This past weekend I was in New Orleans attending what economists refer to as The Meetings. Specifically, I was attending the American Economic Association meetings, which is the main academic gathering in our discipline. In essence, about 18,000 economists descended on New Orleans, much to the chagrin of the fun-loving people of the Big Easy. Fortunately for the city, fans of LSU and Ohio State arrived soon after the AEA meetings began, and when I checked out of my hotel Sunday morning at 4am the party was in full-swing.
Before I left Bakersfield I posted “Sort of Defending Isiah Thomas.” With this column, I had offered at least one posting for 41 consecutive days. Virtually all of these columns have been as least 1,000 words in length; which works out to more than 40,000 words in a little more than a month. To put that number in perspective, The Wages of Wins was only about 119,000 words (including the Table of Contents, Index, and over 40 pages of end notes). So basically I wrote about 33% of a book in about six weeks.
After this quantity of writing, I think I needed a break. So I deliberately left my laptop at home when I went to New Orleans. This meant I did not check in on the Wages of Wins Journal for about 60 hours, an all-time record I think. 
Martin Johnson’s Perspective on the Knicks
When I returned to the forum I saw the following comment from someone named MJ:
Must be something in the air this week. I weighed in on several of these topics in my NY Sun column on Thursday.
Pardon the self-promotion but I think y’all will enjoy this article.
http://www.nysun.com/article/68904
MJ is Martin Johnson, a writer with the New York Sun. Johnson’s column was entitled Poor Risk Assessment Isiah’s Downfall. In this column, Johnson makes the following observations:
After watching this administration for four years, I have to staunchly disagree with the folks who say there wasn’t a plan. There was a plan. It involved remaking the roster by capitalizing on low first-round draft picks (a typically undervalued commodity), and by going aggressively after high-risk players who lacked other options in the free-agent market.
If forwards David Lee and Renaldo Balkman are any indication, then the draft strategy part of the plan is a success (though why Wilson Chandler doesn’t receive more burn is a mystery to me).
…The free agent part of the plan has been an unmitigated disaster to such a degree that not even Dino De Laurentiis or any other Hollywood disaster flick specialist could create anything paralleling the magnitude of what went wrong. Thomas’s free agent signings include centers Jerome James and Eddy Curry, guard Jamal Crawford, and swingman Jared Jeffries, and it illustrates the entirety of the problem with this regime. There’s no attention paid to risk assessment. Each of these players had a considerable downside – a smart GM would have offered each a contract no longer than three years. Three-year deals work well for both sides. For the team, it limits their liability. For a player – if he develops star-level talent – then he’s three years away from a possible nine-figure deal.
If Thomas had signed James to a three-year deal, then he could offer James in trade this season as an expiring contract. He didn’t (James is in year three of a six-year deal), and that illustrates the problem with hiring elite athletes to executive positions: Thomas was a Hall-of-Fame guard in his playing days, and he was accustomed to overcoming long odds every time he drove the lane against bigger opponents. His success on the court gave him a skewed perception of acceptable levels of risk. Thomas essentially bid against himself for Curry, a player with a health issues that prevent his contract from being insured. Crawford and Jeffries also have key weaknesses in their games. Committing to each for six years, as well as trading unprotected lottery picks for Curry was a strategy with a probability of success that was less than one in four. But Thomas, as a player, was used to beating those odds every time he took the court.
In essence, Johnson is making a somewhat similar argument to what I offered. Thomas indeed has a plan. But his past experience causes him to make mistakes. And it’s important to emphasize the word “mistakes.”
Crawford, Curry, and James were all mistakes. I would argue, though, that these were mistakes Thomas could have avoided had he looked past scoring. None of these players had ever been very productive before coming to New York, and it seemed unlikely any of these would become productive in the Big Apple.
Hollinger’s Perspective
I think it’s important to contrast what Johnson and I are saying with another New York Sun columnist, John Hollinger. On the 4th, the day after Johnson’s column appeared, Hollinger wrote the following:
Wednesday night, we saw one of the biggest talent mismatches you’ll ever see in an NBA basketball game.
And the team with all the talent had no chance.
At Madison Square Garden Wednesday night, the Knicks arguably had the seven most talented players in the building in their game against the Sacramento Kings. But they still lost, 107-97, and this result was far, far worse than the final score indicated. First, because the team trailed by 24 points late in the third quarter before some cosmetic scoring closed the deficit, but mostly because of the staggering talent disparity between the two sides.
