The work of Justin Wolfers has been a frequent topic in this forum. Stacey offered two posts examining Wolfers study of point shaving in NCAA men’s basketball.
Point Shaving in NCAA Basketball
Point Shaving Briefly Revisited
And I offered four posts examining the Joe Price–Justin Wolfers study of implicit bias in the NBA.
More on the Price-Wolfers Study
Ian Ayres comments on Price-Wolfers
Each of these studies demonstrates an ability to uncover the unseen – and unexpected – in large data sets. Of course playing with data is not the only skill in Wolfers arsenal. He is also a very talented economist (which means he can do more than just play with spreadsheets). Yesterday he demonstrated his ability to use economics to solve problems with a wonderful op-ed in the New York Times.
In Blow the Whistle on Betting Scandals, Wolfers offers a unique perspective and solution to the negative impact gambling can have on sports. Basically Wolfers argues we need more gambling – with respect to certain aspects of sports – which will reduce gambling on other aspects of sports and hence reduce the likelihood of scandals. If that last sentence didn’t make sense – and I’m not sure it did– go read the column (or read Henry Abbot’s take on this at True Hoop).
Price-Wolfers Explains Win Score
And for more from Wolfers (or is it Joe Price?)…three weeks ago I posted on the Western Economic Association meetings, which has become the annual meeting place for sports economists from around the world. The first paper presented in our first session was the aforementioned Price-Wolfers study. My sense is that this paper has typically been presented in forums where the presenter is given more than 30 minutes to discuss their work. At the WEA, though, you only have 15-20 minutes to explain your paper, consequently Joe Price had to leave out some of the material that Wolfers or he would include in a longer discussion.
Fortunately, though, I was given a copy of their entire power point presentation, and hence I got to see the whole story that they normally tell. Part of this story is a new look at Win Score.
The Win Score formula is as follows:
Win Score = Points + Rebounds + Steals + ½Assists + ½Blocked Shots – Field Goal Attempts – Turnovers – ½Free Throw Attempts – ½Personal Fouls
In the Price-Wolfers power point presentation this basic model was presented as follows:
Points + Possession gained (rebounds, steals) – Possession lost (turnovers, field goal shots, ½ free throws) + ½ Offensive help (assists) + ½ Defensive help (blocks) – ½ Help opponent (fouls)
Let me write this out in words. Basically a player’s value is determined by points scored, possessions gained, possessions lost, help on offense, help on defense, and helping of opponents.
The Price-Wolfers explanation of Win Score seems a bit more intuitive than what we said in The Wages of Wins. As noted in the book, the Win Score model is designed to be a simple (and accurate) measure of player performance, much like OPS in baseball. The intuition of Price-Wolfers, I think, makes the Win Score model easier to understand.
One should note that Win Score was developed to ease research in economics that utilized performance data from the NBA. The Price-Wolfers study demonstrated this aspect of the model. Their work looked at the impact the race of the referee (and the player) had on player performance. Much has been made of the impact uncovered with respect to personal fouls. But Price-Wolfers found much more. This study also found that the race of the referee impacted other aspects of player performance like scoring and turnovers. Utilizing Win Score they were able to estimate the overall impact of a “race-normed refereeing crew.”
Quoting from the paper… consider a game involving five black starters against four blacks and one white. Thus any team-level differences will be driven by the differential treatment of the fifth player, who is black for the home team, and white for their rival. The coefficients in Table 4 suggest that race-norming the refereeing crew would lead the black player to commit around 0.1 fewer fouls per 48 minutes (relative to the change for the white player). Table 5 suggests that he would also score around 0.2 more points and earn 0.05 extra turnovers. Alternatively, using Berri, Schmidt and Brook’s (2006) “Win Score” metric, the black player’s overall contribution to the team winning margin will rise by about one-quarter of a point under a race-normed refereeing crew (relative to his white rival’s contribution). These individual-level estimates are consistent with the estimates of the “direct” effects measured in Table 6. But recall that Table 6 showed that these “direct” effects on fouls committed and points scored are roughly matched by an equal-sized (and opposite signed) “indirect” effect on fouls awarded, turnovers gained, and points conceded. That is, the away team’s boxscore statistics also change in a way that leads further extends the home team’s winning margin by another quarter point.
