Jeremy Britton, an SF Bay Area interaction designer at ZURB, fell in love with NBA basketball as a kid while watching Eric “Sleepy” Floyd drop 28 fourth quarter points on the Lakers in front of a bank of TVs at a Bay Area Price Club with dozens of Warrior faithful. The Warriors took that thrilling game, but dropped the series 4-1. From Don Nelson (part 1), to Run TMC, Chris Webber, Latrell Sprewell attacking PJ Carlesimo with a 2×4, and their amazing “We Believe” run three years ago, Jeremy has inexplicably remained a part of the Warrior faithful but rarely been rewarded.
I wrote Warriors Stumble On 50 Wins? before the season started, confidently stating this year’s edition was within striking distance of 50 wins (OK, I was even more rosy with my prediction). The naysayers came out of the woodwork, quieted down briefly during the Warriors’ 6-2 start, and now are back in full force. Today the Warriors have a losing 22-28 record and dim playoff hopes. So I’m like Linus waiting for the Great Pumpkin.
To put it lightly, I’m taking some heat for that little prediction. From my perspective this Warrior season really sucks; not because it didn’t turn out as I wanted–it rarely does–but for at least three distinct, more painful reasons:
- Low expectations appear justified (‘the Warriors are just losers’)
- The losing appears to validate prior assumptions, especially for David Lee’s naysayers
- It seems to create a red herring suggesting Wins Produced doesn’t work
This blog, sports economics in general, and behavioral economics at large, have established that savvy data analysis can tell us things our lying eyes can’t. Some people just refuse to believe these crazy insights because it runs contrary to the conventional wisdom they’re comfortable with. Often they don’t understand the analysis either.
Baseball management famously showed how judicious use of savvy data analysis can be exploited for tremendous advantage–most notably Billy Beane in Oakland. We have yet to see that applied in practice to professional basketball. Managers and fans prefer to bury their heads in the sand and rely on widely shared truisms (which by its nature cannot be exploited for advantage because everybody believes it).
Before we look at this year’s Warriors I want to reiterate two vital insights Wins Produced teaches us:
- Basketball is a game of possessions (just like baseball is a game of outs), yet people consistently ignore this. Players help you win games note just by scoring points, but by gaining possessions and not losing them.
- Individual scoring totals are vastly overrated. It matters how a player scores their points (i.e. how efficiently they use each possession to do so). This is an easy mistake to make because after all, how do teams win games? By scoring the most points. However, this is not true for individual player contributions.
With that rant behind me, let’s see what the data can tell us about this year’s Warrior team. I’m going to keep this simple by comparing what these players did last year–which is a terrific predictor for what they should do this year–against what they’ve actually done this year.
Cool, we see based on last year the Warriors should have about 30.2 wins, which would project out to about 49.5 wins on the season. Not quite 50, but very close. We even see three players doing better than last year–Monta Ellis, Stephen Curry and Vladimir Radmanovic–to the tune of 5.0 additional wins. That’s awesome.
So what went wrong? It’s clear this squad is only producing 20.3 wins (their actual total is 22 wins). A few disasters that explain this are obvious from the list:
- David Lee is playing the worst basketball of his pro career (5.2 fewer wins). His shooting percentage and defensive rebounding are at career lows. According to Hoopdata.com he is taking far fewer shots at the rim and his jump-shot has fallen apart. Why? Likely injury. Lee suffered a freak accident in New York getting Wilson Chandler’s tooth lodged in his elbow. He was hospitalized for days and continues to wear a large bandage.
- Dorell Wright is also having the worst season of his pro career (2.5 fewer wins). After signing a $12 million contract proclaiming the Warriors would be “a great situation for me to showcase my talent and show everybody what else I bring to the table other than being a spot-up shooter and defender,” Wright has wobbled. He’s continued to improve his three-point shooting, but his mid-range shooting has been terrible (44 of 162 from ten to 23 feet), his rebounding is at a career low, and his turnover rate ties a career high.
- Reggie Williams hasn’t matched flashes we saw late last season (2.0 fewer wins). We only got 779 minutes of Williams in 2009-10, but his play was exciting and very productive for a rookie (.182 WP48 or Wins Produced per 48 minutes). His shooting is different–fewer shots at the rim, more threes–but the difference is a wash. It’s Williams lack of hustle on defense (half the steals) and rebounding (only 76% the defensive boards) that have knocked him down a couple of pegs to a below average producer.
Of the 9.9 expected wins that we have yet to see, 9.8 can be traced to the play of Lee, Wright, and Williams.
Does the Warriors’ losing prove David Lee really wasn’t any good? Not at all. We haven’t even seen the real David Lee (except for flashes during the first eight games before he got hurt). As his elbow heals we should expect to see his shooting and rebounding bounce back.
Dorell Wright is scoring more points in total–no more than we should’ve expected when doubling his minutes on the floor–but is doing so less efficiently. Fans of Wright would point out what a great three point shooter he’s become–which is true–but overlook his less disciplined poor shooting inside the arc (27% on jumpers outside the key).
Reggie Williams may not have been that good after all (remember, his entire NBA career consisted of fewer than 1,000 minutes before this season), or he just needs coaching to remind him to hustle on defense and on the glass while attacking the rim for high percentage shots. He’s a wild card we’ll have to wait and see about.
All in all, the injury to Lee has been the biggest culprit holding the Warriors back this season, with undisciplined play by Wright and Williams adding insult to that injury. As Lee gets healthy, we should expect a late season surge reminiscent of a 50-win team (okay, how about a 40-win team?), but it may be too late to secure a playoff spot this year.
– Jeremy Britton
EntityAbyss
February 8, 2011
Well, to the naysayers’ defense, David Lee’s elbow has completely healed. He doesn’t wear the thing around it anymore. Also, Lousi Williams has had backs spasms and missed plenty of time. He should be playing a lot better. Either way though, It’s an 82 game season, not 50, so who knows what’ll happen.
Moving on though, I hear Carmelo Anthony is going to the lakers for andrew bynum. this is not confirmed, but if it happens, it’ll be the biggest test for WP. If it happens, maybe you could do a post on what you predict will happen the rest of the way.
Devin
February 8, 2011
Good stuff.
Jeremy Britton
February 8, 2011
@EntityAbyss, I wouldn’t be so quick to say 1) that his elbow is “completely healed” or 2) that David Lee is back to 100% physical form. Competing at the NBA level requires peak physical performance. That’s difficult to sustain and very easy to fall off the wagon from. As you say, it’s an 82 game season. We’ll see how this plays out as Lee rounds into shape.
Power Overwhelming
February 8, 2011
Just to make it clear, you predicted a 57-win team and now you are unsure if the team will win 40 games. Further, even using last season’s numbers, the team does not project to win 50 games. That is quite a disparity.
On to your post. You seem quite content repeating the usual straw men. It is getting rather old, and you come off as very condescending when you lecture about fundamental concepts hardcore basketball fans have understood for years. Wow, there is more to the game than scoring points? What insight! Basketball is a game of possessions? This novel ideas shall surely revolutionzie our understanding of basketball.
