I recently asked you about missing data modeling with WLSMV, and you told me that modeling is conducted with a pairwise deletion based model. In factor mixture modeling with MLR, is the missing data method the same?
Hi My question is in relation to missing data modelling in Mplus. I am aiming to estimate a SEMM model, with one categorical latent variable and 4 continuous latent variables. My data consists of 2000 observations, with no missing data on any of the binary indicators for the latent class variable, yet i have missing data on 2 of the continuous latent variables. Hereby, there are a number of cases (approx. 200) who have missing data on every item on a scale (the items indicating a factor), given that respondents had to identify their nationality in order to be routed into a set of questions. In other words, respondents who refused to indicate their nationality did not complete any of the items on the scale, yet there is data available for these respondents on the other continuous latents and the latent class variable of course. WOuld appreciate any help on how Mplus can deal with this issue to help with model estimation. many thanks
Mplus would use all available data in line with handling missing data under the ML-MAR approach (see Little & Rubin's book). The question is if those refusing to give nationality are significantly different on other variables than those who agreed. If those differences can be predicted by variables that are not missing, MAR is motivated. If the reasons for refusing are related to what they would have answered on that set of questions, then you have non-ignorable missingness and MAR is not enough. Then, alternatives are hard to find, although "pattern-mixture" analysis is possible. See the Little-Rubin book again.