I'm just curious, what is the lowest seed or record OKC can have for Russell Westbrook to win the MVP?
As many of you know I did bet on James Harden to win the MVP before the season but before I did that I spent some time looking at MVP trends of the past 36 seasons
-The last player to win the MVP award with his team being seeded less than 2nd in his conference was Michael Jordan in the in 1987-88 season. His team was seeded 3rd (50-32) with him averaging 35, 5.5 rebounds and 5.9 assists with 21.2 win shares, the 3rd highest PER of all time at 31.7 (only 2 Wilt seasons are better), 11.8 VORP and 12.2 BPM. Basically it took one of the greatest individual seasons in NBA history in order for voters to abandon the principle of voting for a player who is a top 2 seed in his conference. This may actually be a testament to MJ's greatness.
-Moses Malone won the 1981-82 MVP as 6th seed (46-36) putting up 31.1 points, 14.7 rebounds and 1.8 assists per game. Definitely an extreme outlier. Russ probably gets here.....
-No one else in the past 36 seasons has won MVP who wasn't a top 2 seed in their conference besides MJ and Moses Malone.
-Lowest 2nd place MVP finishers by seed with record and stats listed
2010-11-Dwight Howard (4th seed: 52-30) 22.9 pts 14.1 rbs, 1.4 asts 1.4 stls, 2.4 blks
PER 26.1, VORP 5, BPM 4.8
2009-10-Kevin Durant (8th seed: 50-32)
30.1 pts, 7.6 rbs, 2.8 asts, 1.4 stls, 1 blks
PER 26.2 VORP 5.8 BPM 5.1
2005-06-LeBron James (4th seed: 50-32)
31.4 pts, 7 rbs, 6.6 asts, 1.6 stls, 0.8 blks
PER 28.1 VORP 9.5 BPM 9.3
2002-03-Kevin Garnett (4th seed: 51-31)
23 pts 13.4 rbs, 6 asts, 1.4 stls, 1.6 blks
PER 26.4 VORP 9 BPM 8.7
1999-00-Kevin Garnett (6th seed: 50-32)
22.9 pts 11.8 rbs,5 asts, 1.5 stls, 1.6 blks
PER 23.6 VORP 6.6 BPM 6.1
1993-94-David Robinson (4th seed: 55-27)
29.8 pts 10.7 rbs, 4.8 asts, 1.7 stls, 3.3 blks
PER 30.7 VORP10.6 BPM 10.9
1988-89-Michael Jordan (6th seed: 47-35)
32.5 pts , 8 rbs, 8 asts, 2.9 stls, 0.8 blks
PER 31.1 VORP 12 BPM 12.6
1986-87-Michael Jordan (8th seed: 40-42)
37.1 pts, 5.2 rbs, 4.6 asts, 2.9 stls, 1.5 blks
PER 29.8 VORP 8.8 BPM 8.6
1985-86-Dominique Wilkins (4th seed: 50-32)
30.3 pts, 7.9 rbs, 2.6 asts, 1.8 stls, 0.6 blks
PER 23.3 VORP 4.3 BPM 3.6
1983-84-Bernard King (5th seed: 47-35)
26.3 pts, 5.1 rbs, 2.1 asts, 1 stls, 0.2 blks
PER 22.7 VORP 3.7 BPM 3.6
Before the season I created both a logistic regression model and a multiple linear regression model to predict who's most likely to win.
A Logistic Regression Model is a regression model where the dependent variable (DV) is categorical. The model takes on binary dependent variables—that is, where it can take only two values, such as pass/fail, win/lose, alive/dead or healthy/sick. In this case..... Win/Lose. (1 for Win, 0 for Lose)
I also ran a multiple linear regression model using voting shares as a defendant variable. I used voting shares because the number of voters changes every year for some odd reason.
I tested PTS, TRB, AST, STL, BLK, FG%, 3P%, FT%, WS, WS/48, Wins, Team win%, Conf Seed, P.E.R (Player Efficiency Rating), V.O.R.P (Value Over Replacement), B.P.M (Box Plus Minus) as independent variables for both models. Speaking to the irrationality of MVP voting, most of these factors had no significance and in the end I was left with:
PTS, TRB, AST, Wins, Team Win%, (necessary to use instead of Wins for the Lockout year in 2011-2012 and 98-99), Conference Seed & V.O.R.P.
The Least Squares Multiple Regression model only captured 51 percent of the variation in MVP Shares. Definitely speaks to the overall inconsistency of MVP voters.
RSquare 0.519715536
RSquare Adj 0.5147384949
Root Mean Square Error 0.1772364251
Mean of Response 0.1527576792
Observations (or Sum Wgts) 586
Intercept -0.70054685 | 0.1178579499 | -5.94 <.0001
V.O.R.P 0.0401937275 | 0.0049151345 | 8.18 <.0001
Cf Seed -0.020582542 | 0.0054760507 | -3.76 0.0002
PTS 0.0168756253 | 0.0017458611 | 9.67 <.0001
TRB 0.0129816115 | 0.0031240918 | 4.16 <.0001
AST 0.0131974466 | 0.003783606 | 3.49 0.0005
Wins 0.0041928452 |0.0016868522 | 2.49 0.0132
I have some ideas on how I can increase the R-square to account for more variability (maybe I might implement them later during the NBA season and post projections based on in season results.)
I do feel as though the impact of team's 2nd highest scorer or best player plays a role in how MVP voters see things (i.e.: Derrick Rose and Allen Iverson.....or in the case of Damian Lillard losing his whole starting lineup and still carrying his team to the playoffs yet getting more MVP votes than Kyle Lowry despite having an inferior statistical season with less wins)
Anyways, Here's some things I found interesting.
-My models (both logistic and multiple regression) does not like Derrick Rose's 2010-11 season and thinks Lebron should have won MVP. Lebron ranked out as the best player in the 2010-2011 NBA season by every metric but conference seeding and wins won out. In a year with a rookie head coach in Tom Thibodeau, a top seed in the conference despite a new formed super team in Miami (although they had plenty of flaws that first season).....someone had to get the credit, just happened to be Derrick Rose. A bit undeserved considering the 2010 Bulls were 1st in Drtg and 11th in Offense Rating. They won games defensively but typically the best offensive player on a team that wins games defensively gets typically overrated. Speaking to the Miami Supeteam backlash, Dwight Howard finished over Lebron James in MVP shares that season. Basically Kevin Durant and Steph Curry almost have no chance if voter backlash to Durant in 2016 is anything like Lebron received. Lebron finished 3rd in voter shares despite being overwhelmingly the best player in the league with he league's 2nd best record.
-Similar to above, my model also prefers Shaq's 2001 season over AI's. Both teams had the same record.