I'm a beginner trying my best to fit a multi-level model using complex survey data (American Community Survey microdata). My outcome is a binary variable at the person level, and my level two variable is metropolitan area (MSA). Ultimately, I hope to incorporate MSA-level predictors, but have not tried that yet.
The following code runs, but I get the error " ". I've gotten that error with a few specifications, which is why I'm using "FASTQUAD," "qpoints=1" and "inititer=10" > these changes were all based on other guidance I found online on how to address this error. Clear that "FASTQUAD" is ignored, though.
proc glimmix data=tmp2.usa1y2021_diss inititer=10 method=quadrature(fastquad qpoints=1)INITGLM empirical=classical;
class MET2013;
model doubledup (descending) =
/dist=binary cl solution link=logit obsweight=PERWT;
random intercept / type=un subject=MET2013 weight=MET2013wgt;
run;
WARNING: Sandwich covariance estimator is not available with the FASTQUAD suboption.The FASTQUAD suboption is ignored.
NOTE: The EMPIRICAL option for METHOD=QUAD adjusts the covariance matrix of the fixed effects and the covariance parameters. It also affects the prediction covariance matrix for the random effectssolutions.
NOTE: The GLIMMIX procedure is modeling the probability that doubledup='1'.
WARNING: The initial estimates did not yield a valid objective function.
--
I also found an example of someone using microdata with the following code, but this doesn't work for me. I thought it might be useful to go this route with something simpler, given the error I was receiving. This ran, but the results don't seem to be weighted to the population. Any help would be much appreciated!
proc glimmix data=tmp2.usa1y2021_diss;
where MET2013 notin(0);
weight perwt;
class MET2013;
model doubledup (desc)=
/dist=binary link=logit ddfm=bw solution;
random intercept/subject= MET2013 solution;
run;
