On Saturday night the Minnesota Timberwolves – the team with the worst record in the Association – defeated the Phoenix Suns (one of the best teams). Judging by the AP story of the game, this victory stunned the winners and losers alike. The Suns are legitimate contenders for the NBA title. And Minnesota’s record is not an accident. Despite the win over Phoenix, this is one of the league’s worst teams.
One suspects that just as the acquisition of Kevin Garnett vaulted the Boston Celtics to the top of the league, the loss of KG has sunk the Timberwolves. And when we look at the numbers, this is true.
But let me hold off on today’s numbers for a moment. What I want to do first is play a game of “what if?” Specifically, what would have happened if Minnesota’s general manager, Kevin McHale, passed on Kevin Garnett in 1995 and instead acquired just an average big man. For example, what if he chose a player like Kurt Thomas or Gary Trent; or what if Joe Smith or Antonio McDyess fell to the T-Wolves [see the review of the 1995 draft for the career numbers of these players]? How much would this have impacted Minnesota’s fortunes over the past twelve years?
Minnesota History Revised
The Minnesota Timberwolves joined the NBA in 1989. In their first six years this team never won more than 29 games and averaged only 21 victories per season. In sum, Minnesota was consistently one of the worst teams in the Association.
And then in 1995 the Timberwolves took a chance on a high school kid named Kevin Garnett. And after a rookie season that was more of the same, the T-Wolves with KG suddenly became respectable.
And yes, respectability for this franchise was tied to the decision to take The Kid.
The Timberwolves won 517 regular season games with Garnett on the roster. Of these, 247.8 – or nearly half — can be connected to the statistical production of Kevin Garnett. To illustrate KG’s impact, consider the numbers in Table One.
Table One: Minnesota With and Without KG
In Table One we see an estimate of how many wins the Timberwolves would have had with KG replaced by an average player (average WP48 is 0.100) each of the past twelve seasons. What we see is rather stunning. With KG replaced by an average player the highest Wins Produced the Timberwolves achieve is 34.3 in 1999-00. The average Wins Produced – in the world without KG – is 27.4.
So let’s review. Before KG this team averaged 21 wins. In the world with KG replaced by an average player, Minnesota only averages 27 wins.
This means that if we leave all of McHale’s decisions in place, but we simply change one of the first he ever made, he would have been presiding over one of the worst NBA teams for more than a decade.
Minnesota Today
Simulations are always neat (if you like such things), but reality is even better. This year we get to see what KG meant to Minnesota.
Table Two reports two projections of the Timberwolves this year. The first projects what would happen if Minnesota’s players (except for the rookies) played as well as they did last year. The second projects what will happen if Minnesota’s players keep playing as well as they have thus far this year.
Table Two: Projecting Minnesota in 2007-08
Heading into this season, here are the four players we would have expected to be the best in terms of WP48: Al Jefferson, Ryan Gomes, Craig Smith, and Marko Jaric. Given what these players did last year, and the minutes they are playing in 2007-08, we would expect these players to produce 26.7 wins this season. As for the rest of the roster (again, taking the rookies as given), we would have expected all of these players to combine to produce -5.2 wins.
When we look at the numbers this season we see our expectations are being confirmed. Jefferson, Gomes, Smith, and Jaric are combining to produce 24.1 wins. The remaining ten players on the roster are producing -5.8 wins.
In sum, this team has very few players who can produce wins in the NBA. And if we are search for “good” performers, our list in 2007-08 includes only Al Jefferson and Marko Jaric.
Reviewing the Trade
When the T-Wolves traded KG to the Celtics, people noted the quantity of players Minnesota received in return. But when we look at Jefferson, Gomes, Sebastian Telfair, Gerald Green, and Theo Ratliff, we see that only Jefferson and Gomes were expected to be above average. And so far, Gomes is not living up to this expectation. Consequently, one could argue (at least, I am going to do this) that this trade is really just KG for Jefferson.
So far, the best year Jefferson has had – in terms of WP48 – was last year. With a mark of 0.252 in his third season, he was clearly above average. And one could note that Garnett’s WP48 in his third season was 0.250. So perhaps Jefferson could develop into a player like Garnett.
Given the similarities in wins production between Jefferson and Garnett in each player’s third season, maybe this specific trade can work out for Minnesota. Still, even if that happens, the Timberwolves are still just one star player and not much else. And this is what Minnesota has been since McHale came to town.
Certainly we see evidence that McHale can identify very productive big men. Okay, we only have two examples, but it’s better than nothing. Unfortunately, I don’t think you need to have years of NBA experience or even a sophisticated statistical model to know that Garnett and Jefferson are good players.
When it comes to finding the rest of a good team, though, McHale comes up short. When we look at the data, we see that how far McHale has comes up short was hidden for years by the brilliance of Kevin Garnett.
And now KG is gone and the T-Wolves are hoping Jefferson develops. And against the Suns on Saturday night, it looks like Jefferson has arrived (I am kidding, but bear with me). In 42 minutes he scored 32 points (on 50% shooting), with 20 rebounds, and four steals. His WP48 for this one game was 0.658. If Jefferson maintained this for an entire season he would produce 47.2 wins and Minnesota would be a playoff team.
If this happened, McHale wouldn’t have to worry about finding “the rest.” Unfortunately, Jefferson is not going to be this good every night. He’s going to need help if Minnesota is ever going to contend. And this means the current cast around Jefferson is going to have to change. In sum, Minnesota is going to need some new faces.
Consequently we come to the big question facing Minnesota: Given McHale’s record, is he really the one you want looking for those new faces?
- DJ
Our research on the NBA was summarized HERE.
The Technical Notes at wagesofwins.com provides substantially more information on the published research behind Wins Produced and Win Score
Wins Produced, Win Score, and PAWSmin are also discussed in the following posts:
Simple Models of Player Performance
What Wins Produced Says and What It Does Not Say
Introducing PAWSmin — and a Defense of Box Score Statistics
Finally, A Guide to Evaluating Models contains useful hints on how to interpret and evaluate statistical models.
80 responses so far ↓
Phil Maymin // December 10, 2007 at 3:35 am
Removing a GM’s best decision and noting the results may not be quite fair. If you take out the single best day per year from the Dow Jones, its annual returns look much worse too.
It would help if you did a similar exercise with a couple other teams. What would the similar numbers be for replacing Tim Duncan with an average big man? (Borderline playoffs?)
Similarly with Detroit. What if they had received an average big man instead of Ben Wallace from Orlando? Or, sticking to #5 picks, what if Miami had gotten an average wing player instead of Wade?
And shouldn’t the replacement value not be average overall but whatever the average was for the relevant draft position? (Though if I recall you had some stats on this and they weren’t far different from average after the top pick or so, right?)
How is this for a metric of a GM’s drafting ability: compare the total production of all his drafted players against the total of the average expected production at those same positions. Then having one big home run like Garnett can make up for oodles of players that didn’t pay off. Or it may still fall short.
