How to do pca for a group of variables?

[Phuc Nguyen]

From my understanding, pca is a method to suppress a couple of variables to one variable standing for all of them in a regression. If it is the case, I want to generate a variable call "present" is pca for 3 variables "a", "b", "c". What I should do then?

is it

pca a b c, components(1)

following this document.

[Nick Cox]

If and only if three predictors are essentially telling the same story you might consider using the first principal component from those three instead as a predictor, in which case what you are missing is in the next manual entry

predict PC1

to put the scores (values) of that in a new variable which you can then use in the next regression.

Many researchers would choose one predictor out of 3 in this circumstance. That's usually a lot easier to explain and to interpret.

Do any mainstream regression texts or econometrics texts recommend this method?

[Phuc Nguyen]

If and only if three predictors are essentially telling the same story you might consider using the first principal component from those three instead as a predictor, in which case what you are missing is in the next manual entry

predict PC1

to put the scores (values) of that in a new variable which you can then use in the next regression.

Many researchers would choose one predictor out of 3 in this circumstance. That's usually a lot easier to explain and to interpret.

Do any mainstream regression texts or econometrics texts recommend this method?

Thanks Nick Cox . I just learn how to do pca, there is no literature thing suggestion here. Thanks a heap.