Much of the discussion of this paper focused solely on the impact on personal fouls. But the finding that players altered other aspects of performance was also quite interesting.
Smush Parker Signs with the Heat
Let me close today’s post with a quick comment on Smush Parker signing with the Miami Heat. This is how this signing was described by Pat Riley:
When we look at Parker’s Win Score per minute, we see that he was at 0.140 in 2005-06 and at 0.090 in 2006-07. Average for a point guard is 0.128, so Parker went from being above average to below average. In other words, just looking at Win Score gives us the impression that Riley is a bit off in his assessment.
If we look deeper into the numbers, though, this signing may make some sense. Parker is above average with respect to turnovers (he commits fewer than an average point guard) and shooting efficiency. Where he declined in 2006-07 is with respect to rebounds and assists. If Riley can change those two aspects of Parker’s performance, he might have acquired an average point guard. And as we all know, average ain’t “bad.”
I bring up this story to emphasize a point made in The Wages of Wins about Win Score and Wins Produced. Looking at the numbers is not the end of analysis. It’s just the beginning. Coaches have to look into why a player achieved those particular numbers. In the case of Parker, the common problem for guards – turnovers and shooting efficiency – are not the problem. If Riley and his staff can fix the problem areas (rebounds and assists), it looks like this signing could pay off. Of course, if they can’t, then the Heat have simply acquired a below average point guard who is not going to help Alonzo Mourning get back to the NBA Finals.
– DJ
t.g. randini
July 29, 2007
Well… well… well…
Player A shoots 15 for 30 and gets 30 points.
He doesn’t commit a foul, make a turnover, grab a rebound, or make an assist.
So his win score is 30 minus 30 equals 0.
Player B comes off the bench and grabs a rebound at the end of the game.
His win score is 1.
Player A is a ZERO on this site even though his shooting percentage is greater than the NBA average AND he did not commit a foul or a turnover.
Player B has a higher win score even though he’s the twelth man on the bench and plays 30 seconds.
GIVE ME A BREAK!!!!!!!!!!!!!!!!!!!!!!!!
Mr. Parker
July 29, 2007
randini,
Billy Knight and Isiah Thomas are looking
for smart people like you to work along with
them. Your point totally makes sense. As long
as a guy has shot the ball a lot he is definitely
helping his team.
Harold Almonte
July 30, 2007
Randini,
As you can see, the basketball game is stepped on one only foot in that simplistic formula. There’s no counterparting for “points made” at the defensive end (points allowed), the possession is gained just once the ball is rebounded and the defense on the field goal shot is not credited, but this same field goal shot is counted as a whole weighted lost no matter is missed or made (a mayor disorder).
Then you adjust for all these defensive stat unavailability, and convert the NBA defense (the most of the time one on one)into a shared defense, but meantime, rebounds still stayed as an isolated defensive action play, what of course it’s not.
Harold Almonte
July 30, 2007
Also in this logic you can see a possession gained (about 1 point) is allways a succesful end, with succesful continuation when the possession is scored. When the shooter scores, he’s only credited with a piece, because the possession gainer already earned his 1 point. But if the shooter misses the shot, he is twice punished while the possession gainer still keeps his 1 point. Of course in these metrics, possession gainers are not punished every time they fail to gain, or attempt to. You can see possessions are lost only at the offensive end.
Harold Almonte
July 30, 2007
And also, given that there’s no way to weight or adjust for difficulty of shots and defense faced, nor help efficiency, we must accept that once with the ball in their hands, all players’s scoring metric is determined only whether or not the ball enters the ring (by shooting only), and a 50% (or 33%) “scoring break even” is an absolute metric for all of them, in all situations. The position adjust arranges only per minute stats, but not per attempts (efficiency) actions.
Mr. Parker
July 30, 2007
Harold,
It seems like you might like the player win
percentage which is posted on basketball-
reference.com for every player who has
ever played.
I used to favor their method until I realized
that what Dean Oliver was using as a ratio
Mr. Berri has computed by addition. The same
amount of credit is given for each stat by
Oliver, but when he calcutes his overall
player rating he weights the defensive stats
because players like Olajowon would have
defensive ratings of 30 out of 100.