No offense to the Kings, but the Knicks should have won this game by 40 points, and if they weren’t so preoccupied with mailing in the season, they would have.
In Hollinger’s view, the Knicks have talent. This team, though, simply has quit on Thomas. Although it may or may not be true that the Knicks are not trying as hard as they could, I don’t think the Knicks would be a very good team even if this team tried very hard. As I noted in my column, other than David Lee and Renaldo Balkman, none of the Knicks players has historically been very productive.
Let me review my argument. If you focus on scoring, many of the Knicks do appear to be productive players. But if you consider the propensity to waste possessions – by missing shots and/or committing turnovers – many of the players on the Knicks have not contributed much to wins in the past. And this, I think, is ultimately why this team does not compete in the Eastern Conference today.
Promises Made in New Orleans
While in New Orleans I had a chance to meet up with several co-authors, fellow economists, and fellow sports economists. The list included Jennifer Van Gilder, Martin Schmidt, Joe Price, Justin Wolfers, Dennis Coates, John Solow, Peter von Allmen, Brad Humphreys, and Victor Matheson.
I would note that since Martin left Colorado State in 1994, Marty and I have rarely been in the same physical location. The last time I had actually seen Marty in person was 2002. As we both observed, for two people who never see each other, we work well together. Perhaps this is because we never see each other?
While talking to Marty (as well as my other co-authors) I frequently made the following promise: In 2008 I will work less on the Wages of Wins Journal and more on my research. This means I need to focus on the sequel to The Wages of Wins (which Marty and I discussed at length), as well as separate projects I am working on with Jennifer and Joe (as well as with other people).
Given this promise, I am going to try and stop posting a column here each and every night. That being said, as long as I make progress on my research program (at least enough to satisfy my co-authors), I think I will be able to offer two or three columns per week.
Perhaps as the quantity dips, the quality can actually increase making this forum even better (or is this just an example of that damn usage argument again?). Of course, this column – that I mostly took from Johnson and Hollinger – hardly represents an increase in quality (at least, not on my part). Still, I hope my future columns are at least worth a read.
Let me close by once again thanking everyone who stops by and reads this blog each day. It’s amazing that nearly 1,000 people seem to make this forum a part of their daily routine. Hopefully, once book II is finished in a few months, I can get back to more regular posting.
- DJ
41 responses so far ↓
magicmerl // January 7, 2008 at 12:25 am
I think that you’re right that Hollinger overrates scoring in PER, but I think that you’re wrong to treat someone who produces 15/30 shooting as having made an equivalent contribution to their team as someone who shoots 1/2.
Hollinger rewards people for every shot taken once they get past 30%. And it’s true, this basically rewards every player simply for jacking up more attempts. But I think that your only rewarding people for shooting over 50% is too harsh.
The truth must lie somewhere in the middle of those two values (maybe 45%?)
The truth must like somewhere in the middle
SHC // January 7, 2008 at 4:15 am
I think that a lot of basketball pundits believe offensive talent is skill whereas defense and rebounding are more results of effort.
People like John Hollinger probably believe that with the right coach who can bring out the maximum effort out of his players, Knicks would be playing much better defense (which in turn could be reflected in increase in rebounds, steals, and probably WP48 of many players).
I think that such a belief holds some truth. Great offensive players usually have tools to be good defenders, rebounders if they tried improving on these areas. But because of popularity of scorers (every fan looks up to scorers, not Ben Wallaces), these guys probably don’t care too much about these ‘hustle’ areas.
GV // January 7, 2008 at 8:58 am
magicmerl, sometimes when someone says 2 + 2 = 6 and someone says 2 + 2 = 4, the truth doesn’t lie in the middle, and there’s no reason to think the truth often lies in the middle when two people disagree.
Harold Almonte // January 7, 2008 at 9:28 am
GV. someone said 4+2=6. What magic is proposing is 3+2=5. Everybody can say whatever they want, the problem is nobody has forceful demonstrated that scoring is (2+2) or (4+2).
Jason // January 7, 2008 at 10:51 am
Whether or not rebounding is more “effort” while scoring is more “skill” doesn’t change how important these tasks are. It also doesn’t necessarily suggest that it’s easier to generate “effort” than it is to refine skill. Rebound rates stay remarkably similar for a player from year to year, even when players change teams, teammates and coaches. If it’s effort, than effort is much more of a constant that isn’t influenced much by coaches, teammates or surroundings such that it starts looking like it’s really something intrinsic to the player.
