Basketball Analytics

A forum for non-baseball sports
Post Reply
User avatar
33anda3rd
Replies Authoritatively
Posts: 8418
Joined: April 7 13, 9:45 am
Location: Chicago, IL

Basketball Analytics

Post by 33anda3rd »

While we're on the topic of defensive metrics in baseball...

Baseball perfected offensive stats by asking what matters. Forget the box score, they said. Let’s find scoring expectation, then create the change to the scoring expectation based on an action and that actions value vs run environment.

Basketball stats are far from perfect because they still ask what matters as told by the box score. Box Plus-Minus is a good example. PER is based on outcomes found in box scores, heavily skewed towards offensive-minded wings who use a lot of possessions. 538's recent RAPTOR stat says the best single seasons of the last 5 years are Curry, CP3, Curry, Harden, CP3, Harden. CP3. Which is hogwash. Basketball has yet to look behind the box score and figure out where the value really lies. They have yet to define the point scoring environment and then work from there the way baseball has and instead keep drawing from the box.

This is way far from perfect. Anyway. To start, one easy example:

Scenario 1: James Harden scores an unassisted two-point basket.
Box score: +2 points for James Harden.
But. The average team in the NBA scores 1.1 points per possession. So if we’re talking about real value added, James Harden has added +0.9 points.
Building from there…

Scenario 2: James Harden scores a two-point basket as a direct result of a pass from Russell Westbrook.
Box Score: +2 points for James Harden, +1 assist for Russell Westrbook.
But. We know that the real added value on the play is +0.9 points over what an average NBA team scores per possession. If we want to use the 2:1 ratio represented in the 2 points/1 assist box score, then real added value is Harden +0.6 and Westbrook +0.3 (realistically, we need to factor likelihood of shooting unassisted and assisted and then weigh how much balance we give James and Russ, this is a simplified example.)

Scenario 3: James Harden scores a two-point basket coming directly off a screen set by Russell Westbrook.
Same as above, except in today’s box score Westbrook gets no assist. The real added value should be, based on Scenario 2: Harden +0.6 and Westbrook +0.1.

Add to Both Scenario 2 and Scenario 3: In each scenario PJ Tucker is standing in the corner ready if needed to receive a pass and shoot a corner three. Tucker’s defensive assignment stays on Tucker. If Tucker’s defensive assignment leaves Tucker he can help on Harden and decrease the chance that Harden scores by 30%.
Now what?
Well, if Harden had a 100% gimme dunk without the help D and the help D drops the +0.9 points added by 30% then PJ, just by standing in the corner, added 0.27 points on the play. So we are now at:
PJ Tucker +0.27
Russell Westbrook +0.24
James Harden +0.39

Add to the Last Scenario: Oh and also Eric Gordon is on the wing ready to launch a three. So he splits PJ’s value.
PJ +0.135
Gordon +0.135
Russ +0.24
Harden +0.39

But then what if: Capela sets an off-ball screen, Russ passes to Harden who directly scores a 2P basket off the pass, PJ and Gordon’s defenders are stuck to their man to protect them getting an open three.
Now how do you divide up the real added value knowing that the team as five men just all contributed to add 0.9 points to the Rockets’ total?
Everyone on the floor has value and is doing something, even the guys who are just standing around.

Scenario A1: Chris Paul steals the ball from Lonzo Ball while Lonzo dribbles in the backcourt. Paul then dribbles to the end of the court and scores unassisted. No other players are involved.
Normal box score is Paul +1 steal, Lonzo +1 turnover.
If the expected score of a possession is +1.1 though, then on the steal it’s:
Paul +1.1
Lonzo -1.1
Then on the unassisted layup its:
Paul +0.9
So on the whole play it’s Paul +2.0, Lonzo -1.1

Scenario A2: Lonzo tries to pass the ball, Terrance Ferguson tips the ball, Chris Paul grabs the loose ball and races for an unassisted layup.
Lonzo -1.1 for turning +1.1 expectation (possession) into 0 (loose ball, neutral possession)
Ferguson +1.1 for the opposite
Paul +1.1 for turning neutral state (loose ball) into possession (+1.1 expected points)
Paul +0.9 for scoring the layup
Paul +2.0, Ferguson +1.1, Lonzo -1.1
Either way, Lonzo lost the ball, Paul gained possession of the ball and scored, added value for Ferguson facilitating for his teammate.