Finally, I wonder how much salary fits into it. If you are a GM with a tight salary restriction because your #1 player is so expensive, you are probably looking for a different kind of player, more long-shot candidates than heavy favorites, because you need more bang per buck. In the case of Sprewell, for example, it paid off for a while. But just the fact that you are trolling for cheap options rather than solid investments means you will get busts more often than the average.
Thanks. I very much enjoy your blog! And being a Celtics fan, I very much enjoy having my favorite and the best player on earth playing on my favorite and the best team on earth.
Best,
Phil
richmindseed // December 10, 2007 at 4:19 am
I thought one of the big things about the KG trade was the cap flexibility obtained by getting a bunch of one-year deals…so you get AlJeff + some cap room. Add a couple decent FAs, a lottery talent, and stir well…maybe it turns into a playoff team? Therefore, McHale’s move might not be as bad as this analysis shows…
Of course, the dude picking those FAs and that lottery talent is Kevin McHale, so…
mikel123 // December 10, 2007 at 6:50 am
Right, not to mention that if they didn’t have Garnett last year, presumably they’d have spent his salary on other players. Maybe not gotten the same return, but a lot better than putting an average player in there.
andrew // December 10, 2007 at 7:21 am
Dave already did a study comparing a duncan-less spurs team to a KG-less wolves. The disparity between quality of teammates was (as expected) staggering.
-Andrew
Patrick // December 10, 2007 at 7:36 am
I think the trade was actually a good one.
If Jefferson =~ Garnett, it would be a fantastic one. In that scenario, you are basically turning your 32-year old star player into a 22-year old star player, and cutting his salary in half. Pretty good deal.
Of course, the jury is out on whether Jefferson is going to be that good. But I suspect that Jefferson is very underpaid (even in his extension years) on a WP-basis. Garnett was certainly not as underpaid. And, I have to say, I think the Celtics would be a very good team if they had not made the trade. Adding Posey and the surprising play of House has a lot to do with their record (as you wrote once, they were on course to be a decent playoff team before Posey was added, and House’s contributions are pretty surprising. This has what transforms them from playoff team to title contender, not so much KG’s play, which is simply exactly what you expected [yes, this is pretty good]).
dbg // December 10, 2007 at 11:13 am
I’m surprised everyone is defending the trade from Minnesota’s perspective. I don’t see that they gained anything at all and they’ve definetly lost quite a bit. As for the cap space wasn’t Garnett on the last year of his contract? So if I remember the details correctly they would have even more cap space by doing absolutely nothing.
As for McHale’s ability as GM I think the Wolves almost 2 decades as a bottom-dwelling club is all that has to be said.
I actually have a question for DB. I think you wrote that usually only 5 teams have an efficiency differencial (off – def efficiency) of 5 points in a season. But looking at Knickerblogger’s site,
http://www.knickerblogger.net/stats/2008/
it looks like 8 or 9 teams beat that mark. Is this just an early season blip or are a lot of teams better this year? If it stays this way we could be watching the best basketball in years this season.
Animal // December 10, 2007 at 12:17 pm
This is a really interesting analysis I saw on the APBR metrics board of shooting by Mike G.
“If 30% of missed FG are offensively rebounded, then even on a 30% FGA the odds of ’something good’ happening are .30 + (.30*.70) = .51
When FT add .044 to a player’s efficiency (league TS%-eFG%), then a 30% shot is 55% ‘good’.”
For the whole post by him look here.
http://sonicscentral.com/apbrmetrics/viewtopic.php?t=1587&postdays=0&postorder=asc&start=15
What do others think about that?
Jason // December 10, 2007 at 12:49 pm
I find it curious that I’ve seen some of the same voices who criticize the assumption of on average a point a possession not having similar problems with the assumption of 30% of misses resulting in an offensive rebound. Anyone who has seen the Warriors play knows you cannot take *any* rebound for granted.
The offensive rebound is a tracked stat. Cutting into the value of it and re-assigning it to someone who missed a shot seems to be rewarding failure and taking it away from someone who did something good.
If getting a shot off–regardless of outcome–was substantially difficult, perhaps there would be logic in this, but I’m not aware of any evidence to show that it’s so difficult. Losing high volume shooters doesn’t seem to result in substantially fewer FG attempts for the team. If it did, then perhaps the “shot creation” logic would dictate that a missed FG (with a modest assumption of an offensive rebound) should bring some credit to the player for being able to shoot, but within the parameters of the majority of NBA players, actually getting a shot off doesn’t seem to be such a challenge such that it should be so rewarded.
Animal // December 10, 2007 at 1:15 pm
I don’t the answer to this, and I am not sure if this is what Dberri was stating when he says teammates don’t have much of an impact on their teammates statistics, but does losing a high volume shooter lower the fg% of teammates. While the shots will taken will be roughly the same, does the % change? For example, last year, what happened to the shooting percentage
of the players whose FGA went up after the Iverson trade? I don’t know who these players are, but I would guess it would Willie Green and Iguodala would have taken most of the shots, although that is clearly speculation and I don’t really know.
Panda Bear // December 10, 2007 at 1:39 pm
Animal, if I recall dberri had a post about how the 76ers didn’t suffer in terms of field goal attempts or shooting percentage after Iverson left.
–Panda Bear
Jason // December 10, 2007 at 1:43 pm
I’ve been looking into the 91 Rockets who lost Olajuwon for a large portion of the year. Short answer: players shot about as well without him as they did with him, his shots being distributed among the remaining starters. Some players actually shot a bit better, though I suspect that with the sample size, the difference wasn’t significant.
Animal // December 10, 2007 at 3:12 pm
Thanks Jason. Also, I don’t think Mike G is necessarily saying rewarding the player who shoots because his team rebounds 30% of his misses, he is just saying because of that, a 30% shooter either produces points or keeps possession of the ball 55% which is more than half. A 30% shooter is still well below average compared to the league, but it still means that more times than not when he shoots the ball something good happens for his team, even if it is not because of his own doing.
Jason // December 10, 2007 at 3:37 pm
I think that interpretation is misleading as it relates to what is ‘good.’
A 30% shooter makes a basket 30% of the time. When the ball goes in, it’s good. Even if we assume that in the 70% of his misses, 30% of those are likely to be rebounded, I don’t know if we can quantify this as “good”. An offensive rebound keeps a possession alive, but the situation before the shot was taken (team has the ball, ergo has *opportunity* to score) and after the offensive rebound (team has the ball, ergo has *opportunity* to score) are identical. I wouldn’t classify this as something good happening, but a completely neutral situation because *nothing* has changed vis a vis the condition of the game save the clock situation. This may be marginally better (fresh 24; closer to end if team is leading) or marginally worse (game closer to end, team trailing) but the equation doesn’t include that factor. In terms of the relative score and opportunity to score in the possession, the missed shot and o-rebound leave things exactly the same which is neither good nor bad.