I don’t see where the evidence is that great scorers could be superior rebounders with effort comes from. I certainly haven’t seen data that suggests that it happens. If it is possible in the abstract, but doesn’t ever happen, evaluating the abstract case in evaluating the player doesn’t seem to hold much value. Perhaps they can do other things defensively with increased effort.
If you look at how WP actually evaluates shooting percentages, the 50% rate is only on 2-pt shots and 50% from 2 point range is about the league average. (~45% is the average for all baskets, but this includes 3 pointers). In actually computing WP, because of the position correction (which uses an average rate of return) the break-even is whatever the actual average at the position is. I’m certain that this has been mentioned before and Dave has posted about it. It would be nice if critics who claim that WP doesn’t reward shooting as it should would be critical of what the formula actually shows rather than what they think it shows.
Harold Almonte // January 7, 2008 at 11:21 am
“It also doesn’t necessarily suggest that it’s easier to generate “effort” than it is to refine skill.” The best sentence I heard today. It isn’t easier for a coach to earn leadership worth when he just passed away the threshold of no reliability either.
You still think like if rebounds is all what matters at the defensive end. And about the WP’s scoring approach: to make the shooter individual responsible for the FGA possession lost is debatable. It’s just a shortcut for not to make things more difficult, and I’ve allways sustained that to take shortcuts at the player level is not the same that to do it at the team level.
Harold Almonte // January 7, 2008 at 11:30 am
Rebounds are more correlated with floor position and height than skills and effort.
SHC // January 7, 2008 at 11:40 am
It doesn’t take a dramatic improvement for players for a team to be successful. Let’s say that a coach gets 2-3 wins out of 5 key players in his team and 1+ wins out of the other 3-4 bench reserves. That’s already 13~18 more wins. I think that you don’t even have to ask people to double their rebounds or steals.
I think that teams really do improve in these areas. Look at Dallas Mavericks, for instance.
They were one of the worse defensive, rebounding teams in the league in 2002-03 (when they still one 60 games). Last year, they ranked in top 5 in both defense and defensive rebounding in the league. Was it addition of players that caused such a dramatic change or was it more about coaching?
Jason // January 7, 2008 at 12:04 pm
Please, Harold, do refrain from telling me what I think. It tends more often than not to be incorrect and merely demonstrates that you do not possess the power to read minds. I do NOT think that rebounds are all that matter at the defensive end, and strongly, strongly suspect that you will be unable to find anything that I’ve said indicating that I thought as much.
What I have said, and continue to believe, is that rebounds are a very, very good indicator of defense. They are not all that matters, but a defensive rebound is an accurate measure of a defensive stop.
A defensive board requires some sort of missed shot. I have said this as well, Harold. (In testing your flawed powers of telepathy to tell me what I think, perhaps you used energy that would have better been tasked to actually reading for comprehension of what I’ve written.) Since missed FTs are the minority of opportunities for such, *most* defensive rebounds indicate a defensive stop after a missed FG attempt. Certainly more went into the miss than just the rebound. However, if you know the number of defensive rebounds, you essentially know the stops as result of failed FG attempt. At the team level, you have most of what you need to know.
While you do not at such a point know how the missed shot occurred or, more specifically, who should get how much credit for the stop, you do know that a stop has occurred. Now is it important to divide up this credit at this point? That’s an empirical question. (Yes, I know I’ve said *this* too, but, Harold, since you clearly seemed to be too busy trying to figure out what I *think* perhaps you missed what I *wrote*.) This is not about fairness. It’s about how well the model holds to accounting for actual team records and towards predicting future results from past results. It may or may not *matter* that the credit for a rebound gets divided *at this point* in the model. And there is reason to say that even though it took something more than the rebound to produce the rebound that it doesn’t really matter as far as using the model to predict future results. This reasoning is supported by the observation that rebounding rates for players are highly *invariable*. They do not vary much from year to year for players regardless of surroundings. This is an indication that using the defensive rebound as proxy for the whole of the defensive stop is *good enough* to make accurate enough predictions and that monkeying around with it and allocating some fraction to someone else isn’t necessarily going to improve the predictive powers. It is not merely a shortcut. It’s drawn from actual observations.
