I have a two-level model where Level-2 is serving strictly as controlling for the higher-order nesting factor. In the Level-1 part of the model, I have a covariate, two Level-1 predictors, and an interaction between the two Level-1 predictors.
In thinking of how to do this, I decided to group-mean center my data prior to entering the data into Mplus so I could create my interaction term from the group-mean centered terms. I am trying to specify for robust-standard errors (I am used to having standard errors for the fixed effects and robust standard errors for the fixed effects in HLM) and to ensure that the syntax has the appropriate code. This is the syntax:
VARIABLE: NAMES ARE store gender IV1 IV2 DV1 DV2; USEVARIABLES ARE store gender IV1 lV2 DV1 IV1xIV2; CLUSTER IS store; within = gender IV1 lV2 DV1 IV1xIV2;
ANALYSIS: type = twolevel; estimator = mlr;
DEFINE: IV1xIV2 = IV1*IV2;
MODEL: %within% DV1 on gender IV1 IV2 IV1xIV2;
OUTPUT: mod stdyx CINTERVAL;
My two questions, therefore, are: 1) Is the syntax correctly specifying robust standard errors with the mlr estimator? 2) Can this syntax be run with low Level-2 group size? There are only 8 stores at Level-2, but again, nothing is being modeled at that level.
Eight clusters are not sufficient for multilevel modeling. You should create 7 dummy variables to take the non-independence of observations into account and use them as covariates in your analysis. Then your model is one level.