Chirag Agrawal

When people object to Drummond's and Chandler's high WP ratings, I think (at least I hope) they simply believe ~.300 is too high. I'm sure they would still agree that both Drummond and Chandler are significantly better than many of the 34 names you listed above. Tyson Chandler's lob dunk may not necessarily be easy, but it's likely easier than most of the shots Carmelo or Kobe puts up. And I definitely agree that Chandler should get credit for taking smart shots while Carmelo or Kobe should get penalized for taking dumb shots, but their roles are definitely different.

Having a player like Melo attract defensive pressure will make it easier for Tyson Chandler. I'm not saying Chandler is as bad as Perkins, Turiaf, Duhon, or Odom; I'm just saying he's not as good as Durant. If there was no shot clock, then yes, every shot other than the ones Chandler (or Novak) shoots is stupid. But there is a 24 second shot clock, and I'm guessing the Knicks don't typically give it to Tyson when the clock is winding down. They probably go to Melo, Smith, or Felton, who are forced to shoot or turn the ball over.

Drummond and Chandler obviously deserve credit for putting up great shots at a high efficiency; but I'm not sure if they'd maintain their efficiency at a higher usage.

129 days ago

Hoopdon

With regards to Drummond, and low usage players in general, there could be a reason they are low usage/don't play a lot.

http://hoopdon.weebly.com/1/post/2013/02/a-lesson-from-andre-drummonds-injury.html

There are some low usage players who are generally valuable (Chandler/Drummond), but there are also those who can be overrated if you only look at WP (Sefalosha).

129 days ago

Xavier Quach

Hoopdon,
Sefalosha shoots almost 40% from 3 on 5 attempts per 48, combined with being a good to great defender and making few mistakes. That's incredibly valuable.

Besides many typos and grammatical mistakes in the blog you posted, it suffers from a logical fallacy called "argument from authority". In effect, you're saying that coaches are professionals and so non-coaches shouldn't question their decision making, because we can't possibly know as much as they do. You assume since someone put them in charge, they MUST be competent or they wouldn't be in charge.

A cursory examination of your HDR metric also leads to some serious questions about your methodology. You more or less seem to be guessing at the weights you are giving different statistics without any sort of analysis to determine their accuracy against historical data. (I am referring specifically to changing the weight of a missed shot from -0.75 to -0.87 to "split the difference" in giving credit for potential offensive rebounds (which is an erroneous assumption, that shooters should get credit for a possible offensive rebound, the value of taking a shot over a turnover is the potential to score, not the potential for an offensive rebound)) Shouldn't you be using some type of regression analysis?

I think it's time WoW and thenbageek.com think about implementing a policy about people promoting their own blogs in the comments section. Discussion and counter-arguments are great, but people seem to think because they managed to get their half-formed thoughts on the internet, that it somehow validates it.

129 days ago

Xavier Quach

Chirag,
Is the issue really that someone like Tyson Chandler NEEDS to increase his usage?

What you're really saying is that Chandler is overrated because he doesn't shoot a lot. Because shooting a lot determines if you're valuable, and therefore a star. And conversely, people who don't shoot a lot are role players and have limited impact on the game.

WP is a cumulative stat, meaning for your FG% to drastically affect your rating, positive or negative, you would have to take a fair number of attempts. One extreme example of this is Miles Plumlee, who has a WP48 of .144 despite shooting 22% from the field. He hasn't shot enough to bottom out his WP. Andrew Nicholson, on the other hand, has a .024 WP48 despite shooting 53.7%. While his shooting percentage is very good, he has taken very few shots, so it doesn't swing his WP far enough to compensate for the rest of his play.

129 days ago

Chirag Agrawal

Xavier,

I'm not sure I understand the relevance of the last paragraph. Regardless, I am not saying Tyson Chandler needs to increase his usage. And I'm not saying Chandler is overrated because he doesn't shoot a lot. Rather, I'm saying the high usage players are sometimes underrated because they may "sacrifice their efficiency" by taking low-percentage shots when none of their teammates are able to create for themselves. The high usage players often face the greatest defensive pressure on their team, giving the lower usage players less coverage. I'd expect a 5 man unit of lower usage players to perform worse than their average efficiencies because they are forced to increase usage and attempt low-percentage shots that would typically be attempted by the high usage players.

129 days ago

Nathan Verney

Have some of these commenters actually seen Chandler and Drummond play? Carmelo doesn't make things easier for Chandler, Chandler makes things easier for Carmelo. Just watch opposing defenses scramble when Chandler or Drummond sets a high pick and rolls to the basket for a lob. It's like a vortex that sucks in help defenders, leaving the point guard to make easy passes to players on the wing or cutting baseline.

Sure there's other low usage bigs that go for love on the high pick and roll, but because Chandler and Drummond are so skilled at it, and pretty much unstoppable in the air on the catch, they draw more intense help on the roll than others do and still come up with sky-high WP stats

129 days ago

Xavier Quach

Chirag,

Ah, I get your point now. But you are operating on the assumption that high usage players "create their own shot" . No player on a basketball court creates a shot for themselves. Not without a pick, or a clear out, or the threat of an outside shooter. Proponents of shot creation want to give all the credit to the wing iso guy because that's what they're conditioned to believe by Sportscenter. Yet when a low usage player scores, people want to give some credit to the Melo's and JR Smith's of the team. You don't see the inherent contradiction there? When Melo scores, it's all Melo. But when Chandler scores, it's because Melo was there. You can't include team dynamics in one instance and ignore it in another.

The play creates the shot, not the player.

I'd point you to last year's Miami Heat as the perfect example. http://articles.sun-sentinel.com/2012-05-29/sports/fl-miami-heat-dwyane-wade-0530-20120529_1_erik-spoelstra-mario-chalmers-role-guys

What I was saying in the last paragraph is this: if a basketball player only took one shot in a year and made it, his FG% would be 100% but it would hardly move his WP48 at all. Low usage players are not overrated in WP because the effect of their shooting is mitigated by being low usage. I gave two examples of low usage players with drastically different FG%, which didn't significantly their WP48.

129 days ago

Xavier Quach

Also, after re-reading your first post, I'm confused as to whether you are saying Drummond and Chandler are rated too high, or that Melo is rated too low.
"When people object to Drummond's and Chandler's high WP ratings, I think (at least I hope) they simply believe ~.300 is too high"

"But there is a 24 second shot clock, and I'm guessing the Knicks don't typically give it to Tyson when the clock is winding down. They probably go to Melo, Smith, or Felton, who are forced to shoot or turn the ball over."

What you're talking about is heroball, and even ESPN has debunked the notion of the last second shot creator (see the article posted above concerning the Heat).

129 days ago

Hoopdon

Xavier,

Apologize for the grammatical mistakes, I accidentally posted an earlier version of the draft.

If we take into account that Sefalosha has an incredibly high % of his shots assisted, and plays on a team with Westbrook and Durant, then no, what he's doing isn't incredibly valuable.

This is where a lot of people have had problems with WP. Yes, Sefalosha is valuable, and much more so then mainstream stats give him credit for. However, he IS NOT the second best shooting guard in the world. His job is to sit in the corner and shoot the handful of opens shots he gets in a game. He makes a lot, which is valuable, but he's not doing much that couldn't be replaced, which is key in determining basketball value. (I'm not acknowledging defense, because neither does WP).

129 days ago

Hoopdon

Regarding my post, there is no logical fallacy. Coaches actually DO know more then fans.

I never said coaches should never be questioned. I said, before "would be" coaches open their mouth and complain, they should consider that just maybe, coaches know something they don't. Like that Drummond was likely to suffer an injury if playing heavy minutes. Or that McGee physically can't play for long periods of time.

I explained HDR on my blog, and have no interest in explaining it to someone who apparently can't read.

Really, trying to portray me as some kind of "spammer" is ridiculous. I have followed and commented on WOW for years, and even tried to bring WP to sites like ESPN and Bleacher Report. Most fans reject it, but I've made a few believers (commentators and columnists).

I started up my own blog to get my ideas down on "paper", and have gotten support from ESPN/Bleacher report columnists and commentators, as well as WOW columnists and commentators. If you don't like it, fine, don't read, but you don't need to act like a spoiled brat who had his favorite TV show insulted.

129 days ago

Chirag Agrawal

Xavier and Nathan,

"Have some of these commenters actually seen Chandler and Drummond play? Carmelo doesn't make things easier for Chandler, Chandler makes things easier for Carmelo."

Of course low-usage players like Jason Kidd and Tyson Chandler help Melo out, but I'm sure Melo helps both Jason Kidd and Tyson Chandler out as well. I don't know which one happens more than the other, and I don't want to assume anything without statistics either.

"Proponents of shot creation want to give all the credit to the wing iso guy because that's what they're conditioned to believe by Sportscenter. Yet when a low usage player scores, people want to give some credit to the Melo's and JR Smith's of the team. You don't see the inherent contradiction there? When Melo scores, it's all Melo. But when Chandler scores, it's because Melo was there."

I think you're taking my point to an extreme. I'm not saying Melo deserves ALL the credit for his baskets. And I'm not saying he deserves credit every time a low usage player scores. However, I do believe teammates can make a significant difference. Chandler's teammates sometimes deserve credit for when he scores, and the same goes for when Melo scores.

"What you're talking about is heroball, and even ESPN has debunked the notion of the last second shot creator (see the article posted above concerning the Heat)."

I gave a "last second shot" as an example; however, I did not mean for that to be the sole case of high usage players benefitting the low usage players.

My hypothesis would be that low-usage players often perform worse than they typically do without high-usage players on the floor with them, and high-usage players perform better than they typically do with other high-usage players on the floor with them.

I decided to test this just now, and I found that it was true. 5-man units composed of low-usage players tend to do worse than expected, and 5-man units composed of high-usage players tend to do better than expected. The relationship was statistically significant at a 99% level.

129 days ago

Patrick Minton

"Have some of these commenters actually seen Chandler and Drummond play? Carmelo doesn't make things easier for Chandler, Chandler makes things easier for Carmelo."

Ding. Ding. Ding.

I think there is a lot of synergy both ways. It's interesting that whenever Points Over Par praises a low-usage player, the chant from the crowd is that you couldn't possibly win basketball games with 5 players like that.

Which, of course, has two problems:

1) that's not what we are saying. To win basketball games, a TEAM needs to do four things: Not turn the ball over, shoot efficiently, rebound, and get and make free throws. That means, of course, you have to have some players who are good at these things.

2) it ignores the fact that you cannot win with 5 players like Carmelo Anthony. It's pretty appropriate that we're talking about the Knicks here, because trying to trot out 5 volume scorers is essentially what Isiah Thomas did. How'd that work out for you, Knicks fans?

I think Tyson Chandler is as good as Durant. That does not mean that I think 5 Tyson Chandler's would beat 5 Durants, because obviously the two poor Chandlers who had to bring the ball past half court would be screwed and commit about 25 turnovers a game. But I do think that 3 Durants and 2 Chandlers would destroy 5 Durants, because the two Durants who had to try to rebound and bang around inside would get mauled.

They have entirely different roles. Part of what Points over Par captures is that these different roles are:

a) equally important to winning
b) based on SKILLS because the data shows that players tend to perform consistently over time, not randomly as one would expect if they weren't skill based

The perception I am trying to fight is those of you who somehow are under the illusion that just because Chandler's role is different from Durant's, it is somehow easier. Which is, frankly, bullshit. If it were at all easy, lots of guys would be doing it, to grab the same huge 15m/year contract for doing what he does.

129 days ago

Patrick Minton

In other words, there are great low usage players (Chandler, Drummond, and YES, Thefolosha) and there are terrible low usage players (Stiemsma, Watson, Biyombo (who might get better because he's young))

Conversely, of course, there are great high usage players (Durant, James, Wade, etc) and there are terrible high usage players (Bargnani, Beasley, Norris Cole, Glen Davis).

Usage in a vacuum is a neutral statistic that means nothing and has no effect on Points over Par.

129 days ago

Chirag Agrawal

Patrick,

"I think there is a lot of synergy both ways. It's interesting that whenever Points Over Par praises a low-usage player, the chant from the crowd is that you couldn't possibly win basketball games with 5 players like that."

You're taking my point to an extreme, just like what Nathan and Xavier did. I never said anything about playing 5 Melos, 5 Durants, or 5 Chandlers.

