To calculate variance partition coefficient (VPC) in three-level logistic model?

[zhdanw]

Dear all,

I used SAS proc glimmix procedure to build a three-level mixed model with a link=logit, and got the variance estimate of second (physician) and third (hospital) level. However, since the variance of level 1 can't be got directly, I tried the simulation method and latent variable (3.29) to calculate the variance partition coefficient (VPC) or intra-class correlation coefficient (ICC), but the results from two methods was quite different.

Does anyone have such experience?