I am looking to collect data for a multilevel CFA (two levels). I am unclear how to determine the number of clusters needed for identification. I know that the number of parameters is k(k+1)/2 multiplied by 2 (within and between portion of the model) plus k for the indicator means. With k = 35 (number of indicators), this is a very large number.
How do I determine the number of clusters to identify this model??
I don't think you have your parameter count correct. That's the number of moments going into the model, not the number of parameters.
Here's how I think about it, for a rule of thumb: At level 2, you're doing a factor analysis with 35 variables. What sample size would you be comfortable using if you were doing a single level factor analysis with 35 variables? I'd say a couple of hundred, at least (and ideally, a lot more). That's how many clusters you need.
Multilevel CFAs are kind of cool, and interesting, but it's really hard to find enough suitable data to run one.