"The perception I am trying to fight is those of you who somehow are under the illusion that just because Chandler's role is different from Durant's, it is somehow easier."

And I don't think Chandler's role is easier. I am saying high-usage players can sometimes benefit their teammates by attracting defensive pressure. Thus, while some high-usage players should be penalized for shooting at lower percentages, they should be given a boost for attracting defensive pressure.

Again, I found a statistically significant relationship at the 99% level demonstrating this. Lineups composed of players whose total usage sums to less than 100% tend to perform worse than expected, and lineups composed of players whose total usage sums to greater than 100% tend to perform better than expected.

Note: I still have Tyson Chandler as a top 15 offensive player when I factor the calculated effect of usage in (as determined by my study above). And I've never attempted to make defensive statistics for players, although I'm sure Chandler would likely be near the top there as well. So please don't exaggerate my argument to be claiming that Chandler is a mediocre player, or Carmelo is absolutely the best.

129 days ago

Hoopdon

Patrick,

Let's use a less extreme example. WP says a team of Jarrett Jack, Thabo Sefalosha, Antawn Jamison, Ed Davis, and Tiago Splitter, would defeat Kyrie Irving, Dwyane Wade, Carmelo Anthony, Blake Griffin, and Joakim Noah. That's just outrageous. They could play 100 games, and the WP team would win maybe 3-4.

Now I'm not saying the WP guys suck, what they do is valuable. However, guys like Thabo and Tyson have a pre-set cap on the volume of production they are capable of. Chandler gets a lot of his points off alley-oops. The knicks would throw that alley-oop every time if they could, but they can't. It takes a series of circumstances for Chandler to get that pass, and they can't be replicated more then a few times a game. Same goes for Sefalosha.

That is why incorporating % of shots assisted, like I do in HDR, may be a way to correct for some of that. For instance, HDR says Sefalosha was the 10th best shooting guard in the league this year. Clearly still a great player, and dozens of spots ahead of where PER ranks him, but not the SECOND BEST shooting guard in the world.

128 days ago

Patrick Minton

Hoopdon,

WP says no such thing. That might be the stupidest straw man argument I have ever heard. For one, you pick two guys who have traditionally sucked or been MEH/average (Jack and Jamison) and who happen to be good this year over a 40 game sample. You really want to pick better sample sizes if you are going to argue with a statistician. I mean, you could have gone with Rondo, or Gerald Wallace, or Matt Barnes, or someone that WP has traditionally "disagreed" with conventional wisdom, and instead you pick those two losers? GTFO.

For another, the "5 Durants/5 Chandlers" argument is one I used to illustrate a simple point that WP does not claim that there are no teammate interactions or that roles are not important. Seriously, this logic is beyond annoying. Read the last chapters of Wages of Wins and Stumbling on Wins, which talk at length of the role of coaching in identifying WHY the metric values certain players.

Bottom line: high usage or low usage is irrelevant: shooting efficiency is what matters. If you shoot efficiently, you are good, and that's if you shoot 6 times a game or 26. Generally if you shoot well, the more you shoot, the better the WP formula will consider your contributions (which should be obvious to anyone with junior high math skills).

128 days ago

Hoopdon

I agree 100% with what you say on the Durant/Chandler argument, and have thought for some time that its a stupid argument to use against WP.

You didn't really address my question though. You just admonished me for using bad examples, even though they aren't, since WP says they are good.

You said despite having higher WP's, team 1 would lose to team 2. I agree, however you hinted at roles and teammate interactions as a cause. What exactly would these be? I mean, that makes sense for the 5/Durants/5Chandler's argument, but the teams I used as an example each have a PG,SG,SF,PF, and C. Why wouldn't they be successful against an inferior WP team?

128 days ago

Kai Xiong

@ Hoopdon, Regarding your 5v5 WP vs ESPN team, I believe Patrick was trying to say (in a rather mean and unnecessary way) that some of those players were having statistical anomaly years. While WP48 is fairly constant over time, as with anything were chance plays a roll, certain players will be exceptions. Jamison has historically been terrible but is having 1 good year. If you took players with say high rolling 5 year average WP vs a team with 5 year high rolling ESPN projections, then yeah, WP would win that game. I mean, I think a team of Calderon, Kidd, Kirilenko, Farried, and Drummond (or Chandler) could give Kyrie Irving, Dwyane Wade, Carmelo Anthony, Blake Griffin, and Joakim Noah a run for their money.

128 days ago

Kevin Sawyer

It might be hero ball, but when the entire team buys into the concept of a hero, it means the hero is going to wind up choosing between a bad shot or a turnover.

It isn't just situations in which the shot clock is winding down. Plays often break down with 10-11 seconds left in the shot clock, often leaving a wing with the choice of a shot he'll hit 2 out of 5 times, or putting his team in a bad situation.

By my lights, this is bad coaching. Teams like the Spurs have guys pop out to the corner for a high-percentage three. When Flip Saunders coached the Pistons, the team was able to work deep into the shot clock to find options.

I could see where this phenomenon results in a player like Sefolosha being overvalued and a LeBron James being undervalued (which I actually think he is by this metric).

However, I wouldn't overstate the effect. It wouldn't be enough to put Carmelo Anthony ahead of Tyson Chandler. After all, we are talking about maybe 3-4 shots per game, with those shots going down 4 times out of 10 instead of 5.

Further, whether or not the player like Melo is a victim of the way teammates perceive him, it certainly isn't beneficial that he takes bad shots. So this doesn't get advanced stats' critics any closer to what they really want to argue, which is that points scored is a great way to assess players.

128 days ago

Kevin Sawyer

"You didn't really address my question though. You just admonished me for using bad examples, even though they aren't, since WP says they are good. "

WP says they have been good for the last 40 games. For their careers, WP says they are not good.

128 days ago

Patrick Minton

What Kai said. The "role" players he picked would be a very, very good team (although, I'd maybe sub out somebody like Tyson Chandler for Drummond because of the sample size involved. Drummond looks like a beast but it's only been 45 games or so).

Basically, Hoopdon, you cherry picked 5 low usage guys that aren't even that good, 2 of whom are actually bad players having a 50-game career stretch. "WP says they are good" is at worst ignorant of how the formula works, and at best an outright lie.

128 days ago

Charles Connaughton

I think there's a lot of misunderstanding about what low usage actually means. In the case of Tyson Chandler, we're talking about a guy who takes 3-4 less shots per game than the average center. There's not a backcourt in the NBA that's going to struggle to get up 3-4 more shots per game - without even mentioning that they could basically throw the ball into the stands with those extra shots and *still* have a net offensive boost from how ridiculously efficient Chandler is.

Usage is not a particularly rare skill. Sure, if you put together a line-up with nothing but low usage guys, that team is going to have a problem, for the same reasons a team of all centers would have a problem. If you have a couple of those guys on your team, however, you're not going to have any trouble whatsoever going out and signing wing players who like to shoot a lot. Sure, a guy like Sefolosha really wants to be surrounded by high usage wings like Durant and Westbrook. At the same time, if you have a pair of high usage wings, you *want* a high efficiency, low usage guy to round out that offense. What's the cost of parking him in the corner to threaten corner 3s if you don't already have a guy doing that?

128 days ago

Jon C

Xavier,

I think calling out HoopDon for mentioning his blog wasn't entirely fair. It is a heavily WP blog. And he has done some work (even if it may not yet be as rigorous as work done on Wages of Win and this site) that readers of the comments sections of both of these sites would find interesting.

I will speculate something. While not shooting bad shots is obviously an easy skill in that any specific player in the league could achieve this if given the right motivation. "Beasley, we have put four snipers in the rafters. If you shoot the ball from over 15 feet from the basket at any point in the game, they are instructed to take you out." However, since Points Per Game is a large factor in how much you get paid. So finding players that are willing to take fewer attempts at the risk of not getting paid is probably the rare "skill". It is an attitude and mind set. And I think guys take a significant risk of being overlooked if they go down the high efficiency and low usage route. Chandler is very lucky that he got a reputation as a defensive force because he is unable to block shots and rebound at a high level. Now we know that he is actually an offensive force. And I'm sure everyone here is familiar with the fact that when he joined Mavericks it is the Maverick's offense that actually improved significantly and not the defense. And while I don't think we saw that same effect last year with the Knicks. We can look at them this year and see that they are an offensive powerhouse team and simply average at defense. It seems unlikely that they would be so exceptional an offense if they guy second on the team in minutes played (Chandler) were not

128 days ago

Julien Rodger

Very nice article Patrick. The evidence for Tyson Chandler being a true star player is overwhelming. Not only are is WS and WP numbers amazing, but the contextual evidence of Dallas' improvement when he joined, the fall-off when he left + New York's improvement adding him, is pretty strong. New Orleans in 08 also played like a team with a 2nd star in Tyson. Finally we know now that Chandler is actually a very rare type of player statistically.

Furthermore from a pragmatic level, I wanted to share how I personally judge players and talent. To me a player can impact a game physically, impact a game with a skill player or can impact it with his IQ/feel. Chandler impacts the game physically by his fierceness rolling to the rim and impacts it with his IQ and sense of angles and space. He's not the most skilled player in the league, but for a center he's somewhere between average and decent for skill, due to how excellent a finisher he is. So in my 3 categories he's near elite in 2, and average to good in the other. This is an elite combination. For example Steve Nash and Dirk Nowitzki in their prime were elite in 2 categories (impacting the game with skill and IQ) and only average in the other (impacting the game physically). This doesn't sound too far off from Chandler's split, mind you they're skill and IQ was a more impressive combo than Tyson's physical impact and IQ combo, IMO. Guys like Lebron, Durant and Harden are elite in all 3 categories so I'm not going to be bullish putting Tyson in that category, but I see no problem putting him in the top 15-20 players in the league

128 days ago

Julien Rodger

What is also comes down to is that C is a unique position. For example in the above post I presumed that Chandler has a great combination of ways to physically impact the game + IQ and an average to good skill level for a C. But at C, this 'average to good' skill shows up much less in the statsheet than at other position. Compare it to the PG position. Compare him to Derrick Rose

Derrick Rose makes most of his plays with his physical tools. He's an incredible mix of speed and strength and slashing. Tyson Chandler is not as impressive physically for a C as Rose is for a PG, but he's not *that* far behind, he's the next level back.

Derrick Rose has a decent IQ for a PG. Tyson Chandler has arguably a better IQ for a C than Rose has for a PG. Nothing against Rose's IQ, but Chandler is one of the smartest Cs in the league and uses it well.

Derrick Rose has skills, but he is not one of the most skilled PGs in the league. By that I mean there are a lot of PGs who are more consistent 3pt and midrange shooters than him. Rose is a player that you live with his jumpshots but don't love him taking them. Tyson Chandler has skills for a C, but isn't one of the most skilled Cs in the league. He can finish around the rim extremely well for a C, but lacks perimeter range and post play.

Using this perspective it becomes a lot easier to see why Chandler may be as comparable a talent at C as Rose is at PG. Both players make a lot of their bread on their physical tools, with Rose having the edge over Chandler compared to his position. Chandler however makes more bread off his IQ and feel than Rose does, the latter is decent in that area but Chandler is a level ahead. So far it looks even. Where it gets interesting is in skill. Rose's offensive skill level looks more impressive, but he's also playing at the most skilled position while Chandler is at the least. In reality, COMPARED TO THEIR POSITIONS, Rose's natural skill level for a PG is arguably in the same category as Chandler's compared to Cs. But the advantage Rose has, is that by playing PG, he gets to put up the raw ppg and apg numbers because he has the ball the most. Interestingly, this also involves a trade-off. Because Rose plays a position where one is expected to dominate the ball and FGA more than at C, he gets to put up more volume ppg, but it comes at the COST of efficiency. Chandler gets to play a role where he gets to put up better efficiency, but it comes at the cost of volume. If PPG and TS% were used equally, this trade-off would be accepted as an even deal. But not only are they not used equally, but volume scoring has a far greater sway than TS%, and thus Rose is an MVP winner and Tyson is considered a fringe all-star

127 days ago

Xavier Quach

Jon C,

I cited Hoopdon specifically because it was in this thread, but he's not the only one to post links to their own blog. Most of what they write would be easily reproducible in the comments. And they KEEP referencing it, over and over, despite the flaws in their analysis. Somehow they believe since it's on the net, that it's validated. I don't know that specifically about Hoopdon, but I was making the statement in general.

