Re: Re: st: baseline adjustment in linear mixed models

[Clyde B Schechter <clyde.schechter@einstein.yu.edu>]

Paul Seed commented on the thread about adjustment for baseline values of outcomes in mixed models.

I don't think he and I actually disagree about anything substantive here.  He states that I suggested the mixed model is inherently superior.  I don't think I said that, and, in any case, I didn't intend my remarks to be taken that way.  In fact, I concluded my initial post in the thread by stating that the choice between the simple mixed model and the model with explicit covariates depends entirely on the science and, specifically, on whether the baseline values are distinguished, or whether they are exchangeable with the others.  And in my second post to the thread, I suggested a way to explore that very question.

Paul has done a nice job of pointing out some circumstances under which one might presume that the baseline values are distinguishable, and I think those are good, instructive examples.

My larger point is that the question often arises, both on Statalist and in daily work life, whether to include baseline variables as explicit covariates in these models and, my experience has been that the person asking the question is often unaware that the mixed model without explicit covariates is equivalent to the ANCOVA approach, if the baseline value is exchangeable with later observations.  I think this fact is under-appreciated, and I have often seen people "routinely" include explicit covariates without considering the issue.

Clyde Schechter
Dept. of Family & Social Medicine
Albert Einstein College of Medicine
Bronx, NY, USA 


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/