I don't know it's even half that. Kardashian is a low upside frontcourt reserve. Those guys grow on trees.BigSoxFan said:Get that 4/32 offer ready, Danny!
I don't know it's even half that. Kardashian is a low upside frontcourt reserve. Those guys grow on trees.BigSoxFan said:Get that 4/32 offer ready, Danny!
He was being factitious.HomeRunBaker said:I don't know it's even half that. Kardashian is a low upside frontcourt reserve. Those guys grow on trees.
I don't know who's what anymore. I've heard some say Kardashian was our best player last year which if accurate would certainly put him in line for that type of deal.bosox79 said:He was being factitious.
HomeRunBaker said:I don't know who's what anymore. I've heard some say Kardashian was our best player last year which if accurate would certainly put him in line for that type of deal.
I don't know one metric that measures spacing (if there is one I'd be VERY interested in reviewing it) which is arguable the most important trait a non-prime scorer can provide to his offense. Kardashian scores average to poorly in this area by my eyes.PedroKsBambino said:
There's guys who score well on efficiency-based metrics because their games involve easy-to-count things like rebounds, assists, etc. and they take only high-percentage shots. This is especially true when the player has more minutes and a larger role (as Kardashian did) because the team stinks. But as you and I know, there's a ton more to basketball (spacing, defense, flexibility about role, matchups, dribbing, mobility) that aren't easy to measure but count a ton. Humphries is a serviceable rotation big guy, but as you note there's a lot of those out there.
HomeRunBaker said:I don't know one metric that measures spacing (if there is one I'd be VERY interested in reviewing it) which is arguable the most important trait a non-prime scorer can provide to his offense. Kardashian scores average to poorly in this area by my eyes.
Defensively.....when Ainge was asked what the teams primary needs was this offseason he didn't hesitate in responding "rim protector" which was a direct hit at what they need out of Kardashian's position. Yet metrics say he was the teams best player? I'm not that sucker who's going to buy that ever.
bowiac said:That is incorrect with respect to RPM and xRAPM. It is only true with respect to raw plus/minus metrics.
You said your concern was that when Humphries leaves and Faverani comes in, "that is a big difference and will make Humphries look 'valuable'." However, xRAPM knows who is coming in.PedroKsBambino said:Please explain....I do not believe you are correct. To keep it simple, I believe they attempt to adjust for actual replacement level, but are limited for two reasons. First, because NBA plays are not independent events. Second, as they compare Humphries on-court to Humphries off court the results are pretty materially impacted by the actual quality of replacement, since we have no basis for assuming what would have happened with a different replacement. They regress to try and manage this, I'd assume, but that's simply not a complete solution to the problem noted.
Yes he was when in Brooklyn and his play resulted in him being benched for Reggie Evans who the Nets couldn't wait to dump less than 12 months later.wutang112878 said:Humphries ceiling is really the Perk role in the 08 Celts minus the rim protecting defense (because he is a little undersized and too bound by gravity) for the 4 but slightly better offense. Basically your 5th best starter or 6th man where his best skill (rebounding) can really help your team and he is decent enough offensively that if he is left alone he will make the opponent pay. The problem is he has never been in one of those situations and has always been asked to play a more important role on a less talented team.
bowiac said:You said your concern was that when Humphries leaves and Faverani comes in, "that is a big difference and will make Humphries look 'valuable'." However, xRAPM knows who is coming in.
Vitor Faverani is graded out as an awful player by xRAPM. When Kris Humphries leaves the floor, and Faverani comes in, xRAPM knows a terrible player is coming in. xRAPM compares the Celtics performance with Humphries out there to their performance with Faverani. To the extent the Celtics are much better with Humphries than Faverani, that's still based off a very low baseline, so a big improvement off Faverani still makes Humphries look like a pretty mediocre player.
And to wit, that's exactly what xRAPM shows: It hates Humphries (actually grades him out identical to Faverani). Precisely because of this base-lining adjustment.
PedroKsBambino said:
I probably oversimplified my first post, and your comments don't respond at all to the challenges I raised in my second post. Obviously, if you are going to do plus/minus you need to track who replaces whom; no question that it does do that. However, doing so across a large set of substitutions is more complicated than you suggest and that's where I note this is still pretty inexact.
First, we need to assess how the change of one spot impacts what people are doing at other spots. If you bring in someone who is horrible (or someone who is slow, etc.) that impacts what happens elsewhere on the floor. That is why these are not truly independent events. Looking at the plus/minus tells you what happened when that substitution happened (since that is a simple math problem) but not the contribution of the player to that outcome. I grant that's a very 'second level' thing.