Sure, it may not seem ‘fair’ since more went into it. This isn’t about fair. It’s about being accurate when it comes to predicting wins. Abandon all notions of fairness; it is not about fairness. It is about generating predictions, testable predictions, and seeing if they hold. It’s empirical. I suspect, Harold, that you’re hung up on some notion of fairness, knowing what you do about basketball and knowing that it took more than the end result marker to get the stop that you must somehow reassign credit on the basis of fairness. But you seem to disregard whether or not this matters in terms of the predictive power of the model.
Westy // January 7, 2008 at 4:12 pm
“This is not about fairness. It’s about how well the model holds to accounting for actual team records and towards predicting future results from past results. “
I agree, but it seems that while this has been argued fairly extensively at the team level, it hasn’t been shown at the individual level as completely (or successfully). I recall you comparing a value of 0.6 and 1.0 for rebounds and showing that 1.0 was slightly higher in predictive power. But really they’re pretty close. So if qualitatively, you decided that 0.6 better defined individual play, you’d have a model that was basically the same, but with a re-sorting of individual player values.
Really, I’d like to see a more thorough examination of what value for a DR is most predictive of team value the next year. A graph showing a correlation range for values for rebounds from 0.3 up to something like 1.1 would be useful.
Jason // January 7, 2008 at 4:59 pm
“Qualitatively?”
In two models that are close, the model that has the slightly higher predictive power is still the better model. Decisions based on some perception (I think you are searching for a different word that ‘qualitatively’) of what a rebound *should* be worth in the formula shouldn’t matter at all in evaluating the models when the aim of the model is to have predictive value, not to recapitulate preconceived notions.
antonio // January 7, 2008 at 5:10 pm
“In two models that are close, the model that has the slightly higher predictive power is still the better model.”
While it may have slightly higher predictive power at the team level, why does that mean it is more accurate at the individual level. Isn’t it possible that .6 for rebounds is more accurate on the individual level, but 1.0 is more accurate on the team level? I am not arguing saying the value should be .6, but merely that just because it is not as good at predicting team wins, does not mean it is not more accurate at ranking individual players
antonio // January 7, 2008 at 5:17 pm
Also, I don’t think Hollinger is really making the argument that the Knicks are a better team. I think talent and productivity get confused by people and many people have different definitions. I think Hollinger would argue that David Lee is not one of the most talented players on the Knicks,but certainly one of the most productive. He would say the exact opposite about Stephon Marbury. So regardless of what your definition of talent is, when reading someone else like Hollinger use the word, you have to think about what he means by it. I think he uses the word talent to mean pure offensive skill and athleticism. Maybe I am wrong, but that is my interpretation
Jason // January 7, 2008 at 5:24 pm
This all goes to what you mean by “at the individual level.” If the purpose is to have a model that can predict future “wins”, and “wins” at the team level are divied up into individual player wins. The value of a rebound at the team or individual level has to be the same, but what you seem to be suggesting isn’t about the value of a rebound, but who gets credit for the rebound. This is where consistency from year to year comes in. The model that provides predictive value for future performance is the better model, regardless of what people feel about how it ranks players.
I’m not sure what you mean by ‘more accurate in ranking individual players’ unless you mean more in line with a subjective evaluation. How are you going to evaluate how accurate it is? If it ranks them in some way such that it detracts from future predictive value, what value does it have over simply saying that X is better than Y on pure speculative assertion?
ty w // January 7, 2008 at 5:36 pm
Professor Berri:
I love the Wages of Wins and I love the work you and your colleagues have done.
I have one question, though. It is possible that your algorithm to a certain extent relies on the very market inefficiencies it seeks to correct?
Here’s what I mean. If NBA players were paid according to their PAWS, rather than their scoring totals, it seems to me that each of them would have an incentive to avoid taking any low percentage shot in any situation. Yet the NBA rules require teams to hoist a shot every 24 seconds or lose possession of the basketball… a penalty that your research proves is devastating to a team’s chance for victory.
I suppose if such a ludicrous situation ever arose, you could penalize each player 1/5th of a turnover… but do you agree with my premise?
ty w // January 7, 2008 at 6:10 pm
Oh by the way…
You should give some sort of “Wages of Wins” Award to the Charlotte Bobcats. Their coaching staff’s incompetence proves the validity of your “Win Score” thesis on a nightly basis.
According to 82games.com… and what I saw last night… they have slightly below average SF Gerald Wallace spending half his time playing as a well below average power forward, they have above average SG Jason Richardson spending half his time as a below average small forward, and they water down PF Emeka Okafor’s PAWS by playing him at the center position.