It is really easy to repeat the now-conventional wisdom of WoW instead of addressing criticisms such as: i) Lee is a poor defensive player; ii) Lee skips boxing out and playing defense to grab rebounds iii) Lee’s teams do not rebound better when he is on the court iv) etc. Your critics clearly recognize that there is more to the game than scoring (defense!), and they recognize the value of possessions, so address address their arguments fairly. Otherwise, you waste your time writing this, my time reading this, and their time arguing in the first place.
That said, I agree that Lee is a good player who has not played well this season, and a lot of that is due to the injury. Prior to the season, I believed an injury-free Warriors were capable of 40 – 45 wins, flirting with 50 if everything went perfectly. I was not one of your critics in the previous thread. I am just irritated at your mindless recitation of the Wins Produced gospel (to be fair, you are not alone). There is absolutely no critical thinking going on in your post. It is just the usual straw man after straw man, backed up by the usual murky statistics.
Nerd Numbers
February 8, 2011
Jeremy,
Awesome article! Keep waiting for the Great Pumpkin. Golden State still looks good going forward and here’s hoping Monta Ellis can pull an Elton Brand and return to his star form from a few years ago. Worst case a medium draft pick may help you out in the offseason.
p.s.
My plate is full but if anyone has time I would curious to see the ratio of stats cited in a comment to the number of insults. It would also be interesting to check stats vs. number of “obvious facts” that “everyone knows” as well.
Italian Stallion
February 8, 2011
Aside from Lee’s injury, I think GS is suffering from some level of diminishing returns on the boards. Lee, Biedrins, Williams, and Dorell Wright all have fewer rebounds per 36 than last year. That makes a lot of sense given that much of Lee’s value in NY came from rebounding and he was moving to a team where the same was true of heavy rebounding Bierdrins.
Lee’s TS% was atrocious after the injury, but it has been climbing steadily over the last few weeks. It wouldn’t surprise me if it continues to rise and gets at least reasonably close to his average in recent years.
But IMO several of the GS’s players are going to continue rebounding below their best levels. There are only so many possessions and missed shots to rebound.
Italian Stallion
February 8, 2011
Power,
“iii) Lee’s teams do not rebound better when he is on the court”
I’m not sure how trustworthy that stat is. I suspect it’s at least partially related to the substitution patterns and the rebounding strength coming off the bench.
If the inferior players on the 2nd unit just happen to be equal or better rebounders than the players on the first unit, it would tend to make the starters look worse than they really are based on the on/off court numbers.
Few would argue with the idea that Dwight Howard is one of the best rebounders in the NBA, but Orlando is a slightly better rebounding team with him off the court this year and was more or less equal last year.
http://www.82games.com/1011/10ORL18.HTM
EntityAbyss
February 8, 2011
Also, Italian Stallion, GS has the worst defensive rebound % in the league. The rebounds are there and available. They’re just not getting them. The knicks last year were a better rebounding team then the warriors are this year. The rebounds are there, they’re just not getting them.
Italian Stallion
February 8, 2011
Entity,
I wonder how much of that has to with Curry/Ellis being a very small back court?
If the rebounds are there for the bigs and they aren’t getting them, then something besides diminishing returns is going on.
dm
February 9, 2011
I want to respond but his power……its….just too……overwhelming…
ohreally
February 9, 2011
David Lee is a non-defender and a fake rebounder. It was all there in the data when he was in NY if you knew how to read it. And it’s there in the data in GSW. He’s perfectly healthy now and doing his thing. His thing just isn’t very good.
This is a textbook case in how WP mis-values players.
ohreally
February 9, 2011
By the way, the “flashes of David Lee” in the first 8 games were against an easy schedule.
ohreally
February 9, 2011
By the way, the elephant in the room on Lee is his minutes. He’s healthy now but his minutes have trended down over the last couple of weeks from earlier in the season. And more importantly, they are lower lately than he averaged last season.
What appears to be going on is that Coach Smart is running Vlad a little longer, and Lee a little less, in an effort to get more defense out of his squad. If this continues, lack of minutes will be a factor in Lee failing to accumulate the WP totals expected of him.
Italian Stallion
February 9, 2011
>What appears to be going on is that Coach Smart is running Vlad a little longer, and Lee a little less, in an effort to get more defense out of his squad. <
Yea, that's going to work. They are 1-8 without Lee.
entityabyss
February 9, 2011
Why would the coach think that vlad is a better defender than lee?
Anyways, even you believe that Lee plays no defense and waits for the rebound, the opportunities are still. GS is a worse defensive rebounding team then new york was last year. Even if Lee’s defense is bad, there’s still plenty of boards to go around.
NeutralCorner
February 9, 2011
It seems to me that there is some sort of rivalry/hostility between the WP crowd and the 82games crowd.
It seems to me that regression analysis is an obviously powerful tool and can produce a quality framework for analysis if we chose the right data to regress.
Has anyone ever tinkered with replacing the team defense data points used in the current WP formula with the individual player’s on/off differentials?
It would seem to me that players who heavily impact the game in areas that the current boxscore stats don’t account for (fake rebounders and heavily substandard individual defenders) simply don’t have a descriptive dataset (in the current boxscore composition) to use in a regression analysis. If the goal is to refine the system, could the inclusion of on/off differentials get us closer to a unifying model?
Or would that be like crossing the streams and reverse the polarity of the known universe?
marparker
February 9, 2011
This case doesn’t prove anything about rebounding. If everyone is rebounding at a lower percentage and the team is rebounding at a worst percentage what exactly does that prove? According to the law of diminishing returns output(i.e team rebounds) should increase while individual productivity(i.e individual player wins produced) should decrease. There is no law to support adding something valuable decreases overall value. Its not far fetched to conclude injury and other outside factors have taken effect on the warriors numbers this year.
The rebounding arguement is getting very old. For such “smart” people how hard is it to understand that rebounding tells so much more about the player than any other box score stat. The best rebounders usually have the best combination of size, speed, quickness, and I.Q. So, the act of grabbing a rebound is not worth nearly as much as is credited by wins produced. However, everything else that you can gather from assumptions you can make about good rebounders is very valuable. What we are left with is outliers. Anyone who is willing to make a derivative number based on wins produced can quickly find out who wins produced is lying about.
The Warriors employ 4 offensive specialists. That is 4 guys who are pretty much useless when the other team has the ball. That is why they suck. My worthless to anyone but me guess is that this team would be able to approach 50 wins produced is they swapped one offensive specialist for one defensive specialist.
The Jazz are having a similar problem. They let Boozer go and replaced him well enough on offense but not defensively. Now everyone’s overall wins produced is suffering but the capability to put up their old numbers is there. They just need a little more defensive mindedness on their roster.
todd2
February 9, 2011
GSW’s rebound% and defensive fg% have both improved over last year. Opponents have almost 400 more fta’s and 100 assists to date. Looks like people are getting beat and/or not helping defensively.
bduran
February 9, 2011
Anyone who wants to use David Lee’s current year as a knock on WP is way off base. I think pretty much any stat/observer/whatever would say that Lee is playing worse this year. He’s rebounding less on a worse rebounding team then last year (so it’s not likely because of diminishing returns). He’s TS% is the lowest of his career and his TO% is the highest and years. So, even if you feel WP over rated him last year, you have to admit that he’s playing worse ball this year.
