Re: [S] Pie diagrams anyone?
Sat, 14 Nov 1998 12:47:07 -0700 (MST)

I think the discussion about pie charts shadows the empirical literature
about pie-charts versus bar charts.

1. If the goal is pure estimation of size, bar charts are often better
(Cleveland etc)
2. If the goal is estimation of proportion of a whole, bar charts and pie
charts may be equal (Simkin & Hastie, Spence and Lewandowsky).
3. If the goal is combination of segments and estimation of proportion,
pie charts may be superior (Spence and Lewandowsky).

These general rules about display efficacy are based on relatively little
data (in the big scheme of things). With experiments that relied
primarily on highly focused tasks (as opposed to the global
pattern-recognition tasks in exploratory phases of analysis). As the many
comments to list atest,
there are times where one format or the other is preferrable because of
either the conceptual system of the viewer, or the pattern of the data. I
think the role of the pattern of data has been greatly neglected in the
statistical and psychological literature (cf Lewandowsky and
Behrens"Statistical Graphs and Maps" in the Handbook of Applied Cognitive

The practical implication is that we dont always know the efficacy
of a plot until we see the exact pattern produced and judge it against the
questions that will be asked of it. Accordingly, the most important thing
is to plot things in different ways.

John T. Behrens Voice (602) 965-3384
Psychology in Education Fax (602) 965-0300
Measurement, Statistics
& Methodological Studies
Arizona State University
Tempe, AZ 85287-0611

This message was distributed by To unsubscribe
send e-mail to with the BODY of the
message: unsubscribe s-news