forum.smartpls.com

Help please!

I want to compare the effect of moderating factors in 2 different groups (male and female), so i separated my data, males (n=49) alone and female (n=69) alone. when i run the bootstrapping procedure i get an error saying a singular matrix occurred during the estimation of the path coefficient....., but when the data is combined the model works very well

Any body with some ideas where can there problem be

Did you try another weighting scheme? Maybe your sample size becomes to small after your splitting.
cn

Cn

i tried a different weighting scheme, but still my model could not work..i was just reading from the this forum and came across the following explanation

"Then it could be that one of your items have to less variance for bootstrapping. That is sometimes happing. To less variance means that for binary items you have to much 0 and to less 1. If I get it right your model could be calculated with PLS but you cannot calculate the s.e.
"
can this be a possible answer to my problem and if yes how can it be solved

Thanks
Regards
Samson

Another mistake is to use the dummy variable (which you use as group variable) in your model after the split. Then a singlur matrix will occure.
cn

Hi,
I have a similar problem. I have 3 independent groups and therefore 3 dummy variables.
As long as I have 2 of them (it does not matter which one) in my model, everything works fine. But when I add the third one I got the "Singular Matrix Problem". In all groups I have around 55 samples.
Does somebody know how to fix this?
Thanks
Ben

This is a standard "problem" in all regression based methods not only PLS.
You cannot include all 3 dummies, because 2 dummies already identify you 3 groups. Thus, you only have to include 2 dummies to get the relevant results.
If you have 0/1 dummies the third group is the reference group where all dummies are 0. The dummies then indicate the change from this reference group to the focal group of the dummy variable (where its value is 1).

Dear Dr. Becker,

thanks a lot for your answer.
Do you know a paper or internet source that can help me to interpret the results I got from using 2 dummies in a path model?
Thanks

Maybe you should not use standard PLS algorithm, but OrdPLSc or non-metric PLS is more appropriate.