Using either metric you actually get the same
results. Some players make so many defensive
stops that they do in fact make their teammates
better by finishing off posessions. These players
are valued highly by both systems.
Who does pw% say the greatest player since
79-80 is….Michael Jordan, followed closely
by Tim Duncan.
With the same inputs I would have to conclude
that Mr. Berri’s system would say the same.
If you create a metric to measure player out-
put and the two best players it spits out
are Tim Duncan and Michael Jordan I would
say that system has merit.
t.g. randini
July 31, 2007
Great comments above Harold and Mr. Parker.
I haven’t had the time to re-read Oliver but I think he values Jordan more highly than Berri does. I’m in the midst of buying/selling homes and I wish I could elucidate further.
Some notes though… concerning Berri’s approach.
WHERE IS THE RISK VS. REWARD VALUATION?
Going back to my first comment… Player A takes all the risk in trying to put the round ball in a cylinder.
Now let’s add two other players. And let’s simplify all three roles.
Player A is the shooter.
Player B is the rebounder (power forward).
Player C is the distributor (point guard).
And let’s hold all their other roles at a constant value. For example, the shooter and the forward could each have two assists and three fouls, etc. etc. For simplicity, let’s hold the other roles at zero. It doesn’t matter if it’s zero or a number as long as non-specialty performance is the same for all of them.
Player A goes 30 for 60 and scores 60 points. A great game. Efficient scoring. No turnovers or fouls (but also no rebounds or assists.) He doesn’t need assists because he’s the guy who can actually shoots and his team, knowing he is an efficient shooter, WANTS him to shoot.
Player A takes ALL THE RISK in trying to score points because in Berri’s system, he LOSES points when the round ball does not go in the hole.
Player B… the rebounder… takes NO risk. He just stands by the hoop and gets the other team’s misses and some of his own team’s misses. He NEVER gets punished for standing around.
B gets 15 rebounds but can’t shoot and doesn’t. He takes no risk. No fouls, assists, etc.
Player C… the distributor… takes NO risk. He dribbles around and passes the ball. It’s a lot easier to pass the ball to a guy who is BIG, a guy who MOVES to catch the ball with LONG arms. That is so much easier than trying to actually put the round ball in a small hole. Player C TAKES NO RISK.
He gets 30 assists because he passes the ball to Player A who takes ALL THE RISK.
What are their win scores?
Player C is 7.5 win score for merely passing the ball to someone who can do actually do something with it. NO RISK.
Player B is 15 win score for merely standing around and grabbing errant shots off the backboard. NO RISK.
Player A gets 0 win score for scoring 60 points efficiently and is the only guy who TAKES ALL THE RISK.
Gentlemen… none of the PASSING AND ‘GETTING’ means jack unless SOMEONE PUTS THE BALL IN THE ROUND HOLE.
Truth through simplification.
If you substitute Dennis Rodman for Player B and the early Jordan for Player A, Berri would prefer Rodman.
I like Dennis. I live in Chicago. He was a stitch. Very enjoyable. But metrics that don’t value RISK overvalue players who don’t take risk…
and UNDERVALUE players who do.
When I get time, maybe I’LL write a book.
t.g. randini
July 31, 2007
P.S. to the above:
CORRECTION… Player C has a win score of 15 just like Player B.
There are SO many things this system doesn’t measure.
It measures blocked shots… but most block shots have only 50% chance of taking the possession away. A ‘hurry’ might actually be better as the defensive team has a greater chance at a rebound (defensive rebounds outnumber offensive rebounds).
A ‘pick’ is as valuable as an ‘assist’.
If we add Player D… the ‘picker’… who frees up Player A (the shooter) and he picks off the defense for 30 baskets scored by Player A… he gets a win score of 0 for ‘throwing the pick’ while Player C gets a win score of 15 for throwing the ball.
And Player D takes more risk! (To his body…)
In summary:
These statisticians should watch the game a little more and the box scores a little less. I’m not ragging on eggheads, either. I have an MBA with straight A’s in Finance and Statistics.
But I’ve played the game.
t.g. randini
July 31, 2007
Last post!
