Hi, I'm planning to run some clustered longitudinal analyses and am looking at my modelling options. I have three levels of data: time, nested in students (n~2000), nested in schools (n~10). The school level clustering is mainly a nuisance - that is, I'm not including any covariates at this level and my goal is just to ensure I get the proper standard error estimates by accounting for this clustering properly.

It seems I have two modelling options. I could specify a two-level model with vce(cluster), as follows:

Code:
 meglm depvar indvar || studentid:, vce(cluster schoolid)
Or I could specify a three-level model, as follows:

Code:
 meglm depvar indvar || studentid: || schoolid:
Could someone explain the differences in these models to me? Is there any difference? I'm struggling to pick one over the other at the moment. Thank you in advance for any clarification.