the user. Some included tips for using the system ( 4.1%), noting how to take advantage of specific features. Overall, 9.0%
of comments referenced the site design, either in the form
of usage tips or feature requests. A few comments included
to-dos for future work ( 2.6%), such as later adding a link to
a relevant wikipedia article. Others served solely as affirmations to another comment ( 1.5%). For example, people stating “I agree with that” to support a hypothesis. In many cases,
study participants would note out loud “that is interesting!”
without posting a comment to the system.
Finally, some comments were social in nature ( 9.0%).
Most pointed out trends in the data, but did so in a joking
manner. One user built a view comparing female lawyers
and bartenders, writing “Women at the bar and behind the
bar.” In the pilot study, one of our lab members annotated a
drop in stock brokers after 1930 with a picture of a person’s
trajectory off a skyscraper (Figure 7). This elicited smiles
and laughter from subjects in the subsequent study, one of
whom replied with an affirmation simply saying “Whoa!”
We also analyzed the structural aspect of comments.
Excluding comments from the pilot study, deleted test
comments, and those written by the paper authors, 195
comments were collected. Of those, 140 ( 71.8%) started new
discussion threads while 55 ( 28.2%) were replies to existing threads. The average thread length was 1. 35 comments
(SD 0.82), with a maximum of 5 comments. In some cases,
discussion spanned multiple threads.
5. 5. Graphical annotation
Next, we wanted to understand how graphical annotations
were used and to what degree they contributed to social data
analysis. Of the 195 nonpilot, nondeleted comments, 68
( 35.9%) included annotations. The vast majority ( 88.6%) of
annotations involved pointing to items or trends of interest.
The others ( 11.4%) involved more playful expression, such as
drawn smiley faces and the visual commentary of Figure 7.
Across these annotations, a total of 179 “shapes” were
drawn, with the options being free-form ink, lines, arrows,
figure 7: annotated view of stock brokers. the attached comment
reads “Great depression ‘killed’ a lot of brokers.”
rectangles, ovals, and text. Arrows were the most popular
shape ( 25.1% of shapes), and were used to point to items
as well as to situate information provided by text captions
( 24.6%). Ovals ( 17.9%) were primarily used to enclose regions
of interest. Free-form ink drawn with the pencil tool ( 16.2%)
was used for pointing, enclosing irregularly shaped regions,
and free-form drawing. Of the rest, lines made up 14.5% of
all shapes and rectangles only 1.7% (Figure 8).
A few users, particularly those with experience in graphic
design, noted that graphical annotations were their favorite
feature. Other users noted that the annotations were often
unnecessary for comments where text could describe the
trend(s) of interest. A few of these users added annotations to
such views anyway, saying the annotations were “surprisingly
satisfying,” enabling “personal expression.” Exit survey results
somewhat reflected these views, as users ranked annotations
more useful for writing their own comments (M = 3.5/5.0, SD
= 0.85) than understanding others’ comments (M = 3.2/5.0,
SD = 0.90). This difference, however, did not reach statistical
significance (t( 23) = − 1. 67, p < 0.108, two-tailed).
5. 6. Visitation and navigation
Our next questions concerned how users navigated the visualizations. Most users began exploring the data directly, starting
from the default overview and drilling down. A few immediately went to the comments listing to see what others had
done. Many participants searched for their own occupations
and those of friends and family. Other strategies included
browsing for items of interest found in the overview (“Wow,
look how the poor farmers died out”) and formulating queries
based on an over-arching interest, such as gender balance.
Looking to the usage logs, navigation by interaction with
the visualization or attached commentary was by far the
most common navigation technique, accounting for 70.5%
of state views. The second most popular was the back and
forward buttons at 17.5%, validating our integration of the
visualization with browser history mechanisms. Following
a link from the comment listings accounted for 8.7% of all
views, while the final 3.3% were due to clicking a bookmark
in the bookmark trail (Figure 9).
figure 8: usage of sense.us graphical annotation tools.
figure 9: usage of sense.us navigation mechanisms.
Comment Listings 8.7%
Bookmark Trail 3.3%
0% 10% 20% 30% 40% 50% 60% 70% 80%