Second, and really more importantly, to do plus/minus you need to make some assessment of the quality of player coming in. You describe above (correctly, I believe) that the metric compares performance with player 1 in vs performance with player 2 in. But that's only part of the measurement challenge, because even if all other players and sets were the same (which, as noted above, we know they are not) we still need to assess how much of the scoring difference between 'lineup with player 1' and 'lineup with player 2' is player 1 being good and how much is player 2 being bad. So, if there's a 10 point difference it might be that Humphries is 10 better than average and Faverani is average, or it might be Humphries is +5 and Faverani is -5, etc. This is essentially a true talent level problem. To simplify, you say that they do this (and I agree) but my belief is that they way they do it leaves real gaps.
The metric tries to do that, I believe effectively by taking the +10 and -10 and comparing it to a large set of substitutions and regressing the changes for different lineups to determine a baseline for each player. That is a reasonable approach; however, it is far from a perfectly accurate one, too. So, it's guess at Faverani's impact is just that---a guess, likely a regressed one. Once we're regressing both parts of the equation, we need to recognize that outliers (such as Faverani's bad-ness) will tend to get washed out to a larger degree than they probably should.
I can't pretend that I've dug into the underlying math of these as I have at times on other stats; however, I have yet to hear (here or elsewhere) answers to either of the above that really address the concerns. And to be clear, the guys doing this are very smart, doing great things that advance our understanding, and some of them fully understand the complexity of the NBA, too. Many of these guys recognize the limitations and the error bars. Sometimes, (just like in baseball sabermetrics) those who don't create the numbers recognizing the gaps is really the bigger challenge
To Wutang's point about 'if we can't see an impact, why should we care?' I think part of the answer is that it's a multivariate problem, and part of the answer is we should remain conscious of the difference betweeen 'we can't yet measure' and 'it has no impact' If you go back and read this very board on the impact of catcher defense in the mid-2000s you'll find hundreds of posts mocking the idea that it matters much; Bill James counseled the sabermetrics community that there's a difference between measurement and impact, and as technology evolved and research continued eventually we learned that just one aspect of catcher defense (pitch framing) has a material impact on runs allowed and were able to clearly document and value that skill. The actual value of that skill was the same in the early 2000s and today---only our ability to measure it (and the number of people who recognized it) changed. I don't know that we are in a similar place in NBA stats, but I think the lesson is one to remember.
HomeRunBaker said:Yes he was when in Brooklyn and his play resulted in him being benched for Reggie Evans who the Nets couldn't wait to dump less than 12 months later.
I don't understand what you're saying here. If you bring in someone who is horrible, that impacts what happens elsewhere on the floor - yes. Everyone needs to adjust for that, and presumably that unit will be less effective. That will be reflected the xRAPM. I honestly do not understand what you're saying. What is the "second level" thing?PedroKsBambino said:First, we need to assess how the change of one spot impacts what people are doing at other spots. If you bring in someone who is horrible (or someone who is slow, etc.) that impacts what happens elsewhere on the floor. That is why these are not truly independent events. Looking at the plus/minus tells you what happened when that substitution happened (since that is a simple math problem) but not the contribution of the player to that outcome. I grant that's a very 'second level' thing.
I do not understand this. Why is Faverani's bad-ness being washed out to too large a degree? I think there's a conflation of terms going on here, between regression analysis, and regression to the mean. The issue you're describing (I think?) is regression to the mean, which isn't what's going on Faverani's rating. Regression to the mean is something that you should probably apply on top of the ridge regression that xRAPM is doing. The ridge regression of xRAPM generates that Earl Clark is the worst player in the NBA at -5.63 points/100 possessions. Regression to the mean on the other hand says we've probably just observed an unusually bad sample of play from Clark, and he's likely to improve. These are distinct issues.Second, and really more importantly, to do plus/minus you need to make some assessment of the quality of player coming in. You describe above (correctly, I believe) that the metric compares performance with player 1 in vs performance with player 2 in. But that's only part of the measurement challenge, because even if all other players and sets were the same (which, as noted above, we know they are not) we still need to assess how much of the scoring difference between 'lineup with player 1' and 'lineup with player 2' is player 1 being good and how much is player 2 being bad. So, if there's a 10 point difference it might be that Humphries is 10 better than average and Faverani is average, or it might be Humphries is +5 and Faverani is -5, etc. This is essentially a true talent level problem. To simplify, you say that they do this (and I agree) but my belief is that they way they do it leaves real gaps.
