognitive

information processing.
7
results of working capacity limits from the literature and of relational complexity proposed by Halford et al. as the
used them as the upper-limits of cognitive complexity for metric to describe how the relation factor of complexity
displays. In the model, the items in maintenance memory affects cognition. The metric is defined as the number
do not need to be retrieved simultaneously but should of independent elements or dimensions of information
be maintained over a period during which the user is that must be simultaneously bonded to use information.
attending to other information. Thus, representational Many cognitive processes, such as selection, manipulation
complexity, defined as the number of functional units, of goal hierarchies, reasoning, and planning, are examples
is in accordance with the operation span of working of processing at high levels of relational complexity.
memory. Conway and Engle (1996) proposed that the Halford argued that the more interacting variables that
operation span is about the limitation in the ability to use have to be processed in parallel, the higher both the
controlled processing to maintain information in an active, cognitive demand and computational cost will be. Since
quickly retrievable state. They found that the number processing memory links pieces of information that are
of items one can reliably recall is typically in a range of simultaneously needed for task performance, relational
5~16, with a mean value of 9~11. Other studies hinted complexity is a straightforward measure of the processing
at similar limitations. For example, Willems, Allen, and memory load of a task.
Stein (1999) reported that a controller could correctly The limit of relational complexity. Halford et al.
recall up to 11 aircraft that he or she operated. further demonstrated that normal adults could reliably
Variety evaluated by cognition. Once information is integrate up to four relations in parallel while children
organized into independent units or chunks and held in can only integrate one or two relations. This quaternary
maintenance memory, the variety of information chunks limitation appears to be consistent with other studies that
does not affect cognitive task performance. Task perfor- demonstrated the capacity limit of processing memory as
mance is not affected because the chunks are meant to 3~5 items (Cowan, 2001). Cowan reviewed the literature
be different. However, the variety factor in the temporal on processing memory and concluded that the human
domain (i.e., dynamic changes of information) affects capacity limit is three to four items on average. Many
information flow between processing and maintenance visual experiments have revealed that the capacity of vi-
memory. Therefore, we proposed that the metric was the sual processing memory is about four items. As another
amount or frequency of unpredictable information onset example, in a single sentence the maximal number of
that demands a change in the contents of maintenance concepts the sentence can contain and still be reliably
memory. We refer to this metric as dynamic complex- interpreted is also four.
ity. Information changes in a display impose cognitive
loads in several ways: 1) increasing maintenance memory Action Complexity
load; sudden onset of visual targets or even changes in We derived the following basic action-related tasks for
luminance of visual patterns (Schmidt, Vogel, Woodman, using ATC displays from ATC task analyses found in the
& Luck, 2002); 2) increasing information flow between literature. The generic tasks include: 1) responding to
maintenance and processing memory; and 3) reducing onset of alert/warning messages and other requests from
the stability of mental representation. To build a mental a display; 2) locating and acquiring specific information;
representation takes time. As a result, if too many enti- and 3) entering data into the display system. The action
ties in maintenance memory are updated at a high rate, functions involved in these tasks include planning eye/
the mental representation tends to deteriorate or even head/hand/body movements, sequential movements, oral
collapse. This corresponds to users’ “losing the picture” communication, etc.
(Hopkin, 1995). Figure 8 is a simplified diagram of action information
The limit of dynamic complexity. While many processing. According to the literature, the parietal and
studies have demonstrated that information can be lost frontal cortices are involved in planning actions, and
if it is rapidly presented or immediately disturbed by the motor cortices are responsible for executing planned
the onset of new stimuli, there are no conclusive results actions (Andersen, Snyder, Bradley, & Xing, 1997). A
in the current literature about the temporal capacity of common feature of these cortices is population coding.
working memory. That is, all the neurons in a cortical area work together
Relation evaluated by cognition. According to the to encode a single action plan, and only after the plan
memory model, the relation between elements of in- is executed do they begin to encode the next (Georgop-
formation is handled by processing memory that binds oulos, Schwartz, & Kettner, 1986; Pouget, Zemel, &
pieces of information simultaneously needed to plan an Dayan, 2000; Xing & Andersen, 2000). Therefore, the
action or make a decision. Thus, this metric measures the brain can only execute one action plan at a time. Using
demand for processing memory. We used the definition a display to perform a specific task may require multiple
8
 Execution

Action planning
Sequenced plans
Requested actions

Figure 8: Diagram of action information processing.

