One possibility (requiring several unofficial Stata packages) would be
to proceed as follows:
1. Use -somersd- to produce a confidence interval and a P-value for
Somers' D comparing rates of return in each pair of groups.
2. Use -parmest- to save each of these confidence intervals in an output
dataset (or resultsset) with 1 observation and variables containing the
estimate, confidence limits and P-value.
3. Use -dsconcat- to concatenate all these output datasets (or
resultssets) into one long output dataset (or resultsset), with 1
observation for each pair of groups compared, and data on the Somers' D
parameter comparing these two groups, with its confidence limits and
P-value.
4. Use -smileplot- to enter all these P-values into a multiple-test
procedure (eg the Simes procedure), and thereby define a subset of
pairwise group differences which are "significant", even considering the
fact that there are multiple comparisons.
The -somersd-, -parmest-, -dsconcat- and -smileplot- packages are
downloadable from SSC if you need the most recent version (for Stata 9
users), or from my website at
http://www.imperial.ac.uk/nhli/r.newson/
(using the -net- command in Stata) if you want an earlier version of the
package in a previous Stata version. More papers and presentations about
resultsset-processing, and about -somersd-, and about multiple-test
procedures and smile plots, can be downloaded from my website at
http://www.imperial.ac.uk/nhli/r.newson/
I hope this helps.
Best wishes
Roger
Roger Newson
Lecturer in Medical Statistics
Respiratory Epidemiology and Public Health Group
National Heart and Lung Institute
Imperial College London
Royal Brompton campus
Room 33, Emmanuel Kaye Building
1B Manresa Road
London SW3 6LR
UNITED KINGDOM
Tel: +44 (0)20 7352 8121 ext 3381
Fax: +44 (0)20 7351 8322
Email: [email protected]
www.imperial.ac.uk/nhli/r.newson/
Opinions expressed are those of the author, not of the institution.
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Nick Cox
Sent: 10 January 2007 13:04
To: [email protected]
Subject: RE: RE: st: RE: RE: -oneway- and unequal variances - thanks
Others have made what look like excellent suggestions, but
with regard to your specific question here:
1. Nothing in Stata is available to follow -kwallis-
that I know of.
2. I am not aware that what you seem to want is defined
in principle, but that means no more than it says,
"I am not aware".
3. But something similar in spirit would be obtainable
by transforming the data into ranks and then feeding
those to some analysis of variance and then following
by multiple comparison stuff. A visceral guess is that
is just going to add complication to your analysis, not
insight. At best you could interpret it in terms of
differences between mean ranks.
I would just be upfront about the non-constant variance
as an intelligible feature of the data, not be paranoid
about doing something technically invalid. It is likely
that your P-values are not literally correct, but then
few are.
Nick
[email protected]
Rajesh Tharyan
> The post was
> referring to tests
> where you could carry out multiple comparision tests using
> nonparametric
> tests the same way as the oneway with the scheffe option
> would do (report
> the significance of the pairwise difference in means between
> the various
> groups).
*
* 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/
*
* 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/
