BCB 520
2024-02-15
These are guidelines and considerations, not really absolute rules:
Common in news and some infographics.
Position channels are very powerful… when they are PLANAR SPATIAL POSITION, not depth!
We don’t really live in 3D: we see in 2.05D.
“Occlusion occurs when one 3D graphic partially blocks another. It is the result of mimicking space in the natural world–where objects have differing X, Y, and Z coordinates. In data visualization, occlusion obscures important data and creates false hierarchies wherein unobstructed graphics appear most important.” … From This BLOG post by M. Bowers.
Interaction can resolve occlusion, but at cost of time and cognitive load.
Perspective in 3D visualizations interferes with all size channel encodings. The power of 2D planar positioning is lost!
“Distortion occurs when 3D graphics recede into or project out from the picture plane through foreshortening. In drawing, foreshortening makes objects seem as though they inhabit three-dimensional space, but in data visualization, it creates more false hierarchies. Foreground graphics appear larger, background graphics smaller, and the relationship between data series is needlessly skewed.”…… From This BLOG post by M. Bowers.
3D bars are very difficult to justify!
Perspective distortion and occlusion make faceting into 2D the better choice in most situations.
Skewed perspective in 3D visualizations does not interact well with fonts.
What information can we decode from this visualization?
This version of the extruded time series visualization uses derived data by computing a cluster hierarchy of power usage patterns. It then juxtaposes multiple views: a calendar and superimposed 2D curves differentiated with a color channel.
The benefits of 3D visualization outweigh its costs when the task is shape perception for 3D spatial data. Interactive navigation often supports synthesis across many viewpoints.
External Cognition vs. Internal Memory
It is easy to compare by moving eyes between side-by-side views.
It is much more difficult to compare a visible item to your memory of what you saw.
Immersion is typically not helpful for abstract data because we do not need a sense of presence or stereoscopic 3D. A desktop view is also usually better for workflow integration.
Resolution in VR is a critical constraint, as pixels are the scarcest resource.
First wave: virtual reality for abstract data! This is difficult to justify.
Second wave: AR/MR (augmented/mixed reality) has more promise.
“Overview first, zoom and filter, details on demand”
This seems simple enough, until you realize that “Overview” is really a microcosm of the full vis design problem space.
Responsiveness (visual feedback to the user) has 3 rough categories:
It is dangerous to start with aesthetics because it is usually impossible to add function retroactively.
Start with focus on functionality because you can improve and refine aesthetics later.
Aesthetics do matter! They are another level of function. Consider working with a graphic designer on important visualizations.
Proximity:
DO group related items together
AVOID equal whitespace between unrelated items
Alignment:
DO find/make a strong line and stick to it
AVOID automatic centering
Repetition:
DO unify by pushing existing consistencies
Contrast:
If not identical, then very different
AVOID not quite the same
Make visualizations as self-documenting as possible!
Meaningful & useful title, labels, legends.
Axes and panes/subwindows should have labels and axes should have good mix/max boundary tick marks.
Everything that’s plotted should have a legend and its own header/labels if not redundant with main title.
Use reasonable numerical format and avoid scientific notation in most cases.
https://xkcd.com/833/