I believe this is his worst year by PER/Win Shares/ and WP48. It’s his best +- year though, must be his D.
ElGee
February 9, 2011
Hi Jeremy,
I don’t understand the reluctance to admit the obvious flaws with WP. (I won’t rehash them as they’ve been spelled out endlessly, but the treatment of scoring and rebounding is the heart of the matter.) Do you not understand why a regression on a box score could produce an undesirable correlative result? Do you not see the problems with the marginal values (eg how rebounders will almost always trump volume-scorers)?
From a scientific perspective, I just don’t get the lack of self-criticism…slamming Dorrell Wright’s career year and blaming David Lee’s injury doesn’t make much sense.
MKSE
February 9, 2011
Getting one of the Wins Produced fans to admit even the possibility they might have been wrong, let alone spectacularly wrong, is like squeezing water from a stone.
bduran
February 9, 2011
“Dorrell Wright’s career year”
?? He’s scoring a lot more points, but that’s mostly due to more minutes. He’s scoring 2.9 more points per 36, but he’s doing it with a lower TS% and higher turnover rate. Also, he rebounded a bit better last year on a much better rebounding team. So, I hope this is not a career year for him. I think he can play better.
thedubfan
February 9, 2011
Jeremy, I watch or listen to every Dubs game. You are way off base here. I believe wholeheartedly in statistical analysis, but even statistical analysis has its flaws. And in the W’s case, it is glaringly obvious. This is a bad basketball team with or without a healthy David Lee. They can play defense in spurts…but those spurts last for 7 minutes and they occur twice a month.
Flawed roster that needs to be overhauled from the front office through the coaching staff and actual players. The Warriors need to:
1. Improve drafting ability
2. Improve interior play (both ends of the court)
3. Break up Curry/Monta (very good players individually; should not be playing together)
thedubfan
ElGee
February 9, 2011
@bduran – You’re overlooking usage. He’s almost DOUBLED his career point total from his first six years in the league. And his drop in efficiency is small – 55.3% TS%, down a whopping 1.4% while doubling his minutes (and nicely above league average). “Higher turnover rate” – you mean below his career average and up 0.6% from last year?
Yes, it’s easily a career year.
Italian Stallion
February 9, 2011
>>GSW’s rebound% and defensive fg% have both improved over last year.<<
That was going to be my next question
How good a rebounding team they are now is somewhat irrelevant. The issue is whether the addition of Lee made them better, worse, or the same.
It's my contention that it almost had to make them better, but probably not
by as much as the simple addition of his rebounds vs. whoever he replaced.
On the Knicks Lee played C with almost no serious rebounding help from guys like Harrington , Jeffries, Gallo, Chandler etc… So he was bound to get an extra board or two here or there.
With GS, Biedrins is good rebounder when he is on the court.
There almost has to be some minor level of diminishing returns (and the evidence of the individual players supports that). The rest could easily be attributed to injury, offensive sets etc…
Italian Stallion
February 9, 2011
I think the view that GS has a flawed team makeup is so obvious I don’t think it even has to be discussed. I think we can sort of assume that they don’t play enough defense and the combination of Ellis/Curry is not an ideal back court.
The important issue is trying to figure out why several players are performing below expectations. I think at least partially, Lee being hurt is one very obvious indisputable factor. The rest of it is debates about diminishing returns on the boards etc…
ohreally
February 9, 2011
“Yea, that’s going to work. They are 1-8 without Lee.”
Check out who the opponents were in those 9 games. And check out who they were in the 6-2 start with Lee.
“Why would the coach think that vlad is a better defender than lee?”
Because he is? It’s pretty clear just from watching the games, but it also shows up in the numbers. It’s especially glaring when the Warriors go zone. Vlad is a pretty good zone defender and Lee is pretty darn awful.
James
February 9, 2011
“I don’t understand the reluctance to admit the obvious flaws with WP. ”
You have to understand that in the economics world there will always be a large number of professors that don’t have the strong analytical skills that you’d see uniformly in a math department. It’s like Berri simply doesn’t have the raw mental firepower to analyze his own work effectively or understand the mechanisms behind it.
Mike
February 9, 2011
I love how people state things with no data eg:
PROVE IT. Show me numbers, demonstrate your position is true with numbers. Is that really so hard?
ElGee
February 9, 2011
@James – I am beginning to see this. ;)
Perhaps I should rephrase then – why the resistance to criticism from WP proponents? You can’t convince me people educated enough to understand regression can’t also see the problems with the marginal values it produces, short of plain old stubbornness. (And to that I say, it’s an anonymous internet post — what’s the big deal?)
ohreally
February 9, 2011
Mike:
I really doubt you need me to show you the numbers. I bet you can do it on your own. Just check out the team rebounding percentages with Lee on vs Lee off at NY the last couple of seasons and at GSW this season. And check out the defensive impact on those teams with him on vs off.
ohreally
February 9, 2011
So Jeremy, if you do a websearch for “FM48 floppymoose” you will find a new boxscore player metric I invented which correlates with wins even better than WP48. It only looks at points and fg% and throws everything else out. But the correlation is better so it must be a better metric, right? I look forward to your next article where you use FM48 instead of WP48 to predict something interesting about GSW.
reservoirgod
February 9, 2011
These Warriors threads are always comical… I don’t think any NBA coach would go on record & say Vlad is a better defender than ANYONE in the league. He may be more familiar w/ the GSW defensive sets than David Lee but that’s the only thing that separates them – experience & familiarity.
Plenty has been said about Lee’s injury but I still think people are understating the severity of what happened. At one point he was playing the game w/ one arm! He stopped using his right arm completely! The doctors were so worried about his infection they thought they might have to remove half his triceps! I think anything they get from Lee this season is gravy, they’ll have to wait until next season to get a double-double season from him.
GSW also has a similar problem the Knicks have – no depth. Their bench is absolutely terrible, so when the starters get injured like Lee, Curry & Biedrins did this season, they’re going to get killed. The only difference between NYK & GSW is that the Knicks have been relatively healthy.
marparker
February 9, 2011
The floppymoose thread was an interesting read. OhReally, I wish you would present evidence in that manner more often.
I like wp48. I believe it ranks player correctly as long as you put each player through a smell test. A distinction needs to be made between specialists and well rounded players.
mystic
February 10, 2011
marparker,
well, WP makes already a distinction between players. It needs a positional adjustment! Why? Which other boxscore metric is doing a positional adjustment? Now you need an additional adjustment for specialist and well rounded players? How do you want to determine “well rounded”? The attributes you are choosing are related to the skills of a player, those are not necessarily showing up in the boxscore. You would need coaches and scouts to determine those correctly, otherwise your metric will not be accurate again.
