xqli
February 16, 2022, 1:27pm
1
Hi everyone. Excuse me. I want to ask you some questions about the covariator model.
Will the order of covariables affect the results of covariable screening? How to include appropriate covariates in covariable filtering?
Why is the CWRES correlation graph not significantly improved after the addition of covariator model?
Thank you for your help. I really appreciate it.
Hi XQLI,
in theory yes the order of covariate has an impact you might miss a covariate at later steps of a stepwise covariate search due to correlation between covariates.
scm: will use the most significant covariate at a given step
shotgun: will test all possible combination ( but not feasible when you have lots of covariates)
a covariate adds at the individual level not the residual level so you will see changed in your eta versus covariate plots, the values of your omega and shrinkage, the PRED might also be improved.
it is not common to plot CWRES vers covariate but that can give you some hints.
xqli
February 21, 2022, 12:11pm
3
smouksassi1:
Hi XQLI,
in theory yes the order of covariate has an impact you might miss a covariate at later steps of a stepwise covariate search due to correlation between covariates.
scm: will use the most significant covariate at a given step
shotgun: will test all possible combination ( but not feasible when you have lots of covariates)
a covariate adds at the individual level not the residual level so you will see changed in your eta versus covariate plots, the values of your omega and shrinkage, the PRED might also be improved.
it is not common to plot CWRES vers covariate but that can give you some hints.
Thank you very much for your reply. Maybe I understand.
xqli
February 26, 2022, 8:25am
4
smouksassi1:
Hi XQLI,
in theory yes the order of covariate has an impact you might miss a covariate at later steps of a stepwise covariate search due to correlation between covariates.
scm: will use the most significant covariate at a given step
shotgun: will test all possible combination ( but not feasible when you have lots of covariates)
a covariate adds at the individual level not the residual level so you will see changed in your eta versus covariate plots, the values of your omega and shrinkage, the PRED might also be improved.
it is not common to plot CWRES vers covariate but that can give you some hints.
Excuse me. Does the following diagnostic diagram suggest that there is an error in my structural model? Look forward to your reply.
CWRES.pdf (31.3 KB)