I'm using Method=quad to obtain AIC values and wondering if a model result that does NOT include a standard error for the covariance param estimate (value = ' . ' ) is still robust, in terms of fixed effect parms and model selection. SE estimate for intercept ("Territory" below) is obtained using Residual PL estimation, but not sure log likelihood estimate is valid for model selection procedure. Any thoughts would be appreciated.
proc glimmix data=BIRDNEST Method=quad;
model Presence = HAB1 HAB2 / link=logit dist=binomial solution cl;
random int/ SUBJECT=Territory ;
Message was edited by: bgdph
Message was edited by: bgdphd