The positional adjustment alone makes WP48 less suitable than other metrics which are not in need of such a thing. Determining the correct position for players is a tough job, especially when you have players who are playing differently on both ends of the floor. Can you really adjust the offensive output of a Stretch-PF by using numbers for center/forwards? They are playing more like a traditional SF on offense, but as a PF on defense. How about perimeter players which like to operate in the post? Can you adjust Magic Johnson in the same fashion as Steve Nash?
Saying that a team just needs an average center (like in the 76ers post) to get to 50 wins and then going on with the note that this is a player with a 0.100 WP48 is completely ignoring context. Will Shelden Williams, the guy is at 0.096, really be enough for this?
And before someone claims the teams would be better overall when they are using a minutes distribution according to WP48 I want to see a proof that WP48 is doing a better job at predicting the outcome of specific lineups than any other metric out here. Looking at the overall team wins and comparing them with the real wins isn’t the test here. Because it just shows that scoring margin is explaining wins very well. We know that already, we don’t need WP48 for this.
ohreally
February 10, 2011
“The floppymoose thread was an interesting read. OhReally, I wish you would present evidence in that manner more often. ”
Thanks. To be clear, I don’t think FM48 is a better metric than WP48. In fact, I think it’s worse. But it correlates with wins very well, so I’m using it to illustrate how getting boxscore data plus a team defensive adjustment to correlate with wins doesn’t really mean you have come up with a good player metric.
Anything that sums up, across the teams players, to basically the scoring differential is going to correlate well, because scoring differential correlates well.
And there are metrics that have that property but that clearly aren’t really assigning the credit properly between players. FM48 is an example of that. I claim that WP48 is another. I’m happy to have a discussion about it – I just don’t want it to be “but WP correlates with wins so well!” because that proves *nothing* about it’s value as a player metric.
Nerd Numbers
February 10, 2011
Oh really,
https://dberri.wordpress.com/2007/11/18/a-guide-to-evaluating-models/ (Thanks to Mosi for reminding me of this)
To clarify why those of us who use the Wins Produced Metric like it
1) It’s based on a good theory. Possessions and Points = Wins.
2) It correlates well with Wins.
3) It is predictable (a players AdjP48 from prior years explains about 80% of future years)
Oh really I think you left out 1 and 3 floppy moose post. Something important any stats person should know is that if you have all the data and wish to make a model that correlates well with all of the data you can. A ton’s been written about this and I’ll forgive you for not knowing it all (I forgot about the post I cited until Mosi reminded me) but you haven’t proven anything except that you’ve only read half the story.
Nerd Numbers
February 10, 2011
On more important news. Golden State just beat Denver after a back to back. Clearly Golden State has to be taken seriously because Denver is a contender right? (Please don’t shatter my hopes as a Denver fan!)
Italian Stallion
February 10, 2011
ohreally,
>Check out who the opponents were in those 9 games. And check out who they were in the 6-2 start with Lee.<
I look at point differential adjusted for schedule home/away and record. IMO, they are better with Lee.
marparker
February 10, 2011
Mystic,
Determinging well rounded players is done by dividing wp48 into defensive contribution and offensive contribution. Its not subjective and doesn’t need any scouts. I try to keep it simple and I don’t expect my answer to a problem to be accepted by anyone else but me.
I also hate that a position adjustment is necessary and don’t use it in my own work. However, the position adjustment sheds light on the huge advantage a guard/small forward who can produce close to what big men produce gives a team. There’s seems to be some tipping point(sorry Gladwell) on a production curve where a guy becomes good enough to make any team good. That bar is much lower for the smallest 3 positions. The position adjustment works from the regression because of this handful of players.(My unverified opinion)
Italian Stallion
February 10, 2011
>I really doubt you need me to show you the numbers. I bet you can do it on your own. Just check out the team rebounding percentages with Lee on vs Lee off at NY the last couple of seasons and at GSW this season. And check out the defensive impact on those teams with him on vs off.<
I already demonstrated the problem with this.
On and off the court doesn't adjust for the fact that several starters often come on and off at the same time . So if the bench subs happen to be better rebounders as a group than the starters, it would reflect badly on Lee as an individual.
As of just a few days ago the MAGIC were a better rebounding team with Howard off the court. They were neutral for all of last year. That is obviously not meaningful since most would disagree with the notion that Howard is among best rebounders in the NBA, if not the best .
There's little doubt that Lee is pretty good rebounder. The issue is how much value to assign each rebound. On NY he was often combined with guys like Gallo, Chandler, Jeffries etc… None of those guys are great rebounders. So Lee was important to the Knicks, but probably got more there than he would on an average team. In GS he's with Biedrins who is also a good rebounder. So he and Biedrins figure to get fewer than previously when they are together (which has happened) and can cover for each other to some extent when one is off.
Italian Stallion
February 10, 2011
ElGee,
>Perhaps I should rephrase then – why the resistance to criticism from WP proponents?<
I accept and point out potential problems all the time.
The problem is that some people come here and exclusively focus on the potential problems with WP and use equally flawed stats to make the case.
That kind of thing causes people to dig in and take sides.
Then we lose any chance for intellectual honesty because everyone is defending their turf instead of trying to get at the truth.
mystic
February 10, 2011
Nerd Numbers,
it starts with picking the right foundation for the model. And the formula in this model for PE and PA are not working for individual players. Check out Dean Oliver’s “Basketball On Paper” and compare the formulas which are used for individual players and the teams. They differ a lot.
A regression on that basis will give you a good estimation for the team’s total wins, but that doesn’t tell a bit more about the players as individuals. Using the team defensive adjustment is the key to the high correlation coefficient, but using that adjustment doesn’t improve the model in terms of individual player evaluation a tiny bit. As Berri noted himself the adjustment doesn’t change the ranking of the players much. Which is understandable, because it basically assume everyone in the league has only a small similar impact on defense. That will not change the ranking of the players much, but that also doesn’t help to evaluate individual player impact on defense.
marparker,
you assume that WP is doing a good job at evaluating both of those things, but are you sure that this is correct? ;)
bduran
February 10, 2011
ElGee,
” You’re overlooking usage. He’s almost DOUBLED his career point total from his first six years in the league.”
Hmm, you say I’m over looking usage, and then make a statement about how much his scoring increseased. It looks like you’re saying that he’s gotten so many more points because of his usage rate. However, he’s increased his career minutes by 49% this year and his career points by 64%. So, looks like most of the increase is primarily due to minutes. Maybe by usage you meant court time?
Yes his usage rate has increased, and yes the drop in his TS% isn’t huge. His TO% increase is also small. Still, it adds up. According to basketball reference his offensive rating is down and his offensive wins shares per 48 is down slightly. So it looks like he’s performing slightly worse on offense, but he’s getting many more minutes. So in that sense it’s a career year on offense I guess.
mystic
February 10, 2011
Italian Stallion,
the real problem comes up, when someone declares his model as perfect and thinks he can give out suggestions for lineups or minutes distributions WITHOUT ANY proof that his model really helps to determine the best possible lineup.
