Growth modelling and observed variables
Message/Author
 Joseph posted on Thursday, March 27, 2014 - 9:18 am
dear Drs,

I need to run a growth model over 7 time points to explore the relationship between the predictor at time t and the outcome at time t+1. My variables are categorical and are all observed.

I'd like to observed whether the type of service received (home; hospital; GP practice) at time t is associated with satisfaction (yes, partly, no) at time t+1 across the 7 time points.

Could you direct me to some relevant reference/material on how to conduct growth analysis with observed and categorical data while accounting for clustering at ID level? can MPLUS handle missing data across time points?

 Linda K. Muthen posted on Thursday, March 27, 2014 - 2:09 pm
See the introduction to Chapter 9 which discusses how clustered data can be handled in Mplus. See Chapter 6 examples particularly Example 6.10 where you would add the CATEGORICAL option and lag the covariates.

Yes, Mplus can handle missing data across time.

You may find the Topic 3 and 4 course videos and handouts on the website helpful.
 Joseph posted on Friday, March 28, 2014 - 2:34 am
Thanks Linda!
 Joseph posted on Friday, March 28, 2014 - 6:52 am
Hi Linda,

after looking at chapter 6 and topic 3, I am thinking I may want to use growth modelling to explore the relationship between service received at time t and satisfaction at time t+1.

I was hoping you could let me know if my code is doing what I need it to do.

I thought I could regress the slope and intercept of satisfaction onto the slope for 'service'.

Also, since satisfaction is assessed at t+1 I set the first satisfaction using @1 - is this OK?

USEVARIABLES ARE
sat1 sat2 sat3 sat4 sat5 sat6
service0 service1 service2 service3 service4 service5;

CATEGORICAL ARE
sat1 sat2 sat3 sat4 sat5 sat6
service0 service1 service2 service3 service4 service5;

ANALYSIS:

MODEL:
isat ssat | sat1@1 sat2@2 sat3@3 sat4@4 sat5@5 sat6@6;
iser sser | service0@0 service1@1 service2@2 service3@3 service4@4 service5@5 ;

isat ssat on sser;

 Linda K. Muthen posted on Friday, March 28, 2014 - 9:54 am
I don't think you can achieve the lagged effect unless you use one process as a time-varying covariate. You cannot do that using the time scores.
 Joseph posted on Friday, March 28, 2014 - 10:16 am
Thanks Linda,

I am really sorry but this is all very new to me. Are the lag covariates covered in any of the Topics online? I found those videos/handouts really helpful.

 Linda K. Muthen posted on Friday, March 28, 2014 - 11:20 am
Example 6.10 shows a time-varying covariate with no lag. You need to change the ON statements to reflect the lag you want, for example,

y2 ON x1;
y3 ON x2;
y4 ON x3;

There is nothing more to it.
 Joseph posted on Monday, March 31, 2014 - 8:03 am
Thanks Linda!!!
 Joseph posted on Wednesday, April 02, 2014 - 2:45 am
Hi Linda,

I had a go at building the model based on your message above and example 6.10 - also accounting for the covariate CENTRE, which is fixed:

USEVARIABLES ARE
sat1 sat2 sat3 sat4 sat5 sat6
service0 service1 service2 service3 service4 service5
CENTRE;

CATEGORICAL ARE
sat1 sat2 sat3 sat4 sat5 sat6;

ANALYSIS:

MODEL:
ii si | sat1@0 sat2@1 sat3@2 sat4@3 sat5@4 sat6@5;
ii si ON CENTRE;
sat1 ON service0;
sat2 ON service1;
sat3 ON service2;
sat4 ON service3;
sat5 ON service4;
sat6 ON service5;

However, I now receive a series of warning messages telling me that there are a number of empty cells. Looking at the output I noticed that this must be down to the fact that only 70 cases have information on (all?) the x-variables (out of the original 600 (!).

Is there a way to handle/impute missing data within MPlus when using the WLSMV estimator for categorical variables? Or do I need to impute the data in a different program and then import it into MPlus?