Active or inactive covariates PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 star09 posted on Wednesday, April 30, 2014 - 4:54 am
Dear Drs.

Can you please advise as to how best to adjust for potential confounding in LCA. I have a covariate that I wish to control for, rather than perform a subsequent sub-group analysis on.

Should I add the covariate into the model as a latent class predictor variable? If I can, I think I can proceed with DCAT to see if class membership predicts the distal outcome.

Or I have read that I should do a R3STEP and DCAT in two separate steps, but in that case, how do the two steps relate to each other? I don't understand.

Thank you.
 Bengt O. Muthen posted on Wednesday, April 30, 2014 - 7:29 pm
I would include the covariate predicting latent class, but then you can't do DCAT. You can do 3-step "by hand" as in our web note 15. You can also do 1-step with covariate and distal.
 star09 posted on Thursday, May 01, 2014 - 1:16 am
Thank you.
So do you mean in step 1 of manual 3 step (appendix E) of web note 15, I should include the covariate in the 'usevar' command. Then use the saved data for step 3 (appendix E) to estimate the distal outcome model? In appendix E, which of the variables is the distal outcome?
Thank you.
 Bengt O. Muthen posted on Thursday, May 01, 2014 - 9:43 pm
Q1. Yes.

Q2. Y
 star09 posted on Thursday, May 01, 2014 - 10:50 pm
Thank you Dr. Muthen
 star09 posted on Sunday, May 04, 2014 - 5:56 am
Can I ask for further clarification please?

You advised that would include the covariate predicting latent class, but then one can't do DCAT.

I tried doing the above and using DCAT and it did work but will the outputs be incorrect?

Variable:
Names are
CaseID Family SClass SchYr SchLoc Nation wddiday wediday QoL QoLQ4
Sleep SelfCare School Studdich Workdich Housdich Reldich Leisure;

Usevariables = CaseID
Sleep SelfCare School Studdich Workdich Housdich Reldich Leisure wediday ;

Categorical = Studdich Workdich Housdich Reldich wediday ;
classes = c(4);
AUXILIARY = QoLQ4 (DCAT);
Missing are all (-9999) ;
IDvariable = CaseID ;
Analysis:
Type = mixture ;

MODEL:
%OVERALL%


%c#1%

[ sleep* ];
[ selfcare* ];
[ school* ];
[ leisure* ];

[ studdich$1* ];
[ workdich$1* ];
[ housdich$1* ];
[ reldich$1* ];

sleep*;
selfcare*;
school*;
leisure*;
...
etc


Also, you write that I could also do 1-step with covariate and distal. In what circumstances would this be acceptable compared to newer 3 step method?

Thank you for your time.
 Linda K. Muthen posted on Sunday, May 04, 2014 - 4:50 pm
I see no covariates in your model. You cannot use DCAT with a covariate and a distal.

You can do one-step if the class formation does not change and if your entropy is greater than .8. 3 step is for entropy between .6 and .8.
 Kari Gloppen posted on Wednesday, July 09, 2014 - 4:10 pm
Is there a citation I could use that the one-step is appropriate if the class formation does not change and entropy is greater than .8?
 Bengt O. Muthen posted on Wednesday, July 09, 2014 - 6:38 pm
The one-step approach is appropriate at any entropy level. If the class formation doesn't change, there is not really an issue in choosing between 1- or 3-step. I don't think there are citations for this.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: