Constantine Daskalakis
>
> At 01:04 PM 5/14/03, Nick Cox wrote:
> >As I recall, Darrell Huff in his 1954 book implied that axes
> >which don't start at zero are deceitful. And I seem
> >to remember my high-school science teachers saying the
> >same thing in the 1960s and recommending a scale break.
> >[snip]
> >
>
> I beg to differ on this one Nick.
> I think being able to "break" the axis is a crucial feature
> in plots.
>
> Suppose I want to plot the percent of adolescents who've
> ever used a
> computer in the US over the past 10 years (estimated from repeated
> cross-sectional samples). During that period, this
> proportion has gone up
> from about 70% or so to over 95%.
>
> Y-axis is "% used computer" and X-axis is calendar time (years).
>
> First, I don't want to plot on the log scale. Why would I?
>
> Second, I don't want to waste three quarters of my graph
> area by using the
> full scale of the Y axis (i.e., 0-100%), when all my
> measurements are
> around 85-95%.
>
> I can start my Y-axis at, say, 60%, but I think that's more
> misleading than
> starting at 0% and having a "break" to jump to, say, 60%.
Your example doesn't convince me, but I think there
are better ones. In principle, I agree with this:
there is sometimes a need for scale breaks.
Evidently I should make my position clearer.
Ernest asked for a scale break indicated
in a particular way by a zig-zag on the axis.
My assertion is that not using that is not bad
practice, contrary to some old advice, and to
your apparent position here.
My question is how far it is really used in
scientific literature(s).
In addition, I drew attention to Cleveland's
suggestion for a full scale break whenever
a scale break is deemed essential.
Graphically this is a very different
practice. It is a matter of dividing the
data region into different panels.
In fact, one objection to the zaggy
axis is that it is not nearly drastic
enough as a way of signalling something
fundamental about the graph. Users too naive
or lazy to look at labels are also likely
to overlook the zags (or perhaps the zigs).
In my previous posting
I referred to an FAQ I wrote giving
a real example and explaining one way to
do it in Stata <8, at least roughly,
and as said the principle could be
applied with Stata 8.
What would do you do with a graph showing
SARS cases against time? Time starts
late 2002, say.
Nick
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
