Geography Reference
In-Depth Information
Participants were also asked, which type of map like they more. The distribution
of the preferences between 2D and 3D maps was almost balanced (19 for 2D
vs. 13 for 3D).
Fixation Detection
One of the most important issues in eye-tracking data analysis is event detection of
recorded data. For almost all analyses, the fixations and saccades are needed.
Eye-tracking data were recorded with sample frequency of 120 Hz, so the disper-
sion algorithm (I-DT), which is more appropriate for the low-frequency data,
was used.
I-DT takes into account the close spatial proximity of the eye position points in
the eye movement trace (Salvucci and Goldberg
2000
). The algorithm defines a
temporal window which moves one point at a time, and the spatial dispersion
created by the points within this window is compared against the threshold. If
such dispersion is below the threshold, the points within the temporal window are
classified as a part of fixation
otherwise, the window is moved by one sample, and
the first sample of the previous window is classified as a saccade (Komogortsev and
Khan
2009
).
The threshold values were set to 80 ms (duration) and 50 px (dispersion). These
values were selected based on the author
;
s unpublished study, which compares four
settings, used in cartographic papers and identified the thresholds, which fits to the
recorded raw data.
For data analysis, open-source software OGAMA was also used. Most important
parameters are “Maximum distance” and “Minimum number of samples”, which
corresponds to dispersion and duration in BeGaze. For optimizing the event detec-
tion parameters in OGAMA, the image of scanpath from BeGaze was used in
OGAMA instead of SlideResource image. The fixations in OGAMA were plotted
over the image of the BeGaze fixations. Event detection parameters in OGAMA
were changed until the scanpath was very similar to the scanpath from BeGaze.
'
Statistical Analysis
For statistical analysis of eye-movement data, several eye-tracking metrics were
calculated. For all metrics, median values for 40 respondents were calculated.
Median was used instead of mean because the data had not normal distribution
and median also filter out the extreme values. Data was analysed with the use of the
Wilcoxon rank sum test and statistically significant difference between 2D and 3D
maps was observed on the significance level
0.05 in all cases.
Analysed metrics were Time to Answer (click), Fixation Count, Fixation Dura-
tion Median and Scanpath Length.
α
¼
