I am attempting to test an observed variable mediation model with all continuous variables, a non-normal mediator, and missing data. I would very much like to use bootstrapping. Yet to my knowledge the non-normal estimators in MPlus that can be used with bootstrapping can't have all continuous variables in the model or do not support accounting for missing data (I have tried WLS/M/V, MLM, MLR, MLF, ULS/M/V). Do you have any suggestion as to how I can analyze this data via bootstrapping while accounting for missing data?
All maximum likelihood estimators obtain the same parameter estimates. The estimators differ only in their standard errors. So if you use ML and BOOTSTRAP, you get maximum likelihood parameters estimates and bootstrap standard errors. In the robust estimators, it is only the standard errors that differ not the parameters estimates. In all cases, missing data are taken into account.