st: gllamm

["Glance, Laurent" <Laurent_Glance@urmc.rochester.edu>]

I am running a 2-level random intercept logistic model using gllamm.   When I use gllapred with the mu option, and then take the mean of the observed and predicted response variable (binary) - the means are the same.  When I use the linpred option with gllapred, and then use convert the linear predictor to a probability, the means of the observed and and the predicted response are slightly different.  According to the gllapred documentation, the linpred option includes both the random and fixed parts. 

Why the difference in means when using the linpred option vs the mu option??

 

Thanks!

 

Larry