i
i
i
i
i
i
i
i
690 27. Visualization
However, a cost to interaction is that it requires human time and attention. If the
user must exhaustively check every possibility, use of the visualization system
may degenerate into human-powered search. Automatically detecting features
of interest to explicitly bring to the user’s attention via the visual encoding is a
useful goal for the visualization designer. However, if the task at hand could be
completely solved by automatic means, there would be no need for a visualization
in the first place. Thus, there is always a tradeoff between finding automatable
aspects and relying on the human in the loop to detect patterns.
27.5.3 Animation
Animation shows change using time. We distinguish animation, where succes-
sive frames can only be played, paused, or stopped, from true interactive control.
There is considerable evidence that animated transitions can be more effective
than jump cuts, by helping people track changes in object positions or camera
viewpoints (Heer & Robertson, 2007). Although animation can be very effec-
tive for narrative and storytelling, it is often used ineffectively in a visualization
context (Tversky et al., 2002). It might seem obvious to show data that changes
over time by using animation, a visual modality that changes over time. How-
ever, people have difficulty in making specific comparisons between individual
frames that are not contiguous when they see an animation consisting of many
frames. The very limited capacity of human visual memory means that we are
much worse at comparing memories of things that we have seen in the past than
at comparing things that are in our current field of view. For tasks requiring com-
parison between up to several dozen frames, side-by-side comparison is often
more effective than animation. Moreover, if the number of objects that change
between frames is large, people will have a hard time tracking everything that
occurs (Robertson et al., 2008). Narrative animations are carefully designed to
avoid having too many actions occurring simultaneously, whereas a dataset being
visualized has no such constraint. For the special case of just two frames with a
limited amount of change, the very simple animation of flipping back and forth
between the two can be a useful way to identify the differences between them.
27.6 Composite and Adjacent Views
A very fundamental visual encoding choice is whether to have a single composite
view showing everything in the same frame or window, or to have multiple views
adjacent to each other.