Basketball Analytics (the book) is available

[steveshea]

Just ordered on Amazon, looking forward to checking it out.

Thanks! I hope you enjoy it.

[nileriver]

Thanks for posting information about this book. I just ordered it from Amazon and can't wait to start reading it.

[steveshea]

Thanks for posting information about this book. I just ordered it from Amazon and can't wait to start reading it.

Thanks. Please let us know what you think.

[Crow]

Bump for discussion of lineup entropy, second look...

[Crow]

http://www.basketballanalyticsbook.com/ ... available/

New / next version.

[steveshea]

http://www.basketballanalyticsbook.com/ ... available/

New / next version.

Thanks for posting the release. If anyone has questions on the book, I'd be happy to answer them here.

[Crow]

My impression from words that I've seen on your blog is that with your offensive and defensive metrics and the Approximate Value roll-up you are focused on biggest activities and not claiming all encompassing and perfectly calibrated one metric. My questions would be: how well does AV explain team performance, do you feel any greater desire to make a comprehensive, near perfect one metric and do you have any opinions about using metric blends over making a singular choice? Any thoughts on Andrew Johnson (Counting Baskets) player tracking plus minus?

[steveshea]

My impression from words that I've seen on your blog is that with your offensive and defensive metrics and the Approximate Value roll-up you are focused on biggest activities and not claiming all encompassing and perfectly calibrated one metric. My questions would be: how well does AV explain team performance, do you feel any greater desire to make a comprehensive, near perfect one metric and do you have any opinions about using metric blends over making a singular choice? Any thoughts on Andrew Johnson (Counting Baskets) player tracking plus minus?

All-in-one metrics have value. They function well as dependent variables in league-wide studies. For example, they can be used as the dependent variable in a regression to project pro performance from college performance.

I have a harder time suggesting that teams should use a comprehensive metric in the evaluation of a player (say to sign as a free agent). I think that type of metric is an efficient way to identify players whose value (according to the metric) is far below "perception" and might be had at a bargain. Beyond that, I think the need to understand the specifics of a player's performance outweighs the simplicity of compressing everything into one number.

So that I'm not misrepresenting the contents of the new book, let me say that approximate value (AV) only gets a brief mention in the new book as an example of an all-in-one metric. The new book does not contain any AV calculations. That said, let me talk about the motivation for AV. I don't believe that a comprehensive, near perfect one metric exists. When using these metrics, it is important to completely understand when they might be misleading. With AV, I prioritized the interpretation. It was intentionally simple in its design. In doing this, I lost accuracy (introduced more misleading cases), but I think I made the cases that were misleading more obvious.

I would not suggest using AV to explain team performance. Half of AV is Defensive Stops Gained (DSG) which is an on/off metric. It naturally pits players against their teammates. Typically, the starting center has a positive DSG when the backup has a negative DSG. It doesn't make much sense to sum AV as a measure of the team's ability. I also wouldn't have much confidence that DSG would transfer with a player to a new team unless the player's lineup entropy were particularly high (which suggests he split his playing time across a variety of teammates and his production is less likely to be influenced by one particular teammate or lineup).

At the beginning of this post, I suggested that the value in an all-in-one metric is in league-wide (high level) studies or as a first level investigation of player evaluation. I believe AV is a good enough approximation and simple enough to interpret to be used in those capacities.

I like Andrew Johnson's player tracking plus/minus. At this point, I find the coefficients in the regressions the most interesting. We'll have to see if these stay relatively constant as more seasons are input. We also don't know how stable these new SportVU metrics will be. For example, does a low Opp. FG% at the rim in one year accurately predict Opp. FG% at the rim the following year (and similarly for contested reb. %, etc)?

[Crow]

Thanks for the responses.

[nrestifo]

Steve,

I ordered a copy a few days ago. Assuming it gets here in time, I'm looking forward to reading it on flights I'm taking flying from home (CT) to Phoenix on Sunday for the holidays. I always wanted to ask you, are you in communication with the basketball staff at St. A's? Feel free to not answer, seeing as we're rivals (University of New Haven)....

[fpliii]

Cool, I'll order it soon. The portions on entropy were an interesting take, looking forward to seeing which topics are elaborated upon in the new book.

[steveshea]

Thanks everyone for the interest in the book. I hope you enjoy it.

I am nothing more than a fan of St. A's men's and women's basketball. Head coaches Keith Dickson and Julie Plant do a great job with their respective teams.

Thanks again,
Steve