Latrinsorm
08-19-2014, 02:32 PM
82games (http://82games.com/index.htm) gives team points scored and allowed while any given player is on and off the court for the period 2005-present, which happens to be 10 years. We can use points scored and allowed to calculate an expected winning % (xw%) by...
________points scored ^ 14__________
points scored ^ 14 + points allowed ^ 14
...and we can therefore estimate the wins produced by any given player by...
82 * (xw% On * mp% On + xw% Off * mp% Off - xw% Off)
...that is, by subtracting off the team's performance when the player wasn't on the court at all, and by prorating his % by the number of minutes he played. We shall call this method Wins Estimated (WE).
.
b-r (http://www.basketball-reference.com/) gives Win Shares, a more arcane method that uses overall team performance, but instead of looking into on/off splits assumes that the player's team performs as well on offense and defense with or without them, relying on box score elements to discern the player's overall contribution. This works reasonably well, as we have seen in the past when we model the MVP with it alone, but obviously there are flaws we can fill in. We would guess that WS will overrate players who (1) get a lot of blocks and/or steals on a good defensive team but (2) due to the infrequency of blocks and steals and their otherwise poor defense don't actually help their team on defense, and underrates the players who play good defense but don't happen to get blocks or steals. (Players who are good at both should be properly rated.) A similar process should take place on the offensive side, but statistics cover so much more on that side that we're only left with things like spacing and screening, which while important are less frequent.
.
One very strong conclusion in general is that the MVP must come from a team with the two best records in either conference, so let's take the WS leaders from those four teams as well as the player with the highest WS otherwise (as an MVP ballot has five votes) and see how they shake out in WE. The following table is sorted by year by WS. WSr divides the player's WS into the highest WS of the candidates from the top four teams, same with WEr and WE. Ratio is obtained by multiplying WSr and WEr, so if a player leads in WS and WE they get a 1.00.
Year Name WS WE WSr WEr Ratio
2005 Amar'e 14.6 18.6 1.00 0.88 0.88
2005 Billups 12.1 16.2 0.83 0.76 0.63
2005 Duncan 11.2 21.2 0.77 1.00 0.77
2005 Wade 11.0 -1.1 0.75 -0.05 -0.04
2005 Garnett 16.1 8.6 1.10 0.40 0.45
2006 Dirk 17.7 17.9 1.00 0.57 0.57
2006 Billups 15.5 23.1 0.88 0.74 0.65
2006 Wade 14.4 31.2 0.81 1.00 0.81
2006 Duncan 10.8 13.7 0.61 0.44 0.27
2006 LeBron 16.3 27.2 0.92 0.87 0.80
2007 Dirk 16.3 22.5 1.00 0.99 0.99
2007 LeBron 13.7 22.7 0.84 1.00 0.84
2007 Nash 12.6 17.4 0.77 0.77 0.59
2007 Billups 11.4 14.3 0.70 0.63 0.44
2007 Duncan 13.0 25.3 0.80 1.12 0.89
2008 Paul 17.8 20.8 1.00 1.00 1.00
2008 Kobe 13.8 17.6 0.78 0.85 0.66
2008 Billups 13.5 10.6 0.76 0.51 0.39
2008 Garnett 12.9 14.1 0.72 0.68 0.49
2008 LeBron 15.2 23.1 0.85 1.11 0.95
2009 LeBron 20.3 40.1 1.00 1.00 1.00
2009 Gasol 13.9 10.9 0.68 0.27 0.19
2009 Allen 11.1 20.4 0.55 0.51 0.28
2009 Billups 9.9 7.7 0.49 0.19 0.09
2009 Paul 18.3 34.2 0.90 0.85 0.77
2010 LeBron 18.5 31.4 1.00 1.00 1.00
2010 Howard 16.1 19.3 0.87 0.62 0.54
2010 Dirk 13.2 24.3 0.71 0.77 0.55
