Thanks for the thanks.
Being rude about pie charts is all good fun and, for many of us, an acceptable variant on blood sports. But if precise decoding of proportions is the aim, then tables beat pie and bar charts hand down, or the actual numbers can just be added to the graph!
One major objection to pie charts is that their effectiveness breaks down very rapidly as displays depart only slightly from extreme simplicity. You need only think about comparing 30 category proportions across 3 pie charts, or 10 categories across 10 pie charts, to realise that.
Irrelevantly, but it just sprang to mind: Some of you may enjoy this algorithm, probably known throughout history, for sharing a pie (cake, whatever) between two children. One child gets to cut and the other gets to choose.
Nick
[email protected]
Ronan Conroy
As well as a thank-you to Nick for his very useful wrapper to Stata's
rather awkward -pie- command, -pieplot-, I thought I might share an
interesting paper which illustrates the history of this humble graph
and reviews the evidence for its supposedly inferior performance to
the bar chart.
The author concludes:
There seems to be little objective basis for a prejudice against the
pie based on considerations of speed or accuracy of estimation—the
pie chart does as well, if not better, on simple tasks such as the
estimation of a single proportion or the com- parison of a small
number of proportions. On the other hand, the natural competitors of
the pie suffer significant disadvantages. For example, the simple bar
chart does not provide an integrated representation of the whole, thus
making part–whole estimation more difficult. This drawback may be
alleviated by providing a reference bar, but the individual
proportions will be at varying distances from the reference bar. The
divided bar chart does provide a pictorial representation of the whole
but it is less desirable than the pie for exactly the same reason that
instru- ments that have circular dials (speedometers, altimeters,
airspeed indicators, clocks, etc.) are generally preferred to those
that use linear representations—they take up less space while
providing the same or better resolution. Finally, the pie provides at
least five natural anchors (0%, 25%, 50%, 75%, 100%) compared to only
two, or at most three, for the divided bar (0%, 50%, 100%). A bar chart
—without a reference bar—affords no natural anchors to assist in
the accurate estimation of proportions
http://www.psych.utoronto.ca/~spence/Spence%202005.pdf
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
