APBRmetrics

Hi everyone,

I am new to the field of Advanced Statistics, and deeply interested in it thus far. I've been playing professional basketball overseas for many years, where advanced statistics isn't very known/used. So far, I've been reading a lot of articles, and just finished Basketball On Paper which I loved (any recommendations for other useful books?).

Currently, as a hobby, I am spending some time in processing my favorite team's, back home, games. I've created a spreadsheet with all the common formulas of advanced statistics according to the team's box score (since it wasn't available online). Now, I am trying to provide a nice, detailed report to the team management (already contacted them, and they expressed interest), which will be beneficial for them as a club.

I was wondering what is the best way for approaching this. What would you consider as the most valuable insights for a club that hasn't been using advanced statistics? Is it more player statistics? is it team's statistics? what findings should I focus on? should I run linear regression, perhaps? I have tried to run a multiple linear regression, with the final score of every game as the dependent variable and all the players' usage rate in those games as the independent one - findings were actually interesting (one of the leading players of the team, which is considered a superstar more or less, had a very negative coefficient).

Again, I'm new to this field (and statistics in general), but am trying to understand it as best as I can. Any advice will be greatly appreciated.

Thanks!

If you can get ideas about the questions / issues management and coach are already interested in those probably would be the most important areas to research. You indicated the report is for management but you will have to decide whether to be an ally, dispassionate, or critic of the coach. Natural tendency may be to be perceived as a critic / opponent regardless of your own perspective though, unless you really sell and prove yourself as an ally.

Read most of the stuff at Nylon Calculus and at least browse the archives here (recent stuff, the short list thread, the links, anything long or that strikes a nerve.)

The usage / results regression sounds interesting but probably should support any argument with other approaches too (lineup level, individual, team four factors).

Pick one or two best analyses to lead with. Probably don't throw 10 things at them at once unless it is likely a one time shot to impress or they are super eager.

Use your experience to focus on the things that you and / or your past good coaches thought had the most impact on winning and / or are easiest to improve.

When you present and get feedback, listen / reflect real hard from every angle possible and try to give them what they want and / or need.

There are some skilled folks here. Figure out who they are and ask them specific questions, here or in private. Show strong effort to learn the skills required but don't take on more than you can handle at the time.

Hope some of this helps.

Omer wrote:I have tried to run a multiple linear regression, with the final score of every game as the dependent variable and all the players' usage rate in those games as the independent one - findings were actually interesting (one of the leading players of the team, which is considered a superstar more or less, had a very negative coefficient).

That's very likely due to selection bias. Generally, I'm sure, good players shoot more in tough games (~better opponent defense) because the not-so-good players deliberately lower their usage once they get the feeling that their actions (drives/shots) are easily stifled.
While, in easy games (~worse opponent defense) the not-so-good-teammates have an easier time getting open etc., and the good player (probably?) won't jack up shot after shot when there's a big lead (unless he's selfish)

My suggestion would be to start with a look at what factors have been most important for team success in the league, then evaluate the team and player statistics in that context. Odds are that won't be much different than in the NBA, but you mind find something surprising.

J.E. wrote:

Omer wrote:I have tried to run a multiple linear regression, with the final score of every game as the dependent variable and all the players' usage rate in those games as the independent one - findings were actually interesting (one of the leading players of the team, which is considered a superstar more or less, had a very negative coefficient).

That's very likely due to selection bias. Generally, I'm sure, good players shoot more in tough games (~better opponent defense) because the not-so-good players deliberately lower their usage once they get the feeling that their actions (drives/shots) are easily stifled.
While, in easy games (~worse opponent defense) the not-so-good-teammates have an easier time getting open etc., and the good player (probably?) won't jack up shot after shot when there's a big lead (unless he's selfish)

Is such a relationship a stylized fact of NBA basketball? It's a story, to be sure (describing sub-optimal play?) but it's not obvious to me that this would be the general case or that the effect would be very large.

If you want, you could post a link to your draft report (and then delete in a couple days) or an outline here for further feedback.