On top of that, they give an unbelievable amount of playing time to the “Human Turnover” PG Jeff McGinnis, whilst treating C Nazr Mohammed like a low minute scrub because he “only rebounds.” Last night he had a WS48 of +0.476, and got 22 minutes. Meanwhile, by my calculation, McGinnis had a WS48 of -0.387, and yet played 28 minutes.
Oh, and the Bobcats lost to a Redd-less Bucks team that started Bobby Simmons and Royal Ivey, and gave significant minutes to Charlie Bell.
Taking Moneyball Beyond Baseball « The Wages of Wins Journal // January 8, 2008 at 12:01 am
[...] The NBA Draft ← Two From the Sun on the Knicks [...]
antonio // January 8, 2008 at 12:27 am
Jason, my question is this. If you were to give the actual rebounder .6 credit for the rebound, and every other player on the floor .1, would this change the team amount of wins produced? And if not, than why are the values the way they are at the individual level? Does this come from a regression?
And don’t misinterpret this as me arguing for the rebounder to get .6 credit and everyone else on the floor .1. This is just a different option that I am stating. I am not saying anything is right or wrong.
Harold Almonte // January 8, 2008 at 5:55 am
Antonio. It’s not in the regressions, it’s in the boxscore. All the boxscore as a whole is a shortcut of what the game is at the player level ( a bad shortcut, since not every stats’s opposite action are boxscored). Since the possession value of every stat is not shredded and distributed to the players according to their game realities and contributions to the stat, all you’ll have is possession aproximations on every stat, but when you add all those aproximations, then you’ll have a deformed look of players and their worth. All of that is supposed to, and can be fixed in a team adjust, providing that you won’t forget anything crucial doing it of course.
Harold Almonte // January 8, 2008 at 6:09 am
Jason. About predictivity, I’ve said before that the metric can be some useful for GMs and probably scouts, even with its flaws, but a gambler wouldn’t need a player rating to find team predictivity. And a coach would have lot of problems with the approachs about the actions of the game at the player level. Pieces of rating would be more useful than an overall shortcut rating.
Jason // January 8, 2008 at 10:46 am
The issue with dividing up a rebound (in any form) vs. giving the whole credit to the guy who gets it is whether or not we get better predictions from one form or another. At the team level, the totals are the same, so it should not have influence on team wins, but it will change how players on the team are ranked relative to one another.
Whether or not this change in rankings is superior or not depends on how well past results predict future results. When I looked at it, dividing up rebounds in the .6vs4x.1 allocation was that this did not improve predictive ability and in fact, made things more variable. It was *not* more accurate at the level of predicting future results for the team based on the performance of players on the team.
It is possible that some other division of credit would be better (and entirely possible that the division of credit varies from team to team and with different player combinations). We could toss out various numbers, but there’s reason to believe that this isn’t necessary and isn’t likely to improve things given the limits of the data collection. There is a strong, strong, strong correlation between a player’s rebounds/min from season to season and this correlation is not diminished by changing teams and teammates at all. This suggests strongly that the biggest single influence on how a player rebounds is that player himself. The influence of teammates may be there–it probably is–but it’s just not variable enough to need to be accounted for in the metric.
Other players may contribute, but if the contributions beyond grabbing the rebound are rather consistent (e.g. there are very few superior boxer-outers) then giving the credit to the rebounder in whole. All players, for the most part, seem to receive the same benefit from teammates.
Harold Almonte // January 8, 2008 at 11:20 am
We honestly have limitations to refute that. That’s in the “Rosembaum Papers”’s level of demonstration, which by the way is not my statistical level. But we don’t need too much statistical to know that DR don’t produce FGMissed, FGMissed are not produced theirselves with or without closed defense, and to rebound with a crowded painted zone is not the same as doing it just against your man behind you, or alone.
Jason // January 8, 2008 at 12:57 pm
True, rebounds do not produce the miss, but that’s not the claim or the question. The question is whether or not the measured statistics capture what is going on *enough* to provide a reliable indicator of player value such that accurate predictions can be made. If the correlation is high enough and consistent enough, the on-court mechanisms don’t really matter. If rebounds stay consistent *somehow*, it doesn’t matter how the miss occurred if the player who got the rebound is going to get rebounds in subsequent contests in subsequent environments. Maybe players on teams with great rebounders are terrified that the rebounder will smash the headlights on their Escalade if they don’t hustle on defense and create missed shots. Maybe great rebounders intimidate the opposition into missing shots. If the defensive rebounds, a clear marker of the desired end result (a missed shot resulting in a change of possession without a score) are consistent, the mechanism isn’t important.