Hoopdon,
"I never said coaches should never be questioned. I said, before "would be" coaches open their mouth and complain, they should consider that just maybe, coaches know something they don't. "

That's kind of the definition of argument from authority.

As for HDR, again I read through your reasoning for changing the value of missed shots and it raised concerns about your methodology. I suggested statistical analysis to determine what your coefficients should be, and somehow that makes me a spoiled brat? As in, being unable to accept criticism of your pet project and lashing out in retaliation? Regardless, this isn't the forum to be discussing it, so I'll stop.

127 days ago

Xavier Quach

Chirag,

I don't think we're as far apart as it initially seemed. Heroball as a concept isn't just last second shots. It's giving the ball to Melo and letting him iso from 20ft out. It's expecting your high usage guys (HUGS) to take over games and bail out the rest of the team. The Heat changed their whole offensive philosophy and it won them a championship. They have two of the most talented high usage, iso players in the league, and they went to a pick n' roll and passing offense.

We all seem to agree on player synergy, and that low usage guys (LUGS) and high usage players benefit from eachother. But the discussion is always how lo LUGS benefit from HUGS and not the other way around. If Sefalosha isn't sitting in the corner waiting for the 3, then the defense can collapse on the lane and Westbrook scores nada. So why is it that Westy gets credit for freeing up Thabo, but not the converse?

Synergy can't really be tracked with readily available stats. Assists are inconsistently tallied and aren't a great measure anyway. Hockey assists and screens leading to a score would go a LONG way to showing people how much more valuable role players are.

127 days ago

Martin Parodi

Great post patrick. What you people seem to be overlooking is the fact that in my perspective WP is not measuring how good durant or chandler are, its measuring how good are they in what they do in the court

127 days ago

Hoopdon

Great distinction Martin, your last statement is key.

Instead of typing pages for a comment, just read my latest post on this exact subject. It discusses the difference between inherent value, and on-court production. The two overlap quite a bit, but there are certain cases where they aren't the same.

Xavier, you have hurt my feelings, I will not speak to you until I can compose myself.

Patrick, I think sometimes we (I in this case) get caught up in what we want WP to be, rather then what it is. Its not a magical number that tells you who the "best" players are. Its a magical number that tells you who is playing the best. The what, but not why, as Dre says.

127 days ago

Adam Tarr

Chirag,

You mention an actual study you did about the performance of high-usage lineups versus low-usage lineups. Do you have a link?

I tend to believe, instinctively, that a lineup composed of players whose usage rates add to less than 100% will tend to shoot less efficiently than their usual rates, since some of the late in the shot clock, forced shots will fall somewhere.

This is at most a very narrow criticism of WP, though. It's an attack on the idea that WP can explain a certain variation for players placed in different environments, as opposed to attacking the idea that WP accurately explains what a player is doing well (or poorly) in a given environment. If you're right, Chirag, it's an interesting insight for sure, but I'm not sure what, if any, changes you would want to make to the WP calculation to accommodate this insight.

I think the mishmashed team data that WP uses to calculate the impact of assists and (especially) team defense seems like a much bigger flaw in the WP calculations than this. Those calculations implicitly assume that players share the court with their backups, which means they are getting assists from each other and are credited with each other's defensive performance.

127 days ago

Chirag Agrawal

Adam,

I think the WP stat is wonderful. After being exposed to WP, I realized how much shooting inefficiently hurts. I do not think WP is perfect, however, and I'm sure the creators do not think so either. I typically use Win Shares instead of WP, primarily because of the team defense adjustment that you mentioned. However, I don't want to come off as if I'm criticizing WP as a bad statistic.

As for the study, I have an excel file I can email you. Right now it's very basic, but I'd like to flesh it out before uploading it anywhere. There is a weak, but statistically significant, relationship that says lineups composed of players whose usages add up to less than 100% score less efficiently than usually (and vice-versa for high-usage players).

127 days ago

Adam Tarr

Chirag, go ahead and polish it up, then link to it. I look forward to seeing the results.

Just for S&G, here's an effort to come up with a version of Hoopdon's somewhat cherry-picked stats. I tried to find a 1-5 where the WP player at each position has done better than the high usage player BOTH this year, AND over his career. I tried to avoid WP players who have gotten a lot of positive press (no Faried) although I included Sefolosha.

So... position, player, career WP48

PG Calderon .203
SG Sefolosha .172
SF Kirelinko .258
PF A. Johnson .205
C Splitter .200

versus...

PG Parker .121
SG Bryant .162
SF Pierce .167
PF Aldridge .099
C Bosh .135

As I understand it, WP would predict that team 1 beats team 2 roughly 60% of the time.

127 days ago

Marshall Rader

Patrick, first of all, great article. Second of all, I have to preface what I'm about to say with the fact that I only read about the first half of the comments on this post, so my apologies if someone has since said something along these lines...

I agree with you that guys like Chandler and Drummond are, in fact, equally valuable to guys like Carmelo, and that just because they play a different position and role on the team doesn't make them any less valuable. But, like you brought up with Durant vs. Chandler somewhat, we all know Durant (in that example) is the more talented player. He just plays a different role on the team (to put it way too simply - the volume scorer), so obviously there may be facets of his game that aren't as efficient as Chandler has been (although don't quote me, because in this example, I wouldn't be shocked if Durant was still just as efficient because he's that good). But anyways, the point I'm trying to make, is that while the two types of players may have similar impact on the game/value to their team, the volume scorer could theoretically play a different role, let's say the Chandler/Drummond role, and excel just as they do.

Yes, it's a common perception, and frankly a myth, that it's EASY to do what Chandler does, like you said. If it were so easy to just catch the ball 1 foot away from the hoop, turn around, and lay it in or dunk, and therefore make 70% of your shots, then every big man in the NBA would be doing this. Clearly there's more to it that maybe we as fans struggle to appreciate. However, who's to say that there aren't plenty of players out there in the league right now that are simply playing the "wrong" role, or perhaps shooting the shots they aren't that good at way too often instead of the shots they are actually good at (think Josh Smith and his spot up shots from deep).

A recent example of uber-efficiency in this sense comes to mind to help prove my point (or at least explain what I'm getting at better). Remember LeBron's 12/13 game from a week or so ago? If I recall correctly, he only missed 1 shot, and all of his shots came in the paint (maybe even all came at the rim, I can't remember). Basically LeBron didn't shoot any midrange jumpers or three pointers. And the result speaks for itself. Obviously it's just one game, but it made me think about this concept, and then I saw your post and it reminded me.

For someone as talented at basketball as LeBron James, and especially for someone who also is elite at scoring at the rim, like James (correct me if I'm wrong, going from memory here), what's to stop them from making a conscious effort to ONLY score in the paint? In a basketball strategy (and not stats) standpoint, guys like LeBron need to shoot the occasional jumper to keep their opponents honest, even if they only score at, say, an average or slightly above average rate. If they never shot the ball from outside the paint, defenders could sag off on them and make it more difficult for them to score in the paint, and it would probably have a negative impact on the team's floor spacing as a whole. So obviously I think we can all agree that for a guy like LeBron, it makes sense to shoot some jumpers here and there, at least to some extent. And surely if it's a truly wide open shot.

But to what extent would defenders really be able to sag off and "cheat" on a guy like LeBron if he made it clear he only wanted to get to the rim? LeBron has been able to score at a very high efficiency while maintaining a very high usage rate, and he's done so in the paint and from deep. He's really damn good at basketball, we know this. But what if he shifted his role to that of someone like Tyson Chandler's? Maybe add in a handful more shots per game than Chandler gets, because it's LeBron after all. And maybe LeBron is getting the ball at the top of the arc, and simply running an iso play and taking it to the hoop, which is obviously different than the put backs and alley oops that Chandler lives off of, but I don't think that's the point. Could LeBron have an even greater impact on the game if he only took shots in the paint, even if it significantly decreased his usage rate? Especially on a team like the Heat who have several other capable players, including guys like Wade who can take on the high usage rates already?

Hopefully you got what I'm trying to say, I know I rambled a bit. And to be clear, I'm not necessarily 100% in support of the theory I just described, it's just been a thought in my head for the last week or so, and I wanted to get your thoughts. Even if there's some credence to it, it might not be that relevant, because it might only apply to the elite players like LeBron and Durant, or who knows. So yeah, let me know what you think! Keep up the great work!

126 days ago

Charles See

Patrick, I'd like to see your response to Chirag. His study is pretty important (and is something that, intuitively, makes a lot of sense).

126 days ago

Charles See

For those commenters interested, I think Kirk Goldberry's posts on the spacial element to FG% is pretty interesting and illustrative of how shot location can be such a key determinant of FG%.

Tyson Chandler is a pretty interesting case: he shoots exclusively at the rim, and excels spectacularly at that one area. It's a really valuable skill! But he's useless away from it. Now, that doesn't really matter for a center. Your team can absolutely survive having one player that's only an offensive threat near the rim. But would I want TWO Tyson Chandlers on the floor simultaneously, one at PF and one at C? No. My spacing would be terrible, and the offense would bog down horrifically. (See: the 2012-2013 Memphis Grizzlies)

My point is this, and I think other posters have alluded to it: player contributions do not occur in a vacuum, but in the context of a larger 5-man lineup. Chandler's work is incredibly valuable, but throw a number at him and say "he produces X number of wins in 48 minutes" is just silly, and this is particularly true for players with very specific skill-sets (and limitations.

126 days ago

Shawn Furyan

Just got around to reading this monstrous thread, but it still seems to be somewhat active, so here goes...

@Patrick - Way to actually evaluate a long standing argument against high efficiency low usage guys, giving strong evidence that it is completely bogus.

@Xavier - Love the acronyms LUGs and HUGs for low and high usage "guys". Especially since it led to the phrase "LUGS benefit from HUGS".

@Hoopdon - I'm going to have to second Xavier's criticisms of your excessive publicizing of HDR (did you know that this is already a very commonly used acronym -- High Dynamic Range?). I took much more than a "cursory glance" at your blog and came to similar conclusions about the metric. Part of the problem is that you don't seem to be specifying/detailing your metric, so referring to how players are ranked by it in the comments of every Wins Produced-oriented blog post lends absolutely nothing of substance to the discussion. I mean, if you made a gross error in calculating HDR for a given player, it would be impossible for anyone else to even catch that, and such miscalculations are far from uncommon. So we can't even trust that the numbers and ranks that you give accurately represent HDR's output. What's more, is that what is out there makes it seem unlikely that the metric is very good. To wit, a couple weeks ago, you wrote a comment here: http://wagesofwins.com/2013/02/03/wages-of-wins-superbowl-sunday/ that indicated that you placed a defensive penalty on Calderon which knocked him down a couple places. This is bad, because Wins Produced captures all of defense. It means that you are double counting. It means that you're no longer just monkeying around with the weights, but now have accounting issues with your metric. If you don't think that Wins Produced handles defense correctly, then you have to undo the way that it handles defense to come up with an at all credible result. The other thing is that you ARE monkeying around with weights. Those weights were discovered, not chosen (this is a gross oversimplification, but the full story has to do with explaining the system first, and only then testing the explanation, this is what makes a model empirical and objective and what guards against bias; incidentally HDR seems to be ABOUT your biases rather than attempting to control for them). That doesn't necessarily mean that the original weights are unassailable, far from it in fact, but it does, in my view command a little respect, and just eyeballing new weights is not very respectful of the work that went into the original metric. In my estimation, you are PER-ing up Wins Produced, which is anathema to Wins Produced itself, as well as its creator's intent. In light of all of this, your comments do feel like spam. You publicize the "product"* a lot, but the product appears to be of low quality. These are two of the defining characteristics of spam, so I don't think that it's an unfair characterization, and would say that it is far from "ridiculous".
All of that being said, I'm not trying to be mean, or discourage you from your pursuits. You certainly seem to have a lot of energy that you want to direct toward evaluating basketball. If you cared for my advice, I would say that it would probably be a good idea to dial down the promotion of HDR for now (can we consider it HDR 1.0?), publish the algorithms you use, and seek criticism of your methods. If your model is thoroughly vetted, then you'll end up being able to defend each of the metric's components. Or maybe you won't. Seeing criticism might discourage you to the point that you give up on the metric. But you seem to want to be a part of the player evaluation discussion, and in my mind, being open to criticism is the cost of admission.

