[R] Choice of notch size in R

hadley wickham h.wickham at gmail.com
Tue Apr 22 21:07:02 CEST 2008

On Tue, Apr 22, 2008 at 2:00 PM, Alex Reynolds
<reynolda at u.washington.edu> wrote:
> Many thanks to you and Peter Dalgaard for your advice. Instead of
>  notches on a classic box plot, I have elected to draw a more
>  "minimalized" box plot along the lines suggested by Tufte, and overlap a
>  custom significance region on this.

Tufte's version are also known as mid-gap plots, and if you're going
to use them, you should be aware of:

	Author = {Stock, William A. and Behrens, John T.},
	Journal = {Journal of Educational Statistics},
	Number = {1},
	Pages = {1-20},
	Title = {Box, Line, and Midgap Plots: Effects of Display
Characteristics on the Accuracy and Bias of Estimates of Whisker
	Volume = {16},
	Year = {1991},
	Abstract = {Examined here were the accuracy and bias of estimates of
whisker length
based on box, line, and midgap plots. For each type of graph, a different
sample of undergraduates viewed 48 single-plot graphs. For each plot,
subjects were given the length of an interquartile spread and asked to esti-
mate the length of a whisker. Plots varied in spatial orientation (horizontal
or vertical), interquartile spread, the ratio of whisker length to
spread, and whisker judged. Overall, estimates of whisker length for box
and line plots were more accurate and less biased than those for midgap
plots. Interquartile spread, the ratio of whisker length to interquartile
spread, and the interaction of these two factors significantly influenced both
accuracy and bias. Boxplots displayed a predicted pattern of over- and
underestimation. We discuss how the present results complement related
work of others (e.g., Cleveland & McGill, 1984) on depiction of quantity.
We conclude that midgap plots are less optimal displays than box and line



More information about the R-help mailing list