2010 Gasol 11.0 5.6 0.59 0.18 0.11
2010 Durant 16.1 35.2 0.87 1.12 0.97
2011 LeBron 15.6 20.2 1.00 0.97 0.97
2011 Gasol 14.7 20.8 0.94 1.00 0.94
2011 Rose 13.1 3.9 0.84 0.19 0.16
2011 Manu 9.9 19.6 0.63 0.94 0.60
2011 Howard 14.4 21.2 0.92 1.02 0.94
2012 LeBron 14.5 27.9 1.00 1.00 1.00
2012 Durant 12.2 8.1 0.84 0.29 0.24
2012 Noah 9.0 -4.5 0.62 -0.16 -0.10
2012 Parker 7.1 12.5 0.49 0.45 0.22
2012 Paul 12.7 27.9 0.88 1.00 0.88
2013 LeBron 19.3 27.6 1.00 1.00 1.00
2013 Durant 18.9 15.8 0.98 0.57 0.56
2013 Carmelo 9.5 7.9 0.49 0.29 0.14
2013 Parker 9.3 10.4 0.48 0.38 0.18
2013 Paul 13.9 8.1 0.72 0.29 0.21
2014 Durant 19.2 14.2 1.00 0.95 0.95
2014 LeBron 15.9 15.0 0.83 1.00 0.83
2014 George 10.8 14.1 0.56 0.94 0.53
2014 Duncan 7.4 -2.5 0.39 -0.17 -0.06
2014 Love 14.3 22.0 0.74 1.47 1.09
And here is the table of the actual winners from each year:
Year Name WS WE WSr WEr Ratio
2005 Nash 10.9 22.4 0.75 0.88 0.66
2006 Nash 12.4 15.4 0.70 0.76 0.53
2007 Dirk 16.3 22.5 1.00 0.99 0.99
2008 Kobe 13.8 17.6 0.78 0.85 0.66
2009 LeBron 20.3 40.1 1.00 1.00 1.00
2010 LeBron 18.5 31.4 1.00 1.00 1.00
2011 Rose 13.1 3.9 0.84 0.19 0.16
2012 LeBron 14.5 27.9 1.00 1.00 1.00
2013 LeBron 19.3 27.6 1.00 1.00 1.00
2014 Durant 19.2 14.2 1.00 0.95 0.95
Okay, first we have to talk about those negative numbers: Wade 05, Noah 12, Duncan 14. Having three outliers in a sample size of fifty is actually right on the money (at 5% * 50 = 2.5), it would be more suspicious if we had none (or six). It's still worth wondering if they reveal something systematically wrong with WE, though. Let's look into the nitty gritty:
Wade 05 made his team better on offense but absolutely destroyed their defense, giving him a net minus in on/off (albeit a small one). He averaged more than 1 block and steal per game and made the All-NBA Defense Second Team, so what gives? What gives is exactly what we talked about before. There are about 100 plays in a basketball game. Making a great play on 3 of those doesn't make up for screwing up the other 97, or a significant fraction of them. As a second year player, it's conceivable that Wade's defense wasn't that great. WS thinks he was worth 4.3 wins on defense alone, with 11.0 WS total his value could change dramatically if his defense is mis-measured.
Noah 12 has the same breakdown, which is a bit more worrisome. He has a decent number of blocks and steals but nothing spectacular, and by all accounts he really is a good defender, so what gives now? The best explanation I can give (beyond a simple fluke) is that his backup that year was Omer Asik, who is also by all accounts really a good defender. My guess is Noah just drew the short statistical straw, though.
Duncan 14 slightly hurt his team on offense and defense, but this is a pretty easy explanation. He's 37, his jump shot deserted him last year (compromising spacing), the Spurs are famously deep which reduces the collinearity of starting players (the traditional weakness of on/off splits).
All in all I'd be comfortable if there was no explanation at all for these three points, but being able to explain away two of them makes me happy too.
.
The winners mostly reinforce what we already knew.
Nash never deserved his, but it's interesting to see how weak the field was in the years he won: the lowest max ratios at .88 and .81 respectively, every other year was .95 or higher.