The metric tries to do that, I believe effectively by taking the +10 and -10 and comparing it to a large set of substitutions and regressing the changes for different lineups to determine a baseline for each player. That is a reasonable approach; however, it is far from a perfectly accurate one, too. So, it's guess at Faverani's impact is just that---a guess, likely a regressed one. Once we're regressing both parts of the equation, we need to recognize that outliers (such as Faverani's bad-ness) will tend to get washed out to a larger degree than they probably should.
What do you mean standard errors? Standard errors are a sampling issue as far as I know, i.e. polling 10,000 Americans and using that sample to represent the entire American population. Basketball statistics are (generally) not doing that. They're not sampling part of the data, and extrapolating to the entire league. They're polling all 300,000,000 people. (Which isn't to say the stats are perfect - problems occur even without sampling errors).luckiestman said:This post is in line with my thinking. Im skeptical of these stats because they are often presented without std. errors so I have no idea if they are meaningful nor do I know how to interpret the marginal effects even if they are meaningful.
bowiac said:What do you mean standard errors? Standard errors are a sampling issue as far as I know, i.e. polling 10,000 Americans and using that sample to represent the entire American population. Basketball statistics are (generally) not doing that. They're not sampling part of the data, and extrapolating to the entire league. They're polling all 300,000,000 people. (Which isn't to say the stats are perfect - problems occur even without sampling errors).
Can you explain more about what this means? This paper is above my "pay grade" so to speak (my mathematical training is mostly in number theory, not pure statistics). What is this paper saying, and what is a standard error outside of a sampling context?luckiestman said:I don't agree with what you are saying exactly, but it is a contested issue (my line of thinking is consistent with the following since we are trying to figure out the relevance of the metrics going forward)
http://www.tinbergen.nl/wp-content/uploads/2014/01/220114_Imbens.pdf
bowiac said:I don't understand what you're saying here. If you bring in someone who is horrible, that impacts what happens elsewhere on the floor - yes. Everyone needs to adjust for that, and presumably that unit will be less effective. That will be reflected the xRAPM. I honestly do not understand what you're saying. What is the "second level" thing?
The broader issue you're describing is comparatively simple mathematically and has been solved for something like 300 years. This is how chess (ELO) ratings work, as well as how Las Vegas generates spreads, and to my knowledge basically every successful method of analyzing team performance.
When two chess players play each other, at first, we don't know if the result reflects the winner being good, the loser being bad, both players being bad (but one better than the other), both players being great (but one better than the other), etc... Then each player plays someone else. We get another W-L-D datapoint. All the players keep playing each other, and we generate a well connected network. With enough of a sample, we get a pretty effective sense of how good each chess player is, and everyone's future wins and losses adjust the entire connected network.
With Las Vegas, they do the same thing, except the result isn't binary, but also builds in a strength of victory adjustment. The Spurs beat the Heat by 10, we don't know if the Spurs are +5 and the Heat -5, if the teams are +20/+10, or whatever. Then the Spurs play the Mavericks, and win by 5, and the Heat play the Blazers and win by 12 and we get new datapoints. As these teams continue to play each other, you get the same well connected network. Then through the method of minimizing the sum of the squares of the errors, you can generate an estimate of team strength that minimizes the sum of the errors.
These are both fairly "robust" mathematical adjustments, and basically what xRAPM is doing. These are issues with xRAPM, but as near as I can understand your question, this isn't one of the major ones. It's not "perfectly accurate", but I honestly have no idea what is in life. That's a useless high standard. We should use NBA scouts either - they're not perfectly accurate, etc...
Sorry, so what's your point here? That xRAPM and its ilk aren't perfect? I think everyone has been very clear about that. I don't see how this generalized concern has anything to do with Humphries or Faverani?PedroKsBambino said:I think you're not really thinking about how team defense works, and that leads you to miss the challenge I'm raising about what the data says about how much we can attribute point difference to a single player. It simply is not the case that all permutations of different matchups are the same. So, if you have a set of players on the court and you change one of them, all of the interactions change in some ways (not just the one player who changed). The method you're describing counts all of that difference as attributable to the matchup which changed. However, that does not mean the true talent level of the guy who changed is equal to the point differential of that second matchup, because if he was in a different configuration he might generate different results (because the matchups may be different, and this will generate different results). To illustrate this, the entire concept of a rim protector is built on the exact same principle I'm noting.
To solve that multivariate problem, you need to do a much more robust analysis, and that analysis is going to require assumptions that themselves have error rates. These multiply when doing the analysis, and that is the challenge to be considered. That is the regression issue I was noting.
I don't really know what to respond to, as you said, your post was a bit rambling. With respect to this question however, is it meaningful, I can say with pretty strong confidence that the answer is yes. The reasons:luckiestman said:I don't think the question is whether or not a stat is perfect, I think it is a question of it being meaningful at all.
