APBRmetrics

EvanZ wrote: What is Dre referring to here?

The retrodiction test made by "Alex".

I dare say we'd be better off ignoring WP... WP is basically a harmless scam and Berri a harmless huckster. It's a miracle tonic, free for all, please buy my book and increase my exposure. It's not up to us to call him out on it, and if some people who buy his books go on to buy Basketball on Paper it has a positive side.

greyberger wrote:It's not up to us to call him out on it

Who else should do it?

As you pointed out earlier, Berri and the WP crowd want to argue with people who don't think about basketball statistics at all. They want to go after conventional wisdom and casual fans. The only way anyone will accept the claims he makes - especially the claim that WP is the true language of wins and losses - is if they're not thinking critically about it. If you're not on-board with its specialness or suspending your disbelief you aren't the right audience.

WP is a bad metric. WP the metric, the commentary, the advertising, and the appeal to authority that backs it up, isn't just bad - it's crazy. Claiming that WP is a player's true contribution and value isn't just absurd on it's own, it also contradicts to the way the metric is set up with elements like positional adjustment. The sales pitch isn't internally consistent, i'd call it borderline dishonest.

He is basically a crazy person, ranting at people who haven't heard his spiel before. Would you stand next to a crazy street prophet, explaining to passer-by that the end is not actually nigh? Anybody who buys into DB's whole argument will be able to reason their way out of it eventually, I don't think it's up to us to help them out and at times that can be counter-productive.

I know a lot of people don't like position adjustment, especially when measuring relative quality across the league, but position adjusted data can serve a purpose, for comparing players at the same position, along with non-adjusted data.

I just think it's not consistent to say that A. box score events like rebounds have concrete, consistent value in terms of wins and losses and B. that contribution is more valuable depending on the 1-5 position that player is assigned. If a steal or ORB is worth one possession, full stop, how can it depend on what position the player plays?

I just think it's not consistent to say that A. box score events like rebounds have concrete, consistent value in terms of wins and losses

Thinking doesn't either make it so or not make it so. Evidence does. Do you have some evidence that rebounds don't have some concrete value? Asked another way, do you have evidence that the values WP regressed to are wrong, or not at least indicative (e.g. if they change game to game, over a season they are not roughly accurate)?

Less thinking, more proving

B. that contribution is more valuable depending on the 1-5 position that player is assigned. If a steal or ORB is worth one possession, full stop, how can it depend on what position the player plays?

That is a fascinating question, and one that I think is answered by a simple mental exercise: would a team of 12 PGs win many games? Would a team of 12 centers? If not, what you have is a situation in which you need a mixture of skills in order to have success, and in which mutually complementing pieces need to work together.

Quoting the WP FAQ:

As noted in The Wages of Wins (and in other writings before that book appeared), centers and power forwards get rebounds and tend not to commit turnovers. Guards are the opposite. The nature of basketball is that teams need guards and big men. Given nature of the game, players should be evaluated relative to their position averages.

I think it is safe to say that:
1. Basketball is a sport in which specific things are demanded of specific positions
2. To win, a team needs to create marginal value at each position.

If every player on your team, at every position, did better than the opposition at the same position, wouldn't it be fair to assume you would win? Starting from there, it isn't hard to move to the idea that these things are fluid, and more net possessions can counteract less efficient shooting, and many other ideas.

The idea that a team needs to create marginal value seems, to me intuitively, to make complete sense.

greyberger wrote:I just think it's not consistent to say that
A. box score events like rebounds have concrete, consistent value in terms of wins and losses
and
B. that contribution is more valuable depending on the 1-5 position that player is assigned. If a steal or ORB is worth one possession, full stop, how can it depend on what position the player plays?

Because this is a clear and concise AND statement -- it's easy to see what greyberger is saying.

If a team of entirely "point guards" can out-rebound their opponent 48-40, they have won the boards.
They haven't done more or less due to the fact they are normally guards. 1 rebound = 1 rebound.

I vote for more thinking! Less misrepresentation and straw-men!

greyberger wrote:As you pointed out earlier, Berri and the WP crowd want to argue with people who don't think about basketball statistics at all. They want to go after conventional wisdom and casual fans. The only way anyone will accept the claims he makes - especially the claim that WP is the true language of wins and losses - is if they're not thinking critically about it. If you're not on-board with its specialness or suspending your disbelief you aren't the right audience.

