james posted on Tuesday, February 10, 2009 - 1:15 pm
Hi, I'm running at Parallel Process model across 4 time points. My sample size is n=451 at w1, dropping to n=322 by w4. When I run the model, it only uses n=275 observations? I'm confused. I obviously have missing data but specified that with missing = 999. My data is also non-normal, so I'm using an MLR estimator. Why is it only using half my sample?
I am fitting an autoregressive model with missingness on time-varying covariates. I have 436 individuals with data across 6 time points. The covariates are complete for time 1, and then have missingness because of loss to follow-up. I am not requesting listwise deletion. The number of observations used in the output is 210.
I would like to use all 436 individuals to estimate the time 1 parameters (I have complete data for this time), but it seems that Mplus is not doing this. It seems that if an individual is missing any covariates (even if they are only missing covariates at time 6), they are dropped from the analysis. Can you advise?
Missing data theory does not apply to independent variables. The model is estimated conditioned on these variables. The only way around this is to mention the variances of the time-varying covariates in the MODEL command. When you do this, they are treated as dependent variables and distributional assumptions are made about them.