You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274979188

Eye Movement-Based Interface Evaluation: What can and Cannot be Assessed?

Article in Proceedings of the Human Factors and Ergonomics Society Annual Meeting · July 2000
DOI: 10.1177/154193120004403721

CITATIONS READS
10 330

1 author:

Joseph H. Goldberg
Meta
78 PUBLICATIONS 4,864 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optimal Product Bundling in Recommender Systems View project

Eye Tracking View project

All content following this page was uploaded by Joseph H. Goldberg on 03 August 2015.

The user has requested enhancement of the downloaded file.


EYE MOVEMENT-BASED INTERFACE EVALUATION:
WHAT CAN AND CANNOT BE ASSESSED?

Joseph H. Goldberg
Pennsylvania State University
University Park, Pennsylvania

Interface evaluation by eye tracking-derived data is discussed in this review and synthesis
paper. While analysis of eye movements during interface use is becoming more popular,
there is little basis for justification of eye tracking methods. A review of traditional
interface assessment methods and criteria is provided, to establish areas where eye
tracking may potentially impact interface evaluations. Studies are then reviewed, that
have used eye tracking-derived measures for performance assessment as interfaces are
manipulated. A synthesis is then provided, suggesting that eye tracking-based analysis
could have a positive impact in evaluations of consistency, resources, visual clarity, and
flexibility, and should have difficulty in ascertaining interface compatibility and locus of
control.

INTERFACE USABILITY ASSESSMENT AND taxonomies exist, several criteria are universally
CRITERIA found (e.g., Shneiderman, 1998; Jordan, 1998; Dix,
et al., 1998), as listed in Table 1. Usability is
A hardware or software interface should allow generally enhanced by the degree to which these are
individuals to effectively achieve their goals with criteria are satisfied.
minimal effort and confusion, while achieving a
high level of satisfaction (Jordan, 1998). The EYE TRACKING ON INTERFACES
assessment of effectiveness, efficiency, and
satisfaction at an interface thus comprises a large Analysis of eye movements during one's
portion of most usability evaluations. Evaluation interaction with an interface can provide otherwise
approaches may be based upon theory, covert information underlying attention and
guidelines/standards, experiments, protocols, decision making strategies. As a methodology, eye
surveys/interviews, or walkthroughs. Each method tracking measures the location of one's gazepoint at
has its advantages and disadvantages (see Jordan, relatively rapid frequency (e.g., 60 Hz) and
1998; Dix et al., 1998). Generally, more subjective
methodologies (e.g., cognitive walkthroughs or Table 1. Common Evaluation Criteria for Interface Usability
theory-based approaches) are applied to early Evaluation Description
concepts, whereas more formalized, objective Criteria
Consistency Similar tasks must be performed in
approaches (e.g., experiments) are applied during similar ways
latter design stages. Many approaches have been Compatibility Interface operation must match user
critiqued due to high resource costs, lengthy expectations
implementation, subjectivity, and/or poor access to Locus of Control Users must feel that they are in control
users' actual performance and behaviors (Jordan, of the interface
1998). Some assessment methods (e.g., task-based Feedback Every interface event must provide
feedback to the user
evaluations) can intrude upon or change the nature Minimum User resource demands (e.g., memory
of the actual tasks under evaluation. Although users Resources and attention) must be controlled
may complete the objectives for a task (high Error Handling User errors must be minimized, and
effectiveness), they may expend tremendous easily handled when they occur
resources in this process (poor efficiency). Visual Clarity Visual components must be easily and
Usability evaluations are usually conducted with rapidly identified
Flexibility Interaction must accommodate broad
respect to specific criteria. While many criteria range of users
reasonable accuracy (usually within 1° of visual labels, and fixation durations were sensitive to
angle). While one's current attentional focus may manipulations of lines separating areas of the labels.
lead or lag the current gaze location, the current Eye tracking as a methodology has also aided
gaze location is usually assumed to reflect what is the interpretation of display hardware quality. For
on top of the stack of cognitive operations (Just and example, Deffner (1995) showed that fixation
Carpenter, 1976; Potter, 1983). locations and durations are sensitive to perceived
The goal of the present work is to determine visual clarity of images presented on alternative TV
whether eye tracking-derived parameters are displays.
sensitive to some aspects of usability at an interface. While the preceding studies have only initiated
Positive indications in this direction can potentially research in this area, it is clear that eye movement
allow the development of more rapid and objective derived parameters are sensitive to at least some
measures of usability than currently exist. A few manipulations of interface quality. Most of the
published studies have relied upon gazepoint data preceding tasks have contained a visual search
from eye tracking methodologies, while component, in which users attempt to locate a
manipulating aspects of interfaces. memorized or known object on a display. Further
Text manipulations during computer interface discussion, below, will consider which types of eye
search have clearly resulted in eye movement- tracking-derived measures should be sensitive to
related effects. Using fixation-related parameters, which aspects of usability on displays.
Kolers, Duchnicky, and Ferguson (1981) noted that
reading rates were faster with more densely packed EYE TRACKING MEASURE CATEGORIES
characters, as they manipulated line spacing,
character density, and text scrolling rate. Yamamoto Although an organized taxonomy of eye
and Kuto (1992) found that more text columns in a movement-derived parameters has never been
display lead to more re-fixations as individuals developed, some classification can be attempted
searched for target words in a matrix. here, in order to provide some generalizations about
Ellis, et al. (1998) evaluated four different eye movement based usability evaluation.
versions of a web page by manipulating the size and The following categories of measures assume
layout of text blocks, inclusion of supporting that samples of gazepoint locations (x, y locations)
graphics, and the percentage of words that were have acquired, then broken into saccades and
hyperlink phrases. Based upon eye tracking data, fixations. Scanpaths are defined by linking
dense text with wide paragraphs and no graphics successive fixation locations from an observer's
supported the fastest search, but was also the least data record. Summaries of these procedures are
preferred of the interfaces. Dense text produced available in Goldberg and Schryver (1995) and
shorter eye fixations to reject irrelevant targets, Goldberg and Kotval (1999).
allowing more information per fixation.
Goldberg and Kotval (1998) manipulated the Scanpaths
grouping of graphical tools in a drawing package-
like interface. A large group of interface users and Consisting of sequences of saccades and
professional designers rated the subjective usability fixations, scanpaths define the serial/temporal
of each interface and grouping. When required to record of attention during interaction with an
rapidly find a particular tool, eye movement interface. While there is a large stochastic element
patterns were strongly related to rated usability, as to scanpath shape, overall cycles can be found in
well as to timed search performance. experienced visual searchers (Goldberg and Kotval,
Goldberg, Probart, and Zak (in press) used eye 1998). Aspects of scanpaths, such as mean fixation
movement-derived parameters to determine duration, ratio of number of saccades to fixation
precisely when desired information is located on duration, and number of acute angles between
presented nutrition labels. More eye movements successive saccades can indicate complexity of
were made in denser information sections of the visual processing and salience of visual targets
(Goldberg and Kotval, 1999; Kotval and Goldberg, SYNTHESIS: EYE MOVEMENT
1998). ASSESSMENT OF EVALUATION CRITERIA