Understanding these mechanisms may be important where there is error in the measurement, but, as oft-attacked as rebounds are in the WP model, rebounds seem to be among the most stable statistic and thus, the most predictable and in the absolutely least need of monkeying with to adjust value.
Westy // January 8, 2008 at 1:22 pm
We could toss out various numbers, but there’s reason to believe that this isn’t necessary and isn’t likely to improve things given the limits of the data collection.
I guess I’m not following why this is the case. If there’s a correlation season to season for players’ rebounds valued at 1.0, doesn’t it mean that no matter what value we give them the fact that they’re consistent means the player should get credit at whatever level is best at predicting team wins?
Jason // January 8, 2008 at 2:58 pm
The value that is best for predicting team wins is the whole rebound. The question is not what the value of a rebound is. That was determined by regression (and, as it turns out, that value makes complete logical sense in the concept of what a possession is and what a defensive rebound undeniably means about the other team’s previous possession). A rebound is not discounted; there is no reason to change a rebound into something else when the value derived in terms of how a rebound affects win probability was derived empirically (as something like 0.034). The coefficient assigned to it might be different if the regression indicated as much, but the way the formula works, a rebound is a rebound is a rebound. There is no way to get the value of a rebound through regression and then make a rebound into something less than a rebound and have it work as well for predicting things at the team level. Since team stats are a sum of the individual player stats, the two must total as well.
The issue isn’t what a rebound is worth, unless you’re trying to make a formula that doesn’t actually look at win probability. The issue is whether or not someone other than the person who was credited with the rebound should get some partial credit for that rebound while deducting that same amount from the guy who got the rebound. e.g. for the rebound, you assign 6/10 of the rebound to the guy who got it and 1/10 each to the other 4 guys–it’s still the same whole rebound as far as the team is concerned and you’ve not changed the regression at all. If you share credit on the same value, team wins will be EXACTLY the same, but the relative player rankings on the team will change. It’s irrelevant if the team remains entirely the same.
However, if the guy who got the rebound was responsible for almost all of the rebound and his rebound rate would remain the same with other teammates if he changes teams, then discounting it will result in a less accurate predictor of the sum wins on the new team. If the other 4 guys who got partial credit were irrelevant, either they did nothing or any other new teammate would have done pretty much the same amount, then the discounted total assigned to the guy who got the rebound is not an accurate assessment of what he will do for a new squad. It can’t be if rebound rates remain the same. You’d underpredict the impact of a great rebounder and overpredict the impact of a guy who played alongside a great rebounder since the former would be expected to bring only a fraction of his “wins” with him while the latter had previously been credited with parts of rebounds that he will not actually be bringing with him.
Rebound correlation from year to year isn’t perfect and PERHAPS some discounted rate better reflects what will happen on a new team, but the high correlation suggests that the influence is small. I suspect it’s much more likely that the imperfections in the rebound rate correlation aren’t going to be fixed by discounting rebounds to the guy who grabbed them but it’s probably more a function of actual variation in player performance (either due to chance, effort, improvement or decline of skill, injury, etc.). Reassigning values won’t fix this.
I think that people are confusing Win Score with wins produced. Win Score gives a rebound a value of 1 and a point a value of 1 and a steal a value of 1. This works because the coefficients derived by regression for all of them are ridiculously close (which, again, makes sense based on what a possession is and what these signify in terms of possessions). In Wins Produced, the rebound is multiplied by this c0efficient. In Win Score, essentially, the coefficients for all these stats were changed to 1.
Westy // January 8, 2008 at 3:04 pm
Rebound correlation from year to year isn’t perfect and PERHAPS some discounted rate better reflects what will happen on a new team, but the high correlation suggests that the influence is small.
This is exactly what I’m suggesting is worth checking.
I was only talking about the individual rebound apportionement.
Jason // January 8, 2008 at 3:11 pm
Have at it! Feel free to try a range of values and see how it affects correlation of WP from year to year.
Harold Almonte // January 8, 2008 at 4:45 pm
Players’s rebounds will be consistent no matter the possession value you give to them, because it just depends of their position, and available FGMissed (also consistent from the Lg. ave.), and after all of that, then you put skill and height in the equation.
