|
Message/Author |
|
|
Hi Dr. Muthen, I would like to ask a question regarding how missing data are handled when Latent Transition Analysis (LTA) is performed. Say I have 100 observations of a latent class indicator variable Y measured at time 1, 2, and 3. This is entered into MPLUS in "wide" format. Now, suppose that 50 observations of Y at time 1 are missing, but it has complete data in time 2 and time 3. Does MPLUS keep all 100 observations when performing LTA or does it only keep the 50 observations with complete data? If it keeps all 100 observations, what does it perform the Full Information Maximum Likelihood (FIML) on to impute the missing data? In contrast, my observation with Latent Class Analysis (LCA) at a specific time point -- say time point 1 -- is that MPLUS performs deletion on all 50 observations thus leaving only 50 to fit the LCA model. However, I think this may not be necessarily how MPLUS handles missing values when LTA is being performed. |
|
|
Mplus uses all available data. ML is used which means that ML estimates are obtained under MAR (some call that FIML). |
|
|
Why, when I perform LCA with FIML, Mplus use only the subjects with no missing on ALL the latent class indicators? (instead of also using the subjects that answered to AT LEAST one indicator) A reviewer is asking me that, and I'm not finding a reference to justify it. Could you help me? |
|
|
This is not the Mplus default. Send your output and data to Support along with your license number. |
|
Back to top |
|
|