Holly Sexton posted on Wednesday, January 30, 2013 - 9:30 am
I'm currently working with a dataset that has classrooms nested in centers. I'm doing a fairly basic single outcome regression model and all of my predictors/covariates are at the classroom level, e.g., no center-level (level-2) variables. I was originally specifying TYPE = COMPLEX with center ID as the cluster variable in order to adjust my standard errors for the nested structure. However, I now also need to add sample weights to my models and the dataset I'm using incorporates stratifcation and PSU variables to make variance estimation corrections instead of using replicate weights. Therefore, I'm adding the STRAT is STRAT command, but because I still want to account for the nested data structure I now have two cluster variables - PSU and center ID, which requires me to specify my model as TYPE = TWOLEVEL COMPLEX and distinguish the within and between levels. Since all of my variables of interest are at the classroom level (Level-1), I'm wondering if I can specify all of my variables at the within level (including the outcome) and not specify a between level model and still have the standard errors be adjusted for center. So, my question is, does Mplus still adjust for the second cluster variable in a TWOLEVEL COMPLEX model if there is no between level model? Or, do I need to partition the within and between level variances of my outcome in order to appropriately take into account the nested structure? Thanks!