|
Message/Author |
|
Resmi Gupta posted on Tuesday, April 05, 2011 - 7:52 am
|
|
|
Hi, I have one latent predictor (y1), one observed predictor (cb), one latent mediator(y2) and a latent outcome variable (e). I think there is an interaction effect between the latent predictor and the observed predictor on both mediator and outcome variable. Following is my code. I am not sure if this is the right code. model: y1 by a b ; y2 by c d ; e1 by f g; e2 by h i; e by e1 e2; y1cb | y1 xwith cb; e on y2 y1 cb y1cb; y2 on y1 cb y1cb; y1@1, y2@1, e1@1, e2@1 ; Would greatly appreciate your input. Also, how is it differ from "mediated moderation" ? Thanks, Resmi Gupta |
|
|
The above looks correct except when you fix factor loadings to one to set the metric of the factor, you must free the first factor loading, for example, y1 BY a* b; The is a type of mediated moderation. |
|
Resmi Gupta posted on Tuesday, April 05, 2011 - 9:42 am
|
|
|
Thanks very much Dr. Muthen. Regards, Resmi Gupta |
|
Resmi Gupta posted on Wednesday, April 06, 2011 - 12:10 pm
|
|
|
Drs. Muthen, I have a quick question on how to calculate indirect effect of a mediator and use Sobel test.I have 2 latent factor predictors (a, b in my code below), one mediator (observed, M ),a continuous outcome (Y). Finally x is my covariate. Following is my code : a by a1* a2; b by b1* b2; y on M a b x; M on f1 f2; f1 with f2; f1@1, f2@1; model indirect: y ind f1 f2; y via f1 f2 M; Do I need to have both "ind" and "via" command ? I am envisioning direct effect of f1 , f2 on Y , and an indirect effect (of f1 and f2 ) via M on Y . Also I would like to get the Magnitude of M (mediator). Greatly appreciate your input. Regards, Resmi Gupta |
|
|
IND and VIA are used for different situations. See the user's guide for a full description. You don't need both. |
|
Drew Fowler posted on Friday, January 06, 2017 - 3:36 pm
|
|
|
Hello Drs. Muthen, I want to test differences in fit between 3 models (a mediation model, a moderation model, and a moderated-mediation model). I’ve created the product term in my data set to test the moderation. The product term is included in both the moderation and moderated mediation models. However, I'm wondering if I should I include the interaction term in the USEVARIABLES line in the mediation model to allow the models to be compared for nesting reasons? |
|
|
Aren't all these model saturated/just-identified? And, using the usual X, M, Y notation with Z as the moderator, is your product variable X*Z or M*Z? |
|
Drew Fowler posted on Saturday, January 07, 2017 - 2:18 pm
|
|
|
Using that notation, our product variable is X*Z, and with multiple IV's, yes, it is over-identified. However, there is only one relationship of interest, with other things in the models. |
|
|
To test for the need to moderate, I would include Z in both of the models you compare and then see if X*Z is significant. I don't think you can compare fit between a model with a mediator and one without. You may want to ask on SEMNET. |
|
Drew Fowler posted on Monday, January 16, 2017 - 12:38 pm
|
|
|
Sorry, I realized my post had an error.... My model includes X, M, and Y, as there is no Z in my model. However, in my model, the IV is the moderator. I want to know: 1. Does M mediate the relationship between X and Y? 2. Does X moderate the relationship between M and Y? 3. Is there omnibus moderated mediation? Do you think there is a way to compare these models using fit statistics (i.e., AIC), or is it only model parameters that are useful in this case? |
|
|
I think only model parameters (and substantive theory) are useful, not fit statistics. X moderating M->Y is discussed as Case 3 in Chapter 2 of our new book http://www.statmodel.com/Mplus_Book.shtml |
|
Back to top |
|
|