Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: paired t-test and clustering

From   "Airey, David C" <david.airey@Vanderbilt.Edu>
To   "" <>
Subject   st: paired t-test and clustering
Date   Wed, 12 May 2010 19:07:31 -0500


The paired t-test is a t-test that the difference between the two score is zero. If you think of your difference measure as your trait, and remind yourself that families resemble each other more within families than between randomly selected families (genetics and shared environment), then the difference may be more alike within family, or clustered by family. So one family may be less likely to show change than another, for example, or the pretest/baseline may cluster regardless of the test effect. You could run -loneway- on your subject different scores by family, say those that have 2 or more subjects per family, -loneway diff family-, and look at the intraclass correlation coefficient and whether it is significantly different from 0. But I would not do this as a litmus to then do a t-test that allows for clustering, as your clustering test may be underpowered to detect it but it may nonetheless affect your ttest. Better to acknowledge biology and your design. There is al!
 so a clustered t-test available at ssc: -cltest-, in addition to the suggestion already made.

> My question concerns paired t-test and clustering.
> I would like to determine if a mean activity score before intervention is different from the mean activity score after intervention. I have used a paired t-test using the pre- and post-intervention activity score from the same person. The activity scores are normally distributed.
> Several reviewers have suggested that my estimates are too optimistic and incorrectly find a significant difference between the pre and post means because I have not controlled for clustering (correlated data) due to including multiple members from families. Measurements from multiple subjects from a family may be similar and not be considered independent observations. I have 300 subjects from 90 families.  
> Does using the individual as their own control in the paired t-test avoid the bias due to clustering? If not, what methods are available to compare the per- and post-intervention means while adjusting for clustering?
> Thank You Bert Stover 

*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index