I'm not so sure. Hibbert did not have a good year at all and makes a ton of money.swingin val said:Gonna have to offer a lot more than just Rondo
What kind of year did Rondo have? And Rajon Rondo makes 13 million, and has een rumored to want a max, or near max deal.Brickowski said:I'm not so sure. Hibbert did not have a good year at all and makes a ton of money.
I doubt it.swingin val said:Gonna have to offer a lot more than just Rondo
swingin val said:Gonna have to offer a lot more than just Rondo
wutang112878 said:Its difficult to judge the value of each of them. Rondo because of his contract demands and injury and Hibbert because no one has much of a clue about his inconsistent play. But because there is a chance, regardless if how small, that Hibbert can be a difference making center he does have positive value
Not just down the stretch. Hibbert disappeared for a couple months mid season as well. And didn't have the excuse of injury.radsoxfan said:30M for 2 more years of Roy Hibbert's lumbering body, after the last few months he just had. I'm not so sure I can say the same. I admit perhaps I'm falling for a bit of recency bias there, but he was just so awful down the stretch this season.
15-20 per what? That's a max deal basically.zenter said:Not just down the stretch. Hibbert disappeared for a couple months mid season as well. And didn't have the excuse of injury.
I get that this board is down on Rondo, but he's one of the few exceptional pass-first players in all of basketball who facilitates an offense and can play good help defense. As I've said elsewhere, reliably near double double. I don't think he deserves a max contract, but I don't think 15-20 is any kind of stretch, given his unique skills and the salary environment. Hibbert's more of a question mark.
radsoxfan said:
Rondo's future contract certainly could alter things, but no one is forced to pay him if they trade for him.
Just looking at the 2014-2015 season, I think Rondo's reputation around the league is such that plenty of teams would be happy to have him at his current salary. If you don't want to pay him the max, and he walks after the year, thats the risk you are taking. But it doesn't change his value for the 2014-2015 season, which I think is worth something decent in a trade.
30M for 2 more years of Roy Hibbert's lumbering body, after the last few months he just had…. I'm not so sure I can say the same. I admit perhaps I'm falling for a bit of recency bias there, but he was just so awful down the stretch this season.
zenter said:Not just down the stretch. Hibbert disappeared for a couple months mid season as well. And didn't have the excuse of injury.
I get that this board is down on Rondo, but he's one of the few exceptional pass-first players in all of basketball who facilitates an offense and can play good help defense. As I've said elsewhere, reliably near double double. I don't think he deserves a max contract, but I don't think 15-20 is any kind of stretch, given his unique skills and the salary environment. Hibbert's more of a question mark.
To be clear, no one has suggested perfection is the goal, or the expectation. The concern expressed by two of us was specific, and likely material (though we can't be sure of course) if it hasn't been addressed. It may have been; you don't seem to know which is fine---I surely don't either.bowiac said:Sorry, so what's your point here? That xRAPM and its ilk aren't perfect? I think everyone has been very clear about that. I don't see how this generalized concern has anything to do with Humphries or Faverani?
As described before, it relates to the assumption made about his performance. I think you will need to understand the concern raised above to see this one, though...because it's in trying to unpack how all of the players on the court relate that I believe one gets to the problem of determining whether to regress their performance and how.What is the regression issue that is causing Faverani's bad-ness to be washed out? I still don't get it. Be specific please.
zenter said:Not just down the stretch. Hibbert disappeared for a couple months mid season as well. And didn't have the excuse of injury.
I get that this board is down on Rondo, but he's one of the few exceptional pass-first players in all of basketball who facilitates an offense and can play good help defense. As I've said elsewhere, reliably near double double. I don't think he deserves a max contract, but I don't think 15-20 is any kind of stretch, given his unique skills and the salary environment. Hibbert's more of a question mark.
The only way a Love deal ever made sense was with an Asik or a Hibbert joining him on the frontline to completely change the dynamics of the team. Love without either of them (or their likeness) is like trading for Ray Allen without adding KG.Brickowski said:According to the Sporting News, The Pacers are "quietly shopping" Roy Hibbert. I realize he makes alot of money and that he sucked for long stretches last year.
Even so, if they could get Hibbert (e.g. for Rondo) I'd be very happy to add Kevin Love. If you put those two together-- one providing defense- the other offense, and both rebounding the ball, you'd really have something. The whole could be greater than the sum of its parts, etc. etc.
knucklecup said:RT @judefox10: Would you trade Goran Dragic for Rajon Rondo?
Verified Phoenix reporter that caught my attention with the mention of Rondo.