APBRmetrics

silverbird5000 · Joined: 18 Jul 2006 Posts: 6

Hey everyone-

I have an off-topic question that I was hoping to get some help with, relating to the general method behind adjusted plus-minus modeling. While I understand adj. plus-minus ratings in the abstract, I don't have much experience with the nitty-gritty regression work, so you'll have to forgive the simplicity of the question. Here goes:

When you first set up your regression model, you have to control for the quality of a player's teammates and opponents. How exactly do you measure their quality? You can't use their adjusted plus-minus rating, because you haven't calculated those yet. So do you use their unadjusted plus-minus ratings. In other words, what exactly are you using for covariates when controlling for teammate/opponent quality?

Again, sorry to intrude with such a basic question. And thanks in advance for the assistance.

Ryan J. Parker · Joined: 23 Mar 2007 Posts: 711 Location: Raleigh, NC

http://www.countthebasket.com/blog/2008/06/01/calculating-adjusted-plus-minus/
_________________
I am a basketball geek.

silverbird5000 · Joined: 18 Jul 2006 Posts: 6

I've read those instructions on counthebasket.com and, if I understand them correctly, the author derives the values of his independent variables using a file downloaded from basketballvalue.com - specifically, the AdjustedPM variable from the playerstats file. In other words, he controls for the quality of teammates and opponents using adjusted plus-minus ratings that have already been calculated. What I want to know is: how do you calculate adj. +/- without knowing what the other players ratings are? Do you use unadjusted +/-, or something else?

Does that make sense?

Ryan J. Parker · Joined: 23 Mar 2007 Posts: 711 Location: Raleigh, NC

You don't get an adjusted +/- rating until you run the regression.

Before you run the regression, all you have is a bunch of data that specifies who is on offense, defense, and what the "lineup" ORtg was during that time period.
_________________
I am a basketball geek.

silverbird5000 · Joined: 18 Jul 2006 Posts: 6

Ok, I see. So, when you say "who is on offense/defense", I assume you mean that each player (P1 through P722 in the counthebasket excel file) is coded as a dummy variable, depending on whether they are on the court. In other words, it's basically the functional equivalent of a linear fixed effects model - it's just a normal OLS regression, but with 722 dummy variables. I thought that might be the case, but I wasn't quite sure. With all that's been written about adj. +/-, there isn't a whole lot out there describing the actual model. At least not in the news coverage and blogs.

Anyway, thanks for the help - it's much appreciated.

Ryan J. Parker · Joined: 23 Mar 2007 Posts: 711 Location: Raleigh, NC

Yeah, you end up with a bunch of zeroes for all the players except for those 10 on the court. Very Happy

_________________
I am a basketball geek.

silverbird5000 · Joined: 18 Jul 2006 Posts: 6

Cool. Thanks again!