*I'm using the word "product" in order to help the analogy. As far as I know, you're not benefiting financially from HDR, so I do not intend to levy a criticism to that effect.

126 days ago

Hoopdon

Excessive publicizing of HDR? Yet you found only one article where I promoted it? I can leave up to a dozen comments a day on various NBA sites, and you found one where I mentioned HDR? Well done.

There is a reason, as I detailed in the blog, why I have only done the HDR for 50 players in one season. I am continually tweaking and researching the weights, as well as determining how well of a predictor it can be. I have hardly rolled out seasons full of data, and challenged Wins Produced to a sword fight. I use Wins produced for quick player evaluations, not PER, HDR, or anything else.

Um, Wins Produced hardly accounts for defense at all. They have a small team defensive adjustment, nothing more. I haven't incorporated that in HDR, and THEN used Defensive Rating. Seeing as how I haven't even made the formula public, I would appreciate you refrain from telling me what's in it.

Monkeying with weights? The only values I've mentioned that I changed were the values of FGA, and Offensive Rebounds. I explained why in my blog, and spoiler alert, it wasn't because I felt like it.

Its been quite a while since I have mentioned HDR, as I am in the process of determining how well it predicted team success.

Dave Berri himself said that if you had an idea about a player metric, to stomp bombing the boards of WP and NBAGeek continuously, make the metric, and see what it does (so you can stop the"it insulted the creators intent"). I haven't taken cheap shots at any other metric, (besides PER, because well, everyone does it) so I would appreciate the same, given I haven't even done, or claimed to do much with it.

I do however appreciate your feedback and criticism. Once I finish the HDR project, make sure to check it out, and see what's what. Until then, don't criticize it, and I will stop publicizing it.

126 days ago

Kai Xiong

@ Hoopdon, You should have ditched the whole defensiveness (wrote it out to vent, then deleted it) and just gone with the last sentence. Just sayin... In either case, I'll be looking forward to seeing a finished "product" and dissecting it. I feel that WP is lacking but hey, its the worst metric besides every other metric I've ever found. Cheers!

126 days ago

Adam Tarr

"I feel that WP is lacking but hey, its the worst metric besides every other metric I've ever found."

That's a great way to put it.

126 days ago

Patrick Minton

Adam,

Yes, I would bet on Team 1 to beat Team 2.

125 days ago

Guy Molyneux

It seems that many low-usage/high-efficiency players receive assists on a high proportion of their shots. But unless I'm mistaken, WP simply assumes all players on a team benefit equally from teammate assists. Now that this data is available, shouldn't you account for that? In Chandler's case, last year he was assisted on 81% of his shots, compared to 56% for the Knicks overall. Doing some back-of-the-envelope estimation, if we debit Chandler for the assists on an additional 25% of his FGA, at .5 points/assist, this would reduce his WP48 by a fairly considerable .025. This seems worth accounting for, no?

And of course, other players benefit from assists much less than their teammates. This year Kobe is assisted on 37% of his shots, vs. 60% for LAL overall. Accounting for this would raise Kobe's WP48 by something like .040, meaning that his value above an average player is twice what WP currently estimates. Again, this seems substantial enough to take account of.

124 days ago

Kai Xiong

@ Guy, Why would that be necessary? Again, we are not going to ever be playing 5 Chandlers. That's not the point? You don't think you could get someone to throw lob passes to Chandler? Ray Felt's got to be like 28 out of 30 in terms of starting point guards only ahead of Mario Chalmers (who thinks he's top ten my god that's funny) and Jamaal tinsley atm. Seriously and he's just fine at tossing that ball up at the basket. Hell even Will Bynum had no problem throwing passes in the general direction of the basket. It's not throwing that pass that's difficult. It's diving to the basket, jumping high, and dunking successfully.

Interesting case study right now in the Lakers, Dwight off of back surgery is clearly no longer able to do the same things physically that he used to (slightly slower, less lift, clumsier, etc). You think Steve Nash is a worse passer than Jameer Nelson? Let's just end the conversation about that at that. Assisting is great and I see the logical behind your reasoning but I don't think there needs to be any debiting. Fundamentally some things that seem difficult may not be and some things go unappreciated (this is one of those cases). I bet if Kobe decided to stop ISO'ing and just cut more, he'd be better. Wait a minute: see 2012 Kobe shot chart vs 2013 Kobe shot chart.

124 days ago

Guy Molyneux

Kai: Are you saying that assists have no value and should be ignored? Perhaps that is so (though it seems unlikely), but that is not how WP is constructed. WP says that assists do have a considerable amount of value, and so players who benefit from assists should not receive "full credit" for their FGs. My point is simply that if you are going to do that, you should debit the players who in fact benefit from the assists. I don't see what the argument against that would be.

124 days ago

Guy Molyneux

Proportion of FGs that were assisted this year:
80% Sefolosha
75% Collison
54% Durant
24% Westbrook

How can it possibly be correct to assume that players on a team benefit equally from their teammates' assists?

124 days ago

Hoopdon

Kai,

I sort of referenced this at my site. There isn't anything inherently wrong with having a high% of your shots assisted. However, that does tend to correlate with players who have a pre-set cap on how much they can produce.

Take Chandler. He is stupid efficient. Looking at his numbers, you would say, hey, he needs to be shooting more then some of these less efficient guys. The problem is, the type of shots he gets are only available a hand full of times per game. If the Knicks could throw a lob to Chandler every time, they would. Just like if OKC could pass to Sefalosha for a wide open 3 every time, they would. They can't, those circumstances are fairly rare. Including % of shots assisted could help account for that, and yes, slightly devalue such players.

I did this with the 2011-2012 HDR Ratings. WP48 said Sefalosha was the 9th best SG in 2011-2012. HDR said he was 11th. Not a huge difference by any means. Regardless, I think the idea has merit.

124 days ago

Kai Xiong

@ Guy Molyneux. I am absolutely not saying that Assists have no value. Where did I say that? I'm saying is scoring the basketball or getting open or taking high percentage shots is harder than throwing someone the ball. An assist by definition are successful FGs. One player gets credit for the assist and one player gets credit for the FG, not some Assists FG hybrid. Each player getting credit for their end of the bargain is fine by me.

I think you are approaching this the wrong way. What I am saying is you have people who are more than capable and more than willing to pass to you if you are in a position to score (ei Open). I believe Drummond is assisted on a high percentage of his shots and hes with some of the biggest scrubs playing with him. Even Will Bynum, Kim English, Jerebko, Villanueva can get him the ball, why would I assume that its more them and less Drummond? Do you see the point I'm making? Being assist a lot is good! Can we agree on that? Players shoot better when they are assisted so why are we assuming it's bad to be assisted often. Lebron started being assisted more and he's went to another level of play. ISO's are bad, it's not the late 90's anymore.

124 days ago

Kai Xiong

@ Hoopdon. Sure, hypothetically this is true. I agree, Chandler has a cap on what he can and cannot do. I'm sure everyone agrees on this. However, he's smart about it. Being limited isn't that big of a problem because he does his role and he does it well. A 1 trick pony isn't a problem if that trick is amazing. Honestly, I think we can agree that James Harden isn't a 1 trick pony, he's a 3 trick pony but by being a 3 trick pony he can create more wins than Kobe even without the flashy post moves, the contested fadeaways, the wannabe Dirklegs. Making good decisions should be applauded. Think about all of the untapped potential and what could have been if players overcame their only deficits. There are reasons why Vince Carter, Beasley, and JR Smith are unstoppable in video games, but just meh in real life.

124 days ago

Guy Molyneux

Kai: You're making this too complicated. If assists have value, then the accounting of WP requires us to adjust the value of FGs (or else things won't add up). The only question is whether we should A) adjust the value of FGs based on whether an assist contributed to that score, or B) adjust the value of all FGs equally, and ignore the information we have on how often players are assisted (as WP now does). Are you really arguing for B, and if so, why?

This isn't about assists being "good" or "bad." It's simply a matter of sharing credit accurately for assisted FGs. And if you feel WP is giving too much credit for assists (because so many of them are easy lobs that anyone could do), that is really a separate issue. My argement is simply that whatever value you want to assign to an assist, let's link that to the scorer who actually benefits, not to scorers who make unassisted FGs.

124 days ago

Adam Tarr

I agree completely with Guy. Currently, WP's calculations just use team data when determining assisted FG% rate and defensive performance. Using the (now available) more accurate data on exactly who is getting and giving assists, and for who is on the court when the team defense is good or bad, are the most obvious ways to make WP more accurate.

These aren't philosophical changes to how WP is calculated. Read http://wagesofwins.com/how-to-calculate-wins-produced/ and you can see that sharing credit for assisted FGs and considering team defense when the player is on the court are already considered. The only adjustment needed is to actually use the data on how often a player got assisted and how the team did defensively when he was on the court, instead of approximating this by assuming every player on the team is identical.

124 days ago

Adam Tarr

And Patrick, FWIW, I agree that I would pick team 1 to beat team 2. Despite believing that there is some value in having a couple guys who can produce a lot of unassisted FGs on a team, the WP difference between those two teams is pretty vast. Even if team 1 under-performs its expected TS%, it's still going to murder team 2 on the glass.

124 days ago

Kai Xiong

@ Guy, if every assist has an associated FG and the assist itself is assigned a regressed value, wouldn't that implicit benefit from the assist already be captured? If you take a portion of every assisted FG's value and more it to the assist, wouldn't you just be increasing the value of assists? I believe assists was regressed to wins already. Wouldn't this break the entire model? I guess the key would be probably rather than reallocating assisted FGs, we should be examining unassisted FGs. I'm not 100% of this, but wouldn't eliminated all of the assisted baskets eliminate somewhere around 60% of all made baskets and leave bunch of misses and a miniscule amount of makes? How would you even being to quantify an "assisted miss." Where would you get the data for that?

124 days ago

Guy Molyneux

Kai: If you read "How To Calculate WP," you will see that there is already an adjustment made for assists. My point is simply that this adjustment should account for the fact that some players receive more help from assists than others, rather than assuming that all players on a team receive (roughly) equal help. That would not be a major structural change to WP, more of a tweak to improve accuracy.

This will tend to reduce WP for lo-usage/hi-efficiency players. For example, fully 95% of Battier's shots are assisted this year. Could these players make as many shots without help? Patrick's comment above seems apt: "I can only presume that they don't do that because they can't."

Conversely, high-usage players tend to benefit far less from assists. Many object to the term "shot creation," so let's just say that these players tend to receive less assistance from teammates when they make FGs, and let's account for that correctly. This really shouldn't be controversial....

124 days ago

Kai Xiong

@ Guy, I have just refreshed myself on the formula and you are correct in that the assist adjustment is indiscriminate. I could see how this would be a reasonable effect to be tested. Whether it would affect the overall picture very much is still to be seen. On average teams have an assist % of 50-60 so it would only affect a little less than half of shots.

124 days ago

Guy Molyneux

"Whether it would affect the overall picture very much is still to be seen."
Well, the exact impact on each player would of course have to be calculated. But we know the impact would be substantial. Roughly speaking, it might lower the WP48 of a lo-usage player by something like .025. And for a hi-usage player it might increase his WP48 by something like .040. It wouldn't make bad players into good ones, or vice-versa, but it would mean a non-trivial change in the value of many important players.

124 days ago

Kai Xiong

"Roughly speaking, it might lower the WP48 of a lo-usage player by something like .025. And for a hi-usage player it might increase his WP48 by something like .040." Is that using your randomly assigned value distribution? Why should we assume its something as large as .5 pts/assist (or as small). You would have to do a detailed analysis where you compare assisted FG% vs unassisted FG%. Again the issue is, where are you going to get data for an "assisted miss." Also with Chandler as an example, if he only takes lobs and put backs wouldn't his unassisted FG% still be pretty high? So why do we deduct so much again.

123 days ago

Jack Adams

I think that a majority of assists baskets occur hen te recipient is open. Why wouldn't they get credit for getting open? At worst they should get partial credit (along with screen setters and play callers and the passers themselves) for getting open. This would mitigate some of the benefit they "receive" on their assisted baskets. But to what extent is the question. I think that the shift in WP48 might be much smaller than you admittedly off the cuff theorized. It's an enormous amount of work for an interesting issue with a tiny payoff (in my opinion). I tend to think that ray Allen standing in the opposite corner deserves some credit for keeping his defender from helping in the paint on the lebron/bosh pick and roll, which is an incredibly difficult thing to quantify, and I can't prove it anyways.

