The Power of Story
Statistics absolutely play a role in determining the NBA’s Most Valuable Player Award each season, but determining exactly how and where they fit can be difficult. Part of the reason is that when we talk about MVP candidates the numbers are always mentioned, but they are never alone -
“LeBron James averaged 26.8 points, 8.0 rebounds, 7.3 assists, AND led the Heat to the second longest winning streak in NBA history.”
“Steve Nash averaged 15.5 points, 11.5 assists, 3.3 rebounds AND ran a revolutionary offense, the most efficient in the league, leading the Phoenix Suns to the NBA’s best record.”
So how do we know which parts of those resumes are most important?
There are no specific statistical criteria which a player must meet to become an MVP candidate, or ultimately win the award. With the final decision resting in the hands of a panel of voters, noise from the way each individual voters assigns value to various statistics can create a confusing picture of what’s most important.
If we average the season-long numbers from the last 20 MVP winners this is the profile we find:
Taken on its own, that line has the distinctive whiff of LeBron’s versatility (and considering that he’s won the award four out of the last five years, that’s not exactly a surprise). But like any average, that line is a combination of values higher and lower. No MVP from the last two decades had a season that was actually a close statistical match to those numbers. Seven of the last 20 MVP seasons were the products of players who produced fewer than 7.0 rebounds per game. Six of the last 20 MVP seasons were the products or players who produced more than 7.0 assists per game. This average hits the middle ground, smoothing out the differences between the unique profiles of each winner, but it doesn’t really tell us many specific details about what to expect from future candidates.
We find the same mix when it comes to looking at winners by position. Of the last 20 MVP Awards, four went to guards, seven went to wings and nine went to front court players. You also don’t need to be the most statistically productive player in the league. Of the last 20 MVP Awards, only 8 went to the player who led the league in John Hollinger’s Player Efficiency Rating and only 12 went to the player who led the league in Basketball-Reference’s overall player metric, Win Shares. If the MVP did go to the most statistically productive player each season Michael Jordan and Kareem Abdul-Jabbar would both be sitting on nine MVP trophies and LeBron would be chasing his sixth, not his fifth.
While it’s clear that there is not one defined template for winning an MVP trophy, statistical or otherwise, we can actually quantify how much statistics explain the annual MVP voting. Last season I did some work with statistical profiles for the last 10 MVP seasons. Using regression analysis, and some basic statistical categories like team win percentage, points per game, minutes per game and games played, I was able to show that about 51% of the variation in MVP voting can be explained by the differences in players’ statistics.
The other 49% of variation in MVP voting can then be attributed to the power of circumstance and how powerfully each player’s narrative captured the collective imagination that season. This is where the “AND . . .” part of each player’s MVP resume comes in.
MVP votes each have a point value – first-place votes are worth 10, second-place votes are worth 7, third-place votes are worth 5, fourth-place votes are worth 3, fifth-place votes are worth one. Each player’s votes are converted into points, with the highest total taking home the award and 1210 points being the maximum that can be earned by any one player.
Using the results of that regression analysis we can project what percentage of that maximum MVP vote point total each player should have received if statistics were the only factor. The table below shows:
- Actual % – This is the percentage of points, out of a maximum of 1210, each player received from MVP voters.
- Projected % – This is the percentage of points, out of a maximum of 1210, each player would be projected to receive, using just their statistics and the output of my regression analysis.
- Difference – This is the difference between the projected percentage of points and the actual percentage of points a player received in MVP voting. Essentially this shows how much more weight voters gave to a player’s story than their statistics.
To be absolutely clear, this does not highlight years where the MVP was given to the wrong player. It highlights, for each season, the relative balance struck by the voters between statistics and narrative
The first thing we see is that even when narrative hasn’t changed the actual order of finish, it definitely affects the percentage of points that are received by the winners. Based on the statistical projection, no player should have expected to earn more than the 72% LeBron earned in 2009 and 2010. But other than Nash’s 74% in 2006, every other MVP winner from the past 10 seasons has earned at least 81% of the maximum points available. These numbers are showing that if we were determining the winners purely by statistics the results would likely be a lot closer and less definitive. Separation is being created by the power of the winners’ individual stories.
We also see some places where using this purely statistical system would have resulted in a different winner than the one the voters gave us. In 2005, Dirk Nowitzki finished third, behind Shaquille O’Neal and Steve Nash, despite have the strongest statistical case by this model. Amazingly, we see the same thing in 2006 when he Dirk finished third behind Nash and LeBron. In 2007, Dirk’s combination of story and statistics finally did the trick and he won his first and only MVP, narrowly edging out Nash. Ironically, this model identifies that as the strongest statistical campaign by Nash, a season where he should have earned a greater percentage of the possible points than in either of the seasons in which he actually won MVP.
The other two seasons where this analysis shows narrative overruling the statistical evidence are 2008 and 2011. Bryant won his first MVP in 2007-08 despite a statistical résumé from Chris Paul that this judges as superior. But there was plenty of public sentiment that an MVP was long-overdue recognition for Bryant’s extended greatness, and that sentiment carried the day. In 2010-11, Derrick Rose exploded with a fantastic individual season. His statistical case was not as strong as LeBron’s, who had won the award the previous two seasons, but it seemed as though voters were intent on recognizing someone new, specifically Rose’s breakout season.
Every year the undefined nature of the league’s MVP Award leads to some sort of disagreement, regardless of the actual results. Whether it’s back-to-back wins for Nash or James just missing out on the first ever sweep of first place votes, there’s always an argument to be made, about something. But that’s part of the fun of this award. Everyone gets to set their own criteria, their own balance, from the voters all the way down to the fans. Whether you value numbers, stories, or some mixture of the two, there’s always something to believe in.