Interpretation of odd error variances in glmer

It is quite legitimate for the ML estimates of a variance component to
be zero.  It simply means that there is not enough variability
accounted for by that term to warrant incorporating the term.

Thanks for the swift response.

Given that the estimate seems to be exactly zero, can I read your answer
to say that the term has somehow been dropped by the algorithm?


I've investigated a bit more and find that the zero variance seems to
occur with the combination of the inclusion of a department-level
covariate (depfemr), and the cross-classifying individual-level random
effect (ulid):


Cross-classified model without dept-level covariate: 

    Generalized linear mixed model fit by the Laplace approximation 
    Formula: gradeA ~ 1 + cao + gender * yr0 + modsize + (1 | deptno) + (1 | modinst) + (1 | ulid) 
       AIC   BIC logLik deviance
     13379 13457  -6681    13361
    Random effects:
     Groups  Name        Variance Std.Dev.
     modinst (Intercept) 1.55918  1.24867 
     ulid    (Intercept) 2.23646  1.49548 
     deptno  (Intercept) 0.37642  0.61353 
    Number of obs: 43955, groups: modinst, 9686; ulid, 1850; deptno, 28
    
    
Cross-classified model with dept-level covariate added:

    Generalized linear mixed model fit by the Laplace approximation 
    Formula: gradeA ~ 1 + cao + gender * yr0 + modsize + depfemr + (1 | deptno) + (1 | modinst) + (1 | ulid) 
       AIC   BIC logLik deviance
     13301 13388  -6640    13281
    Random effects:
     Groups  Name        Variance Std.Dev.
     modinst (Intercept) 3.7718   1.9421  
     ulid    (Intercept) 2.4429   1.5630  
     deptno  (Intercept) 0.0000   0.0000  
    Number of obs: 43955, groups: modinst, 9686; ulid, 1850; deptno, 28


Brendan

Brendan Halpin,  Department of Sociology,  University of Limerick,  Ireland
Tel: w +353-61-213147 f +353-61-202569 h +353-61-338562; Room F2-025 x 3147
mailto:brendan.halpin at ul.ie  http://www.ul.ie/sociology/brendan.halpin.html