Made You Miss: Projecting Rim Protection Metrics

http://nyloncalculus.com/2015/02/04/made-miss-projecting-rim-protection-metrics/




Made You Miss: Projecting Rim Protection Metrics
Posted on February 4, 2015 by Andrew Johnson

Mandatory Credit: Bill Streicher-USA TODAY Sports

“Protecting the Rim” is one of the hot phrases in basketball these days. Every GM is looking for a rim protector, and now with the new SportVU data we can get a much more accurate measure of how well players actually perform that function. Nylon Calculus’s Seth Partnow has been at the forefront with his rim protection metrics adapted from the SportVU data available here.

In developing my Player Tracking Plus Minus (PT-PM). I found that both opponent’s rim protection percentage and the number of contests at the rim were important indicators for defense, more important than the traditional measure of shots blocked.

Now we are in the middle second year of publicly available SportVU data, so we can look at a year over year measures to see how much or how little stability there is in these measures. When I looked at three point shot contesting, basically I found that there was no relationship between a player’s success contesting three pointers last year and the first half of this year.

Luckily, there is more signal in the rim protection numbers than that. However, breaking rim protection into its two very basic component parts, being in position to contest shots near the rim and influencing an opponent’s shot into a miss, the number of contests at the rim, which is significantly a function of role, is the much more stable measure year over year.

Below is the opponent shooting percentage at the rim when the player is in a position to ‘contest’ the shot for last year and the first half of this year, for all players with over 50 contests in each year.



OppPctRim

With that sample there is a modest .12 R2 year over year, or a .35 correlation, about the same level Kevin Pelton found for three-point shooting over a full year.

If one wants to improve the prediction of opponent field goal percentage at the rim there is, in fact, an increase in the variance explained by taking into account simply how often the player contests shots at the basket.  Funnily, though the data to date indicates that blocks as a proxy for rim protection isn’t completely dead as the R^2 between blocks per 40 minutes in the 2014 season and opponent field goal percentage at the rim is slightly higher even than combining contest rate and field goal percentage. Both coming in at about 20% of variance.

On the other hand, the number field goal attempts defended at the rim per forty minutes is very stable, with a R2 of .86, or a .927 correlation, in the ball park of the most stable role-dependent statistics like rebounding and assists. Points saved at the rim per 40 minutes, a measure that combines the number of attempts contested at the rim and the success when doing so, has also been stable from last year to this year, with a R^2 of .79. Interestingly, the outliers there tended to be players like Tyler Zeller playing with a new team or Marreese Speights playing under a new coach.

Ultimately, I think the data indicates some caution in interpreting raw opponent field goal percentage at the rim. It appears to be one of those tricky areas that involves both skill and luck. Sometimes an opponent will miss a bunny with only an ill positioned and half-hearted contest. Sometimes the opponent will make a circus shot that was completely altered by the defender. And there are innumerable cases in-between. The parallel I draw is with three-point shooting, an undoubted skill, but one that takes some time to accurately gauge.




bob



.

When I saw the title of the article, I was hoping for something more than this.

There is some data here, but none of it is associated with players or teams, and there's precious little to tell me what that data actually means. What was most disappointing was this -- after including the undecipherable graph and making my eyes glaze over with stat-geek speak, the author's conclusion is that you can't really learn much from the data. Gee, thanks.

Huh?

Sam