Interaction effect: Othogonalizing (residual centering)

Frequently asked questions about PLS path modeling.
Post Reply
koberrr
PLS Junior User
Posts: 3
Joined: Thu Apr 22, 2010 12:50 pm
Real name and title:

Interaction effect: Othogonalizing (residual centering)

Post by koberrr » Mon Oct 10, 2011 12:39 pm

Hi,

does anybody know how to apply the orthogonalizing approach (i.e. residual centering) for moderating effects?

I only find these three options when creating a moderating effect:
1) use original indicator values
2) standardize indicator values
3) mean-center indicator values

Is option #3 equivalent to residual centering?

Thanks for your help!

mronkko
PLS User
Posts: 14
Joined: Tue Jan 06, 2009 6:45 pm
Real name and title:

Post by mronkko » Thu Oct 13, 2011 12:22 pm

#3 is not orthogonalization.

For orthogonalization, you need to first center (I prefer to standardize)the variables that form the interaction term, then regress the product of these two centered variables on the centered variables. Use the residual as the interaction term.

If you have variables A and B that are centered, you would need to run the following regression equation

A*B=b1*A+b2*B+e

e is the orthogonalized interaction term.

koberrr
PLS Junior User
Posts: 3
Joined: Thu Apr 22, 2010 12:50 pm
Real name and title:

Post by koberrr » Thu Oct 13, 2011 1:27 pm

Thank you mronkko, but i still don't get it. How can i do this with smartpls? where do i find the residuals?

Thanks for your help!!

mronkko
PLS User
Posts: 14
Joined: Tue Jan 06, 2009 6:45 pm
Real name and title:

Post by mronkko » Thu Oct 13, 2011 1:37 pm

Smart PLS does not report residuals. (This is a severe shortcoming, because residual analysis is a standard post-estimation method for all regression analysis based approaches.)

If you want to orthogonalize the indicators, you need to do that with some other statistical software (PASW, Stata, R, etc)

If you want create an orthogonalized interaction term based on the PLS construct scores, you need to export these construct scores from SmartPLS, and then import them to another statistical software and calculate the residuals there.

Post Reply