I’m in no way defending my stuff here, because I’m a boxscore stats guy, my model is not based on +/- numbers.
Italian Stallion
February 10, 2011
reservoirgod,
“Plenty has been said about Lee’s injury but I still think people are understating the severity of what happened. At one point he was playing the game w/ one arm! He stopped using his right arm completely!”
You are definitely correct.
I have been tracking Lee and GS on an almost daily basis.
He was playing WELL below par for a period of time after the injury (like ridiculously bad from the field) and has been improving.
Power Overwhelming
February 10, 2011
marparker – “I like wp48. I believe it ranks player correctly as long as you put each player through a smell test.”
Fair enough, but the story dberri tells and the story this specific post tells is that “savvy data analysis can tell us things our lying eyes can’t. Some people just refuse to believe these crazy insights because it runs contrary to the conventional wisdom they’re comfortable with. Often they don’t understand the analysis either.”
A “smell test” or a “laugh test” is routinely mocked in these parts. The point of this model is to overthrow conventional wisdom and ignore what our “lying eyes” (or other metrics) tell us. So if the model says Marcus Camby is better (more productive) than Dwight Howard, it is correct.
The prevailing message, both in WoW, Stumbling on Wins, and this blog, is that NBA decision makers do not understand the game of basketball. Both in his books and on this blog, dberri and his follower’s analysis of basketball starts and ends with “the numbers”, drinking kool-aid from the holy grail of basketball statistics that is Wins Produced.
So if you want to apply a smell test to teach player individually, go ahead. But why not just skip Wins Produced entirely? I hear there is this new model called FM48. It uses regression analysis and a positional adjustment like Wp48. The only difference? It explains 97% of wins while Wp48 only explains 93 – 95% of wins.
ElGee
February 10, 2011
@bduran – yes, usage as in total load (not rate).
@Italian Stallion – I hear what you’re saying. I just see a lot of resistance to the criticism. Personally, I never use WP because I think it has major flaws that don’t tell us a lot about individual players. Namely, how it treats scoring and rebounding. It’s the same reason I don’t really use PER (treatment of volume scoring).
So in that sense, it’s not really about “sides” or being overly critical, it’s just looking at the model and saying why it doesn’t provide insight. If you can tell me where it does provide useful information beyond a non-regressed box score analysis (box score is inherently limited anyway), I’ll gladly listen…(that’s actually how I “stumbled” into the blog ;) )
mystic
February 10, 2011
Power Overwhelming,
there is no positional adjustment needed for FM48, it works without. FM48 for 2010 explains around 85% of FM48 in 2011. Check out the thread on realgm.
The underlying theory is: Scoring per 100 Possessions + Defense = Wins.
Sounds like a good theory to me.
The model is much easier to apply, has a higher correlation to winning, explains the future years at least as good as WP48. Well, let Occam’s razor decide …
bduran
February 10, 2011
ElGee,
I guess what confused me was you writing this,
“You’re overlooking usage.”
after I wrote this,
“He’s scoring a lot more points, but that’s mostly due to more minutes.”
ohreally
February 10, 2011
“I already demonstrated the problem with this.
On and off the court doesn’t adjust for the fact that several starters often come on and off at the same time . So if the bench subs happen to be better rebounders as a group than the starters, it would reflect badly on Lee as an individual. ”
I accept that on-off stats aren’t perfect. But where does that get us? They are better than nothing. To me, they shift the burden of proof. Just about every premier rebounder in the league makes his team rebound better when he is on the floor, and I’m supposed to believe that it’s just a twist of bad fate that Lee doesn’t for three years running? I need more from you than that if you want me to change my mind.
ElGee
February 10, 2011
@bduran – I can see how that would be confusing. ;) I meant total usage (not usage rate). When you said “mostly due to more minutes” it sounds like you’re overlooking how ramping up minutes (ie raw usage) isn’t an automatic process. Comparing his per36 numbers — up anyway from last year — to say, 2006, when played 132 minutes, would be “ignoring usage.” Maybe I should have said “load?” :/
Apologies if that was unclear.
ohreally
February 10, 2011
“I look at point differential adjusted for schedule home/away and record. IMO, they are better with Lee.”
Hey, I agree. They are better with Lee. Just not *nearly* as better as WP suggests.
Italian Stallion
February 10, 2011
ohreally,
“I accept that on-off stats aren’t perfect. But where does that get us? ”
Not very far. :-)
My own view is that the game is so complex it can’t be reduced to single number that’s going to both capture the values properly and translate well to other situations.
I think you have to look at what the player does well/average/poorly, the circumstances he’s in, and the circumstances he might move to etc… and understand that the values change.
When it comes to Lee, I think at this point in his career he’s an above average rebounder, mildly above average usage/high efficiency scorer (when healthy) that can makes plays for others if he gets touches. He’s a below average defender, but he’s not as bad as he looked in NY when he defended Cs and had absolutely no interior help.
I think he’s in a situation in GS that doesn’t maximize his ability because they have 2 high usage scorers in the back court (neither of which is a true PG that can maximize Lee’s ability to finish on the PnR yet) and because Biedrins is a very good rebounder.
Summed up, he’s a very good player, but a notch below All Star.
bduran
February 10, 2011
ElGee,
Now that makes sense. Yeah I either looked at last year or career average weighted by season minutes when doing comparisons.
ohreally
February 10, 2011
“My own view is that the game is so complex it can’t be reduced to single number that’s going to both capture the values properly and translate well to other situations.
I think you have to look at what the player does well/average/poorly, the circumstances he’s in, and the circumstances he might move to etc… and understand that the values change. ”
I agree with all of that. But that doesn’t mean that I can’t draw some conclusions from what data I have. Given what I have, and allowing for the imperfections in the data, I conclude Lee is not as good a rebounder as his raw rebound totals in the box score would suggest.
And everything that I can find is bearing me out on that. The on/off numbers do. The inability of WP to correctly predict the Warriors improvement – which I called *in advance* – does. My “lying eyes” do. From where I sit, everything points to the same conclusion. Lee isn’t as good as WP thought.
cladden
February 11, 2011
As to the Dorrell Wright discussion I think it’s inherently flawed to think that it’s as easy to put up the same FG% and Points/min on higher minutes and as a starter as it is as a low-minute reserve.
Italian Stallion
February 11, 2011
ohreally,
As noted, SEVERAL GS players are performing below their historical norms and Lee, Curry, and Bierdrins have been hurt at various times.
That’s why GS is not performing as well as some thought.
Exactly what part of that was predictable?
Players get hurt all the time, but the projections here do not try to account for the probability of key players getting hurt (with the rare exception of players with a long history of being hurt).
I think Golden Start has more or less demonstrated that it’s about a .500 team when Lee, Curry, and Biedrins are playing even if several other players are below par. It would be better if Lee was simply scoring as efficiently as he has over many years. It would be even better if the other players were being as productive as expected.