123 days ago

Hoopdon

Jack,

The issue here, is a lot of these players aren't open because of anything they do/possess. Ray Allen is open because teams would rather give up an open 3 from Ray, then an open dunk for LeBron/Wade. Therefore, it would seem fair to give Ray half the credit for hitting that open shot, and LeBron/Wade half the credit for making that pass.

Another issue is there are different types of assisted shots. Slashers like Wade actually DO create their own scoring opportunities, mainly through speed, agility, and knowing when to back-door/random cut a defender. These types of players in theory, should NOT have their assisted shots devalued. Guys like Ray Allen, who stand in the corner waiting for a pass, SHOULD.

Without better data, its difficult to account for all the error anyway. Hockey assists are a huge part of the NBA, but aren't tabulated in the boxscore. Knowing what types of shots players are getting assisted on is also a mystery.

123 days ago

benjamin durham

the flaw with all these arguments is the assumption that being assisted is bad. thats not true at all. not everything a player does on offense requires having the ball. cutting and spotting up and a hundred other basketball activitys are just as much a skill as a "sick crossover"

123 days ago

Jack Adams

Hoopdon,

The ray Allen argument you made seems pretty plausible (w/o data I have no concrete idea). I think while Allen may not deserve as much credit for getting open, that deficiency is made up by the mysterious and elusive floor stretch unicorn. In my example, he gets a piece of the assist pie for occupying the attention of a potential help defender, allowing bosh to roll to the hoop sans impediment and receive the pass for a dunk/layup. Ideally, the quality (location, timing, etc) of the pass would be judged as well, with this data being used to help predict the passes that result most often in assists. But I'll wish in one hand and spit in the other on that one.

123 days ago

Guy Molyneux

Kai/Jack: Accounting for assisted FGs is not complex. Let's take Westbrook as an example. WP now debits Westbrook 96 points, or 3.12 wins, for his teammates' assists. (He receives even more credit for his own assists too, of course, but the issue we're discussing here is the debit applied to shooters. ) That assumes he benefits an average amount from his mates' assists. However, we know that isn't true -- only 24% of his 420 FGs were assisted, which means only 49 of his points were actually produced by his teammates via assists (WP values assists at about .49 points). Crediting Westbrook with the difference, 96-49=47 points, increases his WP by 1.5 and boosts his WP48 by .038. That would increase his WP48 from .129 to .167, a pretty substantial change -- more than doubling his value above an average player.

123 days ago

Lexie Barza

I think we can all agree to acknowledge the fact that the "assist" statistic (and therefore, the % of baskets on which one is assisted) is a remarkably fickle and subjective stat - similar to the "error" statistic in baseball.

Dre gets into that a little bit here and quotes Dean Oliver referencing how "dirty" the statistic is: http://wagesofwins.com/2013/02/15/measure-once-predict-twice-thoughts-on/

It's pretty hard to get an objective sense for which player deserves credit (assister vs. assistee) when the statistic we're relying on is so unreliable.

123 days ago

Xavier Quach

As far as assists go, the conversation seems to be that having a high % of your shots assisted means that someone is helping you, getting the ball in scoring position, and thus deserves some of the credit for your FG. People are not considering that the converse may also be true, that the having a big target like Tyson Chandler makes it EASIER to get an assist. Big men work really hard to get themselves open, and as was mentioned previously, if it weren't a rare skill everyone would do it.

This line of conversation suffers from the same flaw in critical thinking that overrates the Jamal Crawfords and JR Smiths of the league: tunnel vision. With our limited ability to keep track of everything going on, our attention naturally falls to the guy with the ball. So when the ballhandler throws an assist, the inclination is to say the ballhandler created that shot for guy who scored, not that the post player had been fighting for position for the last 15 secs, slipped his defender, and rolled to the basket for the alley-oop. I'd argue that the big man "creates assists" for the perimeter guys by getting himself open, maneuvering for passing lanes so he can get the ball, and having the ability to finish at the rim.

Now is this every assist? Absolutely not. Some passers can get something out of nothing and make seemingly impossible passes. But bringing the ball up the court and giving it up to Kobe is not an impressive assist if he takes two jab steps and shoots a contested fadeaway 20ft shot. To capture the true value of an assist both scenarios have to be taken into consideration. But in all probability the undervalued assist and the overvalued assist mostly cancel eachother out, and the exponential increase in complexity to the forumla wouldn't significantly increase the accuracy of the metric.

123 days ago

Adam Tarr

It's bizarre to me to read all these counterarguments to what Guy is saying. The people making them seem to think the are defending the logic of WP, but in fact what they are doing is arguing against the way WP already handles the assist statistic.

All Guy is arguing is that instead of approximating the portion of a player's FGs that are assisted (which is what WP currently does), we should use the available actual data on how many of a player's FGs actually were assisted.

This is not complicated. It's a way to improve the accuracy of WP without changing any of its underpinning philosophy.

123 days ago

Adam Tarr

And again, the exact same argument applies to team defense. Rather than using the team's defensive stats and giving players a share based on their playing time, we can use the data for exactly how the team has performed defensively while that player was on the court. So Andre Iguodala doesn't get penalized for the fact that nobody on the Nuggets could guard Joe Johnson when Andre Iguodala was hurt.

123 days ago

Guy Molyneux

Thanks, Adam. I'm hoping Patrick will weigh in on this relatively simple fix to a problem with WP. I can't see any reason to keep assuming all players benefit equally from teammate assists, when we now have data telling us the real story.

And the issue is relevant to the value of these low-usage players. The correlation between usage and % of shots assisted (this year) is a strong -.53. Low usage players tend to have a very high assist% (and vice versa). The 14 players who have a usage of 15% or less and an above-average TS% are assisted on 78% of their shots on average (compared to 59.5% for all players).

123 days ago

Shrinidhi Rao

To recap: the current way WP handles assisted shots is to distribute the benefits evenly throughout a team, but it seems to make more sense to use newly available data to distribute the benefits proportional to the percentage of assists each player receives. It says nothing about whether or not an assist is "good" or "bad" and says nothing about changing the way WP currently values assists.

I would definitely like to see Patrick or someone from WoW weigh in on this as well, but I wouldn't hope for any changes to be made. Several of us brought up similar arguments against the uniform defensive adjustment currently in WP. While members of the WP crew actually agree that it is a suboptimal way of accounting for defense, they dismiss the suggestion that they should implement any changes to WP, because they don't think that changing it using available defensive data will improve their metric.

I suspect similar reasoning will be made to discount the assist distribution argument. To put it politely, the WoW crew is very reluctant to consider any changes to WP, and they get touchy when the subject is even broached. I would love to be proven wrong though.

123 days ago

Patrick Minton

Guy,

You want to change the model so that it better aligns with who you think are the better players. I am not interested in that. All I care about is explaining wins.

You seem certain that shooting without being assisted has some extra value, but you provide no evidence to support this conclusion; it just appears to be your intuition about what's important for a basketball game. Why do you think the fact that Kobe is not assisted very much has a positive impact on wins? Why do you think the fact that Chandler is assisted a lot has a negative (or less positive) impact on wins? Where's your evidence?

The WP model doesn't keep track of this because it isn't significantly correlated with wins. We also have data on what color sneakers each player wears, who's birthday it is, and each player's height, skin color, hair color and so on. None of this is correlated with wins, so the model ignores it. There is a reason that the model is called WINS produced.

120 days ago

Patrick Minton

Marshall,

Talent is subjective and I won't argue about Durant vs. Chandler, who are both clearly amazing (in different ways).

But I will also point out that talent, like potential, is worthless unless it is realized. In other words, as far as results are concerned it doesn't really matter whether Josh Smith is untalented or just suffers from boneheadedness; either way, he produces fewer wins than say, Zach Randolph, who is clearly far less athletic.

Now, whether a coach can turn this around or not is an open question; my personal belief is that with young players, a coach can make a difference, with older players, they largely are who they are, although a few coaches like Pop and Phil can make a difference.

120 days ago

Jacob Epstein

I not many people are mentioning this but after glancing through the 50+ comments, I think an clarification has to be made. The idea notion of devaluing a an assisted field goal should is not really devaluing at all but rather a more fair system of sharing credit for points. A player who takes high percentage shots (that usually come with an assist) need the pass to be successful, this should be reflected (to a probably slight extant) in WP.

120 days ago

Adam Tarr

Patrick,

You're missing the content of Guy's suggestion because of the rationale for it, which I agree is debatable.

You say "You seem certain that shooting without being assisted has some extra value, but you provide no evidence to support this conclusion". This is a very bizarre statement, because Wins Produces already does seem to think that.

http://wagesofwins.com/how-to-calculate-wins-produced/

"The results from this model were incorporated into Wins Produced as follows.

1. Calculate for each player his Teammates’ Assists per Minute (TAPM). This is calculated as follows: TAPM = (Team Assists – Player Assists) / (Team Minutes – Player Minutes)
2. Multiply TAPM for each player by the coefficient on TAPM from the above model, or 0.725.
3. Multiply step (2) by 2. This step allows us to see how TAPM impact a player’s points-per-field goal attempts (which is simply ADJ FG * 2).
4. Multiply step (3) by field goal attempts taken. This allows us to see how many points a player scored should be credited to his teammates."

That is what you are doing now, right? You are ALREADY sharing the credit for assisted FGs between the passer and the scorer.

All Guy is suggesting is that you cut out the approximations above, and simply replace the TAPM approximation with the actual assisted FG% for a given player.

Again, this is NOT a change to the underlying philosophy of WP. It's just replacing an approximation with the now available data.

120 days ago

Guy Molyneux

"You seem certain that shooting without being assisted has some extra value, but you provide no evidence to support this conclusion; it just appears to be your intuition about what's important for a basketball game."
Well, this is indeed an odd response. The notion that assists have value is already incorporated into WP. The basis for this, as I understand it, is a regression analysis that indicates assists improve the shooting efficiency of teammates. And if assists provide value, then it logically must follow that the value of shooting the assisted FG declines (because an assisted FG still only gives you the same 2 points on the scoreboard). It can't be the case that "assists have value" and also that "receiving assists does not reduce value." That's a mathematical impossibility. Which is why WP then subtracts from shooters the value of the assists.

The ONLY question here is whether WP will take advantage of data that is now widely available, but was not (I assume) when the metric was first created, and apply this adjustment specifically to the players who actually benefit the most from assists. How can it possibly be right to make this adjustment indiscriminately, when you actually know which players receive the most and least amount of assistance? At that point your error becomes intentional, rather than an unfortunate byproduct of data limitations. I'm sure that's not a direction you want to go.....

120 days ago

Guy Molyneux

"You're missing the content of Guy's suggestion because of the rationale for it, which I agree is debatable."
Adam: My rationale is exactly the same as yours -- we should apply the assist adjustment to FG value as accurately as possible. My comments on low-usage and high-usage players were simply observations about how certain types of players will tend to be impacted by this change (and these were broad generalizations -- some hi-usage players of course have high assist%, and vice-versa). Perhaps I expressed this connection poorly, but the impact on certain types of players is simply a consequence of -- not the rationale for -- making WP less inaccurate.

120 days ago

Hoopdon

Wow Patrick. If what these guys are saying is true, that could have been the worst thing you could have said.

Whether or not being assisted is good/bad is beside the point. If, as many have claimed, WP already adjusts for assisted field goals on a team-wide basis, why not break it down even further by player, since that data is now available in bulk? It has nothing to do with changing the formula to suit "his" desires, its improving the formula.

I have long brushed off the notion that WP guys are arrogant or stubborn. I just thought those people were wound too tight, or simply hating. However, if what these guys are saying has merit, and Patrick's response is what WP has to say, then WOW (pun completely intended).

As an aside, studies have shown potentially assisted filed goals have a significantly higher chance of going in then non-assisted field goals. If 1 player has 80% of his shots assisted, some credit needs to be taken off him, and spread out to either the assister, or the team. From what I have read, this isn't a new idea, this is a WP idea, not sure why it got such a cold, and frankly, thoughtless response.

120 days ago

Adam Tarr

Guy,

I agree with you, but at this point even bringing up low usage versus high usage is sort of muddying the point.

