The world is made up of two kinds of people: those who want to see their information displayed in 3-D and those who want 2-D. Forget religion, politics, Elvis vs. Beatles, cats vs. dogs vs. kids: the real division is extruded shapes vs. plotted numbers.
I've stumbled across this line in the sand on every visualization project I've ever worked on, from the Wal-Mart sustainability site to redesigning my kitchen. (Surprisingly to me, it was the architect who rejected my suggestion of building 3-D models of the kitchen -- "Too hard to understand," he said. Wha?).
The issue is always the same: half the people are attracted by the 'life-like' feel of data rendered in physical form (the 2-D expression of an object so that it looks like it can be picked up and turned over in the hands) and half the people are attracted to the pared-down immediacy of numbers plotted X vs. Y.
Even Grandpa Tufte hems and haws on the rules, falling back on the axiom that says (I paraphrase) "Only use the third dimension if there is actual data that requires a third dimension in order to be visible." Thus the clouds he cleaned up for the folks at Urbana-Champaign retain their lovely 3-D volume (since the storm model in question has legitimate X, Y & Z data), while he's fine with Minard not showing Napoleon's dying troops superimposed on a 3D landscape (altitude being of zero consequence on a march across the utter flatness of western Russia. [I know, I've been there: between Moscow and Poland there is not a hill to be found]).
So I was greatly relieved to find someone with real data to help clear the air on this most essential of all questions. I was given Harvey Smallman's name by aircraft instrument designer Missy Cunningham of MIT. Harvey works at the Pacific Science and Engineering Group, Inc., where he and his co-workers are among the leaders on human perception of information and designers of interfaces for data-rich applications like air traffic control and military invasion planning.
The PSEG Inc. team have completed many studies and published many papers on the differing cognitive values of simulated 3-D vs. 2-D in presenting data sets. In their paper "Designing for the Task: Sometimes 2-D is Just Plane Better" (authored with Michael B. Cowen of the Space and Naval Warfare System Center), there are several insightful findings:
"3-D views are ambiguous along the line of sight, just as 2-D views are ambiguous in the missing dimension."
"Neither type of view is always best. However, we believe that these advantages and limitations are actually complimentary."
"Tasks that require integrating all three dimensions are better performed with a 3-D view, where the dimensions are integrated for the user... 2-D views are superior for relative position judgments."
"Confirming our predictions, we found that participants were three times faster and more accurate at shape understanding when using the 3-D views than the 2-D views."
"... to determine the distances along each dimension, tasks requiring a fair degree of precision for estimating angles and distances... participants were twice as fast and more accurate with the 2-D views than the 3-D views."
In the end, there is one big take-away: 3-D is better for showing shape and 2-D is better for showing measure.
So if you want to show what the resulting shape of the data are (what it looks like), use 3-D. If you want to show how far the data points are from each other, use 2-D.
"Our results would come as no surprise to architects. They would surely advocate showing a client 3-D renderings to visualize what they intend to build. But they would never pass those 3-D views on to the builder to specify the dimensions, rather they pass on 2-D plans."
There. Take that Napoleon.