2
$\begingroup$

I am analysing a dichotomous predicted variable for a cluster randomised trial. The model is below. The predicted variable is binary and there are two measures, pre- and post-intervention. There is a categorical group factor (intervention vs control) and the clustering variable is pracID. The model looks like this:

mod4 <- glmer(irr_fu_therapy_intent_bin ~ group + irr_base_therapy_intent_bin + (1 | pracID), data = dfIntTher, family = binomial(link = "logit"))

and the output looks like this

Random effects:
 Groups Name        Variance Std.Dev.
 pracID (Intercept) 0        0       
Number of obs: 205, groups:  pracID, 20

Fixed effects:
                            Estimate Std. Error z value Pr(>|z|)    
(Intercept)                   -1.536      0.495   -3.10   0.0019 ** 
groupintervention              1.238      0.544    2.27   0.0229 *  
irr_base_therapy_intent_bin    4.816      0.713    6.76  1.4e-11 ***

Is it normal to get a 0 value like this for the between-cluster variance? I am confident that the dataframe itself is sound, so if I were very confident with this I would just assume this means that there is no correlation between individuals in eahc cluster, however I have read that the between-cluster variance is not as straightforward to estimate in a hierarchical logistic regression. I need to calculate the ICC and from here I have learned how to derive the within-cluster component, but I am a very wary of 0 values.

Any advice on this or alternative methods for calculating the ICC would be much appreciated

Improve this question
$\endgroup$
1
  • $\begingroup$ They may just be very close to zero and rounded when printed. Have you tried extracting them and then examining them? Or plotting the random effects? $\endgroup$
    – mdewey
    Commented Jan 9, 2017 at 11:21

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.