In reality, there are three things that can happen when the 30% shooter shoots: it can go in (30% of the time=good), it can be missed and rebounded by the shooter’s team (.7*.3=.21) which is completely neutral, or it can result in a change of possession without a score (.7*.7=.49) More often than not something *bad* does not happen, but this is not the same thing as saying that something *good* actually happens.
I’ll anticipate the counter that WP says that offensive rebounds are good. It does, as they relate to the overall probability of a victory, but such rebounds completely cancel out the missed shots. Missed shot having absolute value equal to the rebound, resulting in a zero sum outcome, it’s neutral to win probability, neither good nor bad, though the *credit* as assessed to the individual goes to the rebounder, the penalty to the shooter. Team effect is the same.
Animal // December 10, 2007 at 3:55 pm
Well yea, when I said that it was good I meant not bad for the team. In other words, the team still has a chance at having a positive outcome on that possession.
So would it be fair to say that if the 30% shooter shoots, 30% of the time he scores + .21* the average shooting % of the nba is the percentage of times a 30% shooter contributes positively. (I guess that would also have to be multiplied by the NBA turnover rate. For example, if on average a team turns it over 4% of the time, than it would have to be multiplied by .96. Would this number be accurate of how often a 30% shooter produces a positive outcome?
And to get it more exact, for each team you would use the team off reb% and team turnover rate. Obviously, it is smart to “jack shots up” on a very good offensive rebounding team compared to a bad one and maybe that should be factored into the percentage.
Animal // December 10, 2007 at 3:55 pm
i meant smarter
Owen // December 10, 2007 at 3:56 pm
Animal – I agree with Jason. Really, I have never understood this argument. I find it totally mystifying. Could you state it more clearly?
To me, it really makes no sense to think about it in any other way than that the credit for the offensive rebound belongs exclusively to the rebounder. Any player in the NBA can perform the service of “generating” an offensive rebounding opportunity. There is no player in the NBA who is so bad that he can’t throw the ball at the rim at any point in the shot clock. Why should this have value?
Animal // December 10, 2007 at 4:08 pm
I am no attributing the credit of the rebound to anybody but the rebounder. I am just stating, by fact, that a 30% shooter only produces a negative play 49% of the time, and that is if you ignore free throws. With free throws it is only 45% of the time. Even though this is not true because of the shooter, it is still true that more likely than not, when a 30% shooter shoots, nothing negative will happen to the team.
So in that case, maybe 30% shooting isn’t all that bad. Obviously compared to every other it is terrible, as I doubt any player with significant minutes actually shoots 30%, but when looked at on its own it is not terrible.
If you want to see more of this argument about what is considered an average shooting percentage I suggest going to this page
http://sonicscentral.com/apbrmetrics/viewtopic.php?t=1587&start=15
Mike G, Harold Almonte, John Hollinger and others put their input into it and it is very insightful I believe. I am not going to restate all of their arguments like I did for Mike G, but I think if you want more opinions on the subject you will look there. I could try and defend their positions, but I don’t know as much as them or understand basketball stats as well as them.
And while you may not like Hollinger’s PER or Mike G’s eWins or Harold Almonte’s philosophy, I am sure you have respect for them as basketball statisticians. They are all intelligent and have something to contribute even if you disagree with them.
Animal // December 10, 2007 at 4:08 pm
I meant compared to every other NBA player in the second paragraph
Panda Bear // December 10, 2007 at 4:11 pm
Owen and Jason,
I think what “Animal” is trying to say is that points are overvalued relative to rebounds in wins produced.
Win score formula is
Points + Rebounds + Steals + ½Assists + ½Blocked Shots – Field Goal Attempts – Turnovers – ½Free Throw Attempts – ½Personal Fouls
In this scheme you have to shoot at least 50% (on two-pointers and assuming unfouled) in order for your shot attempt not to result in negative points.
It is “Animal”’s contention that 30% of shots being offensively rebounded means that the formula should be 0.7*field goal attempts and not 1.0*field goal attempts.
Win score version 2 would be
Points + Rebounds + Steals + ½Assists + ½Blocked Shots – 0.7* Field Goal Attempts – Turnovers – ½Free Throw Attempts – ½Personal Fouls
I agree with “Animal.” A player attempts a shot and misses. There is a 70% probability the possession is lost, in which case the player should be penalized for the shot attempt. There is a 30% probability of no impact, in which case the player should not be penalized for the shot attempt.
Panda Bear // December 10, 2007 at 4:13 pm
I presume the Points – Field Goal Attempts in the win score reflects how teams score 1 point per possession on average?
If so, then I think Points – 0.7*Field Goal Attempts would more accurately assign credit to the individual player.
dberri // December 10, 2007 at 4:19 pm
I will chime in here briefly.
Jason and Owen are correct. In fact, what Jason said earlier explained exactly the problem with using offensive rebound percentages in the evaluation of shooters.
Let me comment briefly on Panda’s model. The big problem with Panda Bear’s reformulation-beyond the fact that this is not how the stats relate to wins –is that it allows an inefficient shooter to increase his value just by taking more shots.
Animal // December 10, 2007 at 4:28 pm
Well I think the argument being made by Mike G is a 30% shooter is efficient because only 49% of the time his shot results in a bad play. Maybe that isn’t what he is saying, but that just seems to me like what his point was on the APBRmetric board.
Panda Bear // December 10, 2007 at 4:33 pm
“Animal,” but doesn’t every player in the NBA have 30%+ shooting percent? As Dberri says, the reformulation of win score I did would reward any player just for increasing his amount of shots. That doesn’t seem intuitively appealing either.
Captain America // December 10, 2007 at 4:55 pm
So much nonsense. First off, the Twolves have not been a playoff caliber team with KG. As Twolves management tried to build talent around KG, a locker room and chemistry emerged. KG liked “his guys” but dissed others. You could see that in the passing patterns that emerged last season. KG wanted to keep Sprewell (since selling boats in Wisconsin and the aging Cassell). You won’t find this reflected in the analysis. Moreover, McHale isn’t the only one involved with player decisions for the Twolves. Owner Glen Taylor, Jim Stack, Fred Hoiberg, and others are equals on the decision team. To wit, the drafting of Corey Brewer and Chris Richards was a collective decision. Besides considerable salary cap flexibility due to trading KG, the Twolves got two first round conditional draft choices. Get real.
DL // December 10, 2007 at 5:21 pm
fyi- Sign is wrong in change column of TWolvesProject0708 table.
dberri // December 10, 2007 at 5:35 pm
DL,
Thanks. I fixed it.
Animal // December 10, 2007 at 5:51 pm
Well maybe what this is saying is that everyone is an efficient shooter in that more than likely when they take a shot it will not result in a bad play, but compared to your peers you are inefficient. But also, there has been nobody in the history who has a been a 30% chucker (someone who shoots a lot) as far as I know. I think what maybe this is saying is that shooting a high volume in the low 40s is not a bad option. Of course it is all comparative to your teammates. If they can do better, than you are hurting the team, but that is only because they are being more efficient than you. That is not to say that you are being inefficient though.