I don’t think GS’s record has much to do with the value of Lee’s rebounding or his deficiencies on defense. Lee’s deficiency on defense was pretty much built into his record fairly well because the Knicks as a group were deficient defensively last year also.
People can debate the value of a rebound, but that’s not why Golden State is disappointing some here.
Jeremy Britton
February 12, 2011
Everybody, thanks for your comments. I’ve been slow to reply this week because I’ve been super sick, but it’s been fun–and in some cases bruising–to read what you’ve had to say.
I’m really surprised. I did not expect such strong resistance to Wins Produced itself. Most of the discussion has had less to do with the analysis I offered about the Warriors than with its underpinnings. Alternate theories, chinks in WP48’s armor, disputing Wins Produced’s validity altogether abound. To those dismissive of the post for these reasons the best I can say is: Guys, Wins Produced is the analytical tool this blog itself is based on. You shouldn’t be surprised to see somebody use it!
I admit I kind of love Wins Produced. I’m excited about the way it has helped me build a better mental model for seeing a game I freakin’ love. I’m thrilled by the detail it addresses. It’s changed the way I read box scores, watch live games, and even play the game. (I mentally track performance when I play every Thursday night. Personally I have focused on better shot selection and rebounding–my teams have been winning more since I’ve done this, which is cool!).
Who is this all for, anyway? Beyond that my response to many of you comes down to a question of audience–who are we writing this stuff for and why? I can see how some of my remarks come off as brash. Fair enough. I’m motivated to try and reach a broader audience. This comes from lots and lots of conversations elsewhere with fans who love basketball, but wouldn’t have a clue what any of you are talking about. I think those fans matter. I love talking to them. I want to reach them with ideas I think are valuable. I’m not interested in amplifying an insider echo chamber. (Yes, you could argue I picked the wrong blog and that I didn’t honor *this* audience, but I don’t think so for other reasons that have to do with moving Wins Produced forward.)
Complicated or conclusive? So much of this discussion is so complicated. It just feels stuck. I admit to @NeutralCorner, @marparker, @ElGee and @mystic, I don’t understand your points much at all. The way they are written I find hard to follow. I’ve had my Linus moments (lots of doubts about WP48), but I keep finding convincing rebuttals from it. I’m totally open to your counter points and rebuttals, but again honestly, I find them very hard to follow.
So far, I think many of your rebuttals have already been answered. Call me an naive acolyte of Wins Produced, but the arguments Berri and others like Arturo and NerdNumbers assemble are well-written and offer clear conclusions. They are approachable (to a casual nerd like me). I trust their approach too–it’s energetic, open-minded, inviting of vigorous debate. I can also point back to several posts that seem to address concerns you have with Wins Produced. Frankly, most of these Berri has already addressed in his frequent replies as well as here:
https://dberri.wordpress.com/frequently-asked-questions-and-comments/
Jeremy Britton
February 12, 2011
My second comment concerns David Lee and the argument that I should pull my head out of the sand already, stop making excuses, and admit Wins Produced has been proven wrong or flawed.
You guys miss the point. Wins Produced shows that David Lee has been performing much worse this year than years past and this in part (-5.2 wins worth) explains why the Warriors are doing worse than I predicted in my first rosy blog post. There is no convincing evidence from this season consistent with Wins Produced that David Lee is overrated by Wins Produced. This season’s data correctly measures both Lee’s lack of production and how the Warriors have suffered from it.
Some of Berri’s past analysis has shown that injury has a very negative impact on player performance. David Lee had a freak, prolonged accident with Wilson Chandler’s tooth in his elbow and an infected open wound.
David Lee said,
Gross! His poor performance is easy to tie to the injury explanation.
But Lee is healthy already! Really? I’ll fall back on personal experience here like you guys do. Anybody who has had an injury like that (I had a nasty one that took a chunk out of my knee) knows that just because you’re back on the floor doesn’t mean you’re 100%. That takes time, even into the off season. Lee has shot far fewer shots than normal at the rim, his jumper went south, and he’s rebounded way less. I think it’s because he’s had 1.5 arms to work with. That’s tough!
dberri
February 12, 2011
Thanks Jeremy. A very good response. I do understand all the critical comments your post inspired. Unfortunately, I find these criticisms to be unconvincing (but also have no desire to continue a back-and-forth with these people).
Mike
February 13, 2011
OMFG – http://nbcprobasketballtalk.files.wordpress.com/2010/12/207801199.jpg?w=204&h=272 OMG! Look at that ad tell me Lee is healed?
Also, is it just me, or is it funny that WP48 gets criticised for saying Lee was really good, and now when he isn’t, getting that right as well? \
marparker
February 13, 2011
Mike,
Golden State was supposed to get much better by adding David Lee. They haven’t been much better and Lee’s Wp48 is less which is what the diminishing returns crowd said would happen.
Complicating matters is an injury which is seen as a convenient excuse for being wrong.
I’m in the camp of David Lee improves their offense and does nothing for their defense and therefore makes them a very good home team and doesn’t help them on the road. They’ll probably win 25 or more home games, which is the equivalent to winning 50 out of 82 games. They probably won’t improve too much on their road record of 8-33. This will land them around 33 wins. That is an improvement on last year of 8 games, but not nearly enough to propel them into the 50 win range. IMO, GS still needs to add a defensive specialist but those guys aren’t being given away by any teams. They’ll have to find one in the draft or a young player who hasn’t blossomed yet.
ElGee
February 13, 2011
Hi Jeremy,
What’s hard to follow? WP is regression-based off a box score. The box score has limitations. Regression in this context has limitations. As a result, the marginal values are problematic, namely in how they treat scoring and rebounding. I’ve read the FAQ and I don’t think it adequately addresses that other than to say “people don’t think about how important rebounding is but regression says it’s important!”
Here’s my specific example. It’s fairly straightforward I hope – please let me know if it’s not: http://elgee35.wordpress.com/2011/01/24/interpreting-advanced-statistics-in-basketball/#WinsProduced
That is why David Lee and Kevin Love and Kris Humphries and every other rebounder looks so good and players like Kobe Bryant and Monta Ellis don’t. Mystic took 10 minutes, regressed on a box score, and came up with a “better” measure. So what exactly is useful about WP?
Yeah, rebounding is important. But it’s not as important as scoring. You know what has a 100% correlation with winning? Scoring more points than the other team. ;)
Anything that goes into generating a possession — including rebounding — to create scoring opportunities is good. But they still have to score! If Minnesota had two No. 1 scorers and Kevin Love, they’d probably be a top 10 offense. (They are 23rd in offense, 27th in defense.) That doesn’t make Kevin Love the team’s best player, and certainly not the league’s best player. Know what I’m sayin?
Mike
February 13, 2011
So your analysis is that the warriors would never have been a 58 win team? And there are no mitigating circumstances? OK.
What about the trade of Iverson for Billups? Of all the trades in the last few years, nothing shows the value of WP better. http://sports.espn.go.com/nba/columns/story?columnist=hollinger_john&page=IversonTrade-Nuggets-081103 Hollinge scored it a Denver win because, and I quote “Iverson is a legend, and Billups isn’t … but who would you rather have going forward?”