actions. Figure 8 provides a framework for multi-step diverse environmental perturbations and reduce the dif-
action planning. The framework is based on the fact that ficulty of decision-making. In such systems, a task of any
the brain is able to hold several action plans and execute complexity can be decomposed into a series of subtasks,
them sequentially. When performing multiple tasks with each represented by a subgoal. Some researchers have
a display, users actually switch action planning back and used the number of serializable subgoals as a measure
forth. The switch can be so quick that they feel that they of complexity for a system with a multi-level structure
are doing multiple things at the same time. However, (Heylighen, 1989).
information can be misinterpreted or lost during these Relation evaluated by action. The relation factor affects
quick switches. Next, we use this framework to evaluate action planning. The brain cortices related to motor planning
the complexity factors and derive the metrics of action can only program one action plan at a time. Therefore, we
complexity. propose that this metric should be the number of simulta-
Quantity evaluated by action. The quantity factor neous action goals required to use displayed information.
affects the first part of the action model, the requested Since the brain can only reliably program one action plan
actions. Performing physical interactions with a display at a time, ideally each action should result in only one ac-
costs time and takes users away from other perceptual and tion goal for the next step. In the case where more than one
cognitive tasks. Therefore, we proposed that the metric action goal needs to be planned simultaneously, the brain
was the minimal amount of keystrokes, mouse move- has to switch back and forth between the goals. Errors may
ments, and transitions of action modes required to use occur when switches of planning occur at a fast pace.
displayed information. Compared with keystrokes and Table 2 summarizes the 3x3 metrics we developed for
mouse movements, the time needed for eye and head generic ATC displays. The metrics describe the objective
movements is negligible. Therefore, we only considered aspects of complexity. So far, we have developed the metric
keystrokes and mouse movements. An action transition definitions, yet the algorithms or methods of computing
is a change of action modes, such as from keystrokes to each metric remain to be developed or implemented from
mouse movements or vice versa. Those transitions also the literature. Next we describe a preliminary case study to
take time and require the brain to coordinate different explore the methods of applying our metrics to computing
action modes. Sears (1994) proposed a layout complexity cognitive complexity of a human-computer interface.
metric as the summed product of the frequency of action
transitions and the cost of transitions. The two factors in A PRELIMINARY CASE STUDY: ASSESSING THE
our metrics, the amount of manual movements and the COGNITIVE COMPLEXITY OF THE MICROSOFT
transitions between the movements, are essentially the POWERPOINT INTERFACE
same as Sears’ metric.
Variety evaluated by action. The variety factor affects The purpose of this case study was to explore how to
storage of action plans. Therefore, we propose that the apply the proposed metrics to evaluate display complex-
metric is action depth, defined as the number of serial ity. Microsoft PowerPoint™ is one of the most popular
steps needed to plan (or select from a number of action software applications for making presentations. Using our
options) to acquire information. Following the need to complexity metrics, we assessed the cognitive complexity
increase the variety of actions, today’s display systems of the PowerPoint interface.
tend to use multi-level structures to cope with more
9
 Table 2: Metrics of information complexity

Perception Cognition Action

Quantity No. of fixation groups No. of functional units Amount of action cost

Variety No. of visual features Dynamic complexity Action depth

Relation Degree of clutter Relational complexity No. of action goals

Methods Processing Complexity

We performed the study in the following steps: Table 3 indicates how we estimated processing com-
1) Assessing representational complexity: 24 Pow- plexity. The table presents the top 11 most frequently
erPoint users were asked to visualize their PowerPoint listed functions, the elements of information required
interface and list the functions that are most essential to access the function, and the indices of processing
and useful (or frequently used). The average number of complexity counted as the number of the elements.
listed functions was used as the index of representational The “*” associated with an index indicates the situation
complexity. where a function can be accessed through a drop-down
2) Assessing processing complexity: We calculated menu or directly through an icon on the top layer of the
the elements of information needed to access each basic interface, depending on individual customization. The
function and took the average number of the elements results show that the indices of processing complexity
as the index of processing complexity. for these individual functions are either zero or one. We
3) Assessing dynamic complexity: We counted the used the average of the indices as a rough estimation of
number of unexpected changes in the interface and used the processing complexity of the interface. The average
it as the index of dynamic complexity. index is 0.66.
In addition, dynamic complexity for the PowerPoint
RESULTS interface is zero because the interface does not have any
unpredicted changes of information or functions.
Representational Complexity Table 4 presents the summarized results of the com-
Figure 9 shows the indices of estimated representational plexity evaluation of the PowerPoint interface. The upper
complexity. The horizontal axis represents the number row displays the three metrics; the middle row indicates
of frequently used functions the users listed. The vertical the estimated indices of the metrics; and the bottom
axis represents the number of users responding. All but row compares the indices to the corresponding capac-
one user listed 5~13 basic functions. The mean number ity limits of working memory. The results indicate that
of listed basic functions was 8.1. the estimated indices of complexity for the PowerPoint
interface are well below the capacity limits.