WP is a bad metric. WP the metric, the commentary, the advertising, and the appeal to authority that backs it up, isn't just bad - it's crazy. Claiming that WP is a player's true contribution and value isn't just absurd on it's own, it also contradicts to the way the metric is set up with elements like positional adjustment. The sales pitch isn't internally consistent, i'd call it borderline dishonest.

He is basically a crazy person, ranting at people who haven't heard his spiel before. Would you stand next to a crazy street prophet, explaining to passer-by that the end is not actually nigh? Anybody who buys into DB's whole argument will be able to reason their way out of it eventually, I don't think it's up to us to help them out and at times that can be counter-productive.

beautiful post

motherwell wrote:Do you have some evidence that rebounds don't have some concrete value?

In which part of his post did greyberger imply that rebounds don't have that? Is that really so hard to understand that greyberger is critizing the inconsistency of WP48 by proclaiming first that rebounds have concrete value and then going on to assume that they have different values depending on the supposed position a player is playing. That is in fact inconsistent.

motherwell wrote: That is a fascinating question, and one that I think is answered by a simple mental exercise: would a team of 12 PGs win many games? Would a team of 12 centers? If not, what you have is a situation in which you need a mixture of skills in order to have success, and in which mutually complementing pieces need to work together.

Only because a team needs different kind of players, doesn't mean that a certain position has to provide a certain trait. Players can provide things from different positions, especially when someone is using anachronistic labels like PG, SG, etc. pp.

And only because a team needs different skilled players, doesn't mean that all players have to be equal in their value or are equal at contributing. That is a wrong assumption, easily to make by someone who has no clue about a real scientific approach. Did Berri ever tested how his metric does in out of sample tests without the positional adjustments and compared that to the performance with positional adjustment? And if that metric needs a positional adjustment, that adjustement would contradict the previous assumption that each entry in the boxscore has a concrete value for the outcome of the game. So, we are again at the point where Berri's approach becomes inconsistent.

motherwell wrote: I think it is safe to say that:
1. Basketball is a sport in which specific things are demanded of specific positions
2. To win, a team needs to create marginal value at each position.

No, it is not safe to say at all. Basketball is a sport in which specific skills are necessary. Those skills are not fixed to a certain position. Magic Johnson was the best playmaker on the Lakers, but Norm Nixon was actually bringing the ball down the court. Was Magic or Norm Nixon the PG? Why did Magic never defended the opponents PG, but Byron Scott did. How do you assign the "specific things" to a "specific position" here?
The Bulls ran the TPO, in which a PG only brought the ball up, made the pass to the wing and then was only providing the necessary spacing in order to give the ball handler the mutliple options. The Bulls constantly lost the "marginal value" battle at the PG position, just due to the structure of their offense. That is the perfect example that your 2nd point is wrong. In order to win a basketball game, a team has to score more points than the opponent by being more efficient. Which position does provide that is irrelavant. Assuming that each position has a certrain concrete value, a value which is average overall, is a wrong assumption.

motherwell wrote: If every player on your team, at every position, did better than the opposition at the same position, wouldn't it be fair to assume you would win?

You are talking a strawman. Nothing else. Nobody is disputing that. Problem is, that a metric which is based on the wrong assumption (the assumption that players are like teams and no interaction happens), will distribute the credit wrongly. That can't be compensated by making the approach inconsistent with a positional adjustment. Two wrongs != correct.

Just going to wade in and make one note here, re Berri.

Wins Produced is about 3% of what he has done as a sports economist. Whatever you may think of his metric, he is a well respected academic with a long list of publications and collaborations and a lot of very interesting things to say on topics like competitive balance, NBA salaries, NBA attendance, race in sports, and the economic impact of publicly funded stadiums, to name just a few.

I guess I can sort of understand the animus towards him, a little bit anyway. And I would agree, Wins Produced isn't perfect (what linear metric is?) But the guy has done a great deal to popularize the building blocks of apbrmetric analysis. It was a revolution that was happening already. But Berri definitely deserves a share of the credit for bringing concepts like rate stats, ts%, and efficiency differential more into the mainstream. Not more than Pelton, Kubatko, Oliver, Beech, Hollinger, and all the others, certainly, but a fair bit.

