Just noticed on page 3 of the new "Introducing S-PLUS 6" brochure the
following historically unsupported claim.
"Note the apparent reversal of the years at the Morris site, a data error
that went undiscovered for decades until the yields were plotted in this
way."
In a previous post (copy below) I gave reasons why I thought this claim was
historically false. It would be overstating the case to call it "false
advertising" but it is surely less than critical history and somewhat
insulting to some very good statisticians.
Surely there must be another less contentious example of the value of
Trellis graphs?
Trust this may be of some interest,
Keith O'Rourke
Statistician, Department of Surgery and Clinical Epidemiology Unit
University of Ottawa/ Ottawa Hospital
2. The following comment may be a "statisticians' " legend
> For an important case study of visualization see Cleveland's 1993, review
of
> the Barley data set in Chap 6 of Visualizing Data. In this case, the
visual
> approach revealed an error in the data that no formal analyses since 1934
had
> never noticed.
> Joe
I believe a claim something like that was made, but it is unlikely to be
true. In a 1938 paper, Yates and Cochran thoroughly evaluated a much smaller
possible error in the same data set. Apparently, the larger possible error
in the claim above had no important effect of the estimates they were
actually interested in!
see
Cox DR. Introduction to: Yates and Cochran (1938) The analysis of groups of
experiments. In:Contributions to Statistics 3. 1998;
[added to this post - Cox called the paper "remarkable"]
or
Yates and Cochran (1938) The analysis of groups of experiments. Collected
works of Cochran.
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|