Message/Author 

anonymous posted on Friday, January 17, 2003  2:00 pm



In Mplus, only complete cases may be used in order to utilize robust standard errors. Is it appropriate to estimate robust standard errors across multiply imputed datasets? 


There are three estimators available for TYPE=MISSING in Mplus: ML, MLR, and MLF. The MLR estimator gives robust standard errors and the chisquare test statistic referred to as the YuanBentler T2* test statistic. This is documented in the Addendum to the Mplus User's Guide which is available at www.statmodel.com under Product Support. 


Hi Dr. Muthen I am using 5 imputed data sets to do a 2class mixture model with an interaction term. i have a fairly small n = 48. is this trustworthy at any possibility or simply just not adequate. 2nd question belonging to the same model(and to the topic): the standard error in the model output for one parameter eg. 64 don't fit to the acov matr. in the output for any of the 5 imputations. here the var of the param = 20. is there one true or both not reliable. thanks Alex 


n=48 is rather small, unless your 2 classes are very well separated, such as means at least 2 SDs apart. With your missing data, this problem is compounded. SEs are computed using a "withinbetween" imputation formula given in books like Schafer's, so not only using the SE for a particular imputation run. 

Back to top 