Jason // December 10, 2007 at 5:55 pm
Animal, Mike G. is wrong if he considers “not bad” to be “good.” Considering “not bad” as “good” is a real, real problem. Yes, 51% of the time a 30% shooter fires up a shot, there will be a “not bad” outcome. However, the situations where the offensive rebound is garnered then need to be evaluated.
If you go through the iterations of what can happen after the offensive rebound (which involves 21% of the cases, excluding fouls), assuming the 30% shooter fires up again, we can go through the same three outcomes. Do this again and after the third shot, more than 99% of all outcomes are accounted for, but theoretically, the cycle can go on and on. The numbers just stop changing.
Assuming the same shooter, the sum of ‘good’ (made basket) outcomes possible vs. the sum of bad outcomes is about 38:62. More often than not, having the 30% shooter fire up shots will eventually end in a bad outcome.
Now there’s always the chance that the next shot will be taken by someone who shoot better. Let’s say that the shot gets rebounded by someone who can hit 50% of his shots (reasonable, since rebounds usually go to big men who are usually higher FG% guys). If this keeps happening,, 42% of the time the possession will end “good” (vs. 58% of the time bad). This is better, but it would have been better if the 50% guy took the first shot, as that has a 50% chance of going in on the first try (and 58.8% with successive attempts at a 30% O-rebound rate). For the *possession* to end ‘good’ (a final outcome where more than 50% of the time the possession ends in a made basket), the successive shots must be taken by someone shooting better than 93% from the floor. Offensive rebounds can lead to easier baskets, but not that easy.
It’s spurious to confuse ‘not bad’ with good. Only when you make that error can you come up with a reasonable situation where a 30% effective fg% nets you positive results anywhere close to half the time.
Animal // December 10, 2007 at 6:20 pm
Thats what I was asking before. So I ask you this, Jason, because you are better than this at me. There is a level (below 50%) where a shooter shoots the ball and more times that not he makes a good play( because of the offensive rebound.) Basically what I am asking you to tell me is the percentage someone has to shot to break even. This is assuming a 30% offensive rebound rate and I think to make the most general statement, instead of saying a 50% shooter or a 30% shooter takes the second shot, that the second shot is taking by somebody shooting the NBA average. I don’t know what that is, nor do I know where to get it. At some percentage, whether it is in the 40s or 30s, a player can be considered to break even. And yes of course it is still better to shoot 50%, but whatever that % is, is still good.
Animal // December 10, 2007 at 6:22 pm
So in conclusion, I agree with you that 30% is inefficient, but I believe there is a number somewhere between 30 and 50 that is considered efficient. And besides shooting percentage on the second and subsequent shots, you would also want to consider NBA turnover percentage. That would reduce the chance of the second possession being positive even more. Regardless, there is some number under 50% where a player “breaks even” because of offensive rebounding
Owen // December 10, 2007 at 6:34 pm
Animal – There is an average ts% in the league. It was around 53% last year I think, while there are different averages by position. TS% is not quite in tune with possession theory, but it’s definitely useful.
Any team that gets more offensive rebounds than turnovers will break even more often than average, if it is shooting at an average rate.
But it’s important to note that it’s not the player who originally shot who breaks even when his shot is offensively rebounded, but the team as whole. That is the point here…
Animal // December 10, 2007 at 6:48 pm
No, the point is that shooting X%, whatever percentage it may be that allows your team to have a positive impact on 50% of your shots (via rebound and putback or making it yourself) is not hurting your team. So if a player shoots X percentage and is than said to be having a negative effect on his team, that isn’t true. He is having a neutral effect, as 50% of the time when he shoots X%, his team does something “good” To penalize a person for shooting this X% is wrong.
Hopefully, Jason will calculate X because I don’t know what the average FG% is and don’t where to get it and i figure he does
Animal // December 10, 2007 at 6:56 pm
Ok, so I looked over the internet, and I found in 2003 the league shooting% was .422. Don’t ask me why I was able to find 2003 but not 2007, I just was. I may be doing this calculation wrong, so if I am correct me, but a very simple version of this calculation (excluding free throws and turnovers and the team getting a third, fourth etc. offensive rebound)
.5 = .3*.442 + x
x = .3674
If a player shots 36.7%, 50% of the times he shoots the team will score. This would indicate anybody who shoots better than this helps their team more than they hurt it by shooting.
And yes, I know they are not the ones who get the offensive rebound and continue the possession. But the fact is, 30% of the time that will happen, and I believe that has to be factored in when considering whether or not the persons shot was a good or bad result.
Animal // December 10, 2007 at 6:57 pm
i meant decision not result
magicmerl // December 10, 2007 at 7:15 pm
My understanding of waht animal is saying is that the ‘break even’ point at which you should be rewarding people for shooting is less than 50%, and the dberri’s model only rewards people who are taking shots at making over 50%.
Is that right?
So if for example, a person was shooting 48%, and we decided that that was a good shooting percentage, then this would be a good person that we want to shoot lots, but because the WOW model rewards people for shooting over 50% and penalises them for shooting less than 50%, this 48% shooter would be effectively penalised for volume shooting even if this is a good thing.
So I guess the question is, ‘what’s the threshold % where you want people above that limit to shoot more and people below that limit to shoot less?’
50% does seem too high. Maybe 45%?
magicmerl // December 10, 2007 at 7:16 pm
Just thinking some more, Probably true shooting percentage is better than just FG%, and maybe 50% is ok if that is the % used.
Animal // December 10, 2007 at 7:21 pm
well my whole calculation above was to prove that is somebody shoots 37%, that is breaking even
Panda Bear // December 10, 2007 at 7:25 pm
A player should be rewarded for shooting above average. “Animal” says the average players makes 43% of shots. If so, then I say wins produced should be roughly …
Points + Rebounds + Steals + ½Assists + ½Blocked Shots – 0.86*Field Goal Attempts – Turnovers – ½Free Throw Attempts – ½Personal Fouls
It’s not exactly 0.86 because it needs to be adjusted (as “magicmerl” says) for “true shooting percentage” but you get the drift. A player that shoots above average should be rewarded for taking a shot. The position adjustment does not fully capture this because as the formula currently stands rebounds are overvalued relative to points.
Panda Bear // December 10, 2007 at 7:28 pm
Why should a player be penalized for taking a shot and making it at a percent that is average?
Jason // December 10, 2007 at 7:34 pm
Your calculations are not correct, Animal. Your equation is not correct. It’s not correct as a simple version or a complex version. The equation cannot be set to .5 and solve for a simple algebraic unknown as you have done.
The ‘break even ‘ when, given a 30% chance of an offensive rebound and the same shooter firing up and discounting turnovers or fouls, a team scores on half their possessions, is between 41 and 42%. Shoot this and your team will lose about 60 games a year, though because breaking even in this simple situation isn’t sufficient.