Scenario B: Andre Drummond rebounds the ball after an opponent’s miss.
The ball has bounded off the rim. At the point in which the ball is in the air on a missed shot no one has possession. The expected scoring value of possession is 1.1 points. By rebounding the ball Drummond is +1.1. It does not matter if it is offensive or defensive, possession has value.

Scenario C1: Klay Thompson draws a shooting foul on a 2P field goal
The league on average shoots 76% from the free throw line. Expected score on two foul shots is 1.52 points because the league shoots that 76%. By getting to the line Klay has taken the possession from 1.1 points on an average possession to 1.52 points, he is +0.42 by getting to the line.

Scenario C2(A): Klay hits both free throws.
Klay scores 2, the expectation was 1.52, so he is +0.48 on the shots and +0.42 for getting to the line.
This is +0.9 the same as Harden hitting a 2P field goal unassisted in Scenario 1

Scenario C2(B): Klay hits one free throw.
Klay scores 1, the expectation was 1.52, so he is -0.52 on the shots and +0.42 for getting to the line.
Klay on this play is -0.10, there was a better chance of scoring had he shot a FG attempt without being fouled.

I think these are the ways the APBRmetricians are going to have to think, and get out of the binary nature of the box score to start really figuring out value, otherwise their work will have the opposite effect of baseball's offensive metrics. They'll muddy the water instead of make it more clear.

User avatar
dan_yall
Everyday Player
Posts: 486
Joined: May 23 12, 3:16 pm

Re: Basketball Analytics

Post by dan_yall »

I find basketball analytics much less interesting than baseball analytics simply because it's so much more obvious in basketball who the best players and teams are because they win a much greater percentage of their games and usually mutliple championships. Baseball's randomness and the relatively smaller individual contributions made by star players means analytics can reveal more interesting insights in baseball that aren't as obvious based on game results. A good example of the difference between the sports is Mike Trout. The best basketball player of his generation is not going to languish on poorly run team that misses the playoffs every year, because if you have that player, even your poorly run franchise is going to make deep playoff runs every year and probably win some titles (see Lebron's entire career in Cleveland.) This means that in basketball, CardsTalk level analysis like "The Lakers should get Lebron and AD" or "The Warriors should sign Durant" is usually the right move.

User avatar
33anda3rd
Replies Authoritatively
Posts: 8418
Joined: April 7 13, 9:45 am
Location: Chicago, IL

Re: Basketball Analytics

Post by 33anda3rd »

IMO it's only semi-obvious who the best players are, and it's not obvious who is creating all the value on the floor.

Best current player is either Kawhi or healthy KD or Giannis or maybe AD or maybe still LeBron or possibly Curry.

Curry? Yes. He creates an extra 80 square feet on the floor the other team has to cover. He disrupts defenses in a way no other player in history has. He can't body everyone in the paint on his way to the rim like LeBron but LeBron can't make 5 other players spread out more on the court creating more space in the lane like Curry and the value Curry provides in doing that relative to the value LeBron provides crashing to the rim is something I'd love to know. If you take Curry's last 5 seasons his cumulative +/- on the floor is 4131 and they are -447 with him on the bench. And the gap from him to next-best is some 2K stuff. It's not just his teammates. They were +19 per 100 with all of the big four last year, +15 with Durant off the floor, +14 with just Steph + Dray, and +14 still with just Curry. Curry--despite the disrespect he gets for being little, for not banging bodies--is just as great as Kawhi and KD on the final score...I think. Silver's RAPTOR score says he is, but it also says that Chris Paul is also better than Kawhi and that's just impossible based on the eye test. It's not IMO obvious that Kawhi is > Curry. Or that Giannis is > Curry. Or AD. I think Curry is > at least two and maybe all three.

There's hidden value for role guys too. Last year there were 38 players who shot between 38.5%-51.5% on 2.5-5.0 spot-up three attempts per game. They range from All-Star level (Karl Towns) to "how do you even say that?" obscurity (Bjelica) to "oh that makes sense" (Korver) to "that's an aspect of his game I didn't know he had" (D-Rose.) Those guys knock down high-return shots at a high rate that provides value while opening up space on the floor for their ball-dominant teammates to create. Should come as no surprise the Spurs had three such guys last year--Bertans, Gay, Forbes--and they exceeded expectations by sliding into the playoffs over LeBron's Lakers and the T-Wolves.