That is all....

You are talking a strawman

Really? I don't think I am (either talking or using a strawman argument hehe). http://en.wikipedia.org/wiki/Straw_man

A straw man is a component of an argument and is an informal fallacy based on misrepresentation of an opponent's position

So the statement that I responded to was that this makes no sense:

that contribution is more valuable depending on the 1-5 position that player is assigned

I asked the question: if you can't win with all of one thing, doesn't that tend to indicate you need a mix? And if you need a mix, won't the value you expect to get from each position, as Berri states in the very helpful FAQ, to be of a very different composition? Guards get assists and TOs. Centers get rebounds and fouls. Doesn't seem controversial to me at all so far.

That is not a misrepresentation of an argument, but a defence of how the opposite - that different positions produce differently - has validity and leads to the value of a thing varying (not really but kind of). I think the idea that the value a player creates is positionally dependant, and based on how far above, or below, their position's average you are at a stat, is a very good starting point.

If you start from the basic "score more points" (the strawman suppossedly) and work forward, to more rebounds, more assists, less TOs etc, the line of thinking, and reasoning, should be a little clearer, and the rationale that a PG that gets more rebounds than another PG (all else being equal) should help his team win, and that even if the PG produces less than his team's center, the PG can produce more wins because the head to head battle was won at PG. I don't think that is shocking or really that difficult to agree with personally, but YMMV.

On the value of something, as always, Shakespeare said it best: "A horse a horse, my kingdom for a horse".

Problem is, that a metric which is based on the wrong assumption (the assumption that players are like teams and no interaction happens), will distribute the credit wrongly

Wrongly, or inaccurately? The two are not the same. The moon is a sqaure is wrong, the moon is a perfect sphere is inaccurate.

I also haven't seen any proof that the interactions can be shown empirically to have an affect that invalidates using them, or is of an orders of magnitude large enough to make the metric useless. Let me use a specific example. In the famous article in the NY TImes on Battier, nowhere was it shown that, although Battier didn't get the rebound, his team got more, or that when Battier left the team, he old teammate's rebounding suffered.

Now, lack of evidence isn't evidence of a lack, but the burden of proof that a stat should not be directly attributed to the player that achieves it is on the claimant I would think.

And, taking it one step further, even if it was true that the player that gets the stat doesn't get 100% of the value, what percentage of a stat that a player collects should be attributed to others? 10%? 120%? 300%? Is Nash hitting a three the result of someone else's effort? Is Nash throwing a wild behind the back pass to Gortat for a dunk something we should give Dudley and Frye value for for spacing the floor? How do we quantify that, and is it likely to be worth closer to 2 or zero of the points scored?

Not being accurate to 19 decimal places is not the same as being invalid or "wrong". If there are interactions (and there no doubt are), but the interactions aren't orders of magnitude in nature, it doesn't invalidate WP, it rather affects the accuracy of WP, simply meaning that it is theoretically possible to create an even more accurate version of WP.

Maybe arguing with oneself is not exactly the same as arguing with a straw-man.
It's close, though.

Metsox wrote: I guess I can sort of understand the animus towards him, a little bit anyway. And I would agree, Wins Produced isn't perfect (what linear metric is?) But the guy has done a great deal to popularize the building blocks of apbrmetric analysis. It was a revolution that was happening already. But Berri definitely deserves a share of the credit for bringing concepts like rate stats, ts%, and efficiency differential more into the mainstream. Not more than Pelton, Kubatko, Oliver, Beech, Hollinger, and all the others, certainly, but a fair bit.

