Principal Components

[gstratis]

Good Morning to everyone,

I have been carrying out PCA on my data in order to reduce the dimensionality and I have concluded to extract 2 Principal Components which explain 85% of the variation in the data set.

My question is, is there a way which I can model these two PC's with my response variable?

Thank you in advance,

George

[PaigeMiller]

Use Partial Least Squares. Do not use the Principal Components for modeling a Y variable.

Why?

PLS extracts components that are as predictive as possible of the Y variable, given the constraint that the components are perpendicular to each other. PCA extracts components which are perpendicular to each other, that ignore the Y variable and so may not be predictive at all.

[gstratis]

Yeah I know that but my task is to compare the models extracted by each method (Principal Components Regression and Partial Least Squares Regression), so both have to be done.

Regards,

George

[PaigeMiller]

---

@gstratis wrote:

My question is, is there a way which I can model these two PC's with my response variable?

---

Didn't you address this in your other thread? You were running PROC PLS with the PCR method selected.

[gstratis]

Yes I did,

I was just wondering if there is another way of doing it so I could verify me results.

Regards,

George

[PaigeMiller]

You save the PCA scores, then use these scores as the X-variables in your regression.

I was just wondering if there is another way of doing it so I could verify me results.

You are implying that the results of PROC PLS with the PCR option need to be verified somehow?? I'm not aware of any such need to verify that PROC PLS is giving the right results, it seems to me that SAS has done that already so you don't have to.