119 days ago

Andy Momplaisir

Hoopdon and Guy. You guys seem to be missing the point. Yes assists do help the people scoring, but that doesn't mean that they should be penalized for getting open. Wins produced gives the passer 2/3 of the credit for a made field goal (assuming it's a 2 point field goal). This means the scorer is getting credit for getting open and taking the good shot. If the passer is not getting full credit for the made field goal, then that means that WP is suggesting that the scorer deserves credit as well.

Consider the games of these players. Player who can shoot (Durant, Novak, Curry, Nash) tend to shoot when they see daylight, so they're more likely to be assisted. I've seen many plays where Kobe would get a good pass, be wide open, and not shoot. He'd hesitate, the defender would come and then he'd go one on one. Ray Allen's game involves multiple screens and getting a pass and then shooting it. He's not even very open all the time. He's a shooter though.

Tyson Chandler is a guy who can finish shots at the rim, and he's good at getting open, and receiving lobs.

The point is this. Guys like Westbrook may not be better because the shots they take are harder. Westbrook is not a guy who benefits from passes, because he's not a shooter or a good cutter. He forces things. He's a one on one guy and that might not benefit the team, so you have to prove that his game helps. Once that is done, then make the argument that WP should account for this. Right now, assumptions are being made.

You want WP to reflect your opinion.

119 days ago

Guy Molyneux

"Right now, assumptions are being made. You want WP to reflect your opinion."

I'm sorry, but this is simply mistaken. WP already gives credit for approximately .5 points (not 2/3) for each assist. This .5 points is then subtracted from the value of ALL FGs recorded by that player's teammates, based on how many FGAs they had. This adjustment currently takes no account at all of how many assisted FGs each player actually had (which varies greatly, and does not in fact increase proportionately to FGAs as WP assumes).

A good analogy to this current method would be if we evaluated players based on how many FGs they had, while assuming that all players took approximately the same number of FGAs. And if we had no data on FGAs, we might well do that. But of course, that leads to evaluating players based only on points scored, which can be very misleading. Since we do have data on how often players shoot, we incorporate that in our evaluations and look at efficiency.

All we are suggesting here is that additional information on assisted shots, which is now available, be used. That would allow WP to correctly distribute credit for assisted FGs -- roughly half to the shooter and half to the passer -- rather than mistakenly adjusting the value of unassisted FGs. Honestly, it is astonishing that anyone would disagree with this.

119 days ago

Patrick Minton

lol guys.

You are obsessed with "fairness"

I do not care about fairness. I care about wins.

Let's assume Chris Paul dishes out X assists per game. Y'all are suggesting that we worry about how we divy up the value that the assist gives to teammates, NOT the total value of the assist. So if he dishes out .7X to DeAndre Jordan, and .3X to the rest of the team, we ding Jordan's WP48.

But what you have failed to do is to SHOW ME THAT THIS IMPACTS WINS. Otherwise, I DO NOT CARE. If it does not impact wins, it is information every bit as useless as the player's hair color.

This would, in fact, be pretty easy for anyone with some knowledge of regression analysis to do. Or, instead, you can keep arguing with me in this thread about "fairness" and "

Furthermore, I think most of you are misreading the instructions for the adjustment we're all talking about. It affects the value of assists, not the value of each players FG% (Hint: if you stopped reading at step 4, like Adam apparently did, you didn't read far enough).

Guy,

As always, you expertly (and presumably willfully) misunderstood. I said:

"You seem certain that shooting without being assisted has some extra value"

And you replied:

"The notion that assists have value is already incorporated into WP."

Which is begging the question.

118 days ago

Shrinidhi Rao

Patrick,

Yes, the way we divide up assists and their impact won't improve WP's correlation with wins at all. But what it will do is (most probably) improve its consistency and its predictive power. All that's being suggested here is that we replace an approximation with more accurate new data. Again, the sums of both the approximation and the new data will be same, but by allocating assists more correctly, we can improve the model in other ways than its correlation with wins. By only caring about correlation with wins, you're missing out on other ways we can improve WP.

The way I understand it, a similar thing happened when WP's treatment of defensive rebounds changed. It actually decreased WP's correlation with wins slightly, but on the whole, it improved the model.

118 days ago

Adam Tarr

Patrick,

If we didn't have individual rebound stats, and we just assumed that each player rebounded a share of the team's rebounds in proportion to their playing time, this wouldn't change the total number of Wins Produced for the team. It would not change the number of wins we would predict for the team. It would not change the correlation of Wins Produced to wins.

If we then started getting individual rebound stats, would you reject using them because it didn't help us predict wins?

118 days ago

Hoopdon

Patrick, you are starting to look like the one who wants to manipulate WP to suit your desires, not the other way around.

As multiple people have iterated, the changes in assists we are talking about shouldn't change the WP totals, just make them more accurate at the player level.

I don't know if you are afraid that the new and improved WP would show low-usage gems, like Chandler and Sefalosha, in a worse light (slightly), or if you can't articulate correctly why we are wrong, or if you are just unbelievably stubborn.

I think you are a cool guy, who writes great articles, but if you are trying to convince people that the changes proposed here are wrong, you are going to have to do a lot better. It honestly appears that random commentators have a much firmer grasp on Wins Produced then you do, and you are simply here to rage at anyone who suggests altering the formula.

118 days ago

Andy Momplaisir

Hoopdon, I think I get where you're coming from, but I think you miss where Patrick is coming from. The idea that the value of assists should be removed from the players who get assisted sounds good, except it isn't proven that it will make the WP metric better. You guys keep saying that it most likely make it better, and we keep saying prove it. Believing that it will, doesn't mean it will.

Saying we make these adjustments, I'm guessing Kobe would have a WP48 greater than that of Sefolosha. Your suggestion for the change would suggest that Kobe is a better player because he creates his own shots. It appears that Patrick is saying that it's not proven.

The alternate argument to your argument appears to be that players who create their own shot do so because they (for whatever reason) don't make plays off ball movement. Their games are not made to be assisted.

Ray Allen would run around multiple screens to get daylight, receive the ball, and then either shoot or drive immediately to take advantage of the step he has on the defender. Kobe Bryant would use a couple screens, but once he gets the ball, he faces up and then goes one on one. He doesn't react immediately towards the basket (definition of a potential assist). It's probably because he can't shoot as well as Ray Allen, and so does things to create his own shot rather than use the advantage he may have gotten. Ray Allen shoots 3s whenever he can. Kobe doesn't. He's not a shooter.

I've noticed that while guys who are beneficiaries of assists get higher percentage shots, they're also good shooters or finishers at the basket. Good at getting themselves open or cutting.

In the WP formula, the idea seems to be that assists add value to the team. The players that get assisted just know how to take advantage of potential assists.

Now, you can possibly prove that if Ray Allen (one examples) created his shots as much as Kobe, he'd be less efficient than Kobe and probably would have a lower WP48. That's nice, but doesn't address the argument being proposed. The argument is not whether the guys being assisted more are more efficient because of it. The argument is do the guys being unassisted help their teams more because they don't get assisted on their shots. You guys don't seem to get that.

118 days ago

Shrinidhi Rao

Andy,

"...The idea that the value of assists should be removed from the players who get assisted sounds good"

That's not what we're suggesting. Let me try to explain. We're not suggesting to alter any of WP's philosophies. We're just suggesting that an approximation is replaced with actual data.

from: http://wagesofwins.com/how-to-calculate-wins-produced

"Specifically, a player’s shooting efficiency is related to the number of assists his teammates accumulate. To see this, the following model was estimated...

...1. Calculate for each player his Teammates’ Assists per Minute (TAPM). This is calculated as follows: TAPM = (Team Assists – Player Assists) / (Team Minutes – Player Minutes)..."

According to the page, this TAPM "allows us to see how many points a player scored should be credited to his teammates." But as you can see, it's an approximation. We're just suggesting that its replaced with the actual assisted FG%.

There isn't anything to prove here. We're just pointing out where the model uses an approximation, and how we can improve on that approximation by using actual newly available data. It doesn't make sense to not take advantage of it. And again, just because it won't necessarily improve correlation with wins is not a good reason.

118 days ago

Guy Molyneux

"I said: "You seem certain that shooting without being assisted has some extra value."
And you replied: "The notion that assists have value is already incorporated into WP."
Which is begging the question."

Patrick: I don't that is fair. I went on to explain why I believe, if assists have value, then it must be true that the value of unassisted FGs is less than the value of assisted FGs. You may disagree with my reasoning, but I was being responsive to your point. This comes down to simple algebra. I think we all agree that every FG, whether assisted or not, has the same value to the team. If so, then:
FG(assist) + Assist = FG(unassist)
And if that is true, then except in the case where the value of Assist = 0 (which no one is arguing), it must be true that:
FG(assist) I still don't understand what part of this simple algebra you disagree with.

And note that this does NOT mean that unassisted FGs -- or those who make them -- get some kind of "extra" credit. The value of FG(unassist) is just what WP has always said it should be (1 point). This is just a question of dividing credit properly between the shooter and the passer on assisted FGs.

This is not about Kobe, or Sefolosha, or any player -- you really shouldn't make decisions about metric construction based on which type of players are "helped" or "hurt" -- it's about math and accuracy. And as others have noted, the underlying idea that passers and shooters should share credit for assisted FGs is already part of WP: "This allows us to see how many points a player scored should be credited to his teammates."

But the hostility of your response suggests you are not yet open to this idea. So that's it for now. But who knows? For many years, concerns about WP's overvaluing of rebounds were met with the exact same hostility. But eventually, changes were made (if incomplete). My guess is the WOW team will eventually come around on this issue too -- hopefully before 2020.

118 days ago

Guy Molyneux

Oops, dropped some text there. I meant to say:
"I think we all agree that every FG, whether assisted or not, has the same total value to the team. If so, then:
FG(assist) + Assist = FG(unassist)
And if that is true, then except in the case where the value of Assist = 0 (which no one is arguing), it must be true that:
FG(assist) I still don't understand what part of this simple algebra you disagree with.

118 days ago

Guy Molyneux

Hmmm, I guess the blog software doesn't like the "less than" character. In any case, IF:
FG(assisted) + Assist = FG(unassisted)
THEN:
FG(assisted) must have less value than FG(unassisted).

118 days ago

Adam Tarr

Just to hammer this point home: the inequality Guy is talking about is not a new property we are arguing should be added to WP. It is an identity that is inherent in the way WP is calculated.

All we are suggesting is that we actually account for how many unassisted and assisted FGs each player had, instead of just using the approximation that everyone on a given team has (after a small adjustment for their share of team assists) the same percentage of each.

118 days ago

joshweil

Adam and Guy are making a lot of sense in this thread. If better data is available, it should be used.

118 days ago

Adam Tarr

Now that I think about it, this more accurate tallying of which field goals are assisted would probably force one other adjustment to the WP formulas. It would lead to smaller position adjustments. The percentage of field goals that are assisted goes 4>5>3>2>>1. Centers and especially power forwards would no longer need such a large negative position adjustment in WP if their higher rate of assisted FGs is properly accounted for. Point guards might not need as much of a bump as the assisted FG ratio would suggest, either, because the current approximation probably does a decent job of reflecting how they receive fewer assists.

So overall, the position adjustments would be flatter than they are now.

Of course, changing the position adjustments, just like this entire change, would have absolutely zero impact on the total WP for each team. It is a way to more accurately distribute WP among a team, as opposed to adjusting how many the team has.

118 days ago

Drew Miller

I started out writing this comment to agree with Adam and Guy, and have gotten to a point where I don't totally understand if WP is correctly valuing assists themselves at all. The model from the calculating wins produced page shows that an overall high number of assists tends to correlate with an increase in field goal percentage from the previous year. Interestingly this model DOESN'T seem to show that an increase in team assists year-over-year increases field goal percentage; the team assists from the previous year just aren't a part of the model at all. It's also a pretty classic case of correlation that could go both ways - if a player improves their jump shot more than average in the off season, more of the passes to them will result in assists.

Is the reason that the model is formulated the way it is addressed somewhere? One of the books maybe?

117 days ago

benjamin durham

Im sorry but as much as I can tell the basis for this argument is that oft assited players should be punished. i think this is rediculous. logically if a teams assist have a positive correlation to winning than being assisted alot should be viewed as a positive. it even carrys the natural reward of better shooting. its just we tunnel vision to ball handlers so we are naturally inclined to believe "creating" is important

117 days ago

Adam Tarr

benjamin im sorry but if thats all you have been able to gather maybe you should gather longer...

