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

Re: st: Help on data analysis strategy

From   David Airey <david.airey@Vanderbilt.Edu>
Subject   Re: st: Help on data analysis strategy
Date   Tue, 29 Jul 2008 14:03:16 -0500


:Ppp <rasberry>.

To get to a simpler model, you could remove the repeated measures aspect of the experiment to go from a mixed model to a fixed effects model. So take a summary statistic of day1 to day7 that best captures the intent of the experiment. Something like the median score or the highest score over the seven day period. Then use ordered logit (ologit in Stata) for each question's endpoint. Be sure to include day0 as a covariate. I prefer this to cutting the ordinal data to binary data or dealing with multiple tests over days (you are not looking for which day shows most improvement etc.). If you have to have multiple tests, I'd probably deal with the 4 questions as a source of multiple tests. (If you have access to a statistician, and the 4 questions index the same construct, a statistician might be able to use GLLAMM or Mplus to use all your data and the 4 questions in one latent variable ordinal mixed model.) A possibility not requiring a statistician, may be to derive a summary statistic for each person as above, and then use a multivariate kruskal-wallis test (see umbrella via ssc). Probably also somersd (via ssc) could be employed here...


On Jul 29, 2008, at 12:50 PM, Lachenbruch, Peter wrote:

While this may be solid advice to experienced statisticians and Stata
users, I would be a bit concerned about letting a novice loose on GLLAMM
- it's tough sledding for most experienced users.


Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001

-----Original Message-----
[] On Behalf Of David Airey
Sent: Tuesday, July 29, 2008 8:09 AM
Subject: re: st: Help on data analysis strategy


My text was stripped in my reply.

There is a chapter (#7) in Multilevel and Longitudinal Modeling Using
Stata (Stata Bookstore) that describes mixed models for ordinal data
using the command GLLAMM from ssc. This is what you need.


Dear subscribers,

I am new to statistics and Stata, and I would like to ask for
advice, if I amy, regarding the type analysis for a clinical

We have 2 groups of patients, 30 in each group that undergo surgery
and receive either standard medication or a new medication to help

Both groups are asked 4 questions regarding for example pain,
inflammation ect and they are required to give an answer that gets a
score from 0 to 5.
All 4 questions are asked repeatedly for day0 (before treatment)
day1, day2, day3, day5 and day7.

The objective of the study is to see if there is a difference
between the control and the experimental group as determined by the
answers to the four questions.

Some of the ideas I have are the following:

1. Perform a Mann Whitney test, ordinal data, between the control
and the experimental group at each day and for each question

2. Define an endpoint per question. For example for the question on
pain define as endpoint when the answer is no pain, and use right
sencoring for persistent pain after day7. Perform a survival
analysis for each question and compare the survival curves for the 2

3. Convert to binary data, for example pain=yes for score 1 to 5,
and pain= no for score 0. Perform logistic regression and evaluate
the effect of treatment separately for every question.

Your advice would be greatly appreciated.

I understand the above questions might be of limited interest to
most subscribers but, anyways, I would like to thank you for your

Best regards,

Nikolaos Pandis
*   For searches and help try:

*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index