Cumulative Time in Areas of Interest (AOIs) A synthesis among the preceding discussions of
usability assessment, eye movements on displays,
An interface can be divided specific areas, with and categories of eye tracking-derived dependent
cumulative visiting (or fixation) time measured in measures can now be attempted, with the objective
each of these areas. Areas may be defined by a of establishing which portions of a usability
regular grid, or by irregular individual objects or assessment could benefit from an eye tracking
features. Referring to defined areas on an interface, approach.
message or interface component salience is
routinely assessed by cumulative time spent within Nature of Task
AOIs. Interference from unintended AOIs may also
be assessed. Because AOIs may differ in size and Candidate interface tasks clearly require active
shape, these approaches are best suited for relative eye movements to be sensitive to eye tracking-
comparison of condition manipulations within a derived measures. These include visual search and
study, as opposed to between-study assessment of visual monitoring tasks, but could also include time
interface usability. Extent and balance of coverage shared tasks, as well as those with some element of
on an interface can be assessed from these stimulus movement. Long duration tasks, with
cumulative times. some element of visual fatigue, should also be
sensitive to these measures.
Transitions Among AOIs Tasks that likely will not provide useful eye-
tracking data include those with little eye
The number and probability of transitions movement, such as pure memory and decision
among AOIs provides information about search making tasks, as well as auditory input tasks. Little
strategies on an interface. Compact representations, is known about the response of eye movements to
such as transition matrices and Markov models are variations in output requirements, such as vocal
feasible. Emphasis on transitions between AOIs versus manual output.
allows an assessment of search efficiency and
complexity on an interface. Strategies and Eye Movements and Usability Criteria
novice/expert differences, as in judging the skill of
airplane pilots, can be obtained from analysis of At this early stage of research predications of
transitions. Scanpaths are another form of transition relationships between eye movement-derived
representation. Variance in transitions are also a parameters and usability criteria are admittedly
meaningful way to judge the usability of an speculative. Each prediction, however, should
interface. Comparison of transitions among AOIs result in testable hypotheses. To consider whether
or portions of an interface can be made by there is a relationship with each of these criteria,
comparing transition probabilities among these one can consider whether quality of each criterion
areas. could be ascertained from eye tracking records.
Similarity among transition records can be done Table 2 lists each of the previously stated usability
by direct transition matrix comparisons, or by the criteria, then assess whether a relationship exists
process of string editing, where the number of between that criterion and eye tracking-derived
deletions and substitutions required to transform measures. Specific predications are then provided
one record to another defines their similarity in the third column.
(Hacisalihzade, Stark, and Allen, 1992). The preceding discussion has focused on the
role of eye movement-based data within interface
evaluations. Using 'classic' criteria, eye tracking
should provide positive contributions to evaluations
of consistency, cognitive resources, visual clarity,
and flexibility. Limited information may be Clearly, a substantial amount of research is required
achieved for interface feedback and error handling. to enable eye tracking to serve as a valid and
Eye tracking should have difficulty in ascertaining reliable interface evaluation tool.
interface compatibility and locus of control.

