Anonymous posted on Thursday, December 13, 2012 - 8:55 am
Is it possible to do multiple group path analysis using the knownclass statement to define the groups? I've tired this, listing the overall and group model statements first before the model indirect statement, and get the message that ind and via are not supported. I can fit the models using the groups option but had been looking to get odds rather than probit estimates.
MODEL INDIRECT is not available when the KNOWNCLASS option is used. You can use MODEL CONSTRAINT to define indirect effects in this case.
Leon Head posted on Saturday, April 15, 2017 - 1:05 am
I have a mediation model with continuous IV (x), a continuous mediator (m), nominal DV (c, 3 groups defined using knownclass), and a distal outcome (p).
The betas in model results between m and c#1/c#2 do not vary but I presume they should as m is predicting different classes of c (c#1 and c#2)?
Could you please advise on the following: 1) Is it correct that c on m shouldn't differ across c#1 and c#2? If so, how would you interpret this in results? 2) Is it possible to test mediation models for different c classes using the syntax below?
VARIABLE: CLASS=c(3); KNOWNCLASS=c(y=1 y=2 y=3); ANALYSIS: TYPE=MIXTURE; ESTIMATOR=MLR; ALGORITHM=INTEGRATION; MODEL: %OVERALL% m ON x (a1); c ON m (b1) x (c1); %c#1% [p] (y1); p; %c#2% [p] (y2); p; %c#3% [p] (y3); MODEL CONSTRAINT: new (diff1 diff2 diff3 inda1b1 dir1); diff1=y1-y2; diff2=y1-y3; diff3=y2-y3; inda1b1=a1*b1; dir1=c1;
1) If m is a predictor of c, it doesn't make sense that c also influences m's prediction. That's circular. So c on m should not differ across c classes and Mplus won't let you.
2) I think this kind of mediation is very complex. You have sequential mediation going on (x->m, m->c, c->p) as well as one of the mediators being nominal. Unless you want to do advanced research on counterfactual causal effects for this case, I recommend staying away from it. It would be easier if c is a moderator; perhaps of all the relations.