Kobe's lack of dessert is even more pronounced than Nash's: his ratio is as low but Chris Paul alone represented a stronger field than Nash burgled.
Rose's win (and to a lesser extent Carmelo's #1 vote) remain an absolute travesty. It is fascinating to wonder how people would look back at this era if Pau Gasol ever got the respect he deserved. He'd clearly have won in 2011 and he'd have at least one FMVP from the Lakers' last two wins.
LeBron being worth 40 wins in 2009 is not a typo. I've often talked about how Carmelo got a rough deal being forever compared to LeBron, but he didn't do Chris Paul any favors either. If not for LeBron's historic year, Chris Paul would have been the runaway statistical candidate for MVP, so much so that it's plausible he would have finished third or second in MVP. As a 7 seed it's hard to imagine him winning outright, but even in our universe's 2009 he got two first place votes. The only players with 18+ WS in the past ten years are LeBron (3 times) Durant (2) Paul (1), and the only players with 30+ WE are LeBron (2) Durant Wade Paul. He follows up the year he deserved MVP with an even better performance but LeBron's was one for the ages so no one will ever remember it. Sorry CP3!
.
Alright, but where does Jordan come into it? Of 50 player-years there are only four where the team would win 70+ games even if the MVP candidate had been on the court for every minute of the season. (Jordan played 78% of the Bulls' minutes in 96, which is pretty typical.) They are:
Year Name onW offW
2005 Duncan 74.7 36.6
2008 Garnett 74.0 50.2
2009 LeBron 72.1 20.2
2007 Duncan 71.8 35.1
Everyone knows that the Celtics were a very deep team, but I think most people underestimate how much the Spurs have reinvented themselves the past three years. They got 30.3 WS from their big three in 05, 33.2 in 07, plummeting to only 21 in 14 (IF you count Kawhi instead of Manu). They were more top heavy than this year's Heat, but nobody would ever say that. The main takeaway though is reminding us all just how bad those Cleveland teams were outside of LeBron. 20 wins would have only gotten them fourth worst record in the NBA that year, but going from 1st to 26th is still pretty dramatic.
________points scored ^ 14__________
points scored ^ 14 + points allowed ^ 14
...and we can therefore estimate the wins produced by any given player by...
82 * (xw% On * mp% On + xw% Off * mp% Off - xw% Off)
...that is, by subtracting off the team's performance when the player wasn't on the court at all, and by prorating his % by the number of minutes he played. We shall call this method Wins Estimated (WE).
.
b-r (http://www.basketball-reference.com/) gives Win Shares, a more arcane method that uses overall team performance, but instead of looking into on/off splits assumes that the player's team performs as well on offense and defense with or without them, relying on box score elements to discern the player's overall contribution. This works reasonably well, as we have seen in the past when we model the MVP with it alone, but obviously there are flaws we can fill in. We would guess that WS will overrate players who (1) get a lot of blocks and/or steals on a good defensive team but (2) due to the infrequency of blocks and steals and their otherwise poor defense don't actually help their team on defense, and underrates the players who play good defense but don't happen to get blocks or steals. (Players who are good at both should be properly rated.) A similar process should take place on the offensive side, but statistics cover so much more on that side that we're only left with things like spacing and screening, which while important are less frequent.
.
One very strong conclusion in general is that the MVP must come from a team with the two best records in either conference, so let's take the WS leaders from those four teams as well as the player with the highest WS otherwise (as an MVP ballot has five votes) and see how they shake out in WE. The following table is sorted by year by WS. WSr divides the player's WS into the highest WS of the candidates from the top four teams, same with WEr and WE. Ratio is obtained by multiplying WSr and WEr, so if a player leads in WS and WE they get a 1.00.