Table 2. Estimates of Eye Tracking Data Relationship by Evaluation Criteria


Evaluation Assessment Predictions and Explanations
Criteria Estimate
Consistency Good Consistent interfaces should result in similar cumulative AOI times, as well as display coverage.
However, AOI transitions are not expected to be similar unless users are highly homogeneous and
similarly trained.
Compatibility Poor Compatibility should be difficult to assess, because of its reliance upon a cognitive expectation. In
novel displays, unexpected areas draw visual attention, but most routine usage has significant context
and expectation by the user.
Locus of Poor While pacing and timing of interface interaction can be assessed from eye movements, it will be
Control difficult to determine whether the user or system is controlling the interaction, from these records.
This could become a promising evaluation approach if this issue can be solved.
Feedback Limited Feedback on an interface must be salient, and must be properly interpreted by the user. Eye
movement data can greatly aid the determination of whether feedback was observed, but cannot
significantly aid whether it has been properly interpreted. Auditory feedback also presents a
challenge to eye movement-based evaluation.
Cognitive Good As more complex decisions and greater focused attention with greater memory loads generally result
Resources in longer fixation durations, resources required by an interface can generally be evaluated by an eye
tracking approach.
Error Limited Eye tracking can provide some strategy analysis underlying an improper choice at an interface, but it
Handling cannot provide information concerning system error handling.
Visual Clarity Excellent One of the primary benefits of an eye tracking evaluation approach is in understanding the visual
clarity and salience of displayed objects. High visual clarity in search tasks should result in directed
scanpaths, sparse AOI transition matrices, and little unnecessary coverage of a display.
Flexibility Good In a similar analysis as for consistency, a flexible interface should result in a large variance of
scanpaths and display coverage between and within users, given that poor visual clarity has been
ruled out. These eye movement data should reflect that interface operations should be easily located
and executed in multiple ways.

REFERENCES Hacisalihzade, S.S., Stark, L.W., and Allen, J.S. (1992), Visual
perception and sequences of eye movement fixations: A
Deffner, G. (1995), Eye movements recordings to study determinants stochastic modeling approach, IEEE Transactions on Systems,
of image quality in new display technology, in Findlay, J.M., et Man, and Cybernetics, 22(3): 474-481.
al. (Eds.), Eye Movement Research, Elsevier Science Publishers. Jordan, P.W. (1998), An Introduction to Usability, Bristol, PA:
Dix, A., Finlay, J., Abowd, G., and Beale, R. (1998), Human- Taylor & Francis.
Computer Interaction, 2nd Ed., London: Prentice Hall Europe. Just, M.A., and Carpenter, P.A. (1976), The role of eye-fixation
Ellis, S., Candrea, R., Misner, J., Craig, C.S., Lankford, C.P., and research in cognitive psychology, Behavior Research Methods
Hutchinson, T.E. (1998), Using eye tracking data to help build & Instrumentation, 8(2): 139-143.
better web pages (abstract), Proc. of the 42nd Annual Meeting of Kolers, P.A., Duchnicky, R.L., and Ferguson, D.C. (1981), Eye
the Human Factors and Ergonomics Society, p. 1613. movement measurement of readability of CRT displays, Human
Goldberg, J.H. and Kotval, X.P. (1998), “Eye Movement-Based Factors, 23(5): 517-527.
Evaluation of the Computer Interface,” in Kumar, S.K. (Ed.), Kotval, X.P., and Goldberg, J.H. (1998), "Eye Movements and
Advances in Occupational Ergonomics and Safety, Amsterdam: Interface Components Grouping: An Evaluation Method," Proc.
IOS Press, pp. 529-532. of the 42nd Annual Meeting of the Human Factors and
Goldberg, J.H., and Kotval, X.P. (1999), "Computer Interface Ergonomics Society, Santa Monica: HFES, pp. 486-490.
Evaluation Using Eye Movements: Methods and Constructs," Potter, M.C. (1983), Representational Buffers: The Eye-Mind
International Journal of Industrial Ergonomics, 24: 631-645. Hypothesis in Picture Perception, Reading, and Visual Search,
Goldberg, J.H., and Schryver, J.C. (1995), “Eye-Gaze Contingent Ch. 24 in Rayner, K. (Ed.), Eye Movements in Reading:
Control of the Computer Interface: Methodology and Example Perceptual and Language Processes, Academic Press, New
for Zoom Detection,” Behavior Research Methods, Instruments York.
& Computers, 27(3): 338-350. Shneiderman, B. (1998), Designing the User Interface, 3rd. Ed.,
Goldberg, J.H., Zak, R.E., and Probart, K. (in press), “Visual Search Reading, MA: Addison Wesley Longman.
of Food Nutrition Labels,” Human Factors. Yamamoto, S., and Kuto, Y. (1992), A method of evaluating VDT
screen layout by eye movement analysis, Ergonomics, 35(5/6):
591-606.
View publication stats

You might also like