s-news
[Top] [All Lists]

Re: paired values and covariates

To: asanquer@virbac.fr
Subject: Re: paired values and covariates
From: Frank E Harrell Jr <f.harrell@vanderbilt.edu>
Date: Tue, 23 Nov 2004 10:03:15 -0500
Cc: s-news <s-news@wubios.wustl.edu>
In-reply-to: <OFE527431A.67305D15-ONC1256F55.004EE596-C1256F55.004F955D@virbac.fr>
References: <OFE527431A.67305D15-ONC1256F55.004EE596-C1256F55.004F955D@virbac.fr>
User-agent: Mozilla Thunderbird 0.8 (X11/20040913)
asanquer@virbac.fr wrote:
Hi,
I need an advice:
I would like to compare paired observations (pre- and post-treatment
measures)to a fixed value: does the change between pre and post values is
significantly higher than 50?

First I thank to use a classical paired t test, but I would like to use an
adjustment covariate (to take into account an important factor).
What is the correct method to use?

If I use an analysis of covariance:

pre = post + covariate

which part of the results will give me the answer related to the fixed
value 50?

Thanks for your replies

Annaële

The pre-test-post-test design has too many problems to trust the result (regression to the mean, etc.). Analysis of covariance is a good idea but you have the pre and post interchanged. And unless you constrain the slope to be one (which is not a good idea) you can't get a simple answer to your question but rather need to get several pre level- specific answers.

If you do want to constrain the slope to one, you can do an analysis not unlike the paired t-test where the response variable is post-pre. If you then have a covariate though, the tests will be conditional on specific covariate values.

You may need a parallel-group design to get a reliable answer to your question.
--
Frank E Harrell Jr   Professor and Chair           School of Medicine
                     Department of Biostatistics   Vanderbilt University

<Prev in Thread] Current Thread [Next in Thread>