I do not believe that the league average was 42%, as in the last couple of years, it’s been about 45% (on all shots, higher on 2pt shots). As of Sat., the league average was 45.2% total (47.9% on 2point FG’s). Because of the increased value of the 3, presently there’s an effective FG% of 49.14%.
It is not a “fact” that 30% of the time a missed shot will wind up as an offensive rebound. That’s a general approximation. In reality, the actual rates vary around that number, but this variation significantly impacts the rate of return on possessions.
Of course what the break-even point for a condition with three possibilities, make, miss and rebound, miss and defensive rebound has very little relevance for evaluating the break-even point for a shooter since more than that can happen in a game. Before the shot there can be a turnover or a foul. In general, teams have to be more successful than 50-50 on their non-turnover-non-foul possessions if they are going to win more often than not.
Panda Bear // December 10, 2007 at 7:39 pm
I retract what I wrote earlier. If the average shooting percentage for 2-point shots is 48% then what I wrote doesn’t apply. My objections were based on a faulty assumption that it was lower.
Jason // December 10, 2007 at 7:43 pm
Pandabear, Animal’s numbers aren’t right. The average effective FG% (exclusive of FTs but counting 3’s as 1.5 baskets) is a shade under 50% currently for this season. It was ~49.7% last year as well. It is real, real close to 50% for all shots. A player is penalized for shooting below this because he’s below average. Those numbers fit real, real close with the coefficients that Dave derived.
magicmerl // December 10, 2007 at 7:49 pm
Mine as well. If the true shooting percentage is 49.14%, then that is so close to 50% that I’d say dberri’s current model is good enough and does not need tobe tweaked.
I mean, the people who are ripping the model for counting a missed shot as costing you something are missing something. It *is* costing you something. You’ve launched the ball into the air!. It’s akin to running a business and ’spending money to make money’. Of course you will pay costs as part of your business, recognising that you have to incur costs as part of yoru attempt to make money. But you don’t ignore those costs when doing up the balance sheet.
That’s the same as saying shot attempt is an investment in trying to get points. Of course you wouldn’t say that nobody should shoot on the team. But the best scorers should be rewarded for scording (just like the best business projects with the highest return on investment should get more funding for subsequent enhancements).
The weaker projects/shooters should be evaluated in a way that penalises them for not being as good an option as their more efficient neighbours.
Panda Bear // December 10, 2007 at 7:49 pm
Jason, thank you very much for the average figures. All my objections were only applicable if the average effective FG% strayed significantly from 50%, which it does not. Thanks.
Animal // December 10, 2007 at 8:00 pm
Ok, I get it. So I was mistaken in using regular field goal%?
magicmerl // December 10, 2007 at 9:05 pm
Yeah.
Shooting 33% from 3pt is functionally the same as shooting 50% from 2pt (you get the same number of points from the same number of field goal attempts).
Shooting 40% from 3 is better than shooting 50% from 2.
Jason // December 10, 2007 at 9:30 pm
You were mistaken *and* the number was wrong, “Animal”. The uncorrected average FG% for the league is about 45%. I do not know where the 42% number came from. It does not seem to be related to anything real.
MC5 // December 10, 2007 at 10:43 pm
Maybe you guys can help me out here, cause the people at APBRmetrics couldn’t/wouldn’t. Why does it make sense to penalize for a FGA at all? Sure, I understand penalizing a missed FGA, but WinsProduced penalizes all shots, presumably under the idea that it represents a lost opportunity cost. But what else are you supposed to do with a possession? Where’s the lost opportunity on a simple FGA? Now, a missed FGA could be seen as a lost opportunity cost, since you’re losing the opportunity to make a FG. But simply penalizing every shot is baffling to me.
Animal // December 10, 2007 at 10:43 pm
that was the league average in 2003. That was the only number I could find when looking it up because I did not know where to look to get the numbers from… sorry to offend you, take it easy. I even said 2003. And if that number is wrong for 2003, than the website where I got that information from was wrong.
DL // December 10, 2007 at 11:14 pm
League average unadjusted FG% in 2002-3 was .442. Looks like a typo just slipped in somewhere along the line.
Here is the year to year raw data
http://www.basketball-reference.com/leagues/league_stats.html
You can calculate it and go to eFG% or TS% if you want.
Panda Bear // December 10, 2007 at 11:33 pm
Ok, I still have one problem wth win score. If a player doesn’t attempt a shot or a player shoots a lot of shots but only makes an average number, win score treats it the same. I don’t like that. I don’t want to seem like Harold A where I’m intransigent about just blindly contending rebounding is overvalued. But this is really intuitively bothering me.
Panda Bear // December 10, 2007 at 11:35 pm
I dont’ like John Hollinger’s system either. It rewards inefficient shot-attempts.
However, I really think win score needs to treat a 15 for 30 shooting performance different than a 1 for 2 shooting performance. Both feed into win score the same given the current composition of the formula.
Panda Bear // December 11, 2007 at 12:06 am
And Animal, don’t worry you didn’t offend me. It was an honest mistake.
dberri // December 11, 2007 at 12:56 am
Panda,
Just a quick thought.
Is a baseball player who hits 1-3 the same as a baseball player who its 100 for 300? Batting average says yes.
Panda Bear // December 11, 2007 at 1:01 am
Dberri, no b/c the 1-3 is so small of a sample as to inhibit any inference about ability.
There was an interesting paper on this–
http://www.amstat.org/pressroom/index.cfm?fuseaction=villanovastudybattingaverages
Jimm // December 11, 2007 at 2:55 am
Admittedly, McHale has been shaky over the years, but Minnesota isn’t actually in bad shape right now.
As you mention, Al Jefferson is a stud, but they have another hidden stud right now by the name of Randy Foye. Foye can definitely ball and score, so that gives Minnesota two potent scorers going forward and a do-everything defensive ace in Corey Brewer to go with them.
The next best thing for them, before I get to the bench, is that they should be well-positioned for a high lottery pick this season, and there could be 3-4 franchise players plus another 4-5 solid stars coming out of this next draft. If the Wolves landed Beasley, Mayo or Rose, look out. All those guys can score and Mayo/Rose can create too.
Looking to the bench, the Wolves have two very solid role player power forwards in Craig Smith and Ryan Gomes, a solid 6th man scorer type in McCants, a solid combo guard in Marko Jaric, an upside potential guy in Gerald Green, and a guy who looks like he could be a decent backup big in this league in Chris Richard.
So that’s their core, the rest is disposable, including Telfair. Three solid (and compatible) starters in Jefferson, Foye and Brewer, with another possible star/starter in Gerald Green, who everyone seems to love for his upside even though he hasn’t really shown it yet. Green is the x-factor in the whole KG trade, as well as Minnesota’s future, because if he did turn into a star, then you have Foye, Green, Brewer and Jefferson as the 4 starters along with the high lottery pick you snag this season (so far in this scenario, I’d have to believe you go with Beasley and move Big Al to C, but you can’t go wrong with Rose/Mayo either).
