Page 407 -

13.7 Summary 397

task. Such questions may involve examination of what the user is doing (which keys
they are pressing, where they are moving the mouse, where they are looking) and
how they are reacting (are they happy, sad, frustrated, or excited)?
Traditional measurement and observation techniques can be used to address these
questions, but they are limited in their applicability. Even the most careful observa-
tions and video recording are very limited in determining which keys a user presses
and how quickly they are pressed. Observation and video tape present similar limi-
tations for tracking mouse movements or eye gazes. Inferring emotional states is
similarly challenging: we may be able to identify excitement simply by watching
someone playing a video game, but more subtle responses such as frustration may
not be apparent. Asking users after the fact provides some detail, but questionnaires
or interviews are limited to details that the participant remembers after the fact, mak-
ing fine-grained data collection difficult, if not impossible.
Automated data collection approaches provide data that are unavailable through
these more traditional approaches. For studies of mice and keyboard usage, ac-
tions that are intrinsically part of user tasks can be recorded for further analy-
sis. Relatively simple data collection software can collect data tracking exactly
what the user did (mouse press, mouse movement, key press) and when she did
it. This information can be used to describe accuracy, identify problems in task
completion, and classify task completion into periods of activity and inactivity.
Combinations of multiple input devices—such as keyboard and mouse—can pro-
vide richer details.
Other interesting sources of human data may require a larger investment, po-
tentially in analysis and possibly in equipment. Costs of eye-tracking systems have
decreased significantly, but data analysis and interpretation can be a challenge. These
concerns are even more pronounced for physiological measurements, which require
equipping participants with electrodes, sensors, gauges, headbands, even helmets, or
even more complex machinery. Interpreting the resulting noisy data is another chal-
lenge that requires substantial experience in signal processing.
You might want to start with simpler, less expensive techniques before you
commit to the expense and difficulty associated with eye-tracking or physiological
approaches. You might try simpler measures such as observation, video recording,
or interviews, to see if they can be used to generate the insights that you need.
Another approach would be to find proxies: although you might be tempted to use
eye gaze to track a user's attention, tracking mouse movements might be a work-
able alternative. Eye tracking and galvanic skin response are tools that (perhaps
with a little help from appropriate experts) many HCI researchers should be able
to adopt for their own work.
For some research problems, the temptation of fine-grained physiological data
using neuroimaging or other advanced techniques may be too great to resist. If
you find yourself faced with such a question, be sure to work with experts: the
assistance of collaborators who are familiar with both the equipment and the data
interpretation challenges will be crucial to your success.