1
$\begingroup$

I am wondering what the best way to proceed in trying to predict the amount of possession a basketball team gets in a game. I have heard that beta regression tends to be good for proportions, but I am wondering how to go forward since I want the possession of both teams. Obviously possesion of one team will be 1 minus the possession of the other team, but will this always be the output of the regression model if it had the structure:

$$ g(p) = x_i^T\beta $$

where $p$ is the possession of a single team.

$\endgroup$
1
$\begingroup$

yes, g() is then the link function of your GLM model. Try package betareg, and this paper might help. https://cran.r-project.org/web/packages/betareg/vignettes/betareg.pdf

$\endgroup$
3
  • $\begingroup$ but im asking if there are any issues with this approach? Is it problematic if we have our response variables $y_1, y_2$ where there is a dependency $y_1 = 1-y_2$ $\endgroup$ Jul 12 '16 at 0:18
  • $\begingroup$ Then you can just model y1. Like logistic regression you always predict for P(X=1), and the probability of X=0 just 1-P(X=1). $\endgroup$
    – Jun Cheng
    Jul 13 '16 at 6:56
  • $\begingroup$ How would you construct your model? Do you want to predict for each team, i.e build a model for each team? Or how would you like to do it? $\endgroup$
    – Jun Cheng
    Jul 13 '16 at 7:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.