Hi, everybody


I am working with student survey data and wondering if anyone had any advice for dealing with data that is cross-classified but with only one unit per crossed pair. Students in my dataset were surveyed once in multiple classrooms (the average is about twice, range 1 to 8), classrooms are nested in teachers, and teachers in schools.


I was hoping to model my dependent variable, which is specific to the classroom a student was surveyed in, using three level models of studentsXclassrooms at level one, teachers at level two, and schools at level three. However, I only have one observation per studentXclassroom pair and my models are having difficulty converging. (They converge fine when I limit the data set to one school and a very small number of observations, but my actual data set is huge). I have read several studies on the consequences of ignoring cross-classification and was hoping someone point me to studies evaluatingany alternatives, especially when there is just one observation per crossed pair? Also, I was wondering if anyone had any thoughts on this potential work around: give each studentXclassroom pair an id variable, nest this pseudo-level in teachers, and include a fixed dummy variable for each classroom at level one?


Thanks so much for your help!