Year Name WS WE WSr WEr Ratio
2005 Amar'e 14.6 18.6 1.00 0.88 0.88
2005 Billups 12.1 16.2 0.83 0.76 0.63
2005 Duncan 11.2 21.2 0.77 1.00 0.77
2005 Wade 11.0 -1.1 0.75 -0.05 -0.04
2005 Garnett 16.1 8.6 1.10 0.40 0.45
2006 Dirk 17.7 17.9 1.00 0.57 0.57
2006 Billups 15.5 23.1 0.88 0.74 0.65
2006 Wade 14.4 31.2 0.81 1.00 0.81
2006 Duncan 10.8 13.7 0.61 0.44 0.27
2006 LeBron 16.3 27.2 0.92 0.87 0.80
2007 Dirk 16.3 22.5 1.00 0.99 0.99
2007 LeBron 13.7 22.7 0.84 1.00 0.84
2007 Nash 12.6 17.4 0.77 0.77 0.59
2007 Billups 11.4 14.3 0.70 0.63 0.44
2007 Duncan 13.0 25.3 0.80 1.12 0.89
2008 Paul 17.8 20.8 1.00 1.00 1.00
2008 Kobe 13.8 17.6 0.78 0.85 0.66
2008 Billups 13.5 10.6 0.76 0.51 0.39
2008 Garnett 12.9 14.1 0.72 0.68 0.49
2008 LeBron 15.2 23.1 0.85 1.11 0.95
2009 LeBron 20.3 40.1 1.00 1.00 1.00
2009 Gasol 13.9 10.9 0.68 0.27 0.19
2009 Allen 11.1 20.4 0.55 0.51 0.28
2009 Billups 9.9 7.7 0.49 0.19 0.09
2009 Paul 18.3 34.2 0.90 0.85 0.77
2010 LeBron 18.5 31.4 1.00 1.00 1.00
2010 Howard 16.1 19.3 0.87 0.62 0.54
2010 Dirk 13.2 24.3 0.71 0.77 0.55
2010 Gasol 11.0 5.6 0.59 0.18 0.11
2010 Durant 16.1 35.2 0.87 1.12 0.97
2011 LeBron 15.6 20.2 1.00 0.97 0.97
2011 Gasol 14.7 20.8 0.94 1.00 0.94
2011 Rose 13.1 3.9 0.84 0.19 0.16
2011 Manu 9.9 19.6 0.63 0.94 0.60
2011 Howard 14.4 21.2 0.92 1.02 0.94
2012 LeBron 14.5 27.9 1.00 1.00 1.00
2012 Durant 12.2 8.1 0.84 0.29 0.24
2012 Noah 9.0 -4.5 0.62 -0.16 -0.10
2012 Parker 7.1 12.5 0.49 0.45 0.22
2012 Paul 12.7 27.9 0.88 1.00 0.88
2013 LeBron 19.3 27.6 1.00 1.00 1.00
2013 Durant 18.9 15.8 0.98 0.57 0.56
2013 Carmelo 9.5 7.9 0.49 0.29 0.14
2013 Parker 9.3 10.4 0.48 0.38 0.18
2013 Paul 13.9 8.1 0.72 0.29 0.21
2014 Durant 19.2 14.2 1.00 0.95 0.95
2014 LeBron 15.9 15.0 0.83 1.00 0.83
2014 George 10.8 14.1 0.56 0.94 0.53
2014 Duncan 7.4 -2.5 0.39 -0.17 -0.06
2014 Love 14.3 22.0 0.74 1.47 1.09
And here is the table of the actual winners from each year:
Year Name WS WE WSr WEr Ratio
2005 Nash 10.9 22.4 0.75 0.88 0.66
2006 Nash 12.4 15.4 0.70 0.76 0.53
2007 Dirk 16.3 22.5 1.00 0.99 0.99
2008 Kobe 13.8 17.6 0.78 0.85 0.66
2009 LeBron 20.3 40.1 1.00 1.00 1.00
2010 LeBron 18.5 31.4 1.00 1.00 1.00
2011 Rose 13.1 3.9 0.84 0.19 0.16
2012 LeBron 14.5 27.9 1.00 1.00 1.00
2013 LeBron 19.3 27.6 1.00 1.00 1.00
2014 Durant 19.2 14.2 1.00 0.95 0.95
Okay, first we have to talk about those negative numbers: Wade 05, Noah 12, Duncan 14. Having three outliers in a sample size of fifty is actually right on the money (at 5% * 50 = 2.5), it would be more suspicious if we had none (or six). It's still worth wondering if they reveal something systematically wrong with WE, though. Let's look into the nitty gritty:
Wade 05 made his team better on offense but absolutely destroyed their defense, giving him a net minus in on/off (albeit a small one). He averaged more than 1 block and steal per game and made the All-NBA Defense Second Team, so what gives? What gives is exactly what we talked about before. There are about 100 plays in a basketball game. Making a great play on 3 of those doesn't make up for screwing up the other 97, or a significant fraction of them. As a second year player, it's conceivable that Wade's defense wasn't that great. WS thinks he was worth 4.3 wins on defense alone, with 11.0 WS total his value could change dramatically if his defense is mis-measured.
