Thanks to Kit Baum, there is now a new upgrade of my -smileplot- package
downloadable on SSC. Type -net desc smileplot-, -ssc desc smileplot- or
-findit smileplot- to find out more.
The -smileplot- package is a data mining tool for use with multiple
parameter estimates. It takes, as input, a data set with 1 obs per
estimated parameter for one or more models, and data on the P-values,
estimates, and other parameter attributes. (Such a data set might be
created by the packages -parmest- and/or -dsconcat-.) The -smileplot-
package contains 2 programs, -multproc- and -smileplot-. -multproc- carries
out multiple test procedures on the set of P-values, with a choice of
methods (eg the Bonferroni method or the much less conservative Simes
method). -smileplot- calls -multproc- and then creates a smile plot,
plotting the P-values on a reverse log scale on the Y-axis against the
estimates on the X-axis, with reference lines on the Y-axis corresponding
to the uncorrected and corrected P-value thresholds. The smile plot
therefore enables a user to see simultaneously both the statistical
significance and the practical significance of each parameter estimate.
(Note, however, that the X-axis does not have to be the parameter estimate.
For instance, in a genome scan, it might be position of a gene on a
chromosome.)
In the new version, both -multproc- and -smileplot- are byable. Therefore,
with a single command line, the user can carry out the same multiple test
procedure on more than one by-group of parameter estimates. These by-groups
might be a set of unadjusted odds ratios and the corresponding set of
confounder-adjusted odds ratios. There are also added optional generated
output variables, containing the uncorrected overall P-value threshold for
the by-group, the corrected overall P-value threshold for the by-group, and
an indicator variable equal to one if the null hypothesis corresponding to
a parameter is rejected and equal to zero if this null hypothesis is not
rejected. Therefore, a user can use the same multiple test procedure on a
large number of unadjusted odds ratios and the corresponding
confounder-adjusted odds ratios, and list the ones that appear
"significant" before and after adjusting for the confounders.
Best wishes
Roger
--
Roger Newson
Lecturer in Medical Statistics
Department of Public Health Sciences
King's College London
5th Floor, Capital House
42 Weston Street
London SE1 3QD
United Kingdom
Tel: 020 7848 6648 International +44 20 7848 6648
Fax: 020 7848 6620 International +44 20 7848 6620
or 020 7848 6605 International +44 20 7848 6605
Email: [email protected]
Opinions expressed are those of the author, not the institution.
