www.cmm.bristol.ac.uk/forum

Hi george,

If I may, an additional question about the multiple membership vs. closs-classificaiton issue.

This is in the manual:

In the last chapter we considered cross-classified models and introduced the concept of a classification. All the classifications we considered were what we would describe as `single membership' classifications . This means that every lowest level unit is a member of one and only one classification unit. For example each pupil in the tutorial.ws dataset belongs to one and only one school and each woman in the bang1.ws dataset belongs to one and only one district.
It is however possible that we cannot (or do not want to) assign each lowest level unit to exactly one classification unit. This may be due to movements between units over the time period for which the data were collected. For example if our response is exam scores at 16 then some pupils will have been educated in more than one school and thus we may want to account for the effects of all schools.

I think we made the choice - based on this information - to go for multiple membership models: Pupils did move between classrooms, so it fits the description here. Did I misunderstand this?

The story continues...

We have looked seriously into cross-classified models now, but still unsure what to do.

What we have is the following:

5 timepoints
nested in students
nested in classrooms
nested in schools

Each timepoint is always nested within the same student.
However, the students are not within the same classroom at each timepoint (i.e., within schools, classroom can be re-arranged each time point, so that students now have different classmates).
The same classrooms are always nested within the same school.

If I would follow the example for cross-clasified, this would mean:

(-> indicating: nested in)
timepoint -> student -> classroomT1-> classroomT2-> classrooMT3 -> classroomT4 -> classroomT5-> school

The examples in the manual of primary schools nested in secondary schools nested in neighborhoods are easier, because the higher levels keep becoming smaller. But the number of classrooms stays the same between timepoints: there is not really nesting from T1 to T2 etc.

Basically we have not found the combination of repeated measures with these clustered analyses in the examples.

Any advice would be appreciated, we would also be open to having a skype call or something if that would help, or sending you more information about the data. Of course we are more than happy to share any outcomes on this forum, in case we find a solution.