Essentially comparing:

glm1 = glmer(Mortality ~ CCI + PatientRace + PatientSex + age_cat + (CCI | FacilityIdentifier), 
             data = tmp, family = binomial, 
             control = glmerControl(optimizer = "bobyqa"), nAGQ = 1)


m1 = glm(Mortality ~ CCI + PatientRace + PatientSex + age_cat, 
         family=binomial, data = tmp)

To determine if the random effect is a significant contributor, hopefully to show that each facility doesn't have varying practices in measuring CCI that may affect interpretation of mortality. Would appreciate any advice.


marked as duplicate by amoeba, kjetil b halvorsen, Peter Flom Mar 6 '18 at 12:08

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ Why not just bootstrap confidence intervals and see if that of the random effect's variance includes zero (non-significant) or not (significant)? You can simply do this with confint(glm1, method = "boot"). $\endgroup$ – Frans Rodenburg Mar 6 '18 at 5:45
  • 1
    $\begingroup$ @amoeba woops, didn't mean to include that $\endgroup$ – Cenoc Mar 6 '18 at 10:13