Testing suppression with SEM

#1
hey everybody

Using multiple regression analysis on my data, I've discovered a suppression effect.
Now I started analyzing the same data with AMOS (mainly because I am testing mediation in the same model) and am not sure how to construct the model to test for the significance of the suppression.
anybody any clue???
Thousand thanks in advance...


PS. I've read this brilliant article by Cheung & Lau ("Testing mediation and Suppression Effects of Latent Variables") which helped me a lot concerning the mediation testing. They also explain WHAT a suppressor is and how to interpret it. However, the do not state explicitely HOW to test it. (all they say is that the procedure is "similar").
I find it unlogical to use the same kind of model as for mediation (ie independent variable -> suppressor -> dependent variable) because - although the suppressor can be defined as an 'intervining' variable - it is not influenced by the independent variable.
 
#2
JUST IN CASE anybody is struggling with the same problem.......
this is Prof. Cheungs reply:

Dear Pascale,

Treating suppressor as an intervening variable is only a statistical treatment, but not theoretical. The effect of X does not go through the suppressor to Y. Yet, the statistical analysis is the same as if the suppressor is a mediator. Only interpretation is different.

The existence of suppressor will cause the effect of X on Y less. If we have controlled the effect of the suppressor, the effect of X on Y will become larger. The suppressor is theoretically related to X, but not to Y. But statistically, when we want to control for the effect of the suppressor, we need to include the suppressor as an IV, together with X, in predicting Y.

Your two approaches for testing suppressor are the same, but the difference approach is theoretically correct. Yet, the difference in direct effect between with and w/o the suppressor is the same as the product term of X --> suppressor and suppressor --> Y.

Good lcuk,

Prof. Cheung