Date: Wed, 3 Sep 2003 14:52:31 -0400
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Arthur J. Kendall" <Art@DrKendall.org>
Organization: Social Research Consultants
Subject: Re: Bad news about SPSS 12.0 - Not a Mandate
Content-Type: text/plain; charset=us-ascii; format=flowed
It is certainly possible to misuse something as powerful as SPSS, but it
is a toolbox, but that is simply the flip side of the fact that it can
I have to admit that it is over 25 years since I looked at the research
behind Cleveland's work. If I recall correctly, the math psychologists
at Bell Labs seemed to have clearly established that people are
decreasingly accurate in comparing:
horizontal distances with a common end point
horizontal distances with no common end point
vertical distances with a common end point
vertical distances with no common end point
The susceptibility of human perception to many optical illusions is also
well established. (think of the St. Louis Arch, what is its aspect ratio? )
The points made in Huff's "how to lie with statistics" are certainly
consistent with what I have seen clients in the public policy sector do.
Brewer and the other consultants on producing the Census Atlas can show
how there can be many artifacts in perception of graphics representing data.
That being said, I believe that the potential for misinterpretation is
well enough established that presentation graphics should be pre-tested
on cold readers to see if the message they report they receive is the
same as what the author intends.
A graphic image is one leg of a three legged stool. It should be
properly labeled, common optical illusions avoided, etc. However, other
aspects of the same information should be communicated in the text, and
in the tables.
Social Research Consultants
University Park, MD USA
Peck, Jon wrote:
> The aversion to pie charts, especially 3D pie charts, among
> statisticians is endemic. Unfortunately, this prevalent belief is not
> supported by research. Writers such as Ed Tufte also assert that
> "chartjunk" features such as pseudo-3D and drop shadows interfere
> with graphical decoding. These assertions, as well, are unsupported
> by research.
> Studies on the perception of pie charts and other graphical forms are
> frequently subject to serious design flaws and other methodological
> problems. The few well-designed studies (such as Simkin and Hastie in
> JASA, 1987) fail to support the pie chart canard. I report these
> results and add further instances in my book, The Grammar of Graphics
> (Springer, 1999).
> Last year, in my graduate class on statistical graphics, we did yet
> another study of pie charts. We used a computer screen to present
> randomized sequences of pie charts. A slice was highlighted in each
> chart and subjects were required to estimate the proportion-of-whole
> for the targeted slice. Position of the target slice was randomized.
> We did this for 2D and 3D pie charts and 2D bar charts. We ran a
> large number of judgments by 8 subjects in a within-subjects design.
> The results showed no significant differences for 2D vs. 3D. Pies
> actually did marginally better than bars. The power of the study was
> high enough for this to be a serious "no-support-for-the-pie-canard"
> As for "chartjunk," the research results fail to support common
> proscriptions. You can take a look at the work of Lewandowsky and
> Spence, for example. They added pseudo-3D effects to bar charts and
> found no degradation in performance. (Spence is a well-known
> cognitive psychologist at Toronto.) Furthermore, there is no research
> support for the "data-ink ratio" as a measure of graphical quality.
> And "moire" chartjunk effects (zigzag, strip, dotted textures) do not
> necessarily bias perception, whatever their aesthetic value.
> This is a rather complex research area in which everyone thinks he or
> she is an expert. Tufte and other gurus have had an enormous
> influence on popular opinions on this subject, but frankly, little of
> what Tufte says is based on research and he has declared more than
> once that he isn't interested in research findings. Fine. He has an
> aesthetic (a nice one) and he is entitled. But before we make
> judgments about aesthetics and the suitability of pie charts or other
> graphical features to the display of data, we need to refer to
> research, not pronouncements of gurus and conventional wisdom.
> One more thing. Chartjunk may not be the kind of stuff you want to
> publish in a scientific journal. But if you need to grab someone's
> attention in a business or media environment, properly-applied
> chartjunk aesthetics can do that without biasing perception. If
> that's the case, why grudge its presence in a package designed for
> both business and research? And, furthermore, since SPSS has a
> world-class visualization team, I think we're in a better position to
> design these graphics (so as not to cause harm) than most companies.
> So we're going to do more of it in the future, as well as add more
> scientific (spartan) charts.
> Leland Wilkinson Sr. VP, SPSS Adjunct Professor of Statistics,