[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Statistical advice about multiple comparison...

From   "Newson, Roger B" <>
To   <>
Subject   RE: st: Statistical advice about multiple comparison...
Date   Mon, 28 Jul 2008 09:36:00 +0100

You might like to look at the somersd package, downloadable from SSC.
The somersd package calculates confidence intervals for a wide range of
rank statistics (Kendall's tau, Somers' D, Harrell's c, and median
slopes, differences or ratios), and provides versions for clustered data
(such as matched pairs). And the package comes with a set of .pdf

To download the somersd package within Stata, type

ssc desc somersd
ssc install somersd, replace

and to copy the manuals to the local folder, type

net get somersd.pdf
net get cendif.pdf
net get censlope.pdf

I hope this helps.

Best wishes


Roger B 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
Tel: +44 (0)20 7352 8121 ext 3381
Fax: +44 (0)20 7351 8322
Web page:
Departmental Web page:

Opinions expressed are those of the author, not of the institution.

-----Original Message-----
[] On Behalf Of jverkuilen
Sent: 25 July 2008 23:52
Subject: RE: st: Statistical advice about multiple comparison...

Nonparametric tests tend to lose a lot of power when you have many ties
as you will have here. 

You may benefit from a model for discete ordinal data such as -ologit-.
However it is not set up for repeated measures analysis. You may find
-gllamm- helpful there. See 

-----Original Message-----
From: "Yupa" <>
To: "Statalist" <>
Sent: 7/25/2008 6:04 PM
Subject: st: Statistical advice about multiple comparison...

Dear statalisters,

I have 9 test items and each item is scored ordinally 1 to 4 points 
(distance between points isn't equal).
The test is administered to two populations; in the first population 
each item is taken two times for each subject (at different ages). The 
second population served as reference.
Some of the test items show a ceiling effect (aren't able to 
discriminate at high end of the scale, and lower scores are not 
I have to compare the results of each item between the two timepoints of

the first population and between each timepoint of the first population 
versus the reference population.
I think that an equality test for matched data is adequate for the first

comparison (signrank) and an equality test for unmatched data (ranksum) 
for the second. 
Am I right or am I missing something?
May I analyze the data in a different way (different tests)?

Thanks in advance!
*   For searches and help try:

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index