12
DISCUSSION

10
This paper presents a framework to decompose infor-
mation complexity and 3x3 complexity metrics for ATC
8 N=24 displays. Previous work has reported measures similar to
No. of responses

the individual metrics we proposed (see Xing & Man-

6 ning for a review). Value can be gained from comparing
our metrics with similar ones found in the literature
4 and by integrating existing complexity measures into
our framework.
2

0 Perceptual Complexity in the Literature

4 6 8 10 12 14 16 18 Many algorithms have been developed to address image
No. of frequently used functions or pattern complexity, but most are based on information
theories and have low correlations with human judgment.
Figure 9: Estimated representational complexity. In contrast, Tullis (1985) developed a set of metrics to
10
 Table 3: Estimated processing complexity for PowerPoint interface

Basic function Elements needed to Processing complexity

access the function of the function
Copy/paste Edit 1

Insert picture Insert 1

Slide layout Format 1 or 0*

Save file File 1

Text box Insert 1

Format text Format 1 or 0*

Open PowerPoint File 1

Slide show 0
Set font
Format 1 or 0*
characteristics
Insert new slide Insert 1

Draw shape / line Draw (Autoshapes) 1 or 0*

Table 4: Summary of the complexity evaluation of PowerPoint

Representational Processing Dynamic

Metric
complexity complexity complexity

Index 8.1 0-1 0

Relative to capacity limits 8.1/(9~11) (0~1)/(3~4) 0

measure display complexity from the perspective of hu- One drawback of Tullis’ metrics is that those measure-
man performance on visual search tasks. The metrics were ments were specified for text-dominant displays but not
comprised of four basic characteristics of display formats for graphical ones. It is hard to define Tullis’ groups in
to describe how well users can extract information from spatially continuous two-dimensional images with vary-
displays. They included a) overall density of displayed ing colors and luminance contrast. On the other hand,
items, b) local density of characters, c) number and aver- Rosenholtz, Li, Mansfield, and Jin (2005) proposed a
age group size, and d) layout complexity, which describes feature congestion measure of display clutter in which clut-
how well the arrangement of items on a display follows ter is considered as the degradation of task performance
a predictable visual scheme. Tullis showed that these caused by the density of visual features such as color or
metrics are highly correlated with subjects’ performance luminance contrast. This measure is related to our clut-
time in visual search tasks. The overall and local density ter metric and can be applied to graphic visual displays.
characteristics, together, can be a measure of our clutter Unfortunately, the algorithm requires displayed materials
metric; the number of groups corresponds to our metric to be converted to digitized images to compute the fea-
of fixation groups; and the layout complexity is some- ture congestion. That is often not practical for dynamic
what related to the variety of visual features. Therefore, displays in which visual images evolve rapidly.
for limited applications, we can use Tullis’ metrics as the
quantitative measurements for the perceptual complexity
we proposed.
11
Comparisons With Other Measures of Cognitive that controllers are forced to manipulate the display fol-
Complexity in the Literature lowing a fixed procedure. That would be contrary to the
Previous studies of cognitive complexity have focused design philosophy.
on text comprehension, creativity, social phenomena, One measure related to action complexity is Sears’ layout
etc. For example, Crokett (1965) used the concept of appropriateness metric (Sears, 1994). Sears proposed this
“level of hierarchic integration of constructs” to define metric to assess users’ performance when using a com-
cognitive complexity. With this definition, cognitive puter interface. The metric was the summed product of
complexity is associated with increasing differentiation the frequency of action transitions and the cost of these
(containing a greater number of constructs), articulation transitions. The two factors comprising our metric of