We can all point to one example of a failure, but the issues with GSW appear to be injuries. That is by far the simplest explanation. When your highest producer of wins is Dorrell Wright, simply because he is the only healthy player – you got issues. And when you have more players with negative wins than positive, and all have played > 100 minutes… it starts to be clear.
Which means more in the same number of possessions. Inefficient scoring hurts, which is why Iverson was not a good as he seemed, and the only reason Leborn outpaces Carmelo.
dberri
February 13, 2011
ElGee,
You simply don’t seem to understand how to evaluate models. I did post a guide at some point. Maybe I need to do this again.
One key point… we don’t judge a model by whether or not it simply confirms what we thought when we started. Throughout your “evaluations” that appears to be a criteria you consider.
As for your observation that points scored and surrendered explain wins… well, that is true. And that is where the derivation of Wins Produced starts. But you have to go beyond that point if you are going to understand how rebounds, turnovers, and shooting efficiency impacts outcomes. In other words, if you stop with the observations that points scored and surrendered determine wins, you are arguing that rebounds, turnovers, and shooting efficiency has no impact on outcomes. You seem to be saying that this isn’t true. So clearly you think that modeling wins as just a function of points scored and defense is not a particularly good approach.
ElGee
February 13, 2011
Hi Dave,
I’m not sure where in your post you illuminate how I “don’t understand models.” My degree was not in math so I always welcome criticism/explanation, but I’m pretty sure I understand the model. ;)
To your point about judging something by looking for confirmation, that’s not what I’m explicitly doing. If the model is showing us something new, then it should be revealed in future predictions. I’ve yet to see that with WP — and feel free to link me to where you think WP has done this, but please don’t be selective since in the last 8 years there have been 240 team seasons and getting one right wouldn’t be a resounding victory. ;) This Warriors team was an obvious opportunity for WP, and to people who have a strong knowledge of basketball — an underlying mental model, if I can borrow from a Chomsky concept — it smacked of ridiculousness for obvious *causal* reasons. (Although I do sympathize with WP a little in such cases bc of diminishing returns.)
And no, a model just on points wouldn’t do much for me. But I have had experience with machine learning, and some level of future performance is relatively easy to achieve depending on we set up the criteria. If we extend it beyond points to rebounds, TOV, and efficiency, dole out the credit to individuals and then sum it back up, then of course the correlation with team results will be really good. But this is individual player analysis, and obviously the credit isn’t being distributed very well or we’d see profound predictive ability outside the generally accepted hierarchy of players. No?
dberri
February 13, 2011
ElGee,
At no point in your discussion of Win Shares (a model you appear to endorse) did you focus on predictions about the future. You did focus on how the evaluations of the model fit your prior beliefs about the players. And this is incorrect (in my view).
As for correlations across time. Win Shares per minute has a correlation across time that is less than 0.7. ADJ P48 has a correlation of about 0.85. So if all you cared about was consistency across time, you would prefer the approach advocated in this forum.
But that approach contradicts your prior beliefs. And when you found an example where a forecast was incorrect, you jumped all over that. Your evaluations, though, are inconsistent. You seem to endorse Win Shares without once discussing the ability to forecast over time.
ElGee
February 13, 2011
Dave,
It’s not inconsistent at all. I endorse WS as a summary, pace-adjusted box score metric. That’s it. It handles that task better than something like PER. I don’t see any great outliers against common basketball knowledge in that stat.
Your criticism about models fitting to prior beliefs is valid, but only to a certain extent. If it doesn’t pass the smell test for an understandable, and predictable reason, it’s a limitation in the model, NOT in the prior belief. (eg Multicollinearity or small sample/noise in Adjusted Plus-minus. It’s possible that the outliers are better than we think, but the second Amir Johnson is asked to do more or Anderson Varejao plays without LeBron James, the results reflect perception and these players struggle again.)
The ability to forecast over time is important in this discussion because it should validate what the model is telling us about these outliers. (As I’ve said, I sympathize to a degree with regards to new team forecasts because of diminishing returns.) You’re simply stating it’s prior beliefs that are wrong and the model that is correct, but how are you substantiating it??
dberri
February 13, 2011
ElGee,
Your argument is quite inconsistent. You acknowledge that prior beliefs shouldn’t be used to evaluate models. Then you want to use prior beliefs to justify the model you prefer, and you then go on to say prior beliefs might be used (but when? and how do you decide?). You say that we have to consider predictions of the future. But in your defense of Win Shares, you never mentioned this issue. So you appear to be picking and choosing your criteria in evaluating the model you prefer (and the one you don’t like).
Again, I wrote a guide to evaluating models (there is a link at NBAEh?). That might help you understand the issues I consider when I evaluate research.
Matt Johnson
February 13, 2011
I feel like there’s probably a more appropriate place to ask my questions, but I don’t know where that would be, and so I’m writing them here. I apologize for my ignorance, and if you’d like me to post elsewhere, just point me in the right direction.
First, “It correlates well with Wins” is often used as a reason for using WP, and the FAQ page talks about Levin/Rosenbaum’s use of adjusted +/- as a barometer (I think that was the term used) only by describing it as odd.
Now, adjusted +/- as a player valuation obviously has huge issues, but I don’t see where Dave gets into what makes it a bad barometer. As a barometer, it’s not competing with WP & PER, it’s competing with wins – and wins are a very coarse thing.
Wins credit a player for what a team does while he’s out. So it would seem like a smarter plan would be to go only by when that player is on the court, right? Well, that’s raw +/-. Adjusted +/- is just a smarter version of that.
Moreover, the issues with adjusted +/- are due to the fact that 82 player games is not nearly enough sample size to reduce its deviation to the quality of something like WP. However when you use adjusted +/- as a barometer, you’re dealing with a sample size of orders of magnitude greater, so why is that issue still pertinent?
With all that in mind, it would seem that it’s at least debatable whether adjusted +/- is a better barometer to use than wins. Yet, Dave dismissed the idea as “odd” and moved on as far as I can tell. I’d think it would at least warrant him illuminating the specific flaws in the thinking, so what am I missing?
2nd general question: What does WP do to address the issue of overfitting? Dave seems to be focused on explaining the current season instead of predicting the next one, and I don’t actually see a problem with that, as long as basic steps are being taken to prevent overfitting – but I haven’t been able to find Dave’s explanation of this.
Cheers,
MJ
Mike
February 14, 2011
MJ,
I have no idea what you mean by “barometer”, but if you read https://dberri.wordpress.com/2007/11/18/a-guide-to-evaluating-models/ (there you go Prof Berri) “Forecasting power out of sample” fits. +/- varies greatly from year to year, whereas WP48 has a .85 correlation year to year.
+/- suffers because how do you account for a player that plays 2 minutes? Or 1? What about 300? What about 500 over 10 years? How do you separate out individual contributions?
