What Do NBA MVP Voters Look For?
This article’s probably a few weeks late, but spare me. I’m a busy fucking man. The MVP conversation was pretty controversial. My conclusion was that Kawhi was MVP this year. I surmised that the Kawhi’s supporting cast as bad as, if not worse, than the Harden’s, and his supporting cast was only a little better than Russ’s. My other justification for picking Kawhi was that his defensive versatility, something that’s in stark contrast to both Russ’s and Harden’s defensive deficiencies. CJ and I talk about the same thing here on Episode 24 of The Crevice:
But, there are justifications for all the MVP candidates: Kawhi had the best two-way season, LeBron is CLEARLY the best player in the league (just take his first seven games of the playoffs as an example), Russ was eternally dominant, Harden was the cog in an all-time top ten offense, and Steph had another amazing shooting season, leading the best team in the NBA. We can go on and on and on, arguing who had the best season. But how do the voters vote based on the stats?
So, what did I do? I went all the way back to LeBron’s first year in the league, took all the players who got at least one MVP vote, and considered the proportion of total MVP points that each player received. I looked at their count stats (i.e. points per game, rebounds per game, etc), their usage and efficiency statistics (i.e. usage rate, shooting percentages, minutes per game), and advanced statistics (i.e. plus/minus stats, PER, etc). All the numbers I used are straight from BasketballReference.
First off, a couple notes on MVP voting through the years:
LeBron’s never finished less than ninth in MVP voting. That includes his rookie year. He’s basically been a top 10 to 15 player for 14 straight years.
On the flip side, Tim Duncan and, to a certain extent, Tony Parker are both sorely underrepresented. After Duncan’s back to back wins in 2001–02 and 2002–03, Duncan finished second to Garnett in 2003–04 and never finished in the top five again. That includes multiple 55+ win seasons.
Nash hosed Shaq. His counting stats were great (15.5 points on 11.5 assists while in the 50/40/90 club), but his advanced stats weren’t anything to write home to, especially given Amar’e Stoudemire’s forgotten contributions in the pick-and-roll, the shooting around said pick-and-roll, and the defensive prowess of Shawn Marion. In fact, Nash and Shaq both hosed Dirk. He put up 26 points and 10 rebounds on 46/40/87 shooting splits and had 5 more win shares than both Nash and Shaq. But alas, gaudy offensive teams are just so shiny, and NBA writers are basically little babies attracted to things that glitter.
P.J. fucking Brown once got votes. At age fucking 35. On fucking 10.6 points and 9.6 rebounds a games on 45% shooting. On a fucking 18 win team. Thanks, Jimmy Smith for your homerism.
We just saw Chris Paul’s team lose to an objectively better Jazz team that matched up so well against the Clippers. And now people have instantly started with the recency biased shit about how Chris Paul isn’t a superstar, isn’t a legend, doesn’t deserve a max contract this summer. Fuck all of you. Let’s take a look at Paul’s 2007–08 season, where he finished second behind Kobe in the MVP race. At only age 22, he put up 21.1 points, 11.6 assists, 2.7 steals on 49/37/85 shooting splits, a crazy 17.8 win shares to lead the league. And who was on that team? The fearsome lineup of David West, Tyson Chandler, Peja Stojakovic, and Bonzi Wells. Hardly a team worth of 56 wins. And what did Paul do the year after? He got better, just check the numbers.
Klay Thompson got a single vote in the 2014–15 MVP race. His PER that year was 14.3. PER is a scale-invariant rating, where the average PER is 15. Klay Thompson continues to be one of the more overrated players in the NBA. His offense is the product of Steph and his defense is the product of Draymond Green and Andrew Bogut.
Hot takes all around.
Alright, now that we’ve dispensed with some historical nuances, let’s get to some sleek graphs. I’m calling this proportion of MVP votes won the MVP Index. The first thing I did was chop off all the pointless players, basically anyone who didn’t garner at least 20% of the total MVP votes. Then, I looked at what stats and variables were highly correlated with this MVP Index. Nothing too interesting, really: wins, points per game, PER, VORP, field goal percentage are all highly correlated with MVP Index.
What’s revealing is when you break down the efficiency and advanced stats down to their offensive and defensive components. I’m not a huge fan of reporting correlation coefficients are just working the taint of their magnitudes. Let’s take a look at a few simple graphs.
These three graphs are telling. Offensive production is positively and clearly associated with MVP votes, but defensive production isn’t quite valued. You see that same damn thing if you break down real plus/minus and box score plus/minus. Hell, the negligible trend actually becomes negative when you look at defensive rating. This is what the stats and voting record say: voters appreciate gaudy offensive numbers and undervalue defense. No wonder Nash won a couple of MVPs.
It’s wrong! There’s something inherently messed up with MVP voting if defense is basically ignored.
What then? I looked at which stats were most predictive of MVP votes. I fit a beta regression on the outcome of MVP Index, a variable that’s between 0 and 1, with a bunch of covariates that we talked about earlier. First off, fuck you Steph for being a unanimous MVP last year. You made me transform my outcome and cost me a few minutes. I weeded down the variables used in the model with an easy backwards variable selection.
Taking all years from 2003–2004 into account, wins, field goal percentage, minutes played, games played, offensive rating and win shares per 48 minutes were the most significant positive predictors of increased MVP Index. Taking only the years on from LeBron’s Miami MVP years (starting in 2011–2012), wins, minutes played, and games played continue to be the most significant positive predictor of MVP votes and all the offensive efficiency and advanced stats become that much more involved in predicting MVP votes. That’s really not unexpected.
The last thing I did was use both my beta regression models to predict who is going to be our 2016–17 MVP. With my model incorporated all the data I collected, our MVP’s not LeBron, not Kawhi, not Russ, and not Harden. It’s Steph! And it’s because of his wins. With the model that only looks at recent voting trends, we get Harden, mostly because of his offensive efficiency and the Rockets’ solid number of wins. I do feel slightly vindicated that both models didn’t pick Russ to be MVP, though I am a little peeved that it didn’t pick Kawhi. But hey, that doesn’t mean I’m an idiot. It just means that all the MVP voters are.
Do I have any profound parting conclusions? No, not really. MVP voting doesn’t always identify the best player in the league in the year, but we knew that. MVP voters aren’t too concerned with defense, but we knew that, too. Wins and efficient scoring are really important, but duh. It’s promising that voters are starting to rely more and more on advanced stats and lineup-based stats and on-off numbers, but there’s a ways to go. Especially if Russ wins this year. It’ll be a step back in the voting process. Can’t believe I wrote that sentence for a story online and it isn’t about the 2016 presidential election.
