Hi Statalist,
I have data on municipalities nested within regions nested within countries countries. I would like to estimate the degree to which variance in my continuous outcome can be explained by clustering within regions within countries. I would normally do this by calculating the intraclass correlation in the following way.
However, the population sizes of the municipalities in question vary widely both within and between regions. As a result I would like to weight the analysis such that larger municipalities contribute more strongly to my results. I would expect to do this using the code below, but stata does not appear to accept analytic weights in mixed models.
As a result, I have the following questions:
1. How can I calculate the ICC taking into account the varying population sizes of my underlying municipalities?
2. Am I even trying to estimate something meaningful? Is there some conceptual reason why analytic weights don't make sense in a multilevel model setting?
Thank you in advance!
Dan
I have data on municipalities nested within regions nested within countries countries. I would like to estimate the degree to which variance in my continuous outcome can be explained by clustering within regions within countries. I would normally do this by calculating the intraclass correlation in the following way.
Code:
mixed y i.country || region: , ml estat icc
Code:
mixed y i.country [aweight=population] || region: , ml estat icc
1. How can I calculate the ICC taking into account the varying population sizes of my underlying municipalities?
2. Am I even trying to estimate something meaningful? Is there some conceptual reason why analytic weights don't make sense in a multilevel model setting?
Thank you in advance!
Dan
