Search for Covariates

From 73532.1742@compuserve.com Wed Aug 9 11:49:15 1995 Subject: Search for Covariates Most of you will have read Stu Beal's answer to my previous question (FO vs FOCE). Many thanks, Stu! My question whether FOCE changes the chances of finding important covariates compared to FO was rather general in nature. When I search with FOCE, I use the individual predicted parameters in plots (or statistical evaluation) of potential covariates. When I search with FO, I use the weighted residuals instead. Given the situation that both FOCE and FO fit the data satisfactorily, can one theoretically claim that the individual predicted parameters are more suited in the search for important covariates? Or are the weighted residuals of FO more suited as they allow the evaluation the entire structural model vis-a-vis covariates? I tend to assume the latter, but I lack the theoretical background to prove it. On a practical note: Yes, the was a data set where FO was leading nowhere due to high standard errors of the predicted parameters (both fixed and random). With FOCE, however, the same kind of models were clearly defined. What was the characteristic of this data set? - It had large numbers of observations per individual (>30) over a long time period and multiple doses... Stu, you guessed right! Thanks again for all your input. Joachim Grevel BAST Inc. 8609 Cross Park Drive Austin, TX 78754 Tel: 512 837-8214 Fax: 512 834-7767 e-mail: 73532.1742@compuserve.com