Of course, the lottery pins might not swing where you want, but the Wolves have at least a legit chance to have a young star squad in a few seasons down the road, maybe even next season if they get Top 3 in the draft, but the max upside and x-factor in all of that (outside the lottery) is Gerald Green.
As for the rest, it seems that the Wolves already have the talent for a fine second unit, it’s just too bad they have to start some of those guys right now (not too, too bad though because this draft is going to be sick).
By the way, I’m a Lakers fan so it’s not easy for me to break it down like this, but even though I have no love for McHale, I did spend a couple years on a contract in Minnesota in 99-00, so only wish the Wolves franchise the best (even more so if they sent McHale packing).
Then again, no matter how great it would have been to get KG, Andrew Bynum is just looking too good to unload now…that kid could be dominating for us until 2020.
Animal // December 11, 2007 at 6:44 am
and jason, I didn’t even say the number was 42%. I actually got it right, as somebody said it was .442 and that is what i wrote. Also, thanks for showing me what was wrong with what I was doing and explaining what is an efficient number, but also thanks for showing me you are also just an ?????
dberri – little known fact about comments. I can actually edit what you say. And on this comment I am removing the name-calling.
Another Pete // December 11, 2007 at 9:16 am
MC5 wrote:
I understand penalizing a missed FGA, but WinsProduced penalizes all shots
I don’t think anyone has addressed this yet…
WinsProduced doesn’t actually penalize made shots. If you look at the related Win Score, and just take the shooting related factors, you’ll see
Points (from field goals) – Field Goal Attempts
For every made 2 point field goal, we get +2 for points and -1 for field goal attempts, for a resulting score of +1.
For every missed 2 point field goal, we get +0 for points and -1 for field goal attempts, for a resulting score of -1.
So, positive for making a field goal and negative for missing, and it balances out so that players with a true shooting percentage of over 50% (right around the league average) get a positive total.
Another Pete // December 11, 2007 at 9:33 am
Panda wrote
I don’t want to seem like Harold A where I’m intransigent about just blindly contending rebounding is overvalued. But this is really intuitively bothering me.
It bothered me for a while at the raw intuitive level as well: an average shooter gets no contribution to Win Score from shooting, while an average rebounder gets quite a bit of contribution to Win Score.
I think the problem comes from using Win Score to compare rebounding to scoring, which I don’t think it’s intended to do… the scales are different. The contribution from scoring typically ranges from -10 to +10, while the contribution from rebounding typically ranges from 0 to 15. Where Win Score is useful is in comparing one player to another, where the different scales don’t matter because each player is measured on the same two scales.
I think that one could revise Win Score to look like
(Points + 10) + Rebounds + Steals + ½Assists + ½Blocked Shots – Field Goal Attempts – Turnovers – ½Free Throw Attempts – ½Personal Fouls
and it would better match intuition about the relative contribution of scoring and rebounding, while not actually changing anything at all about how it compares one player to another. And, since it doesn’t change anything for that purpose, why not just leave it out :->.
Westy // December 11, 2007 at 10:20 am
So in summary, a player who gets a shot off is no more valuable to the team than one who turns it over?
Jason // December 11, 2007 at 10:42 am
Very good point, Another Pete.
There seems to be a whole lot of visceral responses to the model largely because it doesn’t always match intuition. (These manifest from reasonable arguments about who actually effects whether a rebound is grabbed to the near-brain-dead “watch the game” spouting when anyone dare suggest that the popular notion of who is better might be wrong.) Why would one want to simply match intuition? That’s a particularly circular way of making a model.
I think it’s important to remember what Win Score is and where it came from and what its utility is. Win Score does not compare scoring to rebounding. It compares a complete product of a player’s contribution to probability of victory.
Win Score is a simplification of Wins Produced. Wins Produced was not designed to match intuition but designed to model probability of victory. It is an efficiency model taking team efficiency (which has been clearly demonstrated to be what determines the vast majority of victories) and assigning the individual components to players according to their statistical contribution.
Perhaps the way Dave writes about ‘wins’ might put some people off, but what they are is a player’s share towards the *relative probability of a win, weighted for time on the court.* That’s a bit of a mouthful. The WP48 is the factor by which a player’s stats contributed to the summed probability of victory for every minute on the floor. Average has a 50-50% chance of victory by definition in any contest where half of the time
The way WP formula was derived wasn’t by trying to come up with some hypothetical model of fairness, saying a point is worth a rebound or turnover or steal. It was derived by looking at probability of victory (a team’s winning percentage) and regressing the statistical contribution against this probability. The values are empirical. If it says that, because of the ‘penalty’ for a FG attempt, shooting runs you a possibility of a negative contribution and that because of this, rebounds appear to influence things more than people feel they should, that’s what it shows. It’s not based on a rationalization that this is true, but an empirical assessment of the value of a rebound. [I'd note that because of a position correction subtracting the average contribution at a position, there *is* a penalty for not getting rebounds if you do not get as many as the average at that position. I've seen it argued that the correction penalizes shooting a second time. It does not, but it will take much more time in the derivation to demonstrate why this isn't true.]
There is post-facto logic in these values being roughly equal since there is about a point a possession and possessions are equal between teams that points are equal to turnovers are equal to defensive rebounds are equal to missed shots are equal to steal, etc. This *should* be true if there’s on average about a point a possession, but this is an observation after the numbers were derived, not the manner in which they were derived. Since the values are roughly equal, Win Score says they’re equal for the quick-and-dirty calculation and sets them all equal to 1. But do not confuse this “1″ with an actual point or rebound and assume that it was assigned as a straight analogy. It’s for convenience that the unit was used since the isolated probabilities of the stats were all essentially the same (about 0.033 or 0.034). It is not saying that a rebound gets you a point or a turnover gets you a point. It is saying that rebounds affect the probability of victory exactly as much as a point does. That’s not something that intuition has anything to do with
As I see it, the argument of merit is whether or not the exact partitioning of these probabilities derived from team results and team cumulative stats can be assigned to a single player. (In reality, they aren’t because of two corrections for tempo and defense, though the value of these corrections is small, so effectively for relative rankings, this can be ignored.) Perhaps an intuitive approach has some merit here in looking for things to examine, but it is in finality an empirical question and needs to be evaluated empirically. Does losing a player who produced X wins in some number of minutes and replacing those minutes with a player who had previously produced Y wins in the same number of minutes change the team’s total by Y-X wins? If the answer is yes, then empirically the model is correct. Empirically, this seems to be pretty close for the most part for most players. As such, the model is supported by empirical evidence, intuition about player value be damned.
There seems to be plenty of confusion about what the model says when It doesn’t say how this is accomplished at the level of the individual stats. It doesn’t say that a 10 rebound/48 player replacing a 15 rebound/48 player will result in a net increase in 5 rebounds 48. It says that the increase or decrease in wins will be roughly equivalent to the win probability contributed by the player-assigned stats. The model does not say that any isolated components must stay constant. It is important not to confuse the two.