Ahem. Sorry. I found the shift key, and the ability to spell.

Anyway, again, the idea that the Wins Produced by an assisted field goal are shared between the shooter and the passer is ALREADY built into WP. If you hate that idea, you hate WP as it stands. Nobody here (save perhaps Drew, immediately above you) is even suggesting the possibility of changing that.

The only thing we're suggesting is actually counting how many assists go to which players, instead of trying to guess at this as the WP formula currently does.

Drew,

Sorry, I can't help you with the original derivation of the Assist stuff. Intuitively, what WP says makes sense to me, and I would hope (and, given how solid the model seems overall, I would assume) that they looked at the sorts of relationships you're talking about before adding assists into the formula.

117 days ago

Drew Miller

Seems like it should be easily testable - do highly-assistable players (like Chandler) take their high percentage and high assists with them when they move, or do high assist PGs take their high assists with them to new places? Whose year-over-year WP48 is more consistent?

117 days ago

Andy Momplaisir

i feel like this is why people on apbr think the wins produced crowd is arrogant. adam. you, guy, and hoopdon are making the same point over and over, but somehow seem to be missing the point me, patrick and others are making. i'm starting to feel like you guys are just trolling us. the wp says that assists are valuable and help the team. in scoring, to be productive, a player's true shooting % has to be over around 50% (as opposed to 0%) and the average ts% is around 54.5%. in assists, there is something similar. the reason the value of assists are not taken away from the person who benefited is because that would suggest that you could have a team that accumulates 0 assists. the defensive rebound value of .5 suggests that at worse, a team should get at least 50% of the defensive rebounds available (in the nba). same thing with assists. at worst, there will still be assists and guys who benefit from them. guys who get assisted will always get assisted. thats their game and all nba teams have players who can pass. ray allen gets assisted because that's his game and whatever team he goes to, there will be players to pass him the ball. if he gets assisted on 70% of his shots, it makes little sense to take out credit for all the assists he receives because that would suggest that if it wasnt for his teammates, he wouldnt get assisted at all. he plays in the nba. he's always going to have someone to pass him the ball. you guys keep suggesting that if you change how the credit is adjusted, it'll make thd metric better. i disagree because your way suggests that in the nba, you shouldnt expect to get assisted, yet no team has ever been close to averaging 0 assists per game. stop saying that we're opposed to changes and prove that the change would make the metric better. i dont think you can. (i couldnt split this into paragraphs because my phone isnt letting me)

117 days ago

Adam Tarr

Yeah, it's certainly testable, but that last question is the key one, right?

Even if APG is less consistent than assistedFGM/totalFGM, that doesn't tell us about causality or who should get credit. It could be that APG is an inconsistent/noisy stat, but players with a high percentage of assisted FGs are more dependent on high-assist players for their FGAs than the high-assist players are dependent on the highly assistable players for their assists. Or maybe the highly assistable big men like Chandler and Stiemsma work one way, while the Battiers and Novaks work another.

At any rate, just for the casual readers of the thread, I want to reiterate that this discussion is quite distinct from most of what we've been talking about. Drew is suggesting revisiting the entire idea of how WP values assists; all Guy and Shrinidhi and Hoopdon and I are suggesting is that we use now-available data to properly apportion credit for assisted field goals using the existing principles of WP.

117 days ago

Shrinidhi Rao

Andy,

We understand your point; I think you may be misunderstanding ours. We agree with your reasoning; we are not at all advocating changing any of WP's philosophies. If you check our the How Wins Produced Is Calculated page (http://wagesofwins.com/how-to-calculate-wins-produced/) it involves adjusting for assists already, but involves using several approximations. If we now have data to use in place of those approximations, shouldn't we use it? That is all we're saying.

117 days ago

Hoopdon

Andy,

1.) That picture is terrible.

2.) The current argument is not about the value of assists, the merits of those with a high percentage of their shots assisted, or anything else you talked about.

Wins Produced ALREADY assigns credit for assisted field goals. They do this using a team adjustment, assuming everyone on a certain team is assisted at the same rate. There is new data out however, that breaks this down by the individual. So, instead of guessing, we can now know for sure at the individual level.

This isn't a "new' idea to WP. The team total shouldn't change much, if at all, from this data. It will however, make each individual player's WP more accurate.

You, and many others, can't seem to make this distinction, and its quite odd.

Patrick now doesn't seem to be dismissing the idea completely, but wants proof that this adjustment would correlate to wins at the same, or higher rate. A let-down of a response, but a fair response.

117 days ago

Guy Molyneux

"Patrick now doesn't seem to be dismissing the idea completely, but wants proof that this adjustment would correlate to wins at the same, or higher rate. A let-down of a response, but a fair response."
Hoopdon: This is actually an impossible test to meet. Adding assists to the basic productivity metric (PROD) underlying WP does not improve prediction of wins at all. You could skip the Assists component of WP entirely, and it would predict wins exactly as well as it does now (because an assist neither scores points nor creates a possession). As a result, no matter how you adjust for assists within the team, it will have zero impact on win prediction. The reason to include assists is not to predict wins better, but to more accurately distribute credit for wins within a team. And that is the exact same reason to adjust for assists using the best available data.

So Patrick's request for proof that this will better predict wins, which as a WP expert he must know is impossible, is not really a "fair response," or even a "response" at all. It's an attempt to end the discussion.

117 days ago

Adam Tarr

Hoopdon,

This change should have absolutely ZERO impact on WP on the team level. It's a change to how WP is distributed within the team. Even if this necessitates an adjustment to the position adjustments (which is probably would) it's still just a change to in-team WP distribution, because every team plays (nominally) a 1, 2, 3, 4, and 5 at all times.

As such, this would have absolutely zero impact on the correlation of team WP with team wins. It would, I suspect, improve year-to-year correlation of player WP.

If Patrick's argument is that this should do NO WORSE in predicting wins, then we're already there and the change should be tried out. If instead he wants the change to make the model BETTER at predicting wins, then that's not going to happen, and frankly, it's a ridiculous way to dismiss this idea.

As I said earlier, if we only had team rebounding statistics, and we approximated each player's rebounds in proportion to playing time, the WP model would do just as good a job of predicting team wins as it does currently. That's not a good argument for ignoring individual rebound statistics, is it?

117 days ago

Andy Momplaisir

hoopdon, are you sure that the goal is to approximate the value each person is receiving. the page never says that the adjustment was to approximate how the value is being spread. at least i dont remember reading that.

Hoopdon. I don't think that Patrick is trying to end the argument. It just doesn't seem like you guys get the argument. There are other ways to see if the change is better than just explaining wins. Does wp become more consistent and does it become more predictive. I don't think that this change being proposed will improve the model. I also think you guys just can't see why. So to end this, just prove that it will improve wp.

117 days ago

Guy Molyneux

Andy: I don't know what you mean by "more consistent" or "more predictive." But using the assist% data has to make WP more ACCURATE. This is true by definition, because all we're doing is replacing a crude approximation of who receives assists with actual precise data. Adam gives a good analogy in terms of rebounds. Or, imagine we had data on FGs, but had to assume all players took the same number of FGAs. Obviously, using actual FGAs would greatly improve our metric.

117 days ago

Adam Tarr

Andy, it's kind of funny to me to read that we don't get the argument when you are basing this on "i dont remember reading that". Particularly when the WoW page rather carefully explains how it:

1) subtracts WP from each player on the team in proportion to how many assists the model approximates that each player received.
2) redistribuites the WP it just took away to each player on the team in proportion to the number of assists they gave.

All we are suggesting (is that you remove "the model approximates that" from that step 1, by using the real data. That is all.

WP would not become more consistent on the team level after this change, because WP on the team level would not change. WP would, in all likelihood, become more consistent year-to-year on the player level.

If I had access to the code that Patrick uses to access all the data and pump out the current WP data, I would be happy to give this a spin and prove that this improves WP's year-to-year correlation on the player level. I am not interested in trying to code up the entire WP model, database hook-ins and all, to try to prove this one point.

117 days ago

Hoopdon

Guys, we broke the 100 comment barrier. Everyone give yourself a pat on the back.

To those saying that this assist change should be shown to make WP more consistent or accurate. I am reminded of something Andres said on a WOW podcast. If adding more/better data to your model makes it increasingly less effective, then there is a fundamental problem with your model, not a problem with the data.

117 days ago

benjamin durham

We understand what your suggesting. I just don't agree with it. The end result is to take credit away from players who get assited alot thereby finding more value in ball dominate shot "creators". What I'm saying is that if assist have a positive impact on wins which ofcourse they do in the resulting shooting numbers, then players should try to get assisted as much as possible. why shouldnt we put most of the adjustments subtraction on the guys who ARNT being assisted as much because their not putting the team in the best positions to suceed. you dont have data that shows its less valueble for tyson to get free at rim and score that melo to hit an iso jumper. i dont know of any test that proves the point one way or the other, so i actually do think a team wude adj is the fairest way to go.

116 days ago

benjamin durham

I think the issue is were excited because of new data without really considering the application. Assisted rate tells us HOW a player scored, but nothing of the value of it. Cutting, spotting up are skills just like dribbling. Rolling to the rim for 2 is as good as driving. It's unfair to punish a skill set over another and i feep thats why its a team adj. Rather than punishing players whose work off the ball is just as valueble as players with the ball, even if its less noticable.

116 days ago

Drew Miller

It's not totally correct to say that this is just making wp more precise. Right now wp says assists increase FG %, and charges that benefit proportionally to FG attempts whether or not those are assisted or not. The assumption being made by many folks is that assists are exclusively useful to the shots made that are assisted, but another plausible story is that assists are just indicative of good ball movement and that all players on the floor during high assist games benefit relatively equally. Certainly the model does not have any dependence on "assists made to me" currently in assessing the value of assists on FG %. I would be interested in seeing how that data impacts the regressions. If anyone is serious about testing this I am happy to help out.

116 days ago

benjamin durham

i feel like we are taking data intended to provide explanation and trying to use it for evaluation. ast'd% could be useful as a team building tool where n theory high ast players paired with high ast'd% players is good, but if your team isnt goid at being assited rondo prolly wont help. case in poiny the lakers and nash & kobes lack of benefitting each other. but when we look at this data to say player x is inferrior because he gets assited alot, were just buying the confirmation bias that our eyes are good talent evaluators.

116 days ago

Drew Miller

Where is the data on % of shots assisted?

116 days ago

Adam Tarr

Drew,

You can find it at Hoopdata and probably other places.

Benjamin, frankly, the more I read what you're saying the more clear it is that you don't understand how Wins Produced already works.

Again, players are ALREADY docked credit for the shots they get assisted on, and that credit is then given out to the assist men on the team. This is already part of the formula. We are just suggesting dong this more accurately instead of making a guess at how many of a player's shots got assisted.

When you suggest something like "why shouldnt we put most of the adjustments subtraction on the guys who ARNT being assisted as much because their not putting the team in the best positions to suceed"... aside from making my eyes hurt, you are making a basic error that Patrick has pointed out many times in the past. We don't care WHY a player misses shots, we only care whether they do. Why would we penalize a player like Harden, who is extremely efficient despite being assisted on less than a third of his shots?

We don't need to worry about whether getting assisted shots is a better idea or a worse idea. That's already baked into the formula - players who score without missing a lot of shots are rewarded directly. If Harden would be more efficient if he played off the ball more (and this is actually true to some extent - his WP48 is down from last year when he was assisted a lot more), then the penalty is directly reflected in his worse shooting percentage. We don't need to double-count it.

The assist portion is all about sharing the credit for something good that happened... and again, WP already does this.

116 days ago

Andy Momplaisir

Adam,

The "how to calculate wins produced" page says that player's shooting efficiency is affected by the amount of assists his teammates accumulate. It doesn't say by the amount of shots that he has that is assisted.

We are at over 100 comments and this simple point will not be understood. The idea that the assists adjustment was to take credit away from the player being assisted is an assumption. This is not proven. It's not stated on the calculation page and I bet if you ask Dave, he'd tell you himself.

Even if the new data was available before, I don't think it would be used because I don't think it would make the metric better. Every team accumulates assists, so every player that gets assisted will continue to get assisted.