(consisting of more refined and abstract elements), and action cost, number of manual movements, and transi-
hierarchic integration (organized and interconnected). We tions between the movements are essentially the same
identified three metrics of cognitive complexity based on as Sears’ metric. Sears used the distance that users must
working memory: number of functional units, number move the computer mouse and the size of the objects to
or frequency of unpredictable changes, and number of be moved as the cost of a transition. This metric can be
relations. Our metric, the number of functional units, used to evaluate the efficiency of a user-interface layout
is similar to the measure of constructs in the literature. and compute the extent to which a display demands
We also proposed relational complexity as a metric to action. However, it does not apply to ATC displays well
quantify how the relation factor of complexity affects because it primarily emphasizes the effects of mouse
cognition. This measure corresponds to the intercon- movements, while many other kinds of actions are
nected hierarchical integration proposed by Crokett. In involved in using ATC displays (Allendoerfer, Zin-
addition, we proposed to use the frequency of unpredict- gale, Pai, & Willems, 2006) .
able information changes to measure the variety factor of Perhaps the Keystroke Level Model (KLM) proposed
complexity. However, this dynamic aspect of cognitive by Card, Noran, and Newell (1980) is more applicable
complexity has been seldom studied. to measure action complexity within the ATC domain.
Measures of cognitive complexity, explicitly or im- The model measures the sum of the execution time of
plicitly, depend on cognitive task analyses that reveal sequenced operations, including key strokes, mouse
the cognitive aspects of tasks and knowledge needed for movement, switches, and mental preparation for a physi-
situation awareness, decision-making, planning, etc. One cal action. While most of these operations correspond to
popular cognitive task analysis method is GOMS: Goal, our metric of action cost, the last operation — mental
Operator, Methods, and Selection (Card, Moran, & preparation — is somewhat related to the other two ac-
Newell, 1983). This method seeks to analyze and model tion metrics: action depth and simultaneous action goals.
the knowledge and skills a user must develop to perform Compared to Sears’s metric, the KLM model considered
tasks on a device or system. The result is a description the effects of key strokes and mental preparation. Both
of the Goals, Operators, Methods, and Selection rules play crucial roles in controllers’ interacting with ATC
for any task. Currently we are exploring how to calculate displays. Allendoerfer et al. documented the frequency
the three cognitive metrics based on GOMS and other of use of en route controller commands using radar
similar analyses. displays and measured the time and number of key-
strokes or mouse clicks required to perform those
Relevant Measures of Action Complexity in the commands. They found that controllers entered
Literature information on the radar display more frequently
Many methods have been developed to assess the com- than they moved things around. They also proposed
plexity of human-computer interfaces (McCabe, 1976; that the time spent looking at the keyboard or screen
Rauterberg, 1992). Those methods require modeling a while entering commands should be included in the
system’s states and transitions between states. Unfortu- assessment of display usage characteristics. These
nately, it is implausible to directly apply such methods to results suggested that the KLM model is a potential
ATC displays. Controllers use displays adaptively, and no candidate to objectively measure action complexity
standardized procedure has been specified. Thus, there are for ATC displays. Next, we need to explore how to
no clearly defined states and transitions in their interac- calculate action complexity based on the measure-
tions with ATC displays. If the use of a display can be ments like those in the KLM model.
described explicitly with states and transitions, it implies

