Page 400 -

390 CHAPTER 13 Measuring the human

of all data points) can often provide a useful means of reducing data volume without
sacrificing fidelity or accuracy (Rick et al., 2015).
If you are trying to link physiological responses to specific actions or events,
you may face the stream of integrating data streams that are collected separately—
perhaps even on different computers. Although your application data may be fine-
grained logs of individual events, physiological data streams may not have access to
that information. If all data collection is done on one computer, the timestamp might
be used with both data streams. When physiological data is captured on a separate
computer, some clever engineering might be necessary. One set of experimenters
used a modified mouse to solve this problem: in addition to sending control signals to
the computer running the application, the mouse had a second wire that sent a pulse
to the computer collecting physiological data. These pulses were used to synchronize
the two streams (Scheirer et al., 2002).
Appropriate use of tools and validated approaches can simplify matters some-
what. Many eye-tracking systems will come with associated software that will collect
and analyze data, potentially sparing you from the need to clean noisy data streams
and identify fixations. Ideally, such tools will provide access to raw data along with
summarized data, providing you with the means to conduct your own detailed analy-
ses as needed.

13.5.3 DATA INTERPRETATION
Given multiple streams of complex, synchronized data involving one or more physi-
ological signals and interactions with one or more computer programs, potentially
alongside complementary data including survey data and audio/video recording, how
can this data be interpreted?
One initial possibility is manual review. Particularly in earlier stages of inter-
pretation, looking at the signals to find examples of any anomalies, episodes of
interactions that might be informative, or other similar items of interest can often
be a good way to decide where to explore in more details. Tools that facilitate com-
parison and alignment of multiple data streams can be very helpful in this regard.
ChronoViz (Fouse et al., 2011) provides features for alignment and side-by-side re-
view of multiple temporal data streams, allowing users to, for example, review syn-
chronized displays of screen-capture video alongside physiological measurements.
The “LAB-IN-A-BOX” sidebar discusses the use of the ChronoViz tool to analyze
complex, synchronized data streams.
Identification of specific items or actions is often a first step. For example, you
might be interested in seeing how often the user in an eye-tracking study looks in
a certain region of the screen. When criteria are clearly and objectively defined, the
identification of relevant intervals or incidents is generally straightforward.
Data granularity can also influence analysis and interpretation. For simple com-
parisons involving overall responses to differing conditions, averages might be suf-
ficient (Mandryk and Inkpen, 2004). More complex analyses might attempt to model
and classify episodes of emotional reaction (Scheirer et al., 2002), potentially using