Page 1 of 1

Controlled variables in SmartPLS

Posted: Mon Dec 09, 2019 6:41 am
by rebeccachan1111
I am running my model on Smartpls, with 1 DV, 9 IVs and 4 CVs. The 4 CVs are age, income, education and gender. Except the last, all are ordinal (answers are in ranges / categories with orders of low to high).

For the controlled variables, I read many perspectives and ways of doing it but am still confused. Some said just point the CVs to the DV. I did that and only one CV has significant path. Is that all I have to do about controlled variables in Smarpls as I am reporting the rest of my model (the interested relationships) after the effect of the controlled variables? However is it ok to point the ordinal ones to the DV (as some said Smartpls cannot used on ordinal scales)? If it is not ok, what should I do for those ordinal controlled variables?

Another perspective is to prove they don't have effect on DV and continue to run Smartpls without CVs in the model. However I've run F-test (ANOVA) between CVs and IV, and since the ordinal scales come in 5-7 categories, it will be naturally hard to prove all group variances are equal. And in fact it gives contradictory result - gender is not significant in F test but significant in path coefficient.

Last perspective is to run MGA - but I found MGA shows result on each specific path rather than the overall model, it does not serve the purpose. Also it is not multiple but two group analysis only.

Did I miss anything here and what is the right way to do it?

Thanks.

Rebecca

Re: Controlled variables in SmartPLS

Posted: Wed Dec 11, 2019 2:34 pm
by jmbecker
If the categories are not ordered and equidistant you may use a dummy coding scheme to model the ordinal variables.