Multilevel model with longitudinal repeated measures I'm relatively new to mixed linear models. I have a dataset of education data measuring classroom behaviours (continuous dependent variable) during two different learning activities (categorical repeated measure). We collected data over 6 weeks in 9 classrooms, so we have a measure of classroom behaviour for each learning activity over 6 time points for each of the nine classrooms.
Our primary question is whether the number of behaviours (DV) differed between the two learning activities. I figured using mixed effects models (using SPSS) was most appropriate, as our dependent variable is nested within the timepoints, which are nested within the classrooms. We wanted to account for any dependencies between the observations. I've been having a hard time finding similar examples for this kind of design.
So far I have figured to run an analysis using learning activity as a fixed effect, with random intercepts for time and time*classroom, but I'm unsure if this is the correct way to approach this analysis. Any advice is greatly appreciated.
 A: 
So far I have figured to run an analysis using learning activity as a fixed effect, with random intercepts for time and time*classroom

By "time*classroom" I assume you mean the time:classroom interaction. This implies that classroom is nested within time. However, your description of the study design is that time (measurement occasion) is nested within classroom. If that is indeed the case then you want random intercepts for classroom and the also the time:classroom interaction. However, it not obvious to me that you have any nesting here. A measurement occasion does not "belong" to a particular classroom, and neither does a particular classroom "belong" to a particular measurement occasion. Each classroom was measured on multiple occasions, and at each time, multiple classrooms were measured. This it seems  that time is crossed with classroom, so in that case you just want random intercepts for time and also for classroom.  Finally, if you think there may be any systematic change over the time period, it would make sense to treat time as a fixed effect, and not fit random intercepts for it.
