[R-sig-ME] Fwd: Request for help using a generalized linear mixed model in correlated data
D@v|d@Du||y @end|ng |rom q|mrbergho|er@edu@@u
Wed Jan 8 03:12:26 CET 2020
The nature of the GEE (ie marginal) model means that it should agree with the "naive" model ignoring clustering.
One way you can use your logistic-normal GLMM is to predict risk for individuals with comparable covariate values using
predict(mod, type="response"), and calculate the resulting risk difference. Or take your log odds ratio and apply it to a given base rate -
the hypothesis testing done using the logistic link is correctly allowing for the clustering.
Recall that the different links will entail different distributions for the cluster means and correlation - ie a logistic link might be more appropriate for the biology generating your data.See also the zoo of alternatives for glmmTMB (beta, beta-binomial, negative binomial etc) that you could send another few weeks on.
Cheers, David Duffy.
More information about the R-sig-mixed-models