Noah 12 has the same breakdown, which is a bit more worrisome. He has a decent number of blocks and steals but nothing spectacular, and by all accounts he really is a good defender, so what gives now? The best explanation I can give (beyond a simple fluke) is that his backup that year was Omer Asik, who is also by all accounts really a good defender. My guess is Noah just drew the short statistical straw, though.
Duncan 14 slightly hurt his team on offense and defense, but this is a pretty easy explanation. He's 37, his jump shot deserted him last year (compromising spacing), the Spurs are famously deep which reduces the collinearity of starting players (the traditional weakness of on/off splits).
All in all I'd be comfortable if there was no explanation at all for these three points, but being able to explain away two of them makes me happy too.
.
The winners mostly reinforce what we already knew.
Nash never deserved his, but it's interesting to see how weak the field was in the years he won: the lowest max ratios at .88 and .81 respectively, every other year was .95 or higher.
Kobe's lack of dessert is even more pronounced than Nash's: his ratio is as low but Chris Paul alone represented a stronger field than Nash burgled.
Rose's win (and to a lesser extent Carmelo's #1 vote) remain an absolute travesty. It is fascinating to wonder how people would look back at this era if Pau Gasol ever got the respect he deserved. He'd clearly have won in 2011 and he'd have at least one FMVP from the Lakers' last two wins.
LeBron being worth 40 wins in 2009 is not a typo. I've often talked about how Carmelo got a rough deal being forever compared to LeBron, but he didn't do Chris Paul any favors either. If not for LeBron's historic year, Chris Paul would have been the runaway statistical candidate for MVP, so much so that it's plausible he would have finished third or second in MVP. As a 7 seed it's hard to imagine him winning outright, but even in our universe's 2009 he got two first place votes. The only players with 18+ WS in the past ten years are LeBron (3 times) Durant (2) Paul (1), and the only players with 30+ WE are LeBron (2) Durant Wade Paul. He follows up the year he deserved MVP with an even better performance but LeBron's was one for the ages so no one will ever remember it. Sorry CP3!
.
Alright, but where does Jordan come into it? Of 50 player-years there are only four where the team would win 70+ games even if the MVP candidate had been on the court for every minute of the season. (Jordan played 78% of the Bulls' minutes in 96, which is pretty typical.) They are:
Year Name onW offW
2005 Duncan 74.7 36.6
2008 Garnett 74.0 50.2
2009 LeBron 72.1 20.2
2007 Duncan 71.8 35.1
Everyone knows that the Celtics were a very deep team, but I think most people underestimate how much the Spurs have reinvented themselves the past three years. They got 30.3 WS from their big three in 05, 33.2 in 07, plummeting to only 21 in 14 (IF you count Kawhi instead of Manu). They were more top heavy than this year's Heat, but nobody would ever say that. The main takeaway though is reminding us all just how bad those Cleveland teams were outside of LeBron. 20 wins would have only gotten them fourth worst record in the NBA that year, but going from 1st to 26th is still pretty dramatic.