Even if you say that the credits stealed to shot defense include the best rebounders’s shot defenses too, and one thing cancels (or adds to) the other; when everybody talks about weighting the rebounds, is really talking about weghting the penalization of missing a defensive rebound against an offensive rebounder and viceversa, and if you do that by position, which players do you think are the most penalized? yes, deff. rebounders can recuperate part of this penalization offensive rebounding back, but the act of rebound wouldn’t be overrated, because you would need to discount between 20% to 30% of credits already rewarded, due to penalizations. Rebounds are a shortcut stat for all the defense on the FGMissed, but they need to be penalized one against the other, and that is what neither WS, nor WP did.
Then the ilusion of that WP overrates rebounders a lot is magnified by the fact that scorers are also (supposedlly) overpunished another 10% or 15%, by being stealed in credits from opp.FGMade in favor of not-scorers-teammates, given that the attemptor from that not rebounded possession is who deserves all that credit.
Jason // January 8, 2008 at 5:17 pm
While it’s true that position seems to influence rebounds considerably, there’s quite a bit of variation between the abilities of various players at each position. It doesn’t appear from the limited data I looked at that the number of available opportunities was a huge influencer either. Rebound rates were very consistent, even when players changed teams (slightly more consistent actually in this case, which I suspect is just noise based on a smaller sample). Team FG% and defensive FG% (a means of measuring available rebounds) was not nearly as regular. These varied far more than the rebound rates, suggesting that it’s the player ability and skill that is primarily driving the rebound rates, not anything external as the external forces varied you’re hypothesizing influence the rate varied far, far more than the rate.
Harold, you’ve asserted that ‘missed rebounds’ somehow need to be penalized, but you do not support this assertion nor suggest in any way how this is possible. ‘Missed rebounds’ are penalized in the sense of the position correction–players who perform below the average at their position are so penalized for it–and that if a player doesn’t get the rebound, someone else will lowering their relative productivity.
Harold Almonte // January 8, 2008 at 6:11 pm
Jason, that’s old stuff, they aren’t penalized, just not rewarded, and is not the same thing. It’s like you take off FGMissed out of equation and you say players who perform below average in total points at their position are penalized. Rebounding action is not rated, i.e. is not penalized.
The best way to do that better than using the 0.7/0.3 weighting, is including oppOR and oppDR in the team defense adjust. The problem with the team adjust is that stats must be distributed according to its correlated hierarchy of distribution: rebounding by position/attempts, and scoring by scoring option number/ attempts. Minutes is not a good shortcut for that.
The Franchise // January 8, 2008 at 8:25 pm
Harold-
Actually, they are the same thing.
And why don’t we pnalize the players that are willing to give the ball to bad shooters? Shouldn’t their WP be adjusted downward if players that shoot poorly keep getting shot attempts? (tic)
Jason // January 8, 2008 at 9:05 pm
I’m not sure how you define ’scoring option number’ but I am sure that you’re missing that a position correction already accounts for much (if not all) of the inequities of dividing defense.
Harold, I’m forced to conclude that you do not understand the model sufficiently to continue this discussion. You could (hypothetically) refute this, but your regular recurring critiques don’t lead me to believe that you’ve actually figured out how the model arrives at its conclusions sufficiently to make most of your points. WP is a box score method. You appear to be critiquing not the conclusions of the model based on how it works, but some personal problem with it based on how you believe a model (but clearly not the same model) ought to function, that it needs to do something else because, well, I’m not sure why. You aren’t clear enough in your critique beyond parroting someone else’s mention of an opportunity cost. You appear to be critiquing it on the grounds that the model doesn’t incorporate what is beyond the box score rather than asking if this information is sufficient and reasonably accurate in its conclusions. What you appear to be suggesting is something else. It’s like saying a steak doesn’t taste good because it wasn’t also the bottle of wine.
Can you demonstrate that in some way failing to incorporate a ‘missed rebound penalty’ makes the model fatally flawed such that it is coming to inaccurate conclusions with regularity? Hint: “It overvalues rebounds” is not such a demonstration. It’s an opinion.
antonio // January 8, 2008 at 9:20 pm
Jason, while I understand why dberri does not participate in APBR metrics, why do you not participate? You spend a LARGE amount of time defending WP and how it works on this website, yet over on the APBR metrics board there are people who have a great understanding of economics and of WP and criticize it much better than I or Harold Almonte or anybody who posts on this website has. And yes, they have posted on this website, but were ignored and sometimes blocked from posting. Yet on that website you offer no defense. Is there a reason for this?
dberri // January 8, 2008 at 9:32 pm
antonio,
Does the Internet not make it to this forum for the APBR people? I have never understood this request. Why do people have to go to that forum to have a discussion? We are having a discussion here. Everyone who behaves him or herself is invited to join.
