Multilevel Analysis

[Hassen Ali]

Dear all, I am working My Master of Public Health (MPH)Thesis on Factors affecting Nutritional status of under-five children from DHS Data set. I planned to use multilevel mixed effect logistic regression(Two Level) using stata software version 14. My Question is,I have no clear understanding with the difference between Generalized Linear Mixed Model (GLMM) and Generalized Estimation Equation (GEE).Which is best to fit for DHS Data? Which one is better if I want to see the cluster variation of my Dependent variable?
To calculate Intra claster correlation coefficient (ICC)?
Sorry for All,I am a new stata user.
Your Guidance and assistance will improve My MPH Thesis!!
Respectfully,Hassen

[Clyde Schechter]

They are different models and they estimate different things. GEE estimates a population-averaged model. That is, when you use GEE the results are estimates of the difference in the population average outcome associated with a unit difference in the predictor. By contrast in the GLMM, the results are estimates of the average difference in individual outcomes. For linear models these are the same, but in non linear models they are not. So the choice is based on which of these is the answer to your research question. It is not a question of choosing the more convenient one. You have to be clear what your research question is. They will not both be correct answers; at most one of them will.


To my knowledge, you cannot get an ICC after -xtlogit, pa- (the GEE).

[Hassen Ali]

Dear Clyde,Thank you very much for for your assistance and guidance!! It is extremely helpful.
With Best Wishes,Hassen

[Hassen Ali]

Dear all,I have the following difficulties in my MPH Thesis Analysis part.
1) can I know how much is the cluster variability of my dependet variable if I use Generalized estimation equation (GEE)? If it is possible can you help me giving the general command in order to see these variability,I am using Stata version 14.1.
2) Can Stata version 14 produce the output of QIC (Quasi-likelihood Information Criteria) and CIC (Correlation Information Criteria) in order to compare models if I use Generalized Estimation Equation? if possible any command.
Dear all sorry for all,since I am a new Stata user.
Kind regards,Hassen

[Clyde Schechter]

I am not aware of anything for 1).

For 2), there are user written programs at both SSC and Stata journal. Run -findit qic- for links to these. I have no personal familiarity with these programs so I cannot recommend or disrecommend them, though both SSC and the Stata journal generally exhibit care in the programs they curate. I do not know if either of them runs on version 14.1. But you can find those things out easily at the links to the software itself.

[Hassen Ali]

Dear Clyde Schechter,Thank you very much for your assistance and guidance!!
I wish you have an endless happiness and success in your life my sweet.
Respectfully,Hassen

[Hassen Ali]

Dear all,As I describe my title at #1 above,I have planned to use the following code: melogit y x [pweight=wvar1]||psu:,pweight(wvar2) in Stata 14.1/SE. My challenge is,Since I have used Demographic and Health Survey Data (DHS Data),The weight variable are the same for all level for the data. Is there any problem using the same "weight variable " instead of "wvar1" and "wvar2" in the command part? Can it affect fixed and random estimates? I have no clear understanding regarding "svy" command,I need explanation.
Can I use the command for Two-level Mixed effect logistic regression?
With Best Wishes,Hassen

[Hassen Ali]

Dear all, I am waiting reply for #7!! Sorry,I am a beginner of using Stata. I am using Stata 14/SE.
Respectfully,Hassen