Hi all,

I've been trying to better understand the covariance structure to select for a multilevel logistic model that includes two random intercepts and a random slope. The data are not repeated measures and the DV is binary. Essentially, the model is:

melogit dv i.iv1 c.iv2 c.iv3 || clustera: || clusterb: iv1

The random slope (iv1) is a binary variable. I'd like to better understand if the best choice is covariance(unstructured) or the default (independent). Reading Rabe-Hesketh & Skrondal as well as some forum posts, it seems that unstructured may be best because of the number of random parameters, but I'm not quite sure. Is it appropriate to run models with both covariance structures and use lrtest to determine which is preferred?

EDIT: Apologies - the title states "random slopes." It should read "random slope"