We have binary data collected so that we have a complete census of all '1's and subsequent samples of '0's. Inverse probability weights have been calculated. In addition, our data has three levels of clustering (repeated measurements on subjects within households within villages). Our goal is to fit a path model to this data set to explore how the effect of various predictors are mediated by a particular mediator. Is there any way to do this path analysis in MPlus while accounting for the clustering and biased sample simultaneously?
Thank you very much for any input on this problem.
It should be possible to do this. The repeated measure is typically handled by constructing a multivariate model, for example y1, y2, y3, y4, y5 would be the five measurements for one individual. The remaining clustering can be handled with "type=twolevel complex" where the cluster command will take two levels of nesting (households villages). Here is a paper that describes this methodology. http://statmodel.com/download/SurveyJSM1.pdf