Jason // December 11, 2007 at 11:02 am
Westy: a player who *misses* that shot is no more valuable than a player who turned the ball over. The “penalty” for getting the shot off is more than negated by the value of the basket. One could easily adjust to look only at points and missed baskets instead of made vs. missed baskets. The end result doesn’t matter.
It’s also important to remember that it’s not saying that a player who shoots below average doesn’t help their team. He just doesn’t help his team as much as a replacement contributing at the average NBA level. The break-even in the model is set for a team with a .500 record. The shooting component says that if you shoot below average, you are bringing your team’s probability down below this mark. Shooting below average (all else equal) is hurting if you assume that you consider .500 as a baseline, but you will still win *some* games.
Mike H // December 11, 2007 at 11:07 am
Panda, there is no need to treat 1/2 differently from 15/30 in winscore, since winscore is just describing what actually happened. I assume that what you’re really getting at is that an estimate derived from a small sample has a larger margin of error than the same estimate derived from a bigger sample. I’m sure that some sort of error margins could be generated related to W48 and they would be interesting to see. However, dberri generally addresses this by just noting the fact that only a small number of games have been played or that a given player has only done x in limited minutes, which makes sense given the goal of simplicity.
Panda Bear // December 11, 2007 at 12:10 pm
Jason, excellent post (the one at 10:42 am) and explanation of the construction of win score. Thanks.
Westy // December 11, 2007 at 12:15 pm
Jason,
Your second to last post is a very good synopsis of the system we’re discussing. You note, “what [WP's] are is a player’s share towards the *relative probability of a win, weighted for time on the court.*”
That does seem much easier to swallow. The concern many of us have, though, is that while possibly deadly accurate at the team level, there is a breakdown when the system assigns value at that individual level.
I understand that team prediction-wise missing a shot is the same as turning the ball over. But at an individual level, I’d much rather have the player who can advance the ball at least far enough to get the shot off.
You note, “…the argument of merit is whether or not the exact partitioning of these probabilities derived from team results and team cumulative stats can be assigned to a single player.” And therein lies the rub. I am not convinced that we have been shown conclusively that defensive rebounds should not be shared somewhat. The model’s predictive power as players change teams would go down by doing this? And likewise, although it would be bringing the TS% break even point down slightly, giving a shooter 0.1 of the offensive rebound credit would conclusively not improve the individual player evaluation? Do we know this?
Jason // December 11, 2007 at 2:34 pm
Do we know if those modifications would not improve? No. But that’s an empirical question. We don’t know that it will improve the model, we don’t know if it won’t. It can be tested.
I suspect that one of the reasons Dave gets defensive is that the counters that try to logically reason why the model should be adjusted *usually* center on notions that are, to some degree, ways to make the the result more in line with popular opinion.
When you say you’d rather have someone who gets to the point of the bad shot and miss than someone who turns the ball over, the model counters that it doesn’t matter though this may not be true if the ratio of TO: shots changes dramatically and starts to fall outside of the parameters seen when the regression was run. It doesn’t matter, as observed. Gut feelings don’t change this, but gut feelings tend not to care so much about the parameters that were observed and often accompany anecdotal instances that fall outside of the normal parameters or imagine situations outside of the normal parameters.
It’s VERY important to remember that models are created within a realm of the actual and behave best within this realm. The actual is that *most* players who regularly find themselves in the position of advancing the ball to the point of a shot in the NBA do this more often than not. It appears that even the most incompetent point guards get someone the ball for a shot more often than they turn the ball over, at least often enough that the value of getting a shot and missing it isn’t different from the turnover. Perhaps if this was not true, if the NBA was littered with guys who turned the ball over with such frequency that actually getting shots was a real problem, this wouldn’t be true, but that would be reflected in the regression. The model doesn’t try to behave in the realm outside of normal variation and within the realm of normal variation (with turnovers being as infrequent as they are relative to FG attempts), the data say that there’s not a real difference between the missed shot and the turnover relative to win probability.
The better *critiques* I’ve seen do suggest means for testing alternatives (e.g. apportioning some percentage of the offensive rebound to other players on the court). But they still seem to be based on an intuitive notion of what’s right, not on anything more than that? Sure, we can alter coefficients to play with things to see if there’s a better fit, but why 0.1? Why not 0.15? Why not 0.06379? The argument of tossing out a value to see if it’s a better fit seem ad hoc based on a notion of what *should* be, not anything empirical. That intuitive notion brings about models that tell us what we already “know,” whether or not it’s true.
Tossing values out to test doesn’t seem like the way to do it though and it strikes me as an attempt to get at something that is ‘intuitively’ better. There are ways to get at whether or not other players have influence on the rebounds of teammates. My guess is that regression is not the tool for determining this but rather it’s an analysis of variance between rebounds and personnel and points and wins. This would indicate how much of a rebound ‘belongs’ to the player who pulled it down. The null though should be either all of it or none of it (the latter being easy to reject since there is variance between rebounders). Evaluating hypotheses is done by rejecting or failing to reject the null, not by suggesting another *possibility*, not empirically derived, and saying that we don’t know that the other possibility might be better.
Westy // December 11, 2007 at 4:07 pm
Exactly, Jason, I would agree with all that. Your posts are well-reasoned and extremely insightful. Thanks for that.
You note, “…it strikes me as an attempt to get at something that is ‘intuitively’ better.” Yes, this is exactly it. And I wonder, is there a reason for that ‘intuition’? It strikes me that the proponents of this model seem very happy to sit contentedly at the ‘null’ without evaluating the possibility that better individual predictions may be some place below that value.
Rejecting the ‘null’ is impossible without experimenting with values other than it. We all (or at least I) am unable to run such a test. That’s why we’re hoping (wishing?) that folks here would to show us the veracity of the null value.
Part of me wonders if some of this reluctance is because the ’story’ of the WoW would get somewhat less interesting if this [the best value is less than 1.0 for rebounds] were the case. Rebounders would be ranked somewhat less high and scorers somewhat less low. Maybe Rodman is now the 20th most productive player in the league instead of 1st or 2nd and Iverson is like 45th instead of 90th. Still above (and below) what standard media convention might say, but less provocative, and maybe less ‘interesting’ (and attention-grabbing).
It doesn’t take much observation to know that this site grabs the most viewers when claims such as Bynum is better than Kobe (so far this year) or Rodman is better than Pippen (and sometimes Jordan) are made.
dustin // December 11, 2007 at 4:50 pm
I’m pretty sure this site grabs the most viewers when TrueHoop links to it. There is not a whole lot of gain for dberri by having a lot of visitors, there are no ads on this site. And he certainly isn’t enticing the kobe lovers to buy his book by making controversial claims.
dberri // December 11, 2007 at 4:57 pm
I would add to dustin’s observation by noting that when 20,000 people came to the site last week, the Amazon.com sales rank for the paperback dropped. In other words, I have no evidence that increased traffic leads to more book sales.