My assumption is that those guys add value--not a small amount--on every single possession that never makes a box score. TJ Warren makes Malcolm Brogdon better by standing in the corner. Brogdon gets +2s and +3s in the box--and made FT--because Warren is literally just standing in the corner ready to get a pass and hit a three. My guess is that Rudy Gobert is not just +400 in a season because of D, it's because he is an elite pick-setter, something that will never go in a box score. if screen assists counted as assists Gobert would average a triple-double--he gets over 8 screen assists/game--but because we don't count those in the box score, and they don't go in his PER or RAPTOR or whatever, we don't credit Gobert with being an ELITE player we just think of him as very very good. When Brogdon is on the floor with TJ Warren and/or Domantis Sabonis Indy is somewhere between +48 and +58. Without one of those combos, it all goes rapidly downhill. It's because of the interaction of the three and it will never end up in a box score, but I'd guess some enterprising APBRmetrician could get us closer to understanding if they'd just get past box score stats and work from scoring environment changes of actions.

User avatar
33anda3rd
Replies Authoritatively
Posts: 8418
Joined: April 7 13, 9:45 am
Location: Chicago, IL

Re: Basketball Analytics

Post by 33anda3rd »

Yet another thought on this that probably only I find interesting: the box score considers committing an offensive foul a turnover. The box score does not consider drawing a charge a steal. The box score also does not consider tipping a pass a half a steal (changing the scoring expectation for your team from -1.1 to 0) or grabbing a loose ball (changing expectation from neutral to +1.1) to be half a steal, when they really are half-changes in possession.

What if we took that and applied it to everyone? How does that change our perception of volume players when weighed in a manner in which possession is worth +1.1 points of expectation, opponent possession is -1.1, and a loose ball is neutral expectation?

2019 Harden per game:
+2.0 steals / game +4.4 points expectation
-5.0 turnovers / game -11 points expectation
+3.5 tipped passes / game +3.85 points expectation
+1.5 loose balls recovered / game +1.65 points expectation
+0.08 charges taken / game +0.088 points expectation
Harden then is -1.012 points/game expectation in ball-possession plays

Westbrook
+1.9 steals / game +2.09 points expectation
-4.5 turnovers / game -9.9 points expectation
+3.1 tipped passes / game +3.41 points expectation
+1.8 loose balls recovered / game +1.98 points expectation
+0.07 charges taken / game +0.077 points expectation
Russ then is -2.343 points/game expectation in ball-possession plays

Curry
+1.3 steals / game +2.86 points expectation
-2.8 turnovers / game -6.16 points expectation
+2.4 tipped passes / game +2.64 points expectation
+1.5 loose balls recovered / game +1.65 points expectation
+0.16 charges taken / game +0.176points expectation
Curry then is +1.166 points/game expectation in ball-possession plays

Dame Lillard
+1.1 steals / game +2.42 points expectation
-2.7 turnovers / game -5.94 points expectation
+2.1 tipped passes / game +2.31 points expectation
+1.4 loose balls recovered / game +1.54 points expectation
+0.08 charges taken / game +0.088 points expectation
Dame then is+0.418 points/game expectation in ball-possession plays

Curry is worth a couple points a game more than Harden on possession/taking care of the ball, and almost 3.5 better than Westbrook.
Curry's effective FG% last year was 60.4%, to 54.1% for Harden and 46.8% for Westbrook.
The efficiency of Curry in just protecting the ball, taking the ball from the other team, and shooting the basketball is overwhelmingly better than Harden and Russ--like, not particularly close.

Dame is not on Curry's level but he's also WAY better than Russ and Harden at not costing their team possession and points.

But if we look at PER (Player Efficiency Rating) for last year:
Harden 30.6
Curry 24.4
Dame 23.7
Westbrook 21.1

How is the team of Westbrook + Harden doing 9 games into this year? They are combined:
+3.1 steals / game +6.82 points expectation
-10.2 turnovers / game -22.44 points expectation
+4.9 tipped passes / game +5.39 points expectation
+3.4 loose balls recovered / game +3.74 points expectation
+0 charges taken / game +0.0 points expectation
The combo then is -6.49 points/game expectation in ball-possession plays

Of course you'd need to run that v league average or a replacement level but being in the - like that is acceptable for a PG but not great if your team has two PG who sum up to six and a half points a night given away on slop.

Post Reply