As an example, if Lebron played with D Leaguers his whole career, would he have a positive +/-? He would for WP48.
NeutralCorner
February 14, 2011
@ Prove it:
re using individual players on/off differentials to refine the defensive adjustments.
I would expect the correction to be small for such a change, FOR MOST PLAYERS. Most players who rebound well have positive on/off differentials for rebounding. Likewise, most players who execute the team concepts of defense which aren’t captured in blocks or steal boxscore numbers tend to have positive differentials for opponent points per possession, or opponent FG%.
The outliers are the players who play the game in an unconventional way. In most cases I would not expect the incorporation of differential stats to move the needle much, but in the cases where it does I would hypothesize that it will show something revelatory about particular types of players that the current model may be missing.
Part of the Golden State Warrior fan interest in this particular discussion comes from our long standing experience with gruesomely incomplete specialist type of players, the Fortsons, the Troy Murphys the Tawn Jamisons, the Maggettes and now potentially the David Lees. GSW fans have a burr under their saddle over this kind of player. Guys who do impressive things in the boxscore, but not so impressive on the scoreboard. Seeing yet another player who seems to fit this profile regarded highly by the WP metric sets off alarms for some Warriors fans.
There is an obvious flaw with differential stats as well. They have no context. A rebounding differential of -0.5 may not mean much for a guard, or it may mean a lot. I don’t know. I tend to look at differential numbers in a comparison to teammates context. But it would be great if someone would do the work to standardize those numbers across the league.
I think somewhere between the methods of WP and the types of data that differentials are being used to quantify there’s a way to solve the deficiencies and quirks of both.
It may seem like a lot of work for a small yield when the current method works for 85% -90% of players, but if there’s an enhancement that would improve the result for the oddballs, the guys GM’s don’t see coming, as those guys tend to represent major potential free agency and contract landmines, wouldn’t it be good to improve the model to work farther into the margins?
Anyway, my interest wasn’t to wipe my ass with WP. It was to point at a direction that I would personally like to see the analysis go because I hypothesize benefits in that area that would speak directly to the questions surrounding your prediction. The GSW GM certainly thought he was adding 19 or so wins wen he gave Lee $80MM.
Your Dad
February 14, 2011
Your dad’s here to say that until the organization is stripped clean of the Cohan/Rowell residues, this team will not be a winner.
I read the first Warriors will get 50 wins post and I just thought there’s no way in hell. Haven’t you been watching this organization for the last 20 years?? It’s a pitiful excuse for a professional franchise.
We don’t have any players on this team that have leadership, or have earned it. We have no players that have been a valuable asset to a playoff team. We have no elite player like a durant or a LeBron to lead a team lacking leadership, or winning experience out of the catacombs of losing.
WP is a stat that works great in the video game world, but doesn’t ever attest for heart, for losing culture, or for cohesion/chemistry.
WP does great for teams that have an established winning culture, and/or is comprised of solid veterans and predictable stars. That’s why everyone points to WP as predictive when it’s fairly spot on for your Celtics, Spurs, Mavs, etc….but not really at all for the Clips, Warriors, Grizzlies, Nets, Pacers, etc.
There’s an elephant in the room here, and it’s the reason the Warriors don’t have to change, the reason they aren’t winning, and the reason there’s no urgency to change that:
Blindly devoted fans.
Mike
February 14, 2011
Been thinking on a few things and it seems to me that the criticism of WP48 goes like this:
– It undervalues scoring because scoring is hard, and WP48 assumes that someone else would score if inefficient player X didn’t take so many shots.
– It over values rebounding because someone would get the rebounds anyway.
This seems like a contradiction. Either all stats have to be achieved (scoring AND rebounding) or none – the argument that some do and some don’t seems hard to justify.
And the argument for inefficient scoring is an odd one because both Carmelo and Iverson ware inneficient not becaue fo talent but shot selection.
Carmelo has taken 1313 three pointers in his career and hit 409 at .312 and Iverson took 3383 and hit 1059 at .313.
IMHO poor shot selection that leads to inneficient scoring is the biggest issue in mis-evaluating players. That is because people bias for the ability to generate good scoring options, and forget all the contested 19 footers and 3 pointers a player takes and shouldn’t. That is one’s eyes lying to you – plain and simple.
Matt Johnson
February 14, 2011
Hi Mike,
“I have no idea what you mean by ‘barometer’, but if you read https://dberri.wordpress.com/2007/11/18/a-guide-to-evaluating-models/ (there you go Prof Berri) “Forecasting power out of sample” fits. +/- varies greatly from year to year, whereas WP48 has a .85 correlation year to year.”
‘barometer’ was the term Lewin/Rosenbaum used in their paper which Dave responded to in the FAQ. Basically what they mean is that you’re using something as a basis for your actual player metric. In WP, my understanding is that raw team success was used as the “barometer”. Lewin/Rosenbaum argue that it doesn’t seem to make sense to let the relative success of the team when a player is off the court influence matters, so why not just focus on the relationship with team success when players are on the court?
I certainly am not trying saying that that means they are right and Dave’s wrong, it just seems to me that it’s not an absurd line of reasoning and I’ve yet to see Dave really tear it apart despite the fact that he’s dismissed it as wrong.
Fundamentally though what you should understand here is that asserting WP’s superiority to +/- is not related to what I’m talking about, but it is quite understandable why people get confused.
“+/- suffers because how do you account for a player that plays 2 minutes? Or 1? What about 300? What about 500 over 10 years? How do you separate out individual contributions?”
Eh, these don’t seem to me like they’re really on point with what I’m talking about, but I will answer your questions because they are sincere and valid basketball statistical questions generally . Just please, don’t take my answers to mean I’m saying “+/- kicks WP’s butt!”.
I don’t think the mere existence 2 minute thing is an issue particular to +/- (though as I’ve said, I’m completely with Dave that inconsistency is a problem with the stat as a player metric compared to WP, PER, etc). WP48 has to deal with the same thing where a guy could get a lucky 2 minutes and have an unreal score by the metric, but everyone knows it’s silly to judge WP48 on that.
How do you separate out contributions? It’s an unobtainable ideal to think you could ever do this perfectly with any stat. A player’s shooting volume changes based on context. So does his shooting efficiency and his rebounding volume. If you want to come back and say that direct box score stats still don’t have near the problems in this regard as +/-, that’s fine and correct. At the same time, if you’re going to assert assert that we can say nothing about separate contributions based on correlation alone, then you are also taking a smack at any stat that uses regression analysis to help develop its weightings (which obviously WP is one).
“As an example, if Lebron played with D Leaguers his whole career, would he have a positive +/-? He would for WP48.”
When people talk about using +/-, they don’t mean raw +/- (or if they do, it’s very rare). The +/- talked about by Lewin/Rosenbaum is an adjusted +/- that factors how the player’s team has done when he’s on the court while consider every player he’s played with or against. So if LeBron played with D leaguers, and as we’d expect, his team did better with him on the court than with him on the bench, yes he’d have a positive +/- value.
Cheers,
MJ