Jason // January 8, 2008 at 9:44 pm
I read the APBR forum, but do not post, in large part because I already spend far, far too much time posting in various forums.
dberri // January 8, 2008 at 9:48 pm
Antonio,
Let me add… you said you understood why I didn’t post at APBR. I think that’s incorrect.
My reason is the same as Jason’s reason. There is only so much time in a day. It just doesn’t make much sense to wander around the Internet having the same discussion in different places.
antonio // January 8, 2008 at 10:19 pm
All I know is that the best arguments made against WP are at the apbr forum, and there is no response there….
Westy // January 8, 2008 at 10:23 pm
I understand antonio’s point. That is THE location on the internet to discuss basketball statistics. I am not aware of a basketball modeling approach that is not discussed there. For that reason, it only makes sense to at least have those who are most adept at defending this model participate.
And I agree, there’s only so much time. Thus, with articles and a (new) book to write, we tend to understand why you may not participate, Dave. I think antonio’s point was only that Jason is a wonderfully eloquent and thoughtful guy in regards to basketball stats discussion, and his insights would be very welcome on APBR. Maybe split your time?
That many of us float around over here as well is a testament to the interesting posts and discussion happening here as well.
Harold Almonte // January 9, 2008 at 6:04 am
First. This is as an important forum as anyone.
Jason. team’s scoring option chain is the player hierarchy whom the offensive plays are set for, and FGA are distributed. It deppends on scoring skills, not position (that’s the most my poor english can explain).
If you have follow all my post, you may conclude I have personal problems with all the metrics, given I critizice all of them. I’ll honestly confess, and I think Rosembaum realized this in one of his posts, that WP is my favorite of all of them. I can accept a lot of flaws any metric has not solved either (like scoring creation and shot defense), produced by taking shortcuts at the player level, but the lack of rebounding rating is imposible to accept, because is 20% of the game, more than 50% for some players’s game, and overrates at the player level. Every action and his stats have a counterpart, that be counterparting against itself like scoring (Made-Miss) , or against the opponent action (Points-Points allowed), (TO-STL), yes a lot of stats are not rated (AST,BLK, etc) maybe against fouls or TOs, but not to rate rebounds? against an boxscore existent counterpart rebounding stat? and not to attempt anything about that? And that’s nothing to do wether the possession value of a rebound is 1 or not, that’s irrelevant.
Jason // January 9, 2008 at 9:57 am
I am aware of the concept of ’scoring option.’ What I’m questioning is how useful it is in reality and how one determines who the ‘third option’ is. If you assign it simply by FGA and give someone credit for FGA, you are, in essence, rewarding someone for simply jacking up shots, regardless of outcome. Can you demonstrate that this has value? That simply taking shots brings benefit in some way towards aiding win probabilty, or displays value in some way such that that value can be moved elsewhere and bring similar returns in terms of aiding win probability? If you cannot, then the concept is meaningless for predicting team success. You may use it to use numbers to aid in what are objective but rather irrelevant player rankings, but these rankings then don’t have value when determining what a player actually does for his team. It starts becoming a beauty contest.
Again, Harold, you say that rebounds ‘overrates’ players. How exactly is this? How do you arrive at the ‘actual’ rating to show that someone is over or under-rated, or how do you demonstrate that the metric is not accurate?
antonio // January 9, 2008 at 11:30 pm
Maybe Harold is simply saying that he doesn’t have an answer for what to do, but simply that it is flawed. I think Harold, and maybe I am wrong, just doesn’t believe right now we are capable of accurately a players productivity based only on box-score statistics. With the limitations of box-score statistics, while you may get a good estimate of player rankings, you simply can not say player A is better than Player B because his X rating is higher. Maybe Harold doesn’t believe this, but I do. While I think most rankings can help give a general idea of how good a player is, I don’t think any single statistic yet can accurately claim to know exactly how good or productive a player is. Basketball simply is not a box-score game, unlike baseball. There is way too much going on that has yet to be captured by the box-score. Is this a possibility? Do you think this is not at all the case?