Mediation, clusters and longitudinal ... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Eoin O'Connell posted on Wednesday, August 05, 2009 - 12:01 am
I have data that I must analyse using mediation/path analysis. Simple enough. However, each variable was measured four times (at six month intervals). I also wish to use clustering. I know that I can do the longitudinal and path analysis in the 'within' model. I think I can understand how roughly how the model will look but trying to tease out the details is not so easy.

Specifically, I have 'intervention group' as a grouping variable. I want to regress 'physical activity' (measured at four time points) on 'intervention group'. Then I want to use 'physical activity' to predict 'BMI' (again, measured at four time points). I also want to control for effects at the cluster level.

Do I regress the i & s of the 'physical activity' on 'int. group' and then regress the i & s of 'BMI' on both 'int. group' and i & s for 'physical activity'? Or do I create latent categorical variables for each i & s for each longitudinal observed variable and regress those?

Thank you
 Eoin O'Connell posted on Wednesday, August 05, 2009 - 12:33 am
I'm not sure that I've been clear above so I just wanted to clarify.

The simplified model is: Group->PA(4)->BMI(4).
PA (physical activity) and BMI (body mass index) both measured four times.

With y1 = PA, y2 = BMI and u1 = group, the within model I am thinking of would look something like this:

i1 s1 | y11@0 y12@1 y13@2 y14@3
i1 s1 ON u1
i2 s2 | y21@0 y22@1 y23@2 y24@3
i2 s2 ON i1 s1 u1

I presume the between model (to control for clusters) would be similar but am not yet sure.

If I then want to add another mediating variable, would the syntax be more or less the same? Or would I need to add a WITH statement between the two mediating variables?
 Linda K. Muthen posted on Wednesday, August 05, 2009 - 3:44 pm
The model above looks fine. I would suggest however fitting each process separately without any ON statements as a first step. If you add another mediator, you can use WITH to specify a residual covariance.
 Eoin O'Connell posted on Thursday, August 06, 2009 - 10:36 pm
As I understand it then, if I have two mediators (y1 and y2 say) and y3 is the predicted outcome variable. I should model it like this:

i1 s1: y11@0 y12@1 y13@2 y14@3
i2 s2: y21@0 y22@1 y23@2 y24@3
i3 s3: y31@0 y32@1 y33@2 y34@3
i1 s1 ON u1
i2 s2 ON u1
i3 s3 ON i1 s1 i2 s2 u1
y1 WITH y2 (or should it be i1 s1 WITH i2 s2?)

Thank you!
 Linda K. Muthen posted on Friday, August 07, 2009 - 3:44 pm
You mediators are i1 s1 i2 s2. All residual covariances among them should be included in the model. If you add mediators y1 and y2, their residual covariances with the other mediators and themselves should also be included.
 Eoin O'Connell posted on Monday, February 22, 2010 - 11:02 pm
Can I ask why I don't need to use the INDIRECT model to examine mediation effects?

Also, how is it possible to determine if a mediator is significant? Is that done using the INDIRECT model?

Thank you.
 Linda K. Muthen posted on Tuesday, February 23, 2010 - 3:01 pm
I'm not sure what your question is. If you want to estimate an indirect effect, you need to use MODEL INDIRECT or MODEL CONSTRAINT.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message