I don't see it that way at all. Bascially nobody in the bigger basketball community websites is using anything based on the work of Berri. Whenever someone brings up a Berri op-ed, the most people laugh about him and his metric, and pointing out why advanced stats are useless. His behaviour has quite the opposite effect, because people with obvious less knowledge about all that are getting drawn away from advanced metrics. If you want to convince the mainstream to use new methods, you have to show them something they can believe. WP48 isn't a good tool to predict the outcome of the game. And trying to insult the audience by basically calling them dumb, because the mainstream is thinking more about scoring, is not going to convince them that scoring is not that important as they think. Really, I actually agree that the mainstream is much too focussed on scoring instead of other things, but also too much on raw boxscore numbers. Berri's attitude is not helping the cause at all, and that is another issue. I wouldn't care as much about the flaws in WP48, if he wouldn't use it in the way he does with that kind of attitude. Heck, Hollinger's PER is similar "useless" as WP48, but Hollinger is actually communicating much better with his audience and brings them much more to talk about advanced metrics. I guess also, because PER is much more in agreement with the mainstream view. I can live with that, because I see Hollinger is indeed helping our cause, Berri is not.

When I talk to people and they use advanced metrics, they usually have that either from Hollinger, basketball-reference or even Dean Oliver. Nobody ever said to me that Dave Berri was the guy responsible for their interest in advanced metrics.
If you compare "Basketball On Paper" with "Wages Of Wins" in terms of the way it communicates the knowledge, we see Oliver being respectful with a good way of explaining things. Berri on the other hand is rude and basically insults owners, GMs and coaches. The only way to get respect in such a case is to present a really superior tool. But WP48 is not that tool, it is not superior in terms of predicting. Nobody should expect to get more respect with that kind of behaviour.

And regarding his academic work: Publishing all his stuff in basically the same low impact journals while citing himself does not make a good case in terms of the quality of the work. Quantity is mainly the result of effort. Well, Berri puts in the effort, good for him. His effort made it possible to have a good job, earning his money and some sort of "fame". He should enjoy that.

motherwell wrote: I asked the question: if you can't win with all of one thing, doesn't that tend to indicate you need a mix?

No, you actually asked whether it would not lead to a win, if everyone outplays their direct opponents and then proceded to talk as if someone would have said the opposite. That is indeed a strawman, because you misrepresented the point greyberger made. Greybergers point was that it doesn't matter for the team who will get the credit for the rebound in the boxscore as long as the team as the whole ends up with the ball. Did Berri ever show that a rebound by SF is more valuable than the rebound by a PF? In order to justify the positional adjustment, he would need to show that in fact the same amount of points scored on the same efficiency are more valuable, if they come from the SF position than from the PF position. Did he ever do that? No, for sure not.

motherwell wrote: If you start from the basic "score more points" (the strawman suppossedly) and work forward, to more rebounds, more assists, less TOs etc, the line of thinking, and reasoning, should be a little clearer, and the rationale that a PG that gets more rebounds than another PG (all else being equal) should help his team win, and that even if the PG produces less than his team's center, the PG can produce more wins because the head to head battle was won at PG. I don't think that is shocking or really that difficult to agree with personally, but YMMV.

Basketball is a game of 5 players on one team vs. 5 players on the other team. Basketball is NOT a game of 5 different 1on1 games. A point guard can easily lose the "head to head battle" in terms of boxscore stats, but can very well have the bigger positive impact on the game, because he is making the right decisions. The end result, the work of 5 people vs. 5 people is important.

motherwell wrote: I also haven't seen any proof that the interactions can be shown empirically to have an affect that invalidates using them, or is of an orders of magnitude large enough to make the metric useless.

The need for the positional adjustment is proof enough that the original hypothesis is wrong. Either the boxscore entries have intrinsic values, in that case we can use regression in order to determine that or they don't. When you come up with a result which you can't trust at all, because of the way the credit is distributed among players (heavily in favor of bigger players, so much that it would suggest that a team should only use big rebounders), we either can accept the result and go from there or we can dismiss the original hypothesis. Berri is doing something which is basically against all scientific wisdom. He takes the original hypothesis and declares it as "inaccurate", while then procedes with an assumption based on conventional wisdom (a team needs different kind of players for different positions on the court). After that he declares that conventional wisdom is wrong, based on the results coming from an inaccurate hypothesis (something he didn't trust) adjusted for conventional wisdom (something he wants to refute). That is basically completely insane and has nothing to do with science anymore.

motherwell wrote: Not being accurate to 19 decimal places is not the same as being invalid or "wrong". If there are interactions (and there no doubt are), but the interactions aren't orders of magnitude in nature, it doesn't invalidate WP, it rather affects the accuracy of WP, simply meaning that it is theoretically possible to create an even more accurate version of WP.

