I have a data set which has 7 independent variables and 107 dependent varaibles. When i did the PLS, the minimizing number of factors came out to be zero. The minimum RMPRESS is 1.111. However, when i randomly remove some of the Y variables, I got a good fit model with two factors. Could somebody tell why is it happening. What are some of the remdies/counclusions. Thanks
If you could share the data perhaps we can help? Off the top of my head some of the reasons behind why you are observing these phenonemna is I'm presuming you are fitting all 7 response variables at the same time...and the PLS algorithm is having a hard time projecting the correlation structure of all 7 to a single or small number of latent factors for y. I'm basing this idea on the fact that you say when you remove some of the y's you end up with a model that finds some latent factors in x that explain the new set of y's. Have you taken a look at the pairwise or principle components analysis of the y's? This might uncover some of what I'm speculating.