12
 CONCLUSIONS Cowan N (2001). The Magical number 4 in short-term
memory: A reconsideration of mental storage capac-
This paper presents a framework for decomposing ity. Behavioral and Brain Sciences; 24(1): 87-114.
factors of information complexity and a set of metrics Crockett WH (1965). Cognitve complexity and impres-
to measure ATC display complexity. The framework is sion formation. In: Maher BA (Ed.), Progress in ex-
described as follows: 1) information complexity is the perimental personality research. New York: Academic
combination of three basic factors: quantity, variety, and Press; 2: 47-90.
relation; 2) complexity factors need to be evaluated with
the mechanisms of brain information processing at three Crutchfield JP, Young K (1989). Inferring statistical
stages of information processing: perception, cognition, complexity, Physics Review Letters; 63: 105-8.
and action; and 3) the metrics of complexity can be de- Drozdz S, Kwapien J, Speth J, Wojcik M (2002). Iden-
rived by associating task requirements to brain functions. tifying complexity by means of matrices. Physica;
The framework incorporates many human factors studies A314: 355-61.
involving interface evaluation. Within this framework, we
identified a set of complexity metrics for ATC displays. Edmonds B (1999). What is Complexity? - The phi-
Future work will focus on testing the metrics in a real or losophy of complexity per se with application to
simulated ATC work environment and converting the some examples in evolution. In: Heylighen F, Aerts
metrics into easy-to-use products for the design and hu- D (Eds.), The evolution of complexity. Kluwer:
man factors evaluation of new ATC technologies. Dordrecht; 1-18.
Georgopoulos AP, Schwartz AB, Kettner RE (1986).
REFERENCES Neuronal population coding of movement direc-
tion. Science; 233: 1416-9.
Allendoerfer KP, Zingale C, Pai S, Willems B (2006). En Grassberger P (1991). Information and complexity
route air traffic controller commands: Frequency of measures in dynamical systems. In Atmanspacher
use during routine operations. Federal Aviation Ad- H, Scheingraber H (Eds.), Information dynamics.
ministration, William J. Hughes Technical Center; New York: Plenum Press; 15-33.
Technical Report No: DOT/FAA/TC-TN06/04.
Halford GS, Wilson WH, Phillips W (1998). Process-
Andersen RA, Snyder LH, Bradley DC, Xing J (1997). ing capacity defined by relational complexity:
Multimodal representation of space in the posterior Implications for comparative, developmental and
parietal cortex and its use in planning movements. cognitive psychology. Behavioral Brain Sciences;
Annual Review of Neuroscience; 20: 303-30. 21(6): 803-31.
Bennett CH (1990). How to define complexity in phys- Heylighen F (1989). Self-organization, emergence and
ics, and why. In: Zurek WH (Ed.), Complexity, the architecture of complexity. Proceedings of the
entropy and the physics of information. Redwood 1st European Conference on System Science; Paris:
City, California: Addison-Wesley; 137-48. AFCET; 1989: 23-32.
Burleson BR, Caplan SE (1998). Cognitive complexity. Hopkin VD (1995). Human factors in air traffic control.
In: McCroskey JC, Daly JA, Martin MM. & Beatty Bristol, PA: Taylor & Francis.
MJ (Eds.), Communication and personality: Trait
perspectives. Cresskill, NJ: Hampton Press. Klemola T (2000). Cognitive complexity metrics and
requirements comprehension. Australian Conference
Card SK, Moran TP, Newell A (1980). The keystroke- on Software Measurement. Sydney, Australia.
level model for user performance time with interac-
tive systems. Communications of the ACM; 23(7): McCabe (1976). A complexity measure. IEEE Transactions
396-410. on Software Engineering; SE-2: 308-20.

Card SK, Moran TP, Newell, AL (1983). The psychol- Pouget A, Zemel RS, Dayan P (2000). Information
ogy of human computer interaction. Hillsdale, processing with population codes. Nature Review
NJ: Erlbaum. of Neuroscience; 1: 125-32.

Conway AR, Engle RW (1996). Individual differences Rauterberg M (1992). A method of a quantitative mea-
in working memory capacity: more evidence for a surement of cognitive complexity. Human-Com-
general capacity theory. Memory; 4(6): 577-90. puter Interaction: Tasks and Organization - ECCE’92;
Roma, Italy. 295-307.
13
Rosenholtz R, Li Y, Mansfield J, Jin Z (2005). Feature Xing J, Andersen RA (2000). Memory activity of LIP
congestion: A measure of display clutter. Proceedings neurons for sequential eye movements simulated
of the SIGCHI 2005; 761-70. with neural networks. Journal of Neurophysiology;
84(2): 651-65.
Schmidt BK, Vogel EK, Woodman GF, Luck SJ (2002).
Voluntary and automatic attentional control of Xing J, Heeger D (2001). Quantification of contrast-
visual working memory. Perception and Psychophys- dependent center-surround interaction. Vision
ics; 64: 754-63. Research; 41: 571-83.
Sears AL (1994). Automated metrics for user interface Xing J, Manning C (2005). Complexity and automation
design and evaluation. International Journal of displays of air traffic control: Literature review and
Biomedical Computation; 34: 149-57. analysis. Washington DC: Federal Aviation Admin-
istration; Report No: DOT/FAA/AM-05/4.
Tullis TS (1985). A computer-based tool for evaluat-
ing alphanumeric displays. In Shackel B (Ed.), Xing J (2004). Measures of information complexity and
Human-computer interaction: INTERACT ‘84. the implications for automation design. Washington
London, England. DC: Federal Aviation Administration; Report No:
DOT/FAA/AM-04/17.
Wickens CD (1991). Engineering psychology and human
performance. NY: Harper Collins.
Willems B, Allen RC, Stein ES (1999). Air traffic control
specialist visual scanning II: Task load, visual noise,
and intrusions into controlled airspace. Federal
Aviation Administration, William J. Hughes Tech-
nical Center; Technical Report No: DOT/FAA/CT-
TN99/23.

14