topic Re: PCA programming in Statistical Procedures

PCA programming

Aamirkhan — Wed, 08 Nov 2017 14:45:59 GMT

In PCA whether to extract the components of the independent variables or both dependent and independent variables.

Re: PCA programming

hollandnumerics — Wed, 08 Nov 2017 15:15:10 GMT

Aamirkhan,
There is a simple correlation you should understand when posting a code-related question:

The probability of an answer is proportional to the amount of background information, SAS code, log and output you post with your question!

Re: PCA programming

PeterClemmensen — Wed, 08 Nov 2017 15:19:48 GMT

Please provide more information on your problem..

Re: PCA programming

Rick_SAS — Wed, 08 Nov 2017 15:47:12 GMT

Strictly speaking, PCA does not involve a dependent variable. It is a way to choose linear combinations of variables that explain the most variance in the data.

You are probably talking about Principle Component Regression (PCR). Most people use PCR to mean "extract components from only the independent variables. You can read about how to run a principal component regression in SAS.

Although I have answered the question you asked, I want to point out that there are statistical drawbacks with this definition of principal component regression, The better approach, IMHO, is to use PROC PLS to perform partial least squares regression. PLS uses the independent AND the dependent variables to form the components that best explain the variation in BOTH the response and the explanatory variables.

Re: PCA programming

PaigeMiller — Wed, 08 Nov 2017 20:33:31 GMT

It sure sounds to me like the original question leads to the answer of Partial Least Squares regression.

There's no need to perform PCA on the x-variables and a separate PCA on the y-variables, because these two PCAs might find results that are largely uncorrelated with one another. PLS overcomes this problem by finding dimensions of y that are well predicted (at least, as well as the data will allow) by dimensions of x.