See, you don't understand the real issue here. It is not possible to make WP with the current hypothesis (the thing the marginal values are based on, the regression Berri did) more "accurate". With every "adjustment" Berri makes, he goes further away from the original hypothesis. That reminds me on the geoscientists from the beginning to the midst of the last century. When they thought the geological structures on the planet were caused by a shrinkage of the Earth. They refuted Wegener's plate tectonic theory based on their wrong original hypothesis adjusted for the "inaccuracies" of the model. Obviously the shrinking Earth theory wasn't good at predicting, it was good at explaining things in hindsight, especially with the adjustments. Berri is running into the same fallacy those geoscientists had.

greyberger wrote:I just think it's not consistent to say that A. box score events like rebounds have concrete, consistent value in terms of wins and losses and B. that contribution is more valuable depending on the 1-5 position that player is assigned. If a steal or ORB is worth one possession, full stop, how can it depend on what position the player plays?

First off, I'm not a fan of WP for a multitude of reasons discussed on this board, so it surprises me that I'm going to defend one of the WP concepts....

I'm not sure that the execution of the position adjustment was done in the best manner possible, but conceptually there are sound reasons for adjusting production relative to position. The heart of the issue is that box score stats (and even to some extent rates generated from box score stats) do a poor job of telling us how "good" or "valuable" or "productive" someone is at doing Y either individually or in a team context.
Taking reb rates down a couple of levels of specificity:

Level I: total rebounds. Rebounds are nothing more than counting occurences of an event. If player A gets 10 DRB and player B gets 5, saying that player A is a better defensive rebounder isn't necessarily true. You wouldn't say that a player scoring 20 points is a better scorer than one scoring 12 points if you knew the former went 10-45 and the latter went 6-10, would you? ...Assume for a moment that the 6-10 shooter would be able to hit at least 4 shots if he took an extra 35. There is no measurement of opportunity to scale production when it comes to rebounds when using raw totals.

Level II: rebounding rates. Even ORB% and DRB% may not capture the opportunities players have to rebound, which is really what defines rebounding ability. ORB and DRB% measure rebounding frequency in a team context. If 60% of all DRBs happen to fall within an interior player's "rebounding opportunity zone", and if 3 to 4 players (including that player) typically compete for those rebounds, then I would expect that player to have a DRB rate of roughly 15 to 20%. Similarly, if 15% of rebounds happen to fall in a guard's "rebounding zone", and he's only one of 2-3 players competing for those boards, I would expect his DRB rate to be roughly 5% to 7.5%. Rate totals aren't going to provide a good indication of marginal value over some presumed replacement level, which is really what defines contribution.

Assume replacement level equals 90% of expected boards and there is no teammate DRB cannibalization for a moment. A hypothetical comparison:

-PF/C with 20% actual DRB%, an expected DRB% of 17.5%, and a replacement DRB rate of 15.75% generates 4.25 extra DRBs over replacement per 100 team DRB opportunities.
-G with 10% actual DRB%, an expected DRB% of 6.4%, and a replacement DRB rate of 5.76% generates ~ 4.25 extra DRBs over replacement per 100 team DRB opportunities.

In this instance, there is nothing controversial about stating that the G's DRB contribution is equal to that of the big man and that each G DRB is worth twice as much to a team as one DRB from the interior player.

This works the other way too. I don't think it's a stretch to say that five assists from a Sabonis or a Bill Walton in 32 minutes is worth more than 5 assists from Kevin Ollie over the same minutes.

That said, there are two problems with implementing a "vanilla" position adjustment assumption:

1)It doesn't address team-specific offensive and defensive schemes. Just because the average G DRB rate is X%, that doesn't mean it isn't X +/-Y% for certain teams.
2)Replacement level is difficult to determine and can vary widely by position. Even if you could accurately estimate expected boards for a player by DRB rebounding zone opps by players in competition for those rebounds, replacement level will be different based upon position. Ignoring scheme/cannibalization, big men competing among several players under the basket will have higher Reb rate variance (and lower replacement level Reb rates) than a PG standing 25 feet from the basket will. There is more competition under the basket, so superior ability will crowd out inability.