population-average marginal generalized linear models with logarithmic links and robust variancE.

[Malk020]

Hi, I've been feeling extremely stressed and sad while trying to find the best model for this. I've experimented with PROC MIXED, GLM, and GLIMMIX, but none of them provided the results I am looking for.

The outcome variable is continuous.

Thank you,

[sbxkoenk]

Although the information given is minimal,

I am sure this topic belongs to the "Statistical Procedures" -board.

So, I moved it from "Programming" - board to this board!

[sbxkoenk]

Hello,

• Are you having fixed effects only / random effects only / mixed effects?
• Are you having longitudinal data / repeated measurements on same subject?
• Are you having an experiment with a particular design (like split-plot or Cross-Over)?
• Are your data hierarchical?
• Are you having non-linearities in the parameters?
• Are you having convergence problems?

Maybe you need PROC GEE or PROC GENMOD??

Koen

[StatDave · Accepted solution]

The Generalized Estimating Equations (GEE) model is a population-averaged marginal model and supports generalized linear models with various links for clustered data. The procedure to use is PROC GEE. Specify the response distribution in the DIST= option in the MODEL statement. Specify the log link with LINK=LOG option in the MODEL statement. Add the REPEATED statement with the SUBJECT= option to specify the variable whose levels indicate the clusters of correlated observations. See the examples in the PROC GEE documentation.

[Malk020]

Hi all, thank you so much for all your responses; they are truly helpful and make me think more deeply, being aware of all models.

My professor handed over the analysis conducted by the person prior and wants my result analysis to match with that person's.

I tried GEE, and the results became close to the person who conducted the analysis previously. 

In the table, they asked me to fill out the fold change - 95%CI - P-value. 

I have calculated the fold change( ratio= visit6/ baseline). I used the one-sample t-test where H0=1.
Based on the t-test, the results are close to hers, except for the p-value. Then, I used the log, and after that, I redid the t-test to only get the p-value where H0=0, and the result was too far from the person prior.

Now, I am thinking of using GEE with the link=log. From there, I use the p-value result, as the log gives us an idea of how the hypothesis is different from zero. When I did these two steps, first the t-test, then GEE, my results relatively match with that person.

Do you all think this is a valid method to use?

Thanks once again!