As hard as this is to believe, I keep the model the way it is because I think that is what theory and the data says. There is no dark conspiracy to torture fans of Kobe and AI.
Jason // December 11, 2007 at 10:59 pm
I recomputed last year’s wins produced assigning 60% of the value of a rebound to the rebounder and apportioning the rest of it to other players equally. The correlation between this value and the un-altered value is .9722. Even dramatically discounting a player’s share of his own rebound does very, very little to the output of the model.
Panda Bear // December 11, 2007 at 11:58 pm
The correlation between this value and the un-altered value is .9722.
If you can do so easily and don’t mind doing so, it would be interesting to see by comparison what the correlation between the un-altered value is and a new value that ignores individual rebounds completely and just apportions the team total to players equally. I suspect that would be high as well. If so, then the correlation doesn’t necessarily mean that changing the rebound weighting would be irrelevant to use of win score as a predictive measure.
I think the more valid measure is to see which construction of win score is more predictive of future team performance when rosters change. As you’ve suggested it’s purely an empirical issue. For the sake of irony, I hope it turns out that rebounds are underweighted in that sense just to make eat crow all the people that without statistical justification keep coming here and saying rebounds are overweighted. (I have to stick myself in that category a bit, but I’m not dogmatic about it or anything.)
MC5 // December 12, 2007 at 12:36 am
Everyone should read this: http://www.uncg.edu/eco/rosenbaum/nessis.pdf
…and then see if you still really think WinsProduced is such a fantastic metric.
Jason // December 12, 2007 at 12:40 am
It’s not *completely* easy, but I can do the calcs.
Assigning no value of a rebound to a player but apportioning out them according to a fraction of the team total *does* make a significant difference. If you do this, for players with at least 1000 minutes (works out to about 20mpg for 2/3rds of a season), the best player in the league was Brian Scalabrine. Garnett has a negative win score, as does Kidd. Kidd, by this model, cost his team 25 victories and is the worst point guard in the league. There’s an extreme negative correlation. Rebounds matter and some part has to be assigned to the player who gathered them else the model is beyond ridiculous.
I just looked at the explanatory power with the devalued (40% apportioned out) rebound model.
I limited my analysis to players for whom I had 1000 minutes of data in over just two season, and my position correction is cruder than the way Dave does it, but unadjusted (Dave’s model) the correlation for these players was ~.63 while adjusted, correlation in individual score was ~.6. In other words, while the difference is minimal, with this dataset, the original model has more predictive power.
Though the correlation from year to year drops and my data sample was too small to have any real statistical power, this is true for players who change teams as well. The ‘adjustment’ devaluing rebounds does not aid in explaining future performance for players staying on the same team or changing teams.
Giving rebound credit entirely to the player who received it gives greater explanatory power than devaluing it, at least over the last two years for players who have had significant playing time.
Panda Bear // December 12, 2007 at 1:22 am
Jason,
Very interesting! Obviously those empirical results debunk my hypotheses that
(1) Win score with rebounds fully apportioned out would still correlate very highly with win score with individually assigned rebounds. (I had brought that up to question the meaningfulness of your earlier correlation of win score w/ rebounds partly apportioned out.)
(2) Deemphasizing rebounds would make win score more predictive of future performance.
Thank you very much.
Westy // December 12, 2007 at 9:33 am
Thanks for doing this work Jason.
To my way of reading, 0.6 and 0.63 seem very very close. Who’s to say that at say 0.8 for a DR, it is not 0.65? Maybe not, but at least a more thorough study of the possible shared weightings with more years of data seems worth exploring. I personally would be surprised if the data came in showing that a full 1.0 weighting was most predictive. It would be interesting to see the curve showing what predictive power each weighting had. If you have 1.0 at .63, 0.6 at .60, and add a few more points, one could determine whether there was a peak between 0.6 and 1.0 or if it constantly sloped up to 1.
And while there is .9722 correlation, I would anticipate that when looking at individual players it would have a somewhat substantial shift down for players who are exclusively rebounders when ranked against their cohorts. I would suspect that the overall ranking of individual players would move closer to what consensus on who the best players is.
By the way, I enjoyed your new article. Thanks again for the good commenting work you’ve done here. I’m intrigued as well by your company. I’ve personally used FamilyTree DNA, and so definitely have an interest in the field.
Oren // December 12, 2007 at 9:42 am
“Ok, I still have one problem wth win score. If a player doesn’t attempt a shot or a player shoots a lot of shots but only makes an average number, win score treats it the same. I don’t like that. I don’t want to seem like Harold A where I’m intransigent about just blindly contending rebounding is overvalued. But this is really intuitively bothering me.”
I don’t agree with this statement.
A player that shoots more FGs will go to the Free Throw line more often. And the equation for Free Throws is Pts-1/2* FTAs. Most NBA Players hit more then 50% of their FTs. Those that don’t are typically Centers that have high FG shooting percentages.
Consider a player that goes 10-12 from the line. That’s worth 4 WS Points. If you only take six shots a game, you’d need to score ten points to equal those four points.
Jason // December 12, 2007 at 10:08 am
I don’t have the numbers in front of me anymore, but the ‘pure rebounders’ still did pretty well with the ‘adjusted’ (60%) rebounding figures. Tyson Chandler dropped from #12 to #18 IIRC. There wasn’t much movement. I suspect that this is because of the position correction. Overall the productivity at each position shifted, so the corrections changed as well. Certainly the weight of scoring relative to rebounds did change, but it wasn’t huge.
I think it is hard to shine and be substantially above average through shooting and scoring alone via the WP model, but I’m more and more of the mind that this actually accurately depicts value in the game. I suspect that replacing someone’s scoring load is generally easier than replacing someone’s rebounding load. Again, an empirical question to be addressed, but I think that this is what the data suggest.
The players who seemed to suffer the most appeared to be big men who didn’t rebound well and weren’t terribly efficient scorers. (Al Harrington, I’m looking in your direction).
Westy // December 12, 2007 at 10:18 am
Interesting. Yes, that makes sense that the position correction made up for a lot of it (but wouldn’t entirely). I’m sure the position discrepancy between C/PF’s and G’s went down too.
Within a position, there was surely movement as well. For instance, I’d guess a player like Rashard Lewis would show somewhat better in this scenario as compared to other SF’s.
Oren // December 12, 2007 at 10:57 am
“I suspect that replacing someone’s scoring load is generally easier than replacing someone’s rebounding load.”
I wouldn’t disagree with this, but I would note that the Cavs were incapable of replacing Lebron and Hughes scoring load when they were injured for six games.
Jimm // December 12, 2007 at 5:03 pm
The Cavs are just plain horrible without Bron period. Maybe when they get healthier (and Varajeo is playing again)…but right now their depth is laughable.