I am trying to analyze data from a three-group randomized controlled trial, where participants were assessed at baseline, post-treatment, 6-month follow-up, and 12-month follow-up. The same variables were measured at each time point. Total N at baseline was aout 500. There is a fair amount of missing data due to attrition. Some data are relatively normal, other data are very non-normal. There are a few outcomes of interest, some with multiple indicators, others with only one. In addition, some of the outcome variables are continuous, some are categorical. We are interested not just in outcomes, but also several moderator and mediator variables. In earlier work several repeated measures MANCOVAs were used, but I would like to take a latent variable approach like latent growth curve modeling.
Do you have any recommendations based on the info above? Are there particular examples in the literature or elsewhere that could prove useful to follow? Are there any issues to be concerned about that would be unique to the parameters I described above?
You will find several growth models described in Chapter 6 of the Mplus User's Guide. I think all of the situations that you refer to are dealt with. A set of growth models can be estimated together and moderator and mediator variables can be included. See also pages 470-477 of the user's guide where growth models are summarized.
See the papers listed in Recent Papers - Growth Mixtrue Modeling and also the following paper:
Muthén, B. & Curran, P. (1997). General longitudinal modeling of individual differences in experimental designs: A latent variable framework for analysis and power estimation. Psychological Methods, 2, 371-402.
burak aydin posted on Thursday, April 21, 2011 - 1:13 pm
Hello, I am trying to run MANCOVA with mplus. Here is my simplified code: VARIABLE: NAMES ARE cond pre fps2 fps3 fps4; missing are all (-99); analysis: Model: fps2 ON cond pre; fps3 ON cond pre; fps4 ON cond pre; pre; output: tech1 tech4 ;
Pre and FPSs are continuous variables. Condition is a dummy variable. Do you think this is an mancova?
Looks like it. Make sure you have residual covariances between the three outcomes.
burak aydin posted on Tuesday, April 07, 2015 - 7:52 am
Hello again, As a follow up to above question, I now would like to see if there exists an overall condition effect Model: fps2 ON cond (1); fps2 ON pre; fps3 ON cond (1); fps3 ON pre; fps4 ON cond (1); fps4 ON pre;
with this code I could fix the condition effect and its SE. I plan to interpret it as an omnibus test. Then plan to report condition effect separately. Is this defensible in your opinion? Best