My aim of the dissertation is to determine the effect of annual hematopoietic stem cell transplant (HSCT) procedures on overall survival (5 year follow-up) and 30 days readmissions after initial discharge of HSCT by adjusting covariates (organizational, HSCT program, and patient characteristics). My data are population-based with nested characteristics. I am planning to use Mplus for conducting multilevel Cox regression (time to death and time to readmission) or multilevel logistic regression (dead or alive and readmission or no readmission). However, my advisor is concerning about "overexplanation" by adding two organizational characteristics (ownership and geographic location) because multilevel modeling has accounted for clustered characteristics. One of my committees suggested univariate analyses for deleting covariates with p > .15 to .20 before using multilevel modeling. But my concern is that univariate analyses do not explained nested characteristics. I am wondering it is an adequate way to kick out covariates before using multilevel modeling. Or, shall I add one by one covariate in the multilevel modeling and examine its impact on the outcomes? I have seen model command in page 274 of Mplus user¡¦s guide (5 ed.). Could you tell me which page has multilevel logistic regression command in the user¡¦s guide? Thanks.
I am a bit unclear on your situation. Perhaps you are saying that you have 2 level 2 covariates, ownership and geographic location. If that's the case, I would first do a twolevel analysis without covariates to see if there is level 2 variation in random intercepts. And then add the level 2 covariates to explain that variation. The version 5 UG ex 9.3 shows twolevel logistic regression; just get rid of the y mediator.