I was wondering if MPlus could test multilevel modeling data that contains cross-classified random effects (which Raudenbush and Bryk discuss in chapter 12 of their HLM book). I have a data set where children are nested within therapists. However, there are some children that see more than one therapist, so the nesting of children within 1 therapist does not hold for some children. Any advice on how to analyze this data with MPlus would be greatly appreciated.
Mplus does not currently have the facility to do this. This will be added in the future.
Lois Downey posted on Thursday, March 28, 2013 - 9:43 am
I am using type = crossclassified for the first time. I notice that for this type of analysis, unlike other types, Mplus provides one-tailed P-values. May I simply double the values in order to report 2-tailed P-values, or is there some reason that for this type of analysis a 2-tailed P-value is inappropriate?
You are using the Bayes estimator for this analysis. You should be looking at the credibility interval to determine significance. If the credibility interval does not contain zero, there is significance.
Hello, I have imputed 25 datasets with multiple imputation. I have a cross-classified model and so need to use Bayesian estimation, which doesn't allow the use of imputed data. Do you have any advice for how I could examine imputed data with cross-classified models, such as fixing the parameters from an imputed single-level analysis (controlling for clustering) and then running those parameters with a cross-classified model? Also, will Mplus be able to run multiple membership multilevel analysis in the future, or is it able to do so already (and if so, are there any examples for this)?
It is not available yet. It is not clear how to do it theoretically either, but if you know how you want it implemented you can use type=montecarlo to run multiple runs and then manually combine the estimates.