may_k posted on Wednesday, October 24, 2018 - 9:32 am
I am working with data with a three-level structure, with students nested in classes nested in schools.
First, I am interested in conducting two-level analyses with student at level 1 and class at level 2, in order to investigate effects of class-level variables on student variables. Differences between schools are not of interest. Is it correct to use type= twolevel and include school as a stratification variable so that standard errors are corrected for the third level of nesting? Or should I use type=complex two-level with class and school as cluster variables?
I am additionally interested in conducting single-level analyses, where differences between classes and differences between schools are not of interest but I wish to account for the three-level structure of the data in the computation of standard errors. Should I take a different approach in this case?
For your first question - you should use Type = Complex Twolevel where complex refers to the school variable.
The the single-level question, first to a Type = Threelevel Basic run to see at which levels the large variances are. Then do a Type = Complex run where Complex refers to the top level with highest variance.
may_k posted on Thursday, October 25, 2018 - 2:30 am
Sincere thanks for your prompt and clear response.