APBRmetrics

I was having a discussion with a friend the other day about Steven Novak.

IMO his extremely high scoring efficiency and ability to help space the floor adds significant value to an offense. However, it can also be argued he's a very limited offensive player and has to be assisted in order to score.

I've seen various models that subtract value from "assisted shots" to differentiate between players that create their own shots and players that need help like Novak.

Here's my question.

What is the goal of these models?

Is the goal to measure who the most talented offensive player is or who is contributing most to the offense under the current conditions.

Granted, in some circumstances Steve Novak might not be as effective as he has been for the Knicks, but do I really care about that UNLESS the conditions are likely to change? An assisted spot up 3 point basket by Novak puts just as many points on the board as if he handled the ball, did a spectacular crossover, pumped, and created enough to space to knock it down on his own.

I sort of think models should reflect a player's actual production in a given set of circumstances and analyze his skill set and likelihood of fitting and being as productive in another set of circumstances as a separate exercise. Thoughts?

Presumably, the guy who can shake and bake and make his own shot any time could also receive a pass and make an open shot; whereas Novak et al cannot do the shake part.

I'm estimating Novak's unassisted points at less than 5%. As such, I only give him 89% credit for his scoring.
I have Jeremy Lin at 80% unassisted scoring, and he gets a factor of 1.04 toward his points.

While 3 pts = 3 pts, it's reasonable to say 3 pts on 2 shots is better than 3 pts on 3 shots.
Similarly, a player shooting .600 on 18 shots/36 min has more value than the one at .600 on 12 shots/36, if all else is equal.

Mike G wrote:I'm estimating Novak's unassisted points at less than 5%. As such, I only give him 89% credit for his scoring.
I have Jeremy Lin at 80% unassisted scoring, and he gets a factor of 1.04 toward his points.

I agree with everything you said except this.

Assuming all else is equal (same # of shot attempts, same # of 2s vs 3s, same minutes, same TS% etc...) except one gets mostly assisted shots and one gets mostly unassisted shots, I think the offense would essentially be equal in both cases.

If you asked who I would rather have on my team, I would probably say the guy that can create more unassisted shots because I can't be certain the guy getting mostly assisted shots will always have the playmakers available to get him those good looks.

I think modeling the contribution to the offense is different than modeling the skill level that could lead to a different contribution under different circumstances.

statmandu wrote: Assuming all else is equal (same # of shot attempts, same # of 2s vs 3s, same minutes, same TS% etc...) except one gets mostly assisted shots and one gets mostly unassisted shots, I think the offense would essentially be equal in both cases.
.

I agree with this. The assist-dependent scorer is usually going to get fewer good opportunities, though. When you need one successful possession to win the game, the shot-creator is more valuable.

The goals of the models may vary. A model that subtract value from "assisted shots" to differentiate between players that create their own shots and players that need help like Novak appears to be trying to find "who the most talented offensive player is". Your goals are your goals. If you want to know who is contributing most to the offense under the current conditions, you might prefer a different approach. Use models in pursuit of your goals in a way that you can justify. There may not be one right approach. There are wrong ones and certainly wrong ones in the eyes of others.