It's like the defensive rebound adjustment. Why give a player full credit for the defensive rebound if, at the nba level, at worst, you can expect a team to get 50% (or whatever it is) of the rebounds? Why take away the full value of an assist to player if, in the nba, he would've been assisted anyway?

In a sense, WP does give credit for shot creation. On teams with low amount of assists, there's more credit given to the players.

In the end, there is nothing that suggests that the intention of the assist adjustment was to take the full value of an assist away from the guy who got assisted, and there is no proof that this adjustment would make the metric better. I feel like I've said this many different ways. I don't understand why this point can't be understood.

Also, I think Patrick is saying prove it, not to stop the discussion, but actually that you try to prove it. You might be surprised what you find out if you try.

115 days ago

benjamin durham

adam, my poimt that you are not getting is that an assisted basket is equal to an un-assited one and to take away wieghted credit from players wouldnt make things better. where you see a player being helped i see a player putting himself in posistion to score. why should we single out players whose style creates just as many assist and tge passer does. i believe the reason its a team adjustment is you have to have players like that. theres only one ball so yoy need guys who can create chances in ways other than isolation. your wanting to dock them for what is a team effect, and i think its wrong. its not that i dont understand what wp is doing, its that i think its right as is.

115 days ago

Adam Tarr

Benjamin, it's actually very clear that you don't understand what WP is doing. You clearly think the assist adjustment is doing something that it is not.

Andy, it doesn't say it's influenced by the amount of shots that are assisted because it's not! That's the point.

Quiz for both you, since you are making some similar errors.

Teams A, B, and C all have the same number of FGM, FGA, Blocks, Steals, Turnovers, Rebounds, etc. They only differ in one stat: assists.

Team A has not recorded an assist all season - every single made basket was unassisted.
Team B has recorded an assist on 50% of their made baskets.
Team C has recorded an assist on every single made basket all season.

Quiz question: which team will have the most total Wins Produced?

115 days ago

Drew Miller

Adam - I think the basic issue is that there are two valid interpretations of the way WP handles the assist adjustment.

1. Assists are a per-shot metric and their impact should be counted solely in relation to FGs that are assisted. The data to do this wasn't readily available at the time the Wins Produced model was created so some approximations are made.

2. Assists are a general metric and their impact should be considered team-wide, not just in relation to assisted plays. Players benefit from assists solely in terms of their share of FGAs. Wins Produced models this.

You and Guy clearly believe explanation 1, but I think explanation 2 is more plausible given how closely it matches up to the actual method of calculation. In a world where #1 was true, don't you think they would base the penalty ratio on FGs rather than FGAs?

That's not to say that the WP approach is correct - it's definitely possible that factoring in Assist % would improve the model. I now agree with the folks who want to see this demonstrated.

In related news, is there a way to get a dump of all the stats in Hoop Data? I don't feel like writing a screen scraper today.

115 days ago

Adam Tarr

Drew, if the rationale were #2, then there still isn't any argument for modeling it the way it's modeled. The assists themselves are directly credited to the players, after all, and the debit for assisted FGs is taken away from individual players in a proportion that is closely related to their shot attempts. Your rationale #2 doesn't really align with any guiding philosophy for handling assists, and certainly not to what is being done.

Bonus question for Andy and Ben: how would the answer change if what Shrinidhi/Guy/Hoopdon/myself are suggesting were implemented?

115 days ago

Adam Tarr

Sorry Drew, I don't know an easy way to get the data off Hoopdata, but I imagine you could e-mail them and ask.

I am crap with online database management, which is why I can't really test this stuff on my own. If I had access to the existing WP code and a data import of the Hoopdata assist numbers, I could code up the adjustment fairly easily, but coding it from the ground up is beyond my capabilities in the time I'm willing to devote to this.

115 days ago

Hoopdon

Data on % of shots assisted is available at Hoopdata, NBA.com, and Basketball Reference.

To all WOW/NBAGEEK columnists. You guys have access to the WP code. Wouldn't it be a simple matter to just copy and paste the individual % assist numbers, and see how that WP compares to the current WP?

I kind of agree with Adam, in that asking random commentators to "prove this " from the ground up, could be seen as a way to make sure it doesn't get done, and the subject isn't broached anymore. Not saying that's what you are doing (even WP guys aren't that sinister lol), but that will probably be the end result.

114 days ago

benjamin durham

Adam, just because someone disagrees with you doesnt mean they don't know what there talking about. For your lilttle quiz the teams would produce the same number of wins. Doesnt change the fact that what u wanna do is divide credit more favorably to shot creators and "conventional wisdom" stars, and i dont believe that would "improve the model"

114 days ago

Hoopdon

Benjamin,

No one "wants" to do anything. We just want the math to be correct, instead of using outdated data. WP already divides credit based on assists. We just, you know, want the players who actually deserve that credit to get it.

Imagine if Rebounds per Game for each player wasn't available, and all we had was RPG for each team. WP would apply the RPG adjustment to every player on the team, without knowing WHO was actually getting the rebounds. New data comes out, that breaks down the RPG by player, and now we want to improve WP, by using this data.

Would you say "u wanna divide credit more favorably for good rebounders, and I don't believe that would improve the model."? Yes, I want players who get rebounds to be credited for getting rebounds. I want players who garner assists, and benefit from assists, to receive proper credit. What I, you, or Dave "want" or "believe" is irrelevant.

Also, this whole "it has to improve the model", is BS. The statement I made earlier, which I got from Dre, still stands. If adding better and more extensive data to a model makes it less accurate, there is a fundamental problem with the model, not the data. If adding in this assist data makes WP correlate less to wins, or decreases its predictive ability, that is an indictment on the WP, not on assists.

114 days ago

Adam Tarr

Hell yes to all of what Hoopdon just said.

Also, again, what WP currently does really doesn't make sense in support of any underlying philosophy except "we don't have better data". If you really think that the assist is a team stat, then why does it benefit the passer, but we take credit from everyone for the score? How does that make sense?

Imagine a one-possession game is won, 2-0, on an assisted basket. There are no other stats for the game. WP would currently give the scorer positive WP for the basket, the passer positive WP for the assist, and the other four players on the team (including the scorer) one fourth of a negative credit for that assist, spread based on their share of playing time.

There's no underlying philosophy supporting that apportionment. It's not predicting wins better (because it has zero impact on the total WP for the game). It's not rewarding creators or scorers or whatever in some consistent fashion. Can't we just admit this was an approximation? Why the need to invent these elaborate arguments to defend the status quo?

114 days ago

Adam Tarr

EDIT: sorry, that example I gave doesn't actually work. I was trying to come up with a very simple example to make the point, but you would actually need to give everyone a FGA before it would break down that way. If you add that in, the example works... and still doesn't really make any sense as a way to explain what happened in that game.

114 days ago

Thomas Baldwin

Late to the party but I'll wade in... I have done the actual research with my own model, and the result is that assists should be debited from team mates on a minutes-based basis, not a FGA based basis. Doing it on an FGA basis gives inferior results for prediction, which in turn suggests that an assist shouldn't be considered as a measure of ability to create easy shots, but rather a measure of offensive involvement. This is also why the assist*usage interaction is a far better predictor than assists alone, because it gives a stronger indication of the offensive involvement a player has in the game. This isn't news if you spend enough time at APBR.

111 days ago

Adam Tarr

Really interesting stuff Thomas. When you say minutes is superior to FGA, I assume you mean raw FGA, not potentially assisted FGA, right? If so, does debiting directly on the basis of assisted FGM (or potentially assisted FGA, if you have *that* data, which would be awesome) still do worse than a pure minute-based allocation?

If this is the case (which would be surprising to me, but not mind-blowingly surprising), I assume it would still be better to debit the assists based on who was actually on the floor at the time the assist was made (which is also available data at this point). This is more in line with my suggestion for improving defensive stats in WP - rather than just give players a minutes-based share of overall team defensive stats, give them their share of team defensive stats when they were on the floor.

111 days ago

First Last

MINTON YOU A MOTHAFUCKIN' COWARD AND YOU'D BEST NOT SHOW YO FACE ROUND THIS COMMENT THREAD NO MO!

AND DAVE BERRI A BITCH, YALL! DAVE BERRI A BITCH YALL EVERYBODY!

109 days ago

lovethoseknicks

The role of the defense is to try to force bad/tough shots. During the course of a game, the defense will be successful at that many times and force a tough shot. IF all players took the same number of those tough shots, it would be easy to compare them. The thing is, they don't. And more importantly, it's because some of them can make those shots at a 40%-45% clip but others at way less than that. The thing is, taking shots like that (with less than approximately a 50% success rate ) lowers a player's value. So the very skill that makes it desirable for some players to take more of them causes their efficiency and value to fall relative to less skilled players.

Essentially, you can't compare player efficiency without controlling for these differing roles.

On the flip side, there are clearly some players that take more self imposed bad shots, doing the job of the defense on their own. Their value should be punished for that foolishness.

Until we have a method to control for self imposed bad shots vs bad/tough shots that are part of a player's role as a scorer, you can't get accurate values for scoring.

Guys like Chandler are incredibly skilled around the basket, but most players can beat 50% around the basket and create value. He just creates more. But because he's among the least skilled outside shooters, he almost NEVER has the role of taking the tougher mid range & tougher shot that lower efficiency. That makes him look better than he actually is and others look worse.

102 days ago

Omar Kreidieh

Im reading the posts above and I think you keep falling into the same thing. No one is questioning that Drummond and Chandler, theyre good and drummond is unquestionably the best player on the Pistons team in my mind. The question is HOW good are they? Yes there are many good and many bad low usage players. The fact that many of them arent rated highly should not steer you away from the fact that they should be rated even lower, and that those rated so highly should not have that big of a rating. The role that higher usage players play by drawing defenders, the way they sacrifice by taking the lower percentage shots is not accounted for and makes those players underrated by comparison. It is not a myth that certain shots should be created. It takes playing even amature basketball to understand that when the shotclock is winding down, someone needs to create an open look quick, because those screens and plays that you talk about dont work everytime. Anyways I think everyone here is open minded enough to know that we may be wrong but no one has done the statistics or the math to convince the other side. If you really want to sort this out, I suggest you do two things:
A . If possible calculate WP48 for players when taking into account the last 10 seconds of the shot clock and then when you ignore that and see if the changes were significant.
B. Look at specific players from recent trades. Compare their usage rates before and after the trade along with their WP48 and shooting efficiency.

Each of these methods is not perfect and their generalizability will depend on choice of player BUT we will have atleast moved this discussion from just people stubborn about their opinions to hard evidence for or against WP48 being affected by usage.

98 days ago

Omar Kreidieh

A third suggestion is look at the WP48 of players when a high usage teammate is on the floor and again when he is off the floor because a big argument is that the high usage players make it easier for the other players and therefore part of their contribution to winning is present in the calculate WP48 of their teammates ( who benefit from open shots, having worse defenders, and not needing to take as many shots ectt... )

You dont have to do these obviously, but I think such statistics can easily persuade most of us.

98 days ago

Scott Oviedo

In case anyone comes back here, I just want to point something out.
Hoopdon - "If adding better and more extensive data to a model makes it less accurate, there is a fundamental problem with the model, not the data."

I agree. But I think you are skipping a step, which is the step most of the detractors of your proposition have been citing: First you want to be sure that the data you add in is actually measuring something useful, and isn't just noise. Compared to rebounds, that change or preserve a possession, there is more question as to what assists are actually measuring and the impact they have on the game. It has been shown that the number of assists in a game correlates well with shooting efficiency. Now we just need that next step. So let's take that next step, since we have the data! I'd love to see the model improved.

Adam Tarr - "There's no underlying philosophy supporting that apportionment. It's not predicting wins better (because it has zero impact on the total WP for the game). It's not rewarding creators or scorers or whatever in some consistent fashion. Can't we just admit this was an approximation? Why the need to invent these elaborate arguments to defend the status quo?"

This is not accurate. As I noted above, there IS an underlying philosophy. There is a correlation with shooting efficiency and the number of assists produced by the player's teammates. That was the extend of what the authors could prove, and so that is what they incorporated. Now that we have more detailed data of an large number of games, we can perhaps find a more specific correlation. I do expect that there would be a justification for refining the model. But I wouldn't change it based on that assumption. If I like an idea, I do my very best to falsify it.

92 days ago