ANALYZING GRAPHIC AND TEXTUAL
LAYOUTS WITH GOMS
Mei C. Chuah, Bonnie E. John, John Pane
School of Computer Science
Carnegie Mellon University
Pittsburgh, PA, 15213, USA
Tel: +1-412-268-3041
E-mail: {mei+,bej,pane+}@cs.cmu.edu
ABSTRACT
We conducted a preliminary analysis on graphic and textual
layouts using GOMS. In the analysis, an airline
reservation task was modeled and the task performance
times predicted by GOMS were compared with observed
empirical data. The results obtained were very promising
and show that GOMS may turn out to be a very useful tool
for analyzing graphic and textual displays.
INTRODUCTION
Previous studies have shown GOMS to be a powerful and
accurate method of analysis for human performance. Despite
the wealth of studies that have been performed with GOMS,
there has been little or no effort directed towards applying it
as a method of analysis for graphic and textual layouts. The
primary reason for this is because the syntax of GOMS was
not rich enough to model human interaction with the
different layout schemes. Recently however, GOMS has
been expanded to accommodate low level perceptual,
cognitive and motor operators as is shown in [5]. These
developments open up the possibility of using GOMS to
predict and compare the layout of various displays. Such an
analysis would be very useful in testing newly developed
user interfaces. In addition, it can also be used to provide
heuristics and highlight problems in automatic
visualization systems.
KEYWORDS :
GOMS, low level operators, graphic and
textual layout, automatic visualization systems.
THE TASK
GOMS is used here to model users' interaction with various
kinds of graphical presentations of information while they
perform an airline reservation task. The task and graphical
layouts are taken from [3], [4]. The task is to find a pair of
connecting flights that travel from Pittsburgh to Mexico
City, such that the flights are available, the layover is no
longer than four hours, and the combined cost of the flights
do not exceed $500. Casner presents four examples of
graphical layouts that present the required information:
1)tabular layout, where the rows contain information for
each flight and the columns represent origin, destination,
availability, price, departure time and arrival time;
2)graphical layout that displays cost and availability as text
inside rectangles that represent the flights. The right-edge of
the rectangles encode departure times, the left-edge encode
arrival times and the length of the rectangles encode flight
duration; 3)graphical layout that is similar to (2) but uses
shading to indicate availability and finally 4)graphical
layout that is similar to (2) but uses shading to indicate
availability and height to indicate price1.
THE MODEL
There are many different correct algorithms for finishing the
task specified above. In [3], Casner examines eight different
algorithms for the task. These algorithms capture different
types of spatial searches but by no means exhaust all
possible task methods. A complete analysis of the layouts
would try to identify and model a reasonable set of the
possible algorithms as [6] did. However, to limit the scope
of this preliminary analysis, a single straightforward,
efficient algorithm was used for each layout. From the
layout algorithms, we generated a series of CPM-GOMS
operators that the model uses to solve the task. In
particular, the most important task operation, the visual
search, was modeled with the combinations of cognitive,
motor and perceptual operators proposed in [5]. Other task
operations were broken down into CPM-GOMS operators
in similar ways. Each CPM-GOMS operator was assigned a
duration based on previous psychological studies [2].
Assumptions about the necessity of eye-movements were
based on Tullis' proposal that groups of information within
5 degrees of visual angle can be perceived in a single eye
fixation. Casner's users were using a 9x12 inch screen.
Assuming that the user was approximately 15 inches from
the screen, 5 degrees of visual angles is equal to 0.65
inches, which corresponds to approximately six 12 point
characters. In addition it is also assumed that the user
always starts scanning from the top-left corner of the
layout. This is based on Western convention of reading
from left to right, from top to bottom. Such an assumption
may not be suitable for users of other cultures. It is also
assumed that finger pointing used during the task as an aid
to avoid unnecessary visual search. This behavior was
observed by Casner during his empirical study.
RESULTS AND CONCLUSION
The performance time predictions made by the GOMS
model are given in Table1, together with the user
performance times collected by Casner. Using the
percentage difference figures from the last row in Table1,
we calculated the average absolute % error to be 8.31. This
relatively figure indicates that the total task performance
time predicted by the GOMS model match quite closely
with the empirical timings collected by Casner. Figure1
gives a more intuitive representation of this result.
Table 1: Total task performance time as predicted by GOMS
and as obtained by Casner;s empirical test
Figure1: Comparison between the observed performance
time and the GOMS predicted performance time.
Table1 also shows that the predicted performance time is
most inaccurate for Layout1. It is hypothesized that one of
the primary reasons that the GOMS model under predicted
for Layout1 is because it did not take into account problems
that users might have with remembering several
complicated pieces of information at any one time. From
the model, it was determined that in Layout1, a user would
have to keep a maximum of 9 chunks of information (4
chunks for the task constraints and 5 chunks for
intermediate results) at any one time. This exceeds the
average short-term memory capacity that is specified in [2]
to be 7 chunks. In the other three layouts however, the user
only has to keep a maximum of 6 chunks of information (4
chunks for the task constraints and 2 chunks for
intermediate results) in short-term memory throughout the
task. If the procedure of Layout1 were changed so that it is
less taxing to the user's short-term memory, then the
predicted task performance time would be higher and would
better match the empirical results.
It should be noted that the analysis given here is only a
preliminary analysis. First of all, only one procedure was
modeled, even though there are many possible procedures
that users can use to solve the problem. In addition,
Casner's empirical results gave average performance times
of several different data sets, whereas only one such data set
was modeled here. Nevertheless, the predictions of the
GOMS model obtained are very encouraging, With further
modeling and empirical validation, we believe that GOMS
will be able to provide relatively accurate measures for
various graphic and textual layouts.
ACKNOWLEDGMENTS
We would like to thank Steve Roth of CMU for pointing
us to Casner's thesis as a domain of study.
REFERENCES
- Bovair, S, Kieras, D. E., and Polson, P. G. The
acquisition and performance of text-editing skill: A
cognitive complexity analysis. Human Computer
Interaction, 5, pp. 1-48, 1990.
- Card, S.K., Moran, T.P., and Newell, A., The
Psychology of Human-Computer Interaction, Erlbaum,
Hillsdale, NJ, pp. 139--191, 1983.
- Casner, S.M. and Larkin J.H., Cognitive Efficiency
Considerations for Good Graphic Design. In Cognitive
Science Society Proceedings, Ann Arbor, Michigan, pp.
275-282, Erlbaum, Hillsdale, NJ, 1989.
- Casner, S.M., Task Analytic Design of Graphic
Presentations. Submitted to the Graduate Faculty of Arts
and Sciences in partial fulfillment of the requirements for
the degree of Doctor of Philosophy. University of
Pittsburgh, 1990.
- John, B., Extensions of GOMS analyses to expert
performance requiring perception of dynamic visual and
auditory information. In CHI'90 Proceedings, Seattle,
WA, pp. 107-115, ACM, NY, 1990.
- Tullis, T.S., Optimizing the usability of computer-
generated displays. In proceedings of HCIÕ86 Conference
on People and Computers: Designing for Usability,
London, British Computer Society, 1986.
1 Layout 4 was constructed by Casner's automatic visualization
system, BOZ.