Anonymous posted on Wednesday, March 30, 2005 - 8:29 am
We have a design that collects longitudinal data from three tests (A, B, C) of increased difficulty over time. Students self-select to take test C at grade 11 or 12. Students might take test C more than once, but we have decided to take the most recent test date available. Most students take tests A and B at the same two earlier grades. Instead of modeling growth as three time points, we use 4 time points separating those who took test C at grade 11 and those who took test C at grade 12. This results in missing data at the last two time points with zero coverage. The unconditional growth model in Mplus is set up as a two-level model. We also run a conditional model with covariates. Does Mplus run under the assumption of MCAR or MAR for the unconditional and conditional models?
In our Longitudinal data there are 3 timepoints, with ~60% missing data. I am trying to use ML estimator (for first time). My question is if we are assuming the heterogeneous trajectories, Are we justified in using ML for missing values? e.g. if I expect a stable trajectory and an unstable trajectory, but at timepoint 2 and 3 (e.g.) , the imputed values should be a function of trajectory class. Imputing from data at timepoint 1 assuming correlations will most likely make it a stable trajectory? I also want to know the basic idea of how missing data has been imputed in case of trajectories when we are trying to use Growth Mixture Models?
60% missing is a lot and jeopardizes the interpretations even when using missing data techniques.
ML does not impute missing values.
Growth mixture modeling handles missing data using ML under the standard MAR assumption. Other approaches are discussed in the paper on our website:
Muthén, B., Asparouhov, T., Hunter, A. & Leuchter, A. (2011). Growth modeling with non-ignorable dropout: Alternative analyses of the STAR*D antidepressant trial. Psychological Methods, 16, 17-33. Click here to view Mplus outputs used in this paper.
You may want to study Craig Enders' excellent missing data book.
I have missing data in my trajectories. We have four timepoints and have included only those trajectories which have data atleast at 2 timepoints so that (i, s) can be calulated for each trajectory. We want to use LCA to get classes of trajectories without any imputation at missing values. I am running script below. (I want to make sure that I am using trajectories with missing values also and performing no imputations). I want to compare the classification with imputed values also for which I use "Type= Mixture Missing". Am I doing it correct way?
DATA: FILE IS "test.dat"; VARIABLE: NAMES ARE x1 x2 x3 x4 ; USEVARIABLES ARE x1 x2 x3 x4 y; MISSING ARE ALL (-999); CLASSES = c(4); ANALYSIS: TYPE IS mixture ; LRTSTARTS = 0 0 100 20 STARTS 1000 20 MODEL: %OVERALL% i s | x1@email@example.com@firstname.lastname@example.org; OUTPUT: TECH1 TECH8 TECH11 TECH14