Using orthogonal contrast variables in SmartPLS

[Ben]

Hi,
I know the use of orthogonal contrasts in regression analyses for analysing the effects of differences between independent groups (http://www.jds-online.com/files/JDS-563.pdf). Right now, I'm thinking about using a contrast variable in a smart pls path model.

I have data from an experiment with 3 independent groups and want to test how differences between the groups influenced psychological and behavioural outcomes.

The model looks like this:

1.png (103.53 KiB) Viewed 2095 times

Similar to a regression analysis with a contrast variable, I would run the model twice: First with contrast -1, 2, -1 and then with -1,0,1.

My simple question is: Is this a good idea?
I searched in the internet and found no paper about using contrast coding with smartPLS.

[jmbecker]

It should not be a problem as PLS uses regression for the inner model relationships. I also use dummy coded experimental treatments in a study and it works very well.
The only obstacle could be that PLS routinely standardizes all data (also you contrasts). This can change the interpretation of the results (especially if the experimental groups have not the same size). Using the IPMA (which unstandardizes the latent variable scores) could help you in interpreting the results.