|
|
 |
| Model fit and estimating covariates i... |
 
|
| Message/Author |
|
|
|
|
I ran a latent growth curve model with gender as the only time-invariant covariate and the fit was adequate (CFI/TLI: .96/.97) When I added several time-varying covariates to predict the latent outcome variables at various waves, mplus used listwise deletion and my N dropped by 150 participants. I asked for a parameter estimate of the observed covariates to force mplus to use the full N.
The output showed that mlplus did use the full N but the fit became much worse (CFI/TLI: .32/.28). Why did this happen? Can it be fixed? Is there another way to use full estimation of the full sample without sacrificing fit?
Thank you! |
|
|
|
|
| The worse fit is probably due to some of the time-varying covariates having direct effects on some of the indicators of your latent outcome variables. That is, measurement non-invariance. It is likely not a function of how the missing data is handled. |
|
|
|
|
Okay, thank you so much! That is definitely helpful.
In my initial post, I forgot to include that the when I entered the time-varying covariates in the model (without the parameter estimation of the observed x-variables), the fit remained very good (and this was with the reduced N). It was only once I asked for the means of the x-variables (to get the full estimation and deal with the listwise deletion) that the fit substantially worsened.
Do you think this is still due to measurement non-variance or something else?
Thank you again! |
|
|
| Back to top |
|
|
