PLS for experimental design

[ancaagp]

Hello,
I have used an experimental design in my thesis, where I manipulated 6 types of service failure.
My hypotheses say that each type of failure has an impact on a type of attribution (external or internal).

Now I don't know how to include the 6 manipulations in the project -> should I create 6 different projects for each group?
OR should I use Failure type as categorical variable and create 6 groups out of it?

I don't know how to proceed with my model..
Thank you in advance for your help.
Anca

[jmbecker]

The best way is to include the different manipulations (groups) using dummy variables. You should think about, which comparisons are most important and construct corresponding dummy coded variables.

[ancaagp]

Thank you, Jan Michael, for the reply.
Should I then have the 6 variables for failure type, each with values 0,1, and connect them to the attribution variable?
I tried this and I get the Singular Matrix Problem.

[jmbecker]

If you include dummy variables, you alsways have one less then you have groups. The remaining group is the reference group to which differences are calculated (the group has zero on all variables).

[ancaagp]

Thank you. I will look deeper into that.
I thought that PLS doesn't work well with dummy variables, but good to know that I can explore it more.

[jmbecker]

It does not work well if the dummy (binary variable) is a dependent variable as this requires logit models. PLS is based on regressions. Hence, dummies are similar as in a regression analysis (they work well as independent, but not as dependent).

[ancaagp]

Hi,
In my model the dummies are independent and attribution is dependent.