Regressions

[MizzouNoel]

Hi, I am interested in math but am not very savvy with it. I want to know (and please redirect me if this is in the wrong section), if I want to run a regression such as a linear or exponential, could I do it against several x variables? Like, X-1 could be a player's points and X-2 would be his shot attempts while Y would be his team's +/- rating with him.

If it's not possible to get meaningful numbers out of a method like that (if it even exists), could one run the results of several 2-variable regressions together in a formula to get a meaningful display of the information? If so, is there a good method for going about that? Or is that ridiculous?

I'm sorry if its a silly question, I am just curious and hopeful that there are some geniuses on here.

 

[Ishuda]

The short answer is, yes it can be done. But I'm not known for short answers (even when I don't know anything about the subject). The general idea of regression analysis is to define some predictor function. The idea of linear regression is not that the function is linear in the variables, rather it is linear in the coefficients which multiply the variables. So, in your case define your variables as: x₁ = the players points, x₂ = the players shot attempts, x₃ = the players team, ..., up to x_n. Now suppose you want to know the players value (or something else about the player) and have an n of 3. Then let y be that something and define
y = a₁ x₁ + a₂ x₂ + a₃ x₃.

The data you should then have is a set of triplets (x₁, x₂, x₃) along with a set of y's that go along with those x's, i.e. you would have points, shot attempts, and the team of a player and know how valuable the player was for a bunch of players. You could then do a linear regression analysis to find the a's. Other analysis of the a's and input data would tell you if those variables really meant anything (made any difference).

As an example of this kind of thing, watch the movie Moneyball.