Simple Main Effects PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 learningmplus posted on Thursday, May 10, 2012 - 12:51 am
Hi!

I am once again running some moderation models and in addition to moderation I am interested in simple main effects. On the output I get the coefficients for the product term and its components.

I am wondering if the coefficient for components gives the simple main effects or if I need to run separate analysis for this. In other words, I have:

COND
MDD
CONDXMDD

Does the coefficient for COND gives the simple main effect for that variable or is that coefficient "conditional." I ask because when I run simple main effects outside of the context of interaction (aka in separate runs for COND and MDD), I get different mean difference or coefficients for COND and MDD. I hope I am clear in what I am asking. Thanks very much.
 Linda K. Muthen posted on Thursday, May 10, 2012 - 2:29 pm
When you include the main effects and the interaction, the regression coefficients are partial regression coefficients as they should be.
 learningmplus posted on Thursday, May 10, 2012 - 7:50 pm
What do you mean by partial regression coefficients as they should be? If I wanted to report the main effects, do I report these partial ones? Or do I run the model again without the interaction? I ask because the coefficients are different in the two runs and I just don't know which output to report the main effects from. If I use the coefficients from the interaction model the partial coefficient stay significant after i apply the Holm modified Bonferroni to correct for experiment wise error because I am running multiple tests. But I run it without the interaction -- by looking at simple main effects at each level of the moderator variable then the effect is no longer significant when I apply the Holm modified Bonferonni.
 Linda K. Muthen posted on Thursday, May 10, 2012 - 7:51 pm
You report the coefficients from the model that includes both the main effects and the interaction.
 learningmplus posted on Thursday, May 10, 2012 - 8:52 pm
Thank you. Please help me with one more thing. I received the answer below in response to a question I posed regarding moderation. At the time, I thought I "got it" but I'm afraid I'm not certain I am doing this correctly. Esentially, I have 3 regression paths.

COND (treatment condition -- dummy coded 0 1 -- 0=group; 1=family)
MDD (presence of depressive disorders -- dummy coded 0 1 -- 0=absent; 1 = present)
COND X MDD (the interaction).

I did X - 0 = Interaction for GROUP = 0 and MDD = 0. I then replaced X in the equation and got what I think it is the mean for GROUP and MDD present. I don't know how to get the means for FAMILY AND MDD PRESENT AND FAMILY AND MDD ABSENT.


You don't need 4 runs - you have all the estimates you need in the run with the interaction. You just combine the estimates in 4 different ways in line with regular dummy variable regression:

- the intercept gives the mean when both dummies (and therefore their interaction) are zero

- the intercept plus one slope gives the mean for the cells where you have 0, 1 or 1,0

- the intercept plus both means gives you the mean for the cell where both dummies, and therefore their interaction, are 1.
 learningmplus posted on Thursday, May 10, 2012 - 8:54 pm
I don't know what to add where in order to get "the intercept plus one slope gives the mean for the cells where you have 0, 1 or 1,0"
 Linda K. Muthen posted on Friday, May 11, 2012 - 12:25 am
Please send your output and license number to support@statmodel.com.
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: