I am investigating the proficiency developmental pattern for 2000 students by analyzing their test data. Over the three years, a speaking test and a survey were administered to those students 3 times. However, only 10% of the students took all 3. Most of them only participated in the study once. For those who had the data of two years, around a half of them had the data of the first and the last time of measurement. My question is how I can add time-varying predictors given the large amount of missing data. Around 2000 students had at least one test score (DV), but due to the high attrition rate, the number of observations dropped to 31 after I added time-varying variables. Is there a way to trick Mplus to include all the observations?
For subjects who have missing on y at the same time point as they have missing on x, you can change the missing data flag for x to another value. This means that those subjects won't be deleted from the analysis due to missing on x and at the same time they don't contribute to the estimation for parameters at that time point; this is what you want.