CASE STUDY: WILT CHAMBERLAIN IN THE EARLY 1960’S
The Dipper would routinely go 25 for 50 and grab 25 rebounds.
His win score is 25.
His entire value is grabbing the ball.
He gets 0 value for score 50 points at a 50% shooting percentage when the average NBA shooting percentage is 40%.
EVERY TIME HE SHOT AT A 50% PROBABILITY OF SUCCESS, HE WAS HELPING HIS TEAM.
But he gets 0 value for this.
But a 25 win score. (For grabbing the ball).
NOT LET’S SUBSTITUTE ‘LITTLE MINI-ME WILT’… a 5’8” guy who doesn’t get any rebounds.
Same team. Same era.
Goes 25 for 50. Get’s 0 win score.
EVEN THOUGH EVERY TIME HE SHOT AT A 50% PROBABILITY OF SUCCESS, HE WAS HELPING HIS TEAM.
I rest my case.
Mr. Berri… you should spend less time posting new articles on your flawed system and more time learning from us masters.
As Hannibal once said crossing the mountains…
“Correlation is not cause.”
t.g. randini
July 31, 2007
I meant… “NOW let’s substitue Mini-Me Wilt”… not ‘NOT’.
Harold Almonte
July 31, 2007
Complete players with very few shortcommings at both ends will be at the top no matter which metric is used (a not adjusted linear metric can explain about a 70% of reality, ball-on-hands reality). The problem comes with unidimensional players. Metrics will overrate or underrate these players acording to its bias, and that’s produced by using totally different and divorced methods to measure defense and offense, or not using a good solution (or anyone at all) to do counterpartings that boxscore can’t do (points).
WP is a metric that measures and weight “win games” actions, not just individual skills unfolding (that can stay short when you correlate to wins). It’s designed so, and that is the confussion when you try to compare this metric against another. The problem comes when putting worth and price to those skills and players wich possess them, based on the result only, obviating the difficulty of the process. It may be “relatively” easy to gain a possession (even with an inbound pass you get one, and you gain a lot of possessions that you did neither defend nor steal at all), the problem is to convert it in a score, and shooting% might not be the last word here. Shooting defense attempts are not treated like its counterpart at the offensive end in any metric, and all is left to be decided in the battle of rebounds.
Troy
August 7, 2007
I agree that performance should be measured in terms of possessions gained and lost, but I’m not sure that the Win Score formula is the best way of doing it. To see why, imagine a simplified version of basketball in which two players, A and B, take turns shooting from a designated spot on the court. After each shot, two other players, C and D, go up for the rebound. If C rebounds the ball, A takes the next shot, and if D rebounds the ball, B takes the next shot (the first shot is determined by the referee throwing the ball against the backboard). Let’s the say the statistics at the end of the game are as below:
Player A: 8 for 24, 16 points
Player B: 15 for 20, 30 points
Player C: 24 rebounds
Player D: 20 rebounds
It should be pretty clear here that Player B is the most productive player and Player A is the least productive player, with Player C being slightly ahead of Player D. While B and D are both on the winning team, Player B is more productive because he is more effective at doing his job (i.e. scoring) relative to Player A than Player D is doing his job (i.e. rebounding) relative to Player C. If A and B shot with the same accuracy:
Player A: 18 for 24, 36 points
Player B: 15 for 20: 30 points
Player C: 24 rebounds
Player D: 20 rebounds
Here Player C is the most productive player, and Player D is the least productive, with C’s extra rebounds being responsible for getting his team over the line.
Sorry if this line of argument has been brought up before. I don’t know how Win Score could (or should) be adjusted to reflect this, but basically I think something needs to be done to put scorers and rebounders back on an even playing field. Position-Adjusted Win Score could go some way to addressing this, although I don’t think it adjusts for the fact that the role of a player at a particular position may vary from team to team.
Troy
August 7, 2007
Sorry, a friend has pointed out that in my examples rebounds don’t equal missed field goals, but I think you can fix the examples so that my general points hold true.
James Fitzmaurice
December 14, 2009
Hi Troy no worries we all make mistakes. This is a great article